kern_exec.c revision 1.116
1 /* $NetBSD: kern_exec.c,v 1.116 2000/07/26 15:42:09 mycroft 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 "opt_ktrace.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/filedesc.h> 40 #include <sys/kernel.h> 41 #include <sys/proc.h> 42 #include <sys/mount.h> 43 #include <sys/malloc.h> 44 #include <sys/namei.h> 45 #include <sys/vnode.h> 46 #include <sys/file.h> 47 #include <sys/acct.h> 48 #include <sys/exec.h> 49 #include <sys/ktrace.h> 50 #include <sys/resourcevar.h> 51 #include <sys/wait.h> 52 #include <sys/mman.h> 53 #include <sys/signalvar.h> 54 #include <sys/stat.h> 55 56 #include <sys/syscallargs.h> 57 58 #include <uvm/uvm_extern.h> 59 60 #include <machine/cpu.h> 61 #include <machine/reg.h> 62 63 /* 64 * check exec: 65 * given an "executable" described in the exec package's namei info, 66 * see what we can do with it. 67 * 68 * ON ENTRY: 69 * exec package with appropriate namei info 70 * proc pointer of exec'ing proc 71 * NO SELF-LOCKED VNODES 72 * 73 * ON EXIT: 74 * error: nothing held, etc. exec header still allocated. 75 * ok: filled exec package, executable's vnode (unlocked). 76 * 77 * EXEC SWITCH ENTRY: 78 * Locked vnode to check, exec package, proc. 79 * 80 * EXEC SWITCH EXIT: 81 * ok: return 0, filled exec package, executable's vnode (unlocked). 82 * error: destructive: 83 * everything deallocated execept exec header. 84 * non-destructive: 85 * error code, executable's vnode (unlocked), 86 * exec header unmodified. 87 */ 88 int 89 check_exec(p, epp) 90 struct proc *p; 91 struct exec_package *epp; 92 { 93 int error, i; 94 struct vnode *vp; 95 struct nameidata *ndp; 96 size_t resid; 97 98 ndp = epp->ep_ndp; 99 ndp->ni_cnd.cn_nameiop = LOOKUP; 100 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 101 /* first get the vnode */ 102 if ((error = namei(ndp)) != 0) 103 return error; 104 epp->ep_vp = vp = ndp->ni_vp; 105 106 /* check access and type */ 107 if (vp->v_type != VREG) { 108 error = EACCES; 109 goto bad1; 110 } 111 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 112 goto bad1; 113 114 /* get attributes */ 115 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 116 goto bad1; 117 118 /* Check mount point */ 119 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 120 error = EACCES; 121 goto bad1; 122 } 123 if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_flag & P_TRACED)) 124 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 125 126 /* try to open it */ 127 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 128 goto bad1; 129 130 /* unlock vp, since we need it unlocked from here on out. */ 131 VOP_UNLOCK(vp, 0); 132 133 /* now we have the file, get the exec header */ 134 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 135 UIO_SYSSPACE, 0, p->p_ucred, &resid, p); 136 if (error) 137 goto bad2; 138 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 139 140 /* 141 * set up the vmcmds for creation of the process 142 * address space 143 */ 144 error = ENOEXEC; 145 for (i = 0; i < nexecs && error != 0; i++) { 146 int newerror; 147 148 if (execsw[i].es_check == NULL) 149 continue; 150 151 newerror = (*execsw[i].es_check)(p, epp); 152 /* make sure the first "interesting" error code is saved. */ 153 if (!newerror || error == ENOEXEC) 154 error = newerror; 155 if (epp->ep_flags & EXEC_DESTR && error != 0) 156 return error; 157 } 158 if (!error) { 159 /* check that entry point is sane */ 160 if (epp->ep_entry > VM_MAXUSER_ADDRESS) 161 error = ENOEXEC; 162 163 /* check limits */ 164 if ((epp->ep_tsize > MAXTSIZ) || 165 (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur)) 166 error = ENOMEM; 167 168 if (!error) 169 return (0); 170 } 171 172 /* 173 * free any vmspace-creation commands, 174 * and release their references 175 */ 176 kill_vmcmds(&epp->ep_vmcmds); 177 178 bad2: 179 /* 180 * close and release the vnode, restore the old one, free the 181 * pathname buf, and punt. 182 */ 183 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 184 VOP_CLOSE(vp, FREAD, p->p_ucred, p); 185 vput(vp); 186 FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI); 187 return error; 188 189 bad1: 190 /* 191 * free the namei pathname buffer, and put the vnode 192 * (which we don't yet have open). 193 */ 194 vput(vp); /* was still locked */ 195 FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI); 196 return error; 197 } 198 199 /* 200 * exec system call 201 */ 202 /* ARGSUSED */ 203 int 204 sys_execve(p, v, retval) 205 struct proc *p; 206 void *v; 207 register_t *retval; 208 { 209 struct sys_execve_args /* { 210 syscallarg(const char *) path; 211 syscallarg(char * const *) argp; 212 syscallarg(char * const *) envp; 213 } */ *uap = v; 214 int error, i; 215 struct exec_package pack; 216 struct nameidata nid; 217 struct vattr attr; 218 struct ucred *cred = p->p_ucred; 219 char *argp; 220 char * const *cpp; 221 char *dp, *sp; 222 long argc, envc; 223 size_t len; 224 char *stack; 225 struct ps_strings arginfo; 226 struct vmspace *vm; 227 char **tmpfap; 228 int szsigcode; 229 struct exec_vmcmd *base_vcp = NULL; 230 extern struct emul emul_netbsd; 231 232 /* 233 * figure out the maximum size of an exec header, if necessary. 234 * XXX should be able to keep LKM code from modifying exec switch 235 * when we're still using it, but... 236 */ 237 if (exec_maxhdrsz == 0) { 238 for (i = 0; i < nexecs; i++) 239 if (execsw[i].es_check != NULL 240 && execsw[i].es_hdrsz > exec_maxhdrsz) 241 exec_maxhdrsz = execsw[i].es_hdrsz; 242 } 243 244 /* init the namei data to point the file user's program name */ 245 /* XXX cgd 960926: why do this here? most will be clobbered. */ 246 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 247 248 /* 249 * initialize the fields of the exec package. 250 */ 251 pack.ep_name = SCARG(uap, path); 252 MALLOC(pack.ep_hdr, void *, exec_maxhdrsz, M_EXEC, M_WAITOK); 253 pack.ep_hdrlen = exec_maxhdrsz; 254 pack.ep_hdrvalid = 0; 255 pack.ep_ndp = &nid; 256 pack.ep_emul_arg = NULL; 257 pack.ep_vmcmds.evs_cnt = 0; 258 pack.ep_vmcmds.evs_used = 0; 259 pack.ep_vap = &attr; 260 pack.ep_emul = &emul_netbsd; 261 pack.ep_flags = 0; 262 263 /* see if we can run it. */ 264 if ((error = check_exec(p, &pack)) != 0) 265 goto freehdr; 266 267 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 268 269 /* allocate an argument buffer */ 270 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS); 271 #ifdef DIAGNOSTIC 272 if (argp == (vaddr_t) 0) 273 panic("execve: argp == NULL"); 274 #endif 275 dp = argp; 276 argc = 0; 277 278 /* copy the fake args list, if there's one, freeing it as we go */ 279 if (pack.ep_flags & EXEC_HASARGL) { 280 tmpfap = pack.ep_fa; 281 while (*tmpfap != NULL) { 282 char *cp; 283 284 cp = *tmpfap; 285 while (*cp) 286 *dp++ = *cp++; 287 dp++; 288 289 FREE(*tmpfap, M_EXEC); 290 tmpfap++; argc++; 291 } 292 FREE(pack.ep_fa, M_EXEC); 293 pack.ep_flags &= ~EXEC_HASARGL; 294 } 295 296 /* Now get argv & environment */ 297 if (!(cpp = SCARG(uap, argp))) { 298 error = EINVAL; 299 goto bad; 300 } 301 302 if (pack.ep_flags & EXEC_SKIPARG) 303 cpp++; 304 305 while (1) { 306 len = argp + ARG_MAX - dp; 307 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 308 goto bad; 309 if (!sp) 310 break; 311 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 312 if (error == ENAMETOOLONG) 313 error = E2BIG; 314 goto bad; 315 } 316 dp += len; 317 cpp++; 318 argc++; 319 } 320 321 envc = 0; 322 /* environment need not be there */ 323 if ((cpp = SCARG(uap, envp)) != NULL ) { 324 while (1) { 325 len = argp + ARG_MAX - dp; 326 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 327 goto bad; 328 if (!sp) 329 break; 330 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 331 if (error == ENAMETOOLONG) 332 error = E2BIG; 333 goto bad; 334 } 335 dp += len; 336 cpp++; 337 envc++; 338 } 339 } 340 341 dp = (char *) ALIGN(dp); 342 343 szsigcode = pack.ep_emul->e_esigcode - pack.ep_emul->e_sigcode; 344 345 /* Now check if args & environ fit into new stack */ 346 if (pack.ep_flags & EXEC_32) 347 len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(int) + 348 sizeof(int) + dp + STACKGAPLEN + szsigcode + 349 sizeof(struct ps_strings)) - argp; 350 else 351 len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(char *) + 352 sizeof(int) + dp + STACKGAPLEN + szsigcode + 353 sizeof(struct ps_strings)) - argp; 354 355 len = ALIGN(len); /* make the stack "safely" aligned */ 356 357 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 358 error = ENOMEM; 359 goto bad; 360 } 361 362 /* adjust "active stack depth" for process VSZ */ 363 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 364 365 /* 366 * Do whatever is necessary to prepare the address space 367 * for remapping. Note that this might replace the current 368 * vmspace with another! 369 */ 370 uvmspace_exec(p); 371 372 /* Now map address space */ 373 vm = p->p_vmspace; 374 vm->vm_taddr = (char *) pack.ep_taddr; 375 vm->vm_tsize = btoc(pack.ep_tsize); 376 vm->vm_daddr = (char *) pack.ep_daddr; 377 vm->vm_dsize = btoc(pack.ep_dsize); 378 vm->vm_ssize = btoc(pack.ep_ssize); 379 vm->vm_maxsaddr = (char *) pack.ep_maxsaddr; 380 381 /* create the new process's VM space by running the vmcmds */ 382 #ifdef DIAGNOSTIC 383 if (pack.ep_vmcmds.evs_used == 0) 384 panic("execve: no vmcmds"); 385 #endif 386 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 387 struct exec_vmcmd *vcp; 388 389 vcp = &pack.ep_vmcmds.evs_cmds[i]; 390 if (vcp->ev_flags & VMCMD_RELATIVE) { 391 #ifdef DIAGNOSTIC 392 if (base_vcp == NULL) 393 panic("execve: relative vmcmd with no base"); 394 if (vcp->ev_flags & VMCMD_BASE) 395 panic("execve: illegal base & relative vmcmd"); 396 #endif 397 vcp->ev_addr += base_vcp->ev_addr; 398 } 399 error = (*vcp->ev_proc)(p, vcp); 400 #ifdef DEBUG 401 if (error) { 402 if (i > 0) 403 printf("vmcmd[%d] = %#lx/%#lx @ %#lx\n", i-1, 404 vcp[-1].ev_addr, vcp[-1].ev_len, 405 vcp[-1].ev_offset); 406 printf("vmcmd[%d] = %#lx/%#lx @ %#lx\n", i, 407 vcp->ev_addr, vcp->ev_len, vcp->ev_offset); 408 } 409 #endif 410 if (vcp->ev_flags & VMCMD_BASE) 411 base_vcp = vcp; 412 } 413 414 /* free the vmspace-creation commands, and release their references */ 415 kill_vmcmds(&pack.ep_vmcmds); 416 417 /* if an error happened, deallocate and punt */ 418 if (error) { 419 #ifdef DEBUG 420 printf("execve: vmcmd %i failed: %d\n", i-1, error); 421 #endif 422 goto exec_abort; 423 } 424 425 /* remember information about the process */ 426 arginfo.ps_nargvstr = argc; 427 arginfo.ps_nenvstr = envc; 428 429 stack = (char *) (USRSTACK - len); 430 /* Now copy argc, args & environ to new stack */ 431 if (!(*pack.ep_emul->e_copyargs)(&pack, &arginfo, stack, argp)) { 432 #ifdef DEBUG 433 printf("execve: copyargs failed\n"); 434 #endif 435 goto exec_abort; 436 } 437 438 /* copy out the process's ps_strings structure */ 439 if (copyout(&arginfo, (char *) PS_STRINGS, sizeof(arginfo))) { 440 #ifdef DEBUG 441 printf("execve: ps_strings copyout failed\n"); 442 #endif 443 goto exec_abort; 444 } 445 446 /* fill process ps_strings info */ 447 p->p_psstr = PS_STRINGS; 448 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 449 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 450 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 451 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 452 453 /* copy out the process's signal trapoline code */ 454 if (szsigcode) { 455 if (copyout((char *)pack.ep_emul->e_sigcode, 456 p->p_sigacts->ps_sigcode = (char *)PS_STRINGS - szsigcode, 457 szsigcode)) { 458 #ifdef DEBUG 459 printf("execve: sig trampoline copyout failed\n"); 460 #endif 461 goto exec_abort; 462 } 463 #ifdef PMAP_NEED_PROCWR 464 /* This is code. Let the pmap do what is needed. */ 465 pmap_procwr(p, (vaddr_t)p->p_sigacts->ps_sigcode, szsigcode); 466 #endif 467 } 468 469 stopprofclock(p); /* stop profiling */ 470 fdcloseexec(p); /* handle close on exec */ 471 execsigs(p); /* reset catched signals */ 472 p->p_ctxlink = NULL; /* reset ucontext link */ 473 474 /* set command name & other accounting info */ 475 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 476 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 477 p->p_comm[len] = 0; 478 p->p_acflag &= ~AFORK; 479 480 /* record proc's vnode, for use by procfs and others */ 481 if (p->p_textvp) 482 vrele(p->p_textvp); 483 VREF(pack.ep_vp); 484 p->p_textvp = pack.ep_vp; 485 486 p->p_flag |= P_EXEC; 487 if (p->p_flag & P_PPWAIT) { 488 p->p_flag &= ~P_PPWAIT; 489 wakeup((caddr_t) p->p_pptr); 490 } 491 492 /* 493 * deal with set[ug]id. 494 * MNT_NOSUID and P_TRACED have already been used to disable s[ug]id. 495 */ 496 if (((attr.va_mode & S_ISUID) != 0 && p->p_ucred->cr_uid != attr.va_uid) 497 || ((attr.va_mode & S_ISGID) != 0 && p->p_ucred->cr_gid != attr.va_gid)){ 498 p->p_ucred = crcopy(cred); 499 #ifdef KTRACE 500 /* 501 * If process is being ktraced, turn off - unless 502 * root set it. 503 */ 504 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) 505 ktrderef(p); 506 #endif 507 if (attr.va_mode & S_ISUID) 508 p->p_ucred->cr_uid = attr.va_uid; 509 if (attr.va_mode & S_ISGID) 510 p->p_ucred->cr_gid = attr.va_gid; 511 p_sugid(p); 512 } else 513 p->p_flag &= ~P_SUGID; 514 p->p_cred->p_svuid = p->p_ucred->cr_uid; 515 p->p_cred->p_svgid = p->p_ucred->cr_gid; 516 517 doexechooks(p); 518 519 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 520 521 FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI); 522 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 523 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 524 vput(pack.ep_vp); 525 526 /* setup new registers and do misc. setup. */ 527 (*pack.ep_emul->e_setregs)(p, &pack, (u_long) stack); 528 529 if (p->p_flag & P_TRACED) 530 psignal(p, SIGTRAP); 531 532 p->p_emul = pack.ep_emul; 533 FREE(pack.ep_hdr, M_EXEC); 534 535 #ifdef KTRACE 536 if (KTRPOINT(p, KTR_EMUL)) 537 ktremul(p); 538 #endif 539 540 return (EJUSTRETURN); 541 542 bad: 543 /* free the vmspace-creation commands, and release their references */ 544 kill_vmcmds(&pack.ep_vmcmds); 545 /* kill any opened file descriptor, if necessary */ 546 if (pack.ep_flags & EXEC_HASFD) { 547 pack.ep_flags &= ~EXEC_HASFD; 548 (void) fdrelease(p, pack.ep_fd); 549 } 550 /* close and put the exec'd file */ 551 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 552 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 553 vput(pack.ep_vp); 554 FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI); 555 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 556 557 freehdr: 558 FREE(pack.ep_hdr, M_EXEC); 559 return error; 560 561 exec_abort: 562 /* 563 * the old process doesn't exist anymore. exit gracefully. 564 * get rid of the (new) address space we have created, if any, get rid 565 * of our namei data and vnode, and exit noting failure 566 */ 567 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 568 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 569 if (pack.ep_emul_arg) 570 FREE(pack.ep_emul_arg, M_TEMP); 571 FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI); 572 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 573 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 574 vput(pack.ep_vp); 575 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 576 FREE(pack.ep_hdr, M_EXEC); 577 exit1(p, W_EXITCODE(0, SIGABRT)); 578 exit1(p, -1); 579 580 /* NOTREACHED */ 581 return 0; 582 } 583 584 585 void * 586 copyargs(pack, arginfo, stack, argp) 587 struct exec_package *pack; 588 struct ps_strings *arginfo; 589 void *stack; 590 void *argp; 591 { 592 char **cpp = stack; 593 char *dp, *sp; 594 size_t len; 595 void *nullp = NULL; 596 long argc = arginfo->ps_nargvstr; 597 long envc = arginfo->ps_nenvstr; 598 599 #ifdef __sparc_v9__ 600 /* XXX Temporary hack for argc format conversion. */ 601 argc = (argc << 32) | (argc & 0xffffffff); 602 #endif 603 if (copyout(&argc, cpp++, sizeof(argc))) 604 return NULL; 605 606 dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen); 607 sp = argp; 608 609 /* XXX don't copy them out, remap them! */ 610 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 611 612 for (; --argc >= 0; sp += len, dp += len) 613 if (copyout(&dp, cpp++, sizeof(dp)) || 614 copyoutstr(sp, dp, ARG_MAX, &len)) 615 return NULL; 616 617 if (copyout(&nullp, cpp++, sizeof(nullp))) 618 return NULL; 619 620 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 621 622 for (; --envc >= 0; sp += len, dp += len) 623 if (copyout(&dp, cpp++, sizeof(dp)) || 624 copyoutstr(sp, dp, ARG_MAX, &len)) 625 return NULL; 626 627 if (copyout(&nullp, cpp++, sizeof(nullp))) 628 return NULL; 629 630 return cpp; 631 } 632
Indexes created Wed Sep 24 05:09:52 GMT 2025