1 /*- 2 * Copyright (c) 1982, 1986, 1991 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.proprietary.c% 6 * 7 * @(#)kern_exec.c 7.45 (Berkeley) 05/15/91 8 */ 9 10 #include "param.h" 11 #include "systm.h" 12 #include "filedesc.h" 13 #include "kernel.h" 14 #include "proc.h" 15 #include "mount.h" 16 #include "malloc.h" 17 #include "namei.h" 18 #include "vnode.h" 19 #include "seg.h" 20 #include "file.h" 21 #include "acct.h" 22 #include "exec.h" 23 #include "ktrace.h" 24 #include "resourcevar.h" 25 26 #include "machine/cpu.h" 27 #include "machine/reg.h" 28 29 #include "mman.h" 30 #include "vm/vm.h" 31 #include "vm/vm_param.h" 32 #include "vm/vm_map.h" 33 #include "vm/vm_kern.h" 34 #include "vm/vm_pager.h" 35 36 #include "signalvar.h" 37 #include "kinfo_proc.h" 38 39 #ifdef HPUXCOMPAT 40 #include "user.h" /* for pcb */ 41 #include "hp300/hpux/hpux_exec.h" 42 #endif 43 44 #ifdef COPY_SIGCODE 45 extern char sigcode[], esigcode[]; 46 #define szsigcode (esigcode - sigcode) 47 #else 48 #define szsigcode 0 49 #endif 50 51 /* 52 * exec system call 53 */ 54 /* ARGSUSED */ 55 execve(p, uap, retval) 56 register struct proc *p; 57 register struct args { 58 char *fname; 59 char **argp; 60 char **envp; 61 } *uap; 62 int *retval; 63 { 64 register struct ucred *cred = p->p_ucred; 65 register struct nameidata *ndp; 66 register struct filedesc *fdp = p->p_fd; 67 int na, ne, ucp, ap, cc; 68 register char *cp; 69 register int nc; 70 unsigned len; 71 int indir, uid, gid; 72 char *sharg; 73 struct vnode *vp; 74 int resid, error, paged = 0; 75 vm_offset_t execargs; 76 struct vattr vattr; 77 char cfarg[MAXINTERP]; 78 union { 79 char ex_shell[MAXINTERP]; /* #! and interpreter name */ 80 struct exec ex_exec; 81 #ifdef HPUXCOMPAT 82 struct hpux_exec ex_hexec; 83 #endif 84 } exdata; 85 #ifdef HPUXCOMPAT 86 struct hpux_exec hhead; 87 #endif 88 struct nameidata nd; 89 90 ndp = &nd; 91 ndp->ni_nameiop = LOOKUP | FOLLOW | LOCKLEAF | SAVENAME; 92 ndp->ni_segflg = UIO_USERSPACE; 93 ndp->ni_dirp = uap->fname; 94 if (error = namei(ndp, p)) 95 return (error); 96 vp = ndp->ni_vp; 97 indir = 0; 98 uid = cred->cr_uid; 99 gid = cred->cr_gid; 100 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 101 goto bad; 102 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 103 error = EACCES; 104 goto bad; 105 } 106 if ((vp->v_mount->mnt_flag & MNT_NOSUID) == 0) { 107 if (vattr.va_mode & VSUID) 108 uid = vattr.va_uid; 109 if (vattr.va_mode & VSGID) 110 gid = vattr.va_gid; 111 } 112 113 again: 114 if (error = VOP_ACCESS(vp, VEXEC, cred, p)) 115 goto bad; 116 if ((p->p_flag & STRC) && (error = VOP_ACCESS(vp, VREAD, cred, p))) 117 goto bad; 118 if (vp->v_type != VREG || 119 (vattr.va_mode & (VEXEC|(VEXEC>>3)|(VEXEC>>6))) == 0) { 120 error = EACCES; 121 goto bad; 122 } 123 124 /* 125 * Read in first few bytes of file for segment sizes, magic number: 126 * OMAGIC = plain executable 127 * NMAGIC = RO text 128 * ZMAGIC = demand paged RO text 129 * Also an ASCII line beginning with #! is 130 * the file name of a ``shell'' and arguments may be prepended 131 * to the argument list if given here. 132 * 133 * SHELL NAMES ARE LIMITED IN LENGTH. 134 * 135 * ONLY ONE ARGUMENT MAY BE PASSED TO THE SHELL FROM 136 * THE ASCII LINE. 137 */ 138 exdata.ex_shell[0] = '\0'; /* for zero length files */ 139 error = vn_rdwr(UIO_READ, vp, (caddr_t)&exdata, sizeof (exdata), 140 (off_t)0, UIO_SYSSPACE, (IO_UNIT|IO_NODELOCKED), cred, &resid, 141 (struct proc *)0); 142 if (error) 143 goto bad; 144 #ifndef lint 145 if (resid > sizeof(exdata) - sizeof(exdata.ex_exec) && 146 exdata.ex_shell[0] != '#') { 147 error = ENOEXEC; 148 goto bad; 149 } 150 #endif 151 #if defined(hp300) 152 switch ((int)exdata.ex_exec.a_mid) { 153 154 /* 155 * An ancient hp200 or hp300 binary, shouldn't happen anymore. 156 * Mark as invalid. 157 */ 158 case MID_ZERO: 159 exdata.ex_exec.a_magic = 0; 160 break; 161 162 /* 163 * HP200 series has a smaller page size so we cannot 164 * demand-load or even write protect text, so we just 165 * treat as OMAGIC. 166 */ 167 case MID_HP200: 168 exdata.ex_exec.a_magic = OMAGIC; 169 break; 170 171 case MID_HP300: 172 break; 173 174 #ifdef HPUXCOMPAT 175 case MID_HPUX: 176 /* 177 * Save a.out header. This is eventually saved in the pcb, 178 * but we cannot do that yet in case the exec fails before 179 * the image is overlayed. 180 */ 181 bcopy((caddr_t)&exdata.ex_hexec, 182 (caddr_t)&hhead, sizeof hhead); 183 /* 184 * If version number is 0x2bad this is a native BSD 185 * binary created via the HPUX SGS. Should not be 186 * treated as an HPUX binary. 187 */ 188 if (exdata.ex_hexec.ha_version != BSDVNUM) 189 paged |= SHPUX; /* XXX */ 190 /* 191 * Shuffle important fields to their BSD locations. 192 * Note that the order in which this is done is important. 193 */ 194 exdata.ex_exec.a_text = exdata.ex_hexec.ha_text; 195 exdata.ex_exec.a_data = exdata.ex_hexec.ha_data; 196 exdata.ex_exec.a_bss = exdata.ex_hexec.ha_bss; 197 exdata.ex_exec.a_entry = exdata.ex_hexec.ha_entry; 198 /* 199 * For ZMAGIC files, make sizes consistant with those 200 * generated by BSD ld. 201 */ 202 if (exdata.ex_exec.a_magic == ZMAGIC) { 203 exdata.ex_exec.a_text = 204 ctob(btoc(exdata.ex_exec.a_text)); 205 nc = exdata.ex_exec.a_data + exdata.ex_exec.a_bss; 206 exdata.ex_exec.a_data = 207 ctob(btoc(exdata.ex_exec.a_data)); 208 nc -= (int)exdata.ex_exec.a_data; 209 exdata.ex_exec.a_bss = (nc < 0) ? 0 : nc; 210 } 211 break; 212 #endif 213 } 214 #endif 215 switch ((int)exdata.ex_exec.a_magic) { 216 217 case OMAGIC: 218 exdata.ex_exec.a_data += exdata.ex_exec.a_text; 219 exdata.ex_exec.a_text = 0; 220 break; 221 222 case ZMAGIC: 223 paged++; 224 /* FALLTHROUGH */ 225 case NMAGIC: 226 if (exdata.ex_exec.a_text == 0) { 227 error = ENOEXEC; 228 goto bad; 229 } 230 break; 231 232 default: 233 if (exdata.ex_shell[0] != '#' || 234 exdata.ex_shell[1] != '!' || 235 indir) { 236 error = ENOEXEC; 237 goto bad; 238 } 239 for (cp = &exdata.ex_shell[2];; ++cp) { 240 if (cp >= &exdata.ex_shell[MAXINTERP]) { 241 error = ENOEXEC; 242 goto bad; 243 } 244 if (*cp == '\n') { 245 *cp = '\0'; 246 break; 247 } 248 if (*cp == '\t') 249 *cp = ' '; 250 } 251 cp = &exdata.ex_shell[2]; 252 while (*cp == ' ') 253 cp++; 254 ndp->ni_dirp = cp; 255 while (*cp && *cp != ' ') 256 cp++; 257 cfarg[0] = '\0'; 258 if (*cp) { 259 *cp++ = '\0'; 260 while (*cp == ' ') 261 cp++; 262 if (*cp) 263 bcopy((caddr_t)cp, (caddr_t)cfarg, MAXINTERP); 264 } 265 indir = 1; 266 vput(vp); 267 ndp->ni_segflg = UIO_SYSSPACE; 268 if (error = namei(ndp, p)) 269 return (error); 270 vp = ndp->ni_vp; 271 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 272 goto bad; 273 uid = cred->cr_uid; /* shell scripts can't be setuid */ 274 gid = cred->cr_gid; 275 goto again; 276 } 277 278 /* 279 * Collect arguments on "file" in swap space. 280 */ 281 na = 0; 282 ne = 0; 283 nc = 0; 284 cc = NCARGS; 285 execargs = kmem_alloc_wait(exec_map, NCARGS); 286 cp = (char *) execargs; 287 /* 288 * Copy arguments into file in argdev area. 289 */ 290 if (uap->argp) for (;;) { 291 ap = NULL; 292 sharg = NULL; 293 if (indir && na == 0) { 294 sharg = ndp->ni_ptr; 295 ap = (int)sharg; 296 uap->argp++; /* ignore argv[0] */ 297 } else if (indir && (na == 1 && cfarg[0])) { 298 sharg = cfarg; 299 ap = (int)sharg; 300 } else if (indir && (na == 1 || na == 2 && cfarg[0])) 301 ap = (int)uap->fname; 302 else if (uap->argp) { 303 ap = fuword((caddr_t)uap->argp); 304 uap->argp++; 305 } 306 if (ap == NULL && uap->envp) { 307 uap->argp = NULL; 308 if ((ap = fuword((caddr_t)uap->envp)) != NULL) 309 uap->envp++, ne++; 310 } 311 if (ap == NULL) 312 break; 313 na++; 314 if (ap == -1) { 315 error = EFAULT; 316 goto bad; 317 } 318 do { 319 if (nc >= NCARGS-1) { 320 error = E2BIG; 321 break; 322 } 323 if (sharg) { 324 error = copystr(sharg, cp, (unsigned)cc, &len); 325 sharg += len; 326 } else { 327 error = copyinstr((caddr_t)ap, cp, (unsigned)cc, 328 &len); 329 ap += len; 330 } 331 cp += len; 332 nc += len; 333 cc -= len; 334 } while (error == ENAMETOOLONG); 335 if (error) 336 goto bad; 337 } 338 nc = (nc + NBPW-1) & ~(NBPW-1); 339 error = getxfile(p, vp, &exdata.ex_exec, paged, nc + (na+4)*NBPW, 340 uid, gid); 341 if (error) 342 goto bad; 343 vput(vp); 344 vp = NULL; 345 346 #ifdef HPUXCOMPAT 347 /* 348 * We are now committed to the exec so we can save the exec 349 * header in the pcb where we can dump it if necessary in core() 350 */ 351 if (p->p_addr->u_pcb.pcb_flags & PCB_HPUXBIN) 352 bcopy((caddr_t)&hhead, 353 (caddr_t)p->p_addr->u_pcb.pcb_exec, sizeof hhead); 354 #endif 355 356 /* 357 * Copy back arglist. 358 */ 359 ucp = USRSTACK - szsigcode - nc - NBPW; 360 ap = ucp - na*NBPW - 3*NBPW; 361 p->p_regs[SP] = ap; 362 (void) suword((caddr_t)ap, na-ne); 363 nc = 0; 364 cp = (char *) execargs; 365 cc = NCARGS; 366 for (;;) { 367 ap += NBPW; 368 if (na == ne) { 369 (void) suword((caddr_t)ap, 0); 370 ap += NBPW; 371 } 372 if (--na < 0) 373 break; 374 (void) suword((caddr_t)ap, ucp); 375 do { 376 error = copyoutstr(cp, (caddr_t)ucp, (unsigned)cc, 377 &len); 378 ucp += len; 379 cp += len; 380 nc += len; 381 cc -= len; 382 } while (error == ENAMETOOLONG); 383 if (error == EFAULT) 384 panic("exec: EFAULT"); 385 } 386 (void) suword((caddr_t)ap, 0); 387 388 execsigs(p); 389 390 for (nc = fdp->fd_lastfile; nc >= 0; --nc) { 391 if (fdp->fd_ofileflags[nc] & UF_EXCLOSE) { 392 (void) closef(fdp->fd_ofiles[nc], p); 393 fdp->fd_ofiles[nc] = NULL; 394 fdp->fd_ofileflags[nc] = 0; 395 if (nc < fdp->fd_freefile) 396 fdp->fd_freefile = nc; 397 } 398 fdp->fd_ofileflags[nc] &= ~UF_MAPPED; 399 } 400 /* 401 * Adjust fd_lastfile to account for descriptors closed above. 402 * Don't decrement fd_lastfile past 0, as it's unsigned. 403 */ 404 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 405 fdp->fd_lastfile--; 406 setregs(p, exdata.ex_exec.a_entry, retval); 407 #ifdef COPY_SIGCODE 408 /* 409 * Install sigcode at top of user stack. 410 */ 411 copyout((caddr_t)sigcode, (caddr_t)(USRSTACK - szsigcode), szsigcode); 412 #endif 413 /* 414 * Remember file name for accounting. 415 */ 416 p->p_acflag &= ~AFORK; 417 if (ndp->ni_namelen > MAXCOMLEN) 418 ndp->ni_namelen = MAXCOMLEN; 419 bcopy((caddr_t)ndp->ni_ptr, (caddr_t)p->p_comm, 420 (unsigned)(ndp->ni_namelen)); 421 p->p_comm[ndp->ni_namelen] = '\0'; 422 cpu_exec(p); 423 bad: 424 FREE(ndp->ni_pnbuf, M_NAMEI); 425 if (execargs) 426 kmem_free_wakeup(exec_map, execargs, NCARGS); 427 if (vp) 428 vput(vp); 429 return (error); 430 } 431 432 /* 433 * Read in and set up memory for executed file. 434 */ 435 getxfile(p, vp, ep, paged, nargc, uid, gid) 436 register struct proc *p; 437 register struct vnode *vp; 438 register struct exec *ep; 439 int paged, nargc, uid, gid; 440 { 441 segsz_t ts, ds, ss; 442 register struct ucred *cred = p->p_ucred; 443 off_t toff; 444 int error = 0; 445 vm_offset_t addr; 446 vm_size_t size; 447 struct vmspace *vm = p->p_vmspace; 448 449 #ifdef HPUXCOMPAT 450 int hpux = (paged & SHPUX); 451 paged &= ~SHPUX; 452 if (ep->a_mid == MID_HPUX) { 453 if (paged) 454 toff = CLBYTES; 455 else 456 toff = sizeof (struct hpux_exec); 457 } else 458 #endif 459 if (paged) 460 toff = CLBYTES; 461 else 462 toff = sizeof (struct exec); 463 if (ep->a_text != 0 && (vp->v_flag & VTEXT) == 0 && 464 vp->v_usecount != 1) { 465 register struct file *fp; 466 467 for (fp = file; fp < fileNFILE; fp++) { 468 if (fp->f_type == DTYPE_VNODE && 469 fp->f_count > 0 && 470 (struct vnode *)fp->f_data == vp && 471 (fp->f_flag & FWRITE)) { 472 return (ETXTBSY); 473 } 474 } 475 } 476 477 /* 478 * Compute text and data sizes and make sure not too large. 479 * NB - Check data and bss separately as they may overflow 480 * when summed together. 481 */ 482 ts = clrnd(btoc(ep->a_text)); 483 ds = clrnd(btoc(ep->a_data + ep->a_bss)); 484 ss = clrnd(SSIZE + btoc(nargc + szsigcode)); 485 486 /* 487 * If we're sharing the address space, allocate a new space 488 * and release our reference to the old one. Otherwise, 489 * empty out the existing vmspace. 490 */ 491 if (vm->vm_refcnt > 1) { 492 p->p_vmspace = vmspace_alloc(VM_MIN_ADDRESS, 493 VM_MAXUSER_ADDRESS, 1); 494 vmspace_free(vm); 495 vm = p->p_vmspace; 496 } else { 497 #ifdef SYSVSHM 498 if (vm->vm_shm) 499 shmexit(p); 500 #endif 501 (void) vm_map_remove(&vm->vm_map, VM_MIN_ADDRESS, 502 VM_MAXUSER_ADDRESS); 503 } 504 /* 505 * If parent is waiting for us to exec or exit, 506 * SPPWAIT will be set; clear it and wakeup parent. 507 */ 508 if (p->p_flag & SPPWAIT) { 509 p->p_flag &= ~SPPWAIT; 510 wakeup((caddr_t) p->p_pptr); 511 } 512 #ifdef HPUXCOMPAT 513 p->p_addr->u_pcb.pcb_flags &= ~(PCB_HPUXMMAP|PCB_HPUXBIN); 514 /* remember that we were loaded from an HPUX format file */ 515 if (ep->a_mid == MID_HPUX) 516 p->p_addr->u_pcb.pcb_flags |= PCB_HPUXBIN; 517 if (hpux) 518 p->p_flag |= SHPUX; 519 else 520 p->p_flag &= ~SHPUX; 521 #endif 522 p->p_flag |= SEXEC; 523 addr = VM_MIN_ADDRESS; 524 if (vm_allocate(&vm->vm_map, &addr, round_page(ctob(ts + ds)), FALSE)) { 525 uprintf("Cannot allocate text+data space\n"); 526 error = ENOMEM; /* XXX */ 527 goto badmap; 528 } 529 size = round_page(MAXSSIZ); /* XXX */ 530 #ifdef i386 531 addr = trunc_page(USRSTACK - size) - NBPG; /* XXX */ 532 #else 533 addr = trunc_page(USRSTACK - size); 534 #endif 535 if (vm_allocate(&vm->vm_map, &addr, size, FALSE)) { 536 uprintf("Cannot allocate stack space\n"); 537 error = ENOMEM; /* XXX */ 538 goto badmap; 539 } 540 size -= round_page(p->p_rlimit[RLIMIT_STACK].rlim_cur); 541 if (vm_map_protect(&vm->vm_map, addr, addr+size, VM_PROT_NONE, FALSE)) { 542 uprintf("Cannot protect stack space\n"); 543 error = ENOMEM; 544 goto badmap; 545 } 546 vm->vm_maxsaddr = (caddr_t)addr; 547 vm->vm_taddr = (caddr_t)VM_MIN_ADDRESS; 548 vm->vm_daddr = (caddr_t)(VM_MIN_ADDRESS + ctob(ts)); 549 550 if (paged == 0) { 551 /* 552 * Read in data segment. 553 */ 554 (void) vn_rdwr(UIO_READ, vp, vm->vm_daddr, (int) ep->a_data, 555 (off_t)(toff + ep->a_text), UIO_USERSPACE, 556 (IO_UNIT|IO_NODELOCKED), cred, (int *)0, p); 557 /* 558 * Read in text segment if necessary (0410), 559 * and read-protect it. 560 */ 561 if (ep->a_text > 0) { 562 error = vn_rdwr(UIO_READ, vp, vm->vm_taddr, 563 (int)ep->a_text, toff, UIO_USERSPACE, 564 (IO_UNIT|IO_NODELOCKED), cred, (int *)0, p); 565 (void) vm_map_protect(&vm->vm_map, VM_MIN_ADDRESS, 566 VM_MIN_ADDRESS + trunc_page(ep->a_text), 567 VM_PROT_READ|VM_PROT_EXECUTE, FALSE); 568 } 569 } else { 570 /* 571 * Allocate a region backed by the exec'ed vnode. 572 */ 573 addr = VM_MIN_ADDRESS; 574 size = round_page(ep->a_text + ep->a_data); 575 error = vm_mmap(&vm->vm_map, &addr, size, VM_PROT_ALL, 576 MAP_FILE|MAP_COPY|MAP_FIXED, 577 (caddr_t)vp, (vm_offset_t)toff); 578 (void) vm_map_protect(&vm->vm_map, addr, 579 addr + trunc_page(ep->a_text), 580 VM_PROT_READ|VM_PROT_EXECUTE, FALSE); 581 vp->v_flag |= VTEXT; 582 } 583 badmap: 584 if (error) { 585 printf("pid %d: VM allocation failure\n", p->p_pid); 586 uprintf("sorry, pid %d was killed in exec: VM allocation\n", 587 p->p_pid); 588 psignal(p, SIGKILL); 589 p->p_flag |= SKEEP; 590 return(error); 591 } 592 593 /* 594 * set SUID/SGID protections, if no tracing 595 */ 596 if ((p->p_flag&STRC)==0) { 597 if (uid != cred->cr_uid || gid != cred->cr_gid) { 598 p->p_ucred = cred = crcopy(cred); 599 /* 600 * If process is being ktraced, turn off - unless 601 * root set it. 602 */ 603 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) { 604 vrele(p->p_tracep); 605 p->p_tracep = NULL; 606 p->p_traceflag = 0; 607 } 608 } 609 cred->cr_uid = uid; 610 cred->cr_gid = gid; 611 } else 612 psignal(p, SIGTRAP); 613 p->p_cred->p_svuid = cred->cr_uid; 614 p->p_cred->p_svgid = cred->cr_gid; 615 vm->vm_tsize = ts; 616 vm->vm_dsize = ds; 617 vm->vm_ssize = ss; 618 p->p_stats->p_prof.pr_scale = 0; 619 #if defined(tahoe) 620 /* move this when tahoe cpu_exec is created */ 621 p->p_addr->u_pcb.pcb_savacc.faddr = (float *)NULL; 622 #endif 623 return (0); 624 } 625