1 /* $NetBSD: kern_exec.c,v 1.163 2002/11/17 22:53:46 chs 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 1.163 2002/11/17 22:53:46 chs 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/ras.h> 58 #include <sys/signalvar.h> 59 #include <sys/stat.h> 60 #include <sys/syscall.h> 61 62 #include <sys/syscallargs.h> 63 64 #include <uvm/uvm_extern.h> 65 66 #include <machine/cpu.h> 67 #include <machine/reg.h> 68 69 #ifdef DEBUG_EXEC 70 #define DPRINTF(a) uprintf a 71 #else 72 #define DPRINTF(a) 73 #endif /* DEBUG_EXEC */ 74 75 /* 76 * Exec function switch: 77 * 78 * Note that each makecmds function is responsible for loading the 79 * exec package with the necessary functions for any exec-type-specific 80 * handling. 81 * 82 * Functions for specific exec types should be defined in their own 83 * header file. 84 */ 85 extern const struct execsw execsw_builtin[]; 86 extern int nexecs_builtin; 87 static const struct execsw **execsw = NULL; 88 static int nexecs; 89 90 u_int exec_maxhdrsz; /* must not be static - netbsd32 needs it */ 91 92 #ifdef LKM 93 /* list of supported emulations */ 94 static 95 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head); 96 struct emul_entry { 97 LIST_ENTRY(emul_entry) el_list; 98 const struct emul *el_emul; 99 int ro_entry; 100 }; 101 102 /* list of dynamically loaded execsw entries */ 103 static 104 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head); 105 struct exec_entry { 106 LIST_ENTRY(exec_entry) ex_list; 107 const struct execsw *es; 108 }; 109 110 /* structure used for building execw[] */ 111 struct execsw_entry { 112 struct execsw_entry *next; 113 const struct execsw *es; 114 }; 115 #endif /* LKM */ 116 117 /* NetBSD emul struct */ 118 extern char sigcode[], esigcode[]; 119 #ifdef SYSCALL_DEBUG 120 extern const char * const syscallnames[]; 121 #endif 122 #ifdef __HAVE_SYSCALL_INTERN 123 void syscall_intern(struct proc *); 124 #else 125 void syscall(void); 126 #endif 127 128 const struct emul emul_netbsd = { 129 "netbsd", 130 NULL, /* emulation path */ 131 #ifndef __HAVE_MINIMAL_EMUL 132 EMUL_HAS_SYS___syscall, 133 NULL, 134 SYS_syscall, 135 SYS_NSYSENT, 136 #endif 137 sysent, 138 #ifdef SYSCALL_DEBUG 139 syscallnames, 140 #else 141 NULL, 142 #endif 143 sendsig, 144 trapsignal, 145 sigcode, 146 esigcode, 147 setregs, 148 NULL, 149 NULL, 150 NULL, 151 #ifdef __HAVE_SYSCALL_INTERN 152 syscall_intern, 153 #else 154 syscall, 155 #endif 156 NULL, 157 NULL, 158 }; 159 160 #ifdef LKM 161 /* 162 * Exec lock. Used to control access to execsw[] structures. 163 * This must not be static so that netbsd32 can access it, too. 164 */ 165 struct lock exec_lock; 166 167 static void link_es(struct execsw_entry **, const struct execsw *); 168 #endif /* LKM */ 169 170 /* 171 * check exec: 172 * given an "executable" described in the exec package's namei info, 173 * see what we can do with it. 174 * 175 * ON ENTRY: 176 * exec package with appropriate namei info 177 * proc pointer of exec'ing proc 178 * iff verified exec enabled then flag indicating a direct exec or 179 * an indirect exec (i.e. for a shell script interpreter) 180 * NO SELF-LOCKED VNODES 181 * 182 * ON EXIT: 183 * error: nothing held, etc. exec header still allocated. 184 * ok: filled exec package, executable's vnode (unlocked). 185 * 186 * EXEC SWITCH ENTRY: 187 * Locked vnode to check, exec package, proc. 188 * 189 * EXEC SWITCH EXIT: 190 * ok: return 0, filled exec package, executable's vnode (unlocked). 191 * error: destructive: 192 * everything deallocated execept exec header. 193 * non-destructive: 194 * error code, executable's vnode (unlocked), 195 * exec header unmodified. 196 */ 197 int 198 #ifdef VERIFIED_EXEC 199 check_exec(struct proc *p, struct exec_package *epp, int direct_exec) 200 #else 201 check_exec(struct proc *p, struct exec_package *epp) 202 #endif 203 { 204 int error, i; 205 struct vnode *vp; 206 struct nameidata *ndp; 207 size_t resid; 208 209 ndp = epp->ep_ndp; 210 ndp->ni_cnd.cn_nameiop = LOOKUP; 211 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 212 /* first get the vnode */ 213 if ((error = namei(ndp)) != 0) 214 return error; 215 epp->ep_vp = vp = ndp->ni_vp; 216 217 /* check access and type */ 218 if (vp->v_type != VREG) { 219 error = EACCES; 220 goto bad1; 221 } 222 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 223 goto bad1; 224 225 /* get attributes */ 226 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 227 goto bad1; 228 229 /* Check mount point */ 230 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 231 error = EACCES; 232 goto bad1; 233 } 234 if (vp->v_mount->mnt_flag & MNT_NOSUID) 235 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 236 237 /* try to open it */ 238 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 239 goto bad1; 240 241 /* unlock vp, since we need it unlocked from here on out. */ 242 VOP_UNLOCK(vp, 0); 243 244 245 #ifdef VERIFIED_EXEC 246 /* Evaluate signature for file... */ 247 if ((error = check_veriexec(p, vp, epp, direct_exec)) != 0) 248 goto bad2; 249 #endif 250 251 /* now we have the file, get the exec header */ 252 uvn_attach(vp, VM_PROT_READ); 253 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 254 UIO_SYSSPACE, 0, p->p_ucred, &resid, p); 255 if (error) 256 goto bad2; 257 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 258 259 /* 260 * Set up default address space limits. Can be overridden 261 * by individual exec packages. 262 * 263 * XXX probably shoul be all done in the exec pakages. 264 */ 265 epp->ep_vm_minaddr = VM_MIN_ADDRESS; 266 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 267 /* 268 * set up the vmcmds for creation of the process 269 * address space 270 */ 271 error = ENOEXEC; 272 for (i = 0; i < nexecs && error != 0; i++) { 273 int newerror; 274 275 epp->ep_esch = execsw[i]; 276 newerror = (*execsw[i]->es_check)(p, epp); 277 /* make sure the first "interesting" error code is saved. */ 278 if (!newerror || error == ENOEXEC) 279 error = newerror; 280 281 /* if es_check call was successful, update epp->ep_es */ 282 if (!newerror && (epp->ep_flags & EXEC_HASES) == 0) 283 epp->ep_es = execsw[i]; 284 285 if (epp->ep_flags & EXEC_DESTR && error != 0) 286 return error; 287 } 288 if (!error) { 289 /* check that entry point is sane */ 290 if (epp->ep_entry > VM_MAXUSER_ADDRESS) 291 error = ENOEXEC; 292 293 /* check limits */ 294 if ((epp->ep_tsize > MAXTSIZ) || 295 (epp->ep_dsize > 296 (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur)) 297 error = ENOMEM; 298 299 if (!error) 300 return (0); 301 } 302 303 /* 304 * free any vmspace-creation commands, 305 * and release their references 306 */ 307 kill_vmcmds(&epp->ep_vmcmds); 308 309 bad2: 310 /* 311 * close and release the vnode, restore the old one, free the 312 * pathname buf, and punt. 313 */ 314 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 315 VOP_CLOSE(vp, FREAD, p->p_ucred, p); 316 vput(vp); 317 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 318 return error; 319 320 bad1: 321 /* 322 * free the namei pathname buffer, and put the vnode 323 * (which we don't yet have open). 324 */ 325 vput(vp); /* was still locked */ 326 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 327 return error; 328 } 329 330 /* 331 * exec system call 332 */ 333 /* ARGSUSED */ 334 int 335 sys_execve(struct proc *p, void *v, register_t *retval) 336 { 337 struct sys_execve_args /* { 338 syscallarg(const char *) path; 339 syscallarg(char * const *) argp; 340 syscallarg(char * const *) envp; 341 } */ *uap = v; 342 int error; 343 u_int i; 344 struct exec_package pack; 345 struct nameidata nid; 346 struct vattr attr; 347 struct ucred *cred; 348 char *argp; 349 char * const *cpp; 350 char *dp, *sp; 351 long argc, envc; 352 size_t len; 353 char *stack; 354 struct ps_strings arginfo; 355 struct vmspace *vm; 356 char **tmpfap; 357 int szsigcode; 358 struct exec_vmcmd *base_vcp; 359 360 /* 361 * Lock the process and set the P_INEXEC flag to indicate that 362 * it should be left alone until we're done here. This is 363 * necessary to avoid race conditions - e.g. in ptrace() - 364 * that might allow a local user to illicitly obtain elevated 365 * privileges. 366 */ 367 p->p_flag |= P_INEXEC; 368 369 cred = p->p_ucred; 370 base_vcp = NULL; 371 /* 372 * Init the namei data to point the file user's program name. 373 * This is done here rather than in check_exec(), so that it's 374 * possible to override this settings if any of makecmd/probe 375 * functions call check_exec() recursively - for example, 376 * see exec_script_makecmds(). 377 */ 378 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 379 380 /* 381 * initialize the fields of the exec package. 382 */ 383 pack.ep_name = SCARG(uap, path); 384 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 385 pack.ep_hdrlen = exec_maxhdrsz; 386 pack.ep_hdrvalid = 0; 387 pack.ep_ndp = &nid; 388 pack.ep_emul_arg = NULL; 389 pack.ep_vmcmds.evs_cnt = 0; 390 pack.ep_vmcmds.evs_used = 0; 391 pack.ep_vap = &attr; 392 pack.ep_flags = 0; 393 394 #ifdef LKM 395 lockmgr(&exec_lock, LK_SHARED, NULL); 396 #endif 397 398 /* see if we can run it. */ 399 #ifdef VERIFIED_EXEC 400 if ((error = check_exec(p, &pack, 1)) != 0) 401 //if ((error = check_exec(p, &pack, 0)) != 0) 402 #else 403 if ((error = check_exec(p, &pack)) != 0) 404 #endif 405 goto freehdr; 406 407 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 408 409 /* allocate an argument buffer */ 410 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS); 411 #ifdef DIAGNOSTIC 412 if (argp == (vaddr_t) 0) 413 panic("execve: argp == NULL"); 414 #endif 415 dp = argp; 416 argc = 0; 417 418 /* copy the fake args list, if there's one, freeing it as we go */ 419 if (pack.ep_flags & EXEC_HASARGL) { 420 tmpfap = pack.ep_fa; 421 while (*tmpfap != NULL) { 422 char *cp; 423 424 cp = *tmpfap; 425 while (*cp) 426 *dp++ = *cp++; 427 dp++; 428 429 FREE(*tmpfap, M_EXEC); 430 tmpfap++; argc++; 431 } 432 FREE(pack.ep_fa, M_EXEC); 433 pack.ep_flags &= ~EXEC_HASARGL; 434 } 435 436 /* Now get argv & environment */ 437 if (!(cpp = SCARG(uap, argp))) { 438 error = EINVAL; 439 goto bad; 440 } 441 442 if (pack.ep_flags & EXEC_SKIPARG) 443 cpp++; 444 445 while (1) { 446 len = argp + ARG_MAX - dp; 447 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 448 goto bad; 449 if (!sp) 450 break; 451 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 452 if (error == ENAMETOOLONG) 453 error = E2BIG; 454 goto bad; 455 } 456 dp += len; 457 cpp++; 458 argc++; 459 } 460 461 envc = 0; 462 /* environment need not be there */ 463 if ((cpp = SCARG(uap, envp)) != NULL ) { 464 while (1) { 465 len = argp + ARG_MAX - dp; 466 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 467 goto bad; 468 if (!sp) 469 break; 470 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 471 if (error == ENAMETOOLONG) 472 error = E2BIG; 473 goto bad; 474 } 475 dp += len; 476 cpp++; 477 envc++; 478 } 479 } 480 481 dp = (char *) ALIGN(dp); 482 483 szsigcode = pack.ep_es->es_emul->e_esigcode - 484 pack.ep_es->es_emul->e_sigcode; 485 486 /* Now check if args & environ fit into new stack */ 487 if (pack.ep_flags & EXEC_32) 488 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 489 sizeof(int) + sizeof(int) + dp + STACKGAPLEN + 490 szsigcode + sizeof(struct ps_strings)) - argp; 491 else 492 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 493 sizeof(char *) + sizeof(int) + dp + STACKGAPLEN + 494 szsigcode + sizeof(struct ps_strings)) - argp; 495 496 len = ALIGN(len); /* make the stack "safely" aligned */ 497 498 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 499 error = ENOMEM; 500 goto bad; 501 } 502 503 /* adjust "active stack depth" for process VSZ */ 504 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 505 506 /* 507 * Do whatever is necessary to prepare the address space 508 * for remapping. Note that this might replace the current 509 * vmspace with another! 510 */ 511 uvmspace_exec(p, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 512 513 /* Now map address space */ 514 vm = p->p_vmspace; 515 vm->vm_taddr = (caddr_t) pack.ep_taddr; 516 vm->vm_tsize = btoc(pack.ep_tsize); 517 vm->vm_daddr = (caddr_t) pack.ep_daddr; 518 vm->vm_dsize = btoc(pack.ep_dsize); 519 vm->vm_ssize = btoc(pack.ep_ssize); 520 vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr; 521 vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr; 522 523 /* create the new process's VM space by running the vmcmds */ 524 #ifdef DIAGNOSTIC 525 if (pack.ep_vmcmds.evs_used == 0) 526 panic("execve: no vmcmds"); 527 #endif 528 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 529 struct exec_vmcmd *vcp; 530 531 vcp = &pack.ep_vmcmds.evs_cmds[i]; 532 if (vcp->ev_flags & VMCMD_RELATIVE) { 533 #ifdef DIAGNOSTIC 534 if (base_vcp == NULL) 535 panic("execve: relative vmcmd with no base"); 536 if (vcp->ev_flags & VMCMD_BASE) 537 panic("execve: illegal base & relative vmcmd"); 538 #endif 539 vcp->ev_addr += base_vcp->ev_addr; 540 } 541 error = (*vcp->ev_proc)(p, vcp); 542 #ifdef DEBUG_EXEC 543 if (error) { 544 int j; 545 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 546 for (j = 0; j <= i; j++) 547 uprintf( 548 "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n", 549 j, vp[j].ev_addr, vp[j].ev_len, 550 vp[j].ev_offset, vp[j].ev_prot, 551 vp[j].ev_flags); 552 } 553 #endif /* DEBUG_EXEC */ 554 if (vcp->ev_flags & VMCMD_BASE) 555 base_vcp = vcp; 556 } 557 558 /* free the vmspace-creation commands, and release their references */ 559 kill_vmcmds(&pack.ep_vmcmds); 560 561 /* if an error happened, deallocate and punt */ 562 if (error) { 563 DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error)); 564 goto exec_abort; 565 } 566 567 /* remember information about the process */ 568 arginfo.ps_nargvstr = argc; 569 arginfo.ps_nenvstr = envc; 570 571 stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 572 sizeof(struct ps_strings) + szsigcode), 573 len - (sizeof(struct ps_strings) + szsigcode)); 574 #ifdef __MACHINE_STACK_GROWS_UP 575 /* 576 * The copyargs call always copies into lower addresses 577 * first, moving towards higher addresses, starting with 578 * the stack pointer that we give. When the stack grows 579 * down, this puts argc/argv/envp very shallow on the 580 * stack, right at the first user stack pointer, and puts 581 * STACKGAPLEN very deep in the stack. When the stack 582 * grows up, the situation is reversed. 583 * 584 * Normally, this is no big deal. But the ld_elf.so _rtld() 585 * function expects to be called with a single pointer to 586 * a region that has a few words it can stash values into, 587 * followed by argc/argv/envp. When the stack grows down, 588 * it's easy to decrement the stack pointer a little bit to 589 * allocate the space for these few words and pass the new 590 * stack pointer to _rtld. When the stack grows up, however, 591 * a few words before argc is part of the signal trampoline, 592 * so we have a problem. 593 * 594 * Instead of changing how _rtld works, we take the easy way 595 * out and steal 32 bytes before we call copyargs. This 596 * space is effectively stolen from STACKGAPLEN. 597 */ 598 stack += 32; 599 #endif /* __MACHINE_STACK_GROWS_UP */ 600 601 /* Now copy argc, args & environ to new stack */ 602 error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp); 603 if (error) { 604 DPRINTF(("execve: copyargs failed %d\n", error)); 605 goto exec_abort; 606 } 607 /* Move the stack back to original point */ 608 stack = (char *)STACK_GROW(vm->vm_minsaddr, len); 609 610 /* fill process ps_strings info */ 611 p->p_psstr = (struct ps_strings *)STACK_ALLOC(vm->vm_minsaddr, 612 sizeof(struct ps_strings)); 613 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 614 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 615 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 616 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 617 618 /* copy out the process's ps_strings structure */ 619 if ((error = copyout(&arginfo, (char *)p->p_psstr, 620 sizeof(arginfo))) != 0) { 621 DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n", 622 &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo))); 623 goto exec_abort; 624 } 625 626 /* copy out the process's signal trampoline code */ 627 if (szsigcode) { 628 p->p_sigctx.ps_sigcode = STACK_ALLOC(STACK_MAX(p->p_psstr, 629 sizeof(struct ps_strings)), szsigcode); 630 if ((error = copyout((char *)pack.ep_es->es_emul->e_sigcode, 631 p->p_sigctx.ps_sigcode, szsigcode)) != 0) { 632 DPRINTF(("execve: sig trampoline copyout failed\n")); 633 goto exec_abort; 634 } 635 #ifdef PMAP_NEED_PROCWR 636 /* This is code. Let the pmap do what is needed. */ 637 pmap_procwr(p, (vaddr_t)p->p_sigctx.ps_sigcode, szsigcode); 638 #endif 639 } 640 641 stopprofclock(p); /* stop profiling */ 642 fdcloseexec(p); /* handle close on exec */ 643 execsigs(p); /* reset catched signals */ 644 p->p_ctxlink = NULL; /* reset ucontext link */ 645 646 /* set command name & other accounting info */ 647 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 648 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 649 p->p_comm[len] = 0; 650 p->p_acflag &= ~AFORK; 651 652 /* record proc's vnode, for use by procfs and others */ 653 if (p->p_textvp) 654 vrele(p->p_textvp); 655 VREF(pack.ep_vp); 656 p->p_textvp = pack.ep_vp; 657 658 p->p_flag |= P_EXEC; 659 if (p->p_flag & P_PPWAIT) { 660 p->p_flag &= ~P_PPWAIT; 661 wakeup((caddr_t) p->p_pptr); 662 } 663 664 /* 665 * deal with set[ug]id. 666 * MNT_NOSUID has already been used to disable s[ug]id. 667 */ 668 if ((p->p_flag & P_TRACED) == 0 && 669 670 (((attr.va_mode & S_ISUID) != 0 && 671 p->p_ucred->cr_uid != attr.va_uid) || 672 673 ((attr.va_mode & S_ISGID) != 0 && 674 p->p_ucred->cr_gid != attr.va_gid))) { 675 /* 676 * Mark the process as SUGID before we do 677 * anything that might block. 678 */ 679 p_sugid(p); 680 681 /* Make sure file descriptors 0..2 are in use. */ 682 if ((error = fdcheckstd(p)) != 0) 683 goto exec_abort; 684 685 p->p_ucred = crcopy(cred); 686 #ifdef KTRACE 687 /* 688 * If process is being ktraced, turn off - unless 689 * root set it. 690 */ 691 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) 692 ktrderef(p); 693 #endif 694 if (attr.va_mode & S_ISUID) 695 p->p_ucred->cr_uid = attr.va_uid; 696 if (attr.va_mode & S_ISGID) 697 p->p_ucred->cr_gid = attr.va_gid; 698 } else 699 p->p_flag &= ~P_SUGID; 700 p->p_cred->p_svuid = p->p_ucred->cr_uid; 701 p->p_cred->p_svgid = p->p_ucred->cr_gid; 702 703 #if defined(__HAVE_RAS) 704 /* 705 * Remove all RASs from the address space. 706 */ 707 ras_purgeall(p); 708 #endif 709 710 doexechooks(p); 711 712 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 713 714 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 715 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 716 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 717 vput(pack.ep_vp); 718 719 /* notify others that we exec'd */ 720 KNOTE(&p->p_klist, NOTE_EXEC); 721 722 /* setup new registers and do misc. setup. */ 723 (*pack.ep_es->es_emul->e_setregs)(p, &pack, (u_long) stack); 724 if (pack.ep_es->es_setregs) 725 (*pack.ep_es->es_setregs)(p, &pack, (u_long) stack); 726 727 if (p->p_flag & P_TRACED) 728 psignal(p, SIGTRAP); 729 730 free(pack.ep_hdr, M_EXEC); 731 732 /* 733 * Call emulation specific exec hook. This can setup setup per-process 734 * p->p_emuldata or do any other per-process stuff an emulation needs. 735 * 736 * If we are executing process of different emulation than the 737 * original forked process, call e_proc_exit() of the old emulation 738 * first, then e_proc_exec() of new emulation. If the emulation is 739 * same, the exec hook code should deallocate any old emulation 740 * resources held previously by this process. 741 */ 742 if (p->p_emul && p->p_emul->e_proc_exit 743 && p->p_emul != pack.ep_es->es_emul) 744 (*p->p_emul->e_proc_exit)(p); 745 746 /* 747 * Call exec hook. Emulation code may NOT store reference to anything 748 * from &pack. 749 */ 750 if (pack.ep_es->es_emul->e_proc_exec) 751 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack); 752 753 /* update p_emul, the old value is no longer needed */ 754 p->p_emul = pack.ep_es->es_emul; 755 756 /* ...and the same for p_execsw */ 757 p->p_execsw = pack.ep_es; 758 759 #ifdef __HAVE_SYSCALL_INTERN 760 (*p->p_emul->e_syscall_intern)(p); 761 #endif 762 #ifdef KTRACE 763 if (KTRPOINT(p, KTR_EMUL)) 764 ktremul(p); 765 #endif 766 767 #ifdef LKM 768 lockmgr(&exec_lock, LK_RELEASE, NULL); 769 #endif 770 p->p_flag &= ~P_INEXEC; 771 772 if (p->p_flag & P_STOPEXEC) { 773 int s; 774 775 sigminusset(&contsigmask, &p->p_sigctx.ps_siglist); 776 SCHED_LOCK(s); 777 p->p_stat = SSTOP; 778 mi_switch(p, NULL); 779 SCHED_ASSERT_UNLOCKED(); 780 splx(s); 781 } 782 783 return (EJUSTRETURN); 784 785 bad: 786 p->p_flag &= ~P_INEXEC; 787 /* free the vmspace-creation commands, and release their references */ 788 kill_vmcmds(&pack.ep_vmcmds); 789 /* kill any opened file descriptor, if necessary */ 790 if (pack.ep_flags & EXEC_HASFD) { 791 pack.ep_flags &= ~EXEC_HASFD; 792 (void) fdrelease(p, pack.ep_fd); 793 } 794 /* close and put the exec'd file */ 795 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 796 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 797 vput(pack.ep_vp); 798 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 799 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 800 801 freehdr: 802 p->p_flag &= ~P_INEXEC; 803 #ifdef LKM 804 lockmgr(&exec_lock, LK_RELEASE, NULL); 805 #endif 806 807 free(pack.ep_hdr, M_EXEC); 808 return error; 809 810 exec_abort: 811 p->p_flag &= ~P_INEXEC; 812 #ifdef LKM 813 lockmgr(&exec_lock, LK_RELEASE, NULL); 814 #endif 815 816 /* 817 * the old process doesn't exist anymore. exit gracefully. 818 * get rid of the (new) address space we have created, if any, get rid 819 * of our namei data and vnode, and exit noting failure 820 */ 821 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 822 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 823 if (pack.ep_emul_arg) 824 FREE(pack.ep_emul_arg, M_TEMP); 825 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 826 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 827 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 828 vput(pack.ep_vp); 829 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 830 free(pack.ep_hdr, M_EXEC); 831 exit1(p, W_EXITCODE(error, SIGABRT)); 832 833 /* NOTREACHED */ 834 return 0; 835 } 836 837 838 int 839 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo, 840 char **stackp, void *argp) 841 { 842 char **cpp, *dp, *sp; 843 size_t len; 844 void *nullp; 845 long argc, envc; 846 int error; 847 848 cpp = (char **)*stackp; 849 nullp = NULL; 850 argc = arginfo->ps_nargvstr; 851 envc = arginfo->ps_nenvstr; 852 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) 853 return error; 854 855 dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen); 856 sp = argp; 857 858 /* XXX don't copy them out, remap them! */ 859 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 860 861 for (; --argc >= 0; sp += len, dp += len) 862 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 863 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 864 return error; 865 866 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 867 return error; 868 869 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 870 871 for (; --envc >= 0; sp += len, dp += len) 872 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 873 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 874 return error; 875 876 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 877 return error; 878 879 *stackp = (char *)cpp; 880 return 0; 881 } 882 883 #ifdef LKM 884 /* 885 * Find an emulation of given name in list of emulations. 886 * Needs to be called with the exec_lock held. 887 */ 888 const struct emul * 889 emul_search(const char *name) 890 { 891 struct emul_entry *it; 892 893 LIST_FOREACH(it, &el_head, el_list) { 894 if (strcmp(name, it->el_emul->e_name) == 0) 895 return it->el_emul; 896 } 897 898 return NULL; 899 } 900 901 /* 902 * Add an emulation to list, if it's not there already. 903 */ 904 int 905 emul_register(const struct emul *emul, int ro_entry) 906 { 907 struct emul_entry *ee; 908 int error; 909 910 error = 0; 911 lockmgr(&exec_lock, LK_SHARED, NULL); 912 913 if (emul_search(emul->e_name)) { 914 error = EEXIST; 915 goto out; 916 } 917 918 MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry), 919 M_EXEC, M_WAITOK); 920 ee->el_emul = emul; 921 ee->ro_entry = ro_entry; 922 LIST_INSERT_HEAD(&el_head, ee, el_list); 923 924 out: 925 lockmgr(&exec_lock, LK_RELEASE, NULL); 926 return error; 927 } 928 929 /* 930 * Remove emulation with name 'name' from list of supported emulations. 931 */ 932 int 933 emul_unregister(const char *name) 934 { 935 const struct proclist_desc *pd; 936 struct emul_entry *it; 937 int i, error; 938 struct proc *ptmp; 939 940 error = 0; 941 lockmgr(&exec_lock, LK_SHARED, NULL); 942 943 LIST_FOREACH(it, &el_head, el_list) { 944 if (strcmp(it->el_emul->e_name, name) == 0) 945 break; 946 } 947 948 if (!it) { 949 error = ENOENT; 950 goto out; 951 } 952 953 if (it->ro_entry) { 954 error = EBUSY; 955 goto out; 956 } 957 958 /* test if any execw[] entry is still using this */ 959 for(i=0; i < nexecs; i++) { 960 if (execsw[i]->es_emul == it->el_emul) { 961 error = EBUSY; 962 goto out; 963 } 964 } 965 966 /* 967 * Test if any process is running under this emulation - since 968 * emul_unregister() is running quite sendomly, it's better 969 * to do expensive check here than to use any locking. 970 */ 971 proclist_lock_read(); 972 for (pd = proclists; pd->pd_list != NULL && !error; pd++) { 973 LIST_FOREACH(ptmp, pd->pd_list, p_list) { 974 if (ptmp->p_emul == it->el_emul) { 975 error = EBUSY; 976 break; 977 } 978 } 979 } 980 proclist_unlock_read(); 981 982 if (error) 983 goto out; 984 985 986 /* entry is not used, remove it */ 987 LIST_REMOVE(it, el_list); 988 FREE(it, M_EXEC); 989 990 out: 991 lockmgr(&exec_lock, LK_RELEASE, NULL); 992 return error; 993 } 994 995 /* 996 * Add execsw[] entry. 997 */ 998 int 999 exec_add(struct execsw *esp, const char *e_name) 1000 { 1001 struct exec_entry *it; 1002 int error; 1003 1004 error = 0; 1005 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1006 1007 if (!esp->es_emul) { 1008 esp->es_emul = emul_search(e_name); 1009 if (!esp->es_emul) { 1010 error = ENOENT; 1011 goto out; 1012 } 1013 } 1014 1015 LIST_FOREACH(it, &ex_head, ex_list) { 1016 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1017 if (it->es->es_check == esp->es_check 1018 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1019 && it->es->es_emul == esp->es_emul) { 1020 error = EEXIST; 1021 goto out; 1022 } 1023 } 1024 1025 /* if we got here, the entry doesn't exist yet */ 1026 MALLOC(it, struct exec_entry *, sizeof(struct exec_entry), 1027 M_EXEC, M_WAITOK); 1028 it->es = esp; 1029 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1030 1031 /* update execsw[] */ 1032 exec_init(0); 1033 1034 out: 1035 lockmgr(&exec_lock, LK_RELEASE, NULL); 1036 return error; 1037 } 1038 1039 /* 1040 * Remove execsw[] entry. 1041 */ 1042 int 1043 exec_remove(const struct execsw *esp) 1044 { 1045 struct exec_entry *it; 1046 int error; 1047 1048 error = 0; 1049 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1050 1051 LIST_FOREACH(it, &ex_head, ex_list) { 1052 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1053 if (it->es->es_check == esp->es_check 1054 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1055 && it->es->es_emul == esp->es_emul) 1056 break; 1057 } 1058 if (!it) { 1059 error = ENOENT; 1060 goto out; 1061 } 1062 1063 /* remove item from list and free resources */ 1064 LIST_REMOVE(it, ex_list); 1065 FREE(it, M_EXEC); 1066 1067 /* update execsw[] */ 1068 exec_init(0); 1069 1070 out: 1071 lockmgr(&exec_lock, LK_RELEASE, NULL); 1072 return error; 1073 } 1074 1075 static void 1076 link_es(struct execsw_entry **listp, const struct execsw *esp) 1077 { 1078 struct execsw_entry *et, *e1; 1079 1080 MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry), 1081 M_TEMP, M_WAITOK); 1082 et->next = NULL; 1083 et->es = esp; 1084 if (*listp == NULL) { 1085 *listp = et; 1086 return; 1087 } 1088 1089 switch(et->es->es_prio) { 1090 case EXECSW_PRIO_FIRST: 1091 /* put new entry as the first */ 1092 et->next = *listp; 1093 *listp = et; 1094 break; 1095 case EXECSW_PRIO_ANY: 1096 /* put new entry after all *_FIRST and *_ANY entries */ 1097 for(e1 = *listp; e1->next 1098 && e1->next->es->es_prio != EXECSW_PRIO_LAST; 1099 e1 = e1->next); 1100 et->next = e1->next; 1101 e1->next = et; 1102 break; 1103 case EXECSW_PRIO_LAST: 1104 /* put new entry as the last one */ 1105 for(e1 = *listp; e1->next; e1 = e1->next); 1106 e1->next = et; 1107 break; 1108 default: 1109 #ifdef DIAGNOSTIC 1110 panic("execw[] entry with unknown priority %d found", 1111 et->es->es_prio); 1112 #endif 1113 break; 1114 } 1115 } 1116 1117 /* 1118 * Initialize exec structures. If init_boot is true, also does necessary 1119 * one-time initialization (it's called from main() that way). 1120 * Once system is multiuser, this should be called with exec_lock held, 1121 * i.e. via exec_{add|remove}(). 1122 */ 1123 int 1124 exec_init(int init_boot) 1125 { 1126 const struct execsw **new_es, * const *old_es; 1127 struct execsw_entry *list, *e1; 1128 struct exec_entry *e2; 1129 int i, es_sz; 1130 1131 if (init_boot) { 1132 /* do one-time initializations */ 1133 lockinit(&exec_lock, PWAIT, "execlck", 0, 0); 1134 1135 /* register compiled-in emulations */ 1136 for(i=0; i < nexecs_builtin; i++) { 1137 if (execsw_builtin[i].es_emul) 1138 emul_register(execsw_builtin[i].es_emul, 1); 1139 } 1140 #ifdef DIAGNOSTIC 1141 if (i == 0) 1142 panic("no emulations found in execsw_builtin[]"); 1143 #endif 1144 } 1145 1146 /* 1147 * Build execsw[] array from builtin entries and entries added 1148 * at runtime. 1149 */ 1150 list = NULL; 1151 for(i=0; i < nexecs_builtin; i++) 1152 link_es(&list, &execsw_builtin[i]); 1153 1154 /* Add dynamically loaded entries */ 1155 es_sz = nexecs_builtin; 1156 LIST_FOREACH(e2, &ex_head, ex_list) { 1157 link_es(&list, e2->es); 1158 es_sz++; 1159 } 1160 1161 /* 1162 * Now that we have sorted all execw entries, create new execsw[] 1163 * and free no longer needed memory in the process. 1164 */ 1165 new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK); 1166 for(i=0; list; i++) { 1167 new_es[i] = list->es; 1168 e1 = list->next; 1169 FREE(list, M_TEMP); 1170 list = e1; 1171 } 1172 1173 /* 1174 * New execsw[] array built, now replace old execsw[] and free 1175 * used memory. 1176 */ 1177 old_es = execsw; 1178 execsw = new_es; 1179 nexecs = es_sz; 1180 if (old_es) 1181 free((void *)old_es, M_EXEC); 1182 1183 /* 1184 * Figure out the maximum size of an exec header. 1185 */ 1186 exec_maxhdrsz = 0; 1187 for (i = 0; i < nexecs; i++) { 1188 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1189 exec_maxhdrsz = execsw[i]->es_hdrsz; 1190 } 1191 1192 return 0; 1193 } 1194 #endif 1195 1196 #ifndef LKM 1197 /* 1198 * Simplified exec_init() for kernels without LKMs. Only initialize 1199 * exec_maxhdrsz and execsw[]. 1200 */ 1201 int 1202 exec_init(int init_boot) 1203 { 1204 int i; 1205 1206 #ifdef DIAGNOSTIC 1207 if (!init_boot) 1208 panic("exec_init(): called with init_boot == 0"); 1209 #endif 1210 1211 /* do one-time initializations */ 1212 nexecs = nexecs_builtin; 1213 execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK); 1214 1215 /* 1216 * Fill in execsw[] and figure out the maximum size of an exec header. 1217 */ 1218 exec_maxhdrsz = 0; 1219 for(i=0; i < nexecs; i++) { 1220 execsw[i] = &execsw_builtin[i]; 1221 if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz) 1222 exec_maxhdrsz = execsw_builtin[i].es_hdrsz; 1223 } 1224 1225 return 0; 1226 1227 } 1228 #endif /* !LKM */ 1229