1 /* $OpenBSD: kern_exec.c,v 1.254 2024/01/17 18:56:13 deraadt Exp $ */ 2 /* $NetBSD: kern_exec.c,v 1.75 1996/02/09 18:59:28 christos Exp $ */ 3 4 /*- 5 * Copyright (C) 1993, 1994 Christopher G. Demetriou 6 * Copyright (C) 1992 Wolfgang Solfrank. 7 * Copyright (C) 1992 TooLs GmbH. 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by TooLs GmbH. 21 * 4. The name of TooLs GmbH may not be used to endorse or promote products 22 * derived from this software without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 25 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 26 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 27 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 29 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 30 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 31 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 32 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 33 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 34 */ 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/filedesc.h> 39 #include <sys/proc.h> 40 #include <sys/user.h> 41 #include <sys/mount.h> 42 #include <sys/malloc.h> 43 #include <sys/pool.h> 44 #include <sys/namei.h> 45 #include <sys/vnode.h> 46 #include <sys/fcntl.h> 47 #include <sys/file.h> 48 #include <sys/acct.h> 49 #include <sys/exec.h> 50 #include <sys/exec_elf.h> 51 #include <sys/ktrace.h> 52 #include <sys/resourcevar.h> 53 #include <sys/mman.h> 54 #include <sys/signalvar.h> 55 #include <sys/stat.h> 56 #include <sys/conf.h> 57 #include <sys/pledge.h> 58 #ifdef SYSVSHM 59 #include <sys/shm.h> 60 #endif 61 62 #include <sys/syscallargs.h> 63 64 #include <uvm/uvm_extern.h> 65 #include <machine/tcb.h> 66 67 #include <sys/timetc.h> 68 69 struct uvm_object *sigobject; /* shared sigcode object */ 70 vaddr_t sigcode_va; 71 vsize_t sigcode_sz; 72 struct uvm_object *timekeep_object; 73 struct timekeep *timekeep; 74 75 void unveil_destroy(struct process *ps); 76 77 const struct kmem_va_mode kv_exec = { 78 .kv_wait = 1, 79 .kv_map = &exec_map 80 }; 81 82 /* 83 * Map the shared signal code. 84 */ 85 int exec_sigcode_map(struct process *); 86 87 /* 88 * Map the shared timekeep page. 89 */ 90 int exec_timekeep_map(struct process *); 91 92 /* 93 * If non-zero, stackgap_random specifies the upper limit of the random gap size 94 * added to the fixed stack position. Must be n^2. 95 */ 96 int stackgap_random = STACKGAP_RANDOM; 97 98 /* 99 * check exec: 100 * given an "executable" described in the exec package's namei info, 101 * see what we can do with it. 102 * 103 * ON ENTRY: 104 * exec package with appropriate namei info 105 * proc pointer of exec'ing proc 106 * NO SELF-LOCKED VNODES 107 * 108 * ON EXIT: 109 * error: nothing held, etc. exec header still allocated. 110 * ok: filled exec package, one locked vnode. 111 * 112 * EXEC SWITCH ENTRY: 113 * Locked vnode to check, exec package, proc. 114 * 115 * EXEC SWITCH EXIT: 116 * ok: return 0, filled exec package, one locked vnode. 117 * error: destructive: 118 * everything deallocated except exec header. 119 * non-destructive: 120 * error code, locked vnode, exec header unmodified 121 */ 122 int 123 check_exec(struct proc *p, struct exec_package *epp) 124 { 125 int error, i; 126 struct vnode *vp; 127 struct nameidata *ndp; 128 size_t resid; 129 130 ndp = epp->ep_ndp; 131 ndp->ni_cnd.cn_nameiop = LOOKUP; 132 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 133 if (epp->ep_flags & EXEC_INDIR) 134 ndp->ni_cnd.cn_flags |= BYPASSUNVEIL; 135 /* first get the vnode */ 136 if ((error = namei(ndp)) != 0) 137 return (error); 138 epp->ep_vp = vp = ndp->ni_vp; 139 140 /* check for regular file */ 141 if (vp->v_type != VREG) { 142 error = EACCES; 143 goto bad1; 144 } 145 146 /* get attributes */ 147 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 148 goto bad1; 149 150 /* Check mount point */ 151 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 152 error = EACCES; 153 goto bad1; 154 } 155 156 /* SUID programs may not be started with execpromises */ 157 if ((epp->ep_vap->va_mode & (VSUID | VSGID)) && 158 (p->p_p->ps_flags & PS_EXECPLEDGE)) { 159 error = EACCES; 160 goto bad1; 161 } 162 163 if ((vp->v_mount->mnt_flag & MNT_NOSUID)) 164 epp->ep_vap->va_mode &= ~(VSUID | VSGID); 165 166 /* check access. for root we have to see if any exec bit on */ 167 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 168 goto bad1; 169 if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 170 error = EACCES; 171 goto bad1; 172 } 173 174 /* try to open it */ 175 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 176 goto bad1; 177 178 /* unlock vp, we need it unlocked from here */ 179 VOP_UNLOCK(vp); 180 181 /* now we have the file, get the exec header */ 182 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 183 UIO_SYSSPACE, 0, p->p_ucred, &resid, p); 184 if (error) 185 goto bad2; 186 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 187 188 /* 189 * set up the vmcmds for creation of the process 190 * address space 191 */ 192 error = ENOEXEC; 193 for (i = 0; i < nexecs && error != 0; i++) { 194 int newerror; 195 196 if (execsw[i].es_check == NULL) 197 continue; 198 newerror = (*execsw[i].es_check)(p, epp); 199 /* make sure the first "interesting" error code is saved. */ 200 if (!newerror || error == ENOEXEC) 201 error = newerror; 202 if (epp->ep_flags & EXEC_DESTR && error != 0) 203 return (error); 204 } 205 if (!error) { 206 /* check that entry point is sane */ 207 if (epp->ep_entry > VM_MAXUSER_ADDRESS) { 208 error = ENOEXEC; 209 } 210 211 /* check limits */ 212 if ((epp->ep_tsize > MAXTSIZ) || 213 (epp->ep_dsize > lim_cur(RLIMIT_DATA))) 214 error = ENOMEM; 215 216 if (!error) 217 return (0); 218 } 219 220 /* 221 * free any vmspace-creation commands, 222 * and release their references 223 */ 224 kill_vmcmds(&epp->ep_vmcmds); 225 226 bad2: 227 /* 228 * close the vnode, free the pathname buf, and punt. 229 */ 230 vn_close(vp, FREAD, p->p_ucred, p); 231 pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf); 232 return (error); 233 234 bad1: 235 /* 236 * free the namei pathname buffer, and put the vnode 237 * (which we don't yet have open). 238 */ 239 pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf); 240 vput(vp); 241 return (error); 242 } 243 244 /* 245 * exec system call 246 */ 247 int 248 sys_execve(struct proc *p, void *v, register_t *retval) 249 { 250 struct sys_execve_args /* { 251 syscallarg(const char *) path; 252 syscallarg(char *const *) argp; 253 syscallarg(char *const *) envp; 254 } */ *uap = v; 255 int error; 256 struct exec_package pack; 257 struct nameidata nid; 258 struct vattr attr; 259 struct ucred *cred = p->p_ucred; 260 char *argp; 261 char * const *cpp, *dp, *sp; 262 #ifdef KTRACE 263 char *env_start; 264 #endif 265 struct process *pr = p->p_p; 266 long argc, envc; 267 size_t len, sgap, dstsize; 268 #ifdef MACHINE_STACK_GROWS_UP 269 size_t slen; 270 #endif 271 char *stack; 272 struct ps_strings arginfo; 273 struct vmspace *vm = p->p_vmspace; 274 struct vnode *otvp; 275 276 /* 277 * Get other threads to stop, if contested return ERESTART, 278 * so the syscall is restarted after halting in userret. 279 */ 280 if (single_thread_set(p, SINGLE_UNWIND | SINGLE_DEEP)) 281 return (ERESTART); 282 283 /* 284 * Cheap solution to complicated problems. 285 * Mark this process as "leave me alone, I'm execing". 286 */ 287 atomic_setbits_int(&pr->ps_flags, PS_INEXEC); 288 289 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 290 nid.ni_pledge = PLEDGE_EXEC; 291 nid.ni_unveil = UNVEIL_EXEC; 292 293 /* 294 * initialize the fields of the exec package. 295 */ 296 pack.ep_name = (char *)SCARG(uap, path); 297 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 298 pack.ep_hdrlen = exec_maxhdrsz; 299 pack.ep_hdrvalid = 0; 300 pack.ep_ndp = &nid; 301 pack.ep_interp = NULL; 302 pack.ep_args = NULL; 303 pack.ep_auxinfo = NULL; 304 VMCMDSET_INIT(&pack.ep_vmcmds); 305 pack.ep_vap = &attr; 306 pack.ep_flags = 0; 307 pack.ep_pins = NULL; 308 pack.ep_npins = 0; 309 310 /* see if we can run it. */ 311 if ((error = check_exec(p, &pack)) != 0) { 312 goto freehdr; 313 } 314 315 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 316 317 /* allocate an argument buffer */ 318 argp = km_alloc(NCARGS, &kv_exec, &kp_pageable, &kd_waitok); 319 #ifdef DIAGNOSTIC 320 if (argp == NULL) 321 panic("execve: argp == NULL"); 322 #endif 323 dp = argp; 324 argc = 0; 325 326 /* 327 * Copy the fake args list, if there's one, freeing it as we go. 328 * exec_script_makecmds() allocates either 2 or 3 fake args bounded 329 * by MAXINTERP + MAXPATHLEN < NCARGS so no overflow can happen. 330 */ 331 if (pack.ep_flags & EXEC_HASARGL) { 332 dstsize = NCARGS; 333 for(; pack.ep_fa[argc] != NULL; argc++) { 334 len = strlcpy(dp, pack.ep_fa[argc], dstsize); 335 len++; 336 dp += len; dstsize -= len; 337 if (pack.ep_fa[argc+1] != NULL) 338 free(pack.ep_fa[argc], M_EXEC, len); 339 else 340 free(pack.ep_fa[argc], M_EXEC, MAXPATHLEN); 341 } 342 free(pack.ep_fa, M_EXEC, 4 * sizeof(char *)); 343 pack.ep_flags &= ~EXEC_HASARGL; 344 } 345 346 /* Now get argv & environment */ 347 if (!(cpp = SCARG(uap, argp))) { 348 error = EFAULT; 349 goto bad; 350 } 351 352 if (pack.ep_flags & EXEC_SKIPARG) 353 cpp++; 354 355 while (1) { 356 len = argp + ARG_MAX - dp; 357 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 358 goto bad; 359 if (!sp) 360 break; 361 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 362 if (error == ENAMETOOLONG) 363 error = E2BIG; 364 goto bad; 365 } 366 dp += len; 367 cpp++; 368 argc++; 369 } 370 371 /* must have at least one argument */ 372 if (argc == 0) { 373 error = EINVAL; 374 goto bad; 375 } 376 377 #ifdef KTRACE 378 if (KTRPOINT(p, KTR_EXECARGS)) 379 ktrexec(p, KTR_EXECARGS, argp, dp - argp); 380 #endif 381 382 envc = 0; 383 /* environment does not need to be there */ 384 if ((cpp = SCARG(uap, envp)) != NULL ) { 385 #ifdef KTRACE 386 env_start = dp; 387 #endif 388 while (1) { 389 len = argp + ARG_MAX - dp; 390 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 391 goto bad; 392 if (!sp) 393 break; 394 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 395 if (error == ENAMETOOLONG) 396 error = E2BIG; 397 goto bad; 398 } 399 dp += len; 400 cpp++; 401 envc++; 402 } 403 404 #ifdef KTRACE 405 if (KTRPOINT(p, KTR_EXECENV)) 406 ktrexec(p, KTR_EXECENV, env_start, dp - env_start); 407 #endif 408 } 409 410 dp = (char *)(((long)dp + _STACKALIGNBYTES) & ~_STACKALIGNBYTES); 411 412 /* 413 * If we have enabled random stackgap, the stack itself has already 414 * been moved from a random location, but is still aligned to a page 415 * boundary. Provide the lower bits of random placement now. 416 */ 417 if (stackgap_random == 0) { 418 sgap = 0; 419 } else { 420 sgap = arc4random() & PAGE_MASK; 421 sgap = (sgap + _STACKALIGNBYTES) & ~_STACKALIGNBYTES; 422 } 423 424 /* Now check if args & environ fit into new stack */ 425 len = ((argc + envc + 2 + ELF_AUX_WORDS) * sizeof(char *) + 426 sizeof(long) + dp + sgap + sizeof(struct ps_strings)) - argp; 427 428 len = (len + _STACKALIGNBYTES) &~ _STACKALIGNBYTES; 429 430 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 431 error = ENOMEM; 432 goto bad; 433 } 434 435 /* adjust "active stack depth" for process VSZ */ 436 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 437 438 /* 439 * we're committed: any further errors will kill the process, so 440 * kill the other threads now. 441 */ 442 single_thread_set(p, SINGLE_EXIT); 443 444 /* 445 * Prepare vmspace for remapping. Note that uvmspace_exec can replace 446 * ps_vmspace! 447 */ 448 uvmspace_exec(p, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 449 450 vm = pr->ps_vmspace; 451 /* Now map address space */ 452 vm->vm_taddr = (char *)trunc_page(pack.ep_taddr); 453 vm->vm_tsize = atop(round_page(pack.ep_taddr + pack.ep_tsize) - 454 trunc_page(pack.ep_taddr)); 455 vm->vm_daddr = (char *)trunc_page(pack.ep_daddr); 456 vm->vm_dsize = atop(round_page(pack.ep_daddr + pack.ep_dsize) - 457 trunc_page(pack.ep_daddr)); 458 vm->vm_dused = 0; 459 vm->vm_ssize = atop(round_page(pack.ep_ssize)); 460 vm->vm_maxsaddr = (char *)pack.ep_maxsaddr; 461 vm->vm_minsaddr = (char *)pack.ep_minsaddr; 462 463 /* create the new process's VM space by running the vmcmds */ 464 #ifdef DIAGNOSTIC 465 if (pack.ep_vmcmds.evs_used == 0) 466 panic("execve: no vmcmds"); 467 #endif 468 error = exec_process_vmcmds(p, &pack); 469 470 /* if an error happened, deallocate and punt */ 471 if (error) 472 goto exec_abort; 473 474 #ifdef MACHINE_STACK_GROWS_UP 475 pr->ps_strings = (vaddr_t)vm->vm_maxsaddr + sgap; 476 if (uvm_map_protect(&vm->vm_map, (vaddr_t)vm->vm_maxsaddr, 477 trunc_page(pr->ps_strings), PROT_NONE, 0, TRUE, FALSE)) 478 goto exec_abort; 479 #else 480 pr->ps_strings = (vaddr_t)vm->vm_minsaddr - sizeof(arginfo) - sgap; 481 if (uvm_map_protect(&vm->vm_map, 482 round_page(pr->ps_strings + sizeof(arginfo)), 483 (vaddr_t)vm->vm_minsaddr, PROT_NONE, 0, TRUE, FALSE)) 484 goto exec_abort; 485 #endif 486 487 memset(&arginfo, 0, sizeof(arginfo)); 488 489 /* remember information about the process */ 490 arginfo.ps_nargvstr = argc; 491 arginfo.ps_nenvstr = envc; 492 493 #ifdef MACHINE_STACK_GROWS_UP 494 stack = (char *)vm->vm_maxsaddr + sizeof(arginfo) + sgap; 495 slen = len - sizeof(arginfo) - sgap; 496 #else 497 stack = (char *)(vm->vm_minsaddr - len); 498 #endif 499 /* Now copy argc, args & environ to new stack */ 500 if (!copyargs(&pack, &arginfo, stack, argp)) 501 goto exec_abort; 502 503 pr->ps_auxinfo = (vaddr_t)pack.ep_auxinfo; 504 505 /* copy out the process's ps_strings structure */ 506 if (copyout(&arginfo, (char *)pr->ps_strings, sizeof(arginfo))) 507 goto exec_abort; 508 509 free(pr->ps_pin.pn_pins, M_PINSYSCALL, 510 pr->ps_pin.pn_npins * sizeof(u_int)); 511 if (pack.ep_npins) { 512 pr->ps_pin.pn_start = pack.ep_pinstart; 513 pr->ps_pin.pn_end = pack.ep_pinend; 514 pr->ps_pin.pn_pins = pack.ep_pins; 515 pack.ep_pins = NULL; 516 pr->ps_pin.pn_npins = pack.ep_npins; 517 pr->ps_flags |= PS_PIN; 518 } else { 519 pr->ps_pin.pn_start = pr->ps_pin.pn_end = 0; 520 pr->ps_pin.pn_pins = NULL; 521 pr->ps_pin.pn_npins = 0; 522 pr->ps_flags &= ~PS_PIN; 523 } 524 if (pr->ps_libcpin.pn_pins) { 525 free(pr->ps_libcpin.pn_pins, M_PINSYSCALL, 526 pr->ps_libcpin.pn_npins * sizeof(u_int)); 527 pr->ps_libcpin.pn_start = pr->ps_libcpin.pn_end = 0; 528 pr->ps_libcpin.pn_pins = NULL; 529 pr->ps_libcpin.pn_npins = 0; 530 pr->ps_flags &= ~PS_LIBCPIN; 531 } 532 533 stopprofclock(pr); /* stop profiling */ 534 fdcloseexec(p); /* handle close on exec */ 535 execsigs(p); /* reset caught signals */ 536 TCB_SET(p, NULL); /* reset the TCB address */ 537 pr->ps_kbind_addr = 0; /* reset the kbind bits */ 538 pr->ps_kbind_cookie = 0; 539 arc4random_buf(&pr->ps_sigcookie, sizeof pr->ps_sigcookie); 540 541 /* set command name & other accounting info */ 542 memset(pr->ps_comm, 0, sizeof(pr->ps_comm)); 543 strlcpy(pr->ps_comm, nid.ni_cnd.cn_nameptr, sizeof(pr->ps_comm)); 544 pr->ps_acflag &= ~AFORK; 545 546 /* record proc's vnode, for use by sysctl */ 547 otvp = pr->ps_textvp; 548 vref(pack.ep_vp); 549 pr->ps_textvp = pack.ep_vp; 550 if (otvp) 551 vrele(otvp); 552 553 if (pack.ep_flags & EXEC_NOBTCFI) 554 atomic_setbits_int(&p->p_p->ps_flags, PS_NOBTCFI); 555 else 556 atomic_clearbits_int(&p->p_p->ps_flags, PS_NOBTCFI); 557 558 atomic_setbits_int(&pr->ps_flags, PS_EXEC); 559 if (pr->ps_flags & PS_PPWAIT) { 560 atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT); 561 atomic_clearbits_int(&pr->ps_pptr->ps_flags, PS_ISPWAIT); 562 wakeup(pr->ps_pptr); 563 } 564 565 /* 566 * If process does execve() while it has a mismatched real, 567 * effective, or saved uid/gid, we set PS_SUGIDEXEC. 568 */ 569 if (cred->cr_uid != cred->cr_ruid || 570 cred->cr_uid != cred->cr_svuid || 571 cred->cr_gid != cred->cr_rgid || 572 cred->cr_gid != cred->cr_svgid) 573 atomic_setbits_int(&pr->ps_flags, PS_SUGIDEXEC); 574 else 575 atomic_clearbits_int(&pr->ps_flags, PS_SUGIDEXEC); 576 577 if (pr->ps_flags & PS_EXECPLEDGE) { 578 pr->ps_pledge = pr->ps_execpledge; 579 atomic_setbits_int(&pr->ps_flags, PS_PLEDGE); 580 } else { 581 atomic_clearbits_int(&pr->ps_flags, PS_PLEDGE); 582 pr->ps_pledge = 0; 583 /* XXX XXX XXX XXX */ 584 /* Clear our unveil paths out so the child 585 * starts afresh 586 */ 587 unveil_destroy(pr); 588 pr->ps_uvdone = 0; 589 } 590 591 /* 592 * deal with set[ug]id. 593 * MNT_NOEXEC has already been used to disable s[ug]id. 594 */ 595 if ((attr.va_mode & (VSUID | VSGID)) && proc_cansugid(p)) { 596 int i; 597 598 atomic_setbits_int(&pr->ps_flags, PS_SUGID|PS_SUGIDEXEC); 599 600 #ifdef KTRACE 601 /* 602 * If process is being ktraced, turn off - unless 603 * root set it. 604 */ 605 if (pr->ps_tracevp && !(pr->ps_traceflag & KTRFAC_ROOT)) 606 ktrcleartrace(pr); 607 #endif 608 p->p_ucred = cred = crcopy(cred); 609 if (attr.va_mode & VSUID) 610 cred->cr_uid = attr.va_uid; 611 if (attr.va_mode & VSGID) 612 cred->cr_gid = attr.va_gid; 613 614 /* 615 * For set[ug]id processes, a few caveats apply to 616 * stdin, stdout, and stderr. 617 */ 618 error = 0; 619 fdplock(p->p_fd); 620 for (i = 0; i < 3; i++) { 621 struct file *fp = NULL; 622 623 /* 624 * NOTE - This will never return NULL because of 625 * immature fds. The file descriptor table is not 626 * shared because we're suid. 627 */ 628 fp = fd_getfile(p->p_fd, i); 629 630 /* 631 * Ensure that stdin, stdout, and stderr are already 632 * allocated. We do not want userland to accidentally 633 * allocate descriptors in this range which has implied 634 * meaning to libc. 635 */ 636 if (fp == NULL) { 637 short flags = FREAD | (i == 0 ? 0 : FWRITE); 638 struct vnode *vp; 639 int indx; 640 641 if ((error = falloc(p, &fp, &indx)) != 0) 642 break; 643 #ifdef DIAGNOSTIC 644 if (indx != i) 645 panic("sys_execve: falloc indx != i"); 646 #endif 647 if ((error = cdevvp(getnulldev(), &vp)) != 0) { 648 fdremove(p->p_fd, indx); 649 closef(fp, p); 650 break; 651 } 652 if ((error = VOP_OPEN(vp, flags, cred, p)) != 0) { 653 fdremove(p->p_fd, indx); 654 closef(fp, p); 655 vrele(vp); 656 break; 657 } 658 if (flags & FWRITE) 659 vp->v_writecount++; 660 fp->f_flag = flags; 661 fp->f_type = DTYPE_VNODE; 662 fp->f_ops = &vnops; 663 fp->f_data = (caddr_t)vp; 664 fdinsert(p->p_fd, indx, 0, fp); 665 } 666 FRELE(fp, p); 667 } 668 fdpunlock(p->p_fd); 669 if (error) 670 goto exec_abort; 671 } else 672 atomic_clearbits_int(&pr->ps_flags, PS_SUGID); 673 674 /* 675 * Reset the saved ugids and update the process's copy of the 676 * creds if the creds have been changed 677 */ 678 if (cred->cr_uid != cred->cr_svuid || 679 cred->cr_gid != cred->cr_svgid) { 680 /* make sure we have unshared ucreds */ 681 p->p_ucred = cred = crcopy(cred); 682 cred->cr_svuid = cred->cr_uid; 683 cred->cr_svgid = cred->cr_gid; 684 } 685 686 if (pr->ps_ucred != cred) { 687 struct ucred *ocred; 688 689 ocred = pr->ps_ucred; 690 crhold(cred); 691 pr->ps_ucred = cred; 692 crfree(ocred); 693 } 694 695 if (pr->ps_flags & PS_SUGIDEXEC) { 696 cancel_all_itimers(); 697 } 698 699 /* reset CPU time usage for the thread, but not the process */ 700 timespecclear(&p->p_tu.tu_runtime); 701 p->p_tu.tu_uticks = p->p_tu.tu_sticks = p->p_tu.tu_iticks = 0; 702 703 memset(p->p_name, 0, sizeof p->p_name); 704 705 km_free(argp, NCARGS, &kv_exec, &kp_pageable); 706 707 pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf); 708 vn_close(pack.ep_vp, FREAD, cred, p); 709 710 /* 711 * notify others that we exec'd 712 */ 713 knote_locked(&pr->ps_klist, NOTE_EXEC); 714 715 /* map the process's timekeep page, needs to be before exec_elf_fixup */ 716 if (exec_timekeep_map(pr)) 717 goto free_pack_abort; 718 719 /* setup new registers and do misc. setup. */ 720 if (exec_elf_fixup(p, &pack) != 0) 721 goto free_pack_abort; 722 #ifdef MACHINE_STACK_GROWS_UP 723 setregs(p, &pack, (u_long)stack + slen, &arginfo); 724 #else 725 setregs(p, &pack, (u_long)stack, &arginfo); 726 #endif 727 728 /* map the process's signal trampoline code */ 729 if (exec_sigcode_map(pr)) 730 goto free_pack_abort; 731 732 #ifdef __HAVE_EXEC_MD_MAP 733 /* perform md specific mappings that process might need */ 734 if (exec_md_map(p, &pack)) 735 goto free_pack_abort; 736 #endif 737 738 if (pr->ps_flags & PS_TRACED) 739 psignal(p, SIGTRAP); 740 741 free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen); 742 743 p->p_descfd = 255; 744 if ((pack.ep_flags & EXEC_HASFD) && pack.ep_fd < 255) 745 p->p_descfd = pack.ep_fd; 746 747 if (pack.ep_flags & EXEC_WXNEEDED) 748 atomic_setbits_int(&p->p_p->ps_flags, PS_WXNEEDED); 749 else 750 atomic_clearbits_int(&p->p_p->ps_flags, PS_WXNEEDED); 751 752 atomic_clearbits_int(&pr->ps_flags, PS_INEXEC); 753 single_thread_clear(p, P_SUSPSIG); 754 755 /* setregs() sets up all the registers, so just 'return' */ 756 return EJUSTRETURN; 757 758 bad: 759 /* free the vmspace-creation commands, and release their references */ 760 kill_vmcmds(&pack.ep_vmcmds); 761 /* kill any opened file descriptor, if necessary */ 762 if (pack.ep_flags & EXEC_HASFD) { 763 pack.ep_flags &= ~EXEC_HASFD; 764 fdplock(p->p_fd); 765 /* fdrelease unlocks p->p_fd. */ 766 (void) fdrelease(p, pack.ep_fd); 767 } 768 if (pack.ep_interp != NULL) 769 pool_put(&namei_pool, pack.ep_interp); 770 free(pack.ep_args, M_TEMP, sizeof *pack.ep_args); 771 free(pack.ep_pins, M_PINSYSCALL, pack.ep_npins * sizeof(u_int)); 772 /* close and put the exec'd file */ 773 vn_close(pack.ep_vp, FREAD, cred, p); 774 pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf); 775 km_free(argp, NCARGS, &kv_exec, &kp_pageable); 776 777 freehdr: 778 free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen); 779 atomic_clearbits_int(&pr->ps_flags, PS_INEXEC); 780 single_thread_clear(p, P_SUSPSIG); 781 782 return (error); 783 784 exec_abort: 785 /* 786 * the old process doesn't exist anymore. exit gracefully. 787 * get rid of the (new) address space we have created, if any, get rid 788 * of our namei data and vnode, and exit noting failure 789 */ 790 uvm_unmap(&vm->vm_map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 791 if (pack.ep_interp != NULL) 792 pool_put(&namei_pool, pack.ep_interp); 793 free(pack.ep_args, M_TEMP, sizeof *pack.ep_args); 794 pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf); 795 vn_close(pack.ep_vp, FREAD, cred, p); 796 km_free(argp, NCARGS, &kv_exec, &kp_pageable); 797 798 free_pack_abort: 799 free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen); 800 exit1(p, 0, SIGABRT, EXIT_NORMAL); 801 802 /* NOTREACHED */ 803 atomic_clearbits_int(&pr->ps_flags, PS_INEXEC); 804 805 return (0); 806 } 807 808 809 int 810 copyargs(struct exec_package *pack, struct ps_strings *arginfo, void *stack, 811 void *argp) 812 { 813 char **cpp = stack; 814 char *dp, *sp; 815 size_t len; 816 void *nullp = NULL; 817 long argc = arginfo->ps_nargvstr; 818 int envc = arginfo->ps_nenvstr; 819 820 if (copyout(&argc, cpp++, sizeof(argc))) 821 return (0); 822 823 dp = (char *) (cpp + argc + envc + 2 + ELF_AUX_WORDS); 824 sp = argp; 825 826 /* XXX don't copy them out, remap them! */ 827 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 828 829 for (; --argc >= 0; sp += len, dp += len) 830 if (copyout(&dp, cpp++, sizeof(dp)) || 831 copyoutstr(sp, dp, ARG_MAX, &len)) 832 return (0); 833 834 if (copyout(&nullp, cpp++, sizeof(nullp))) 835 return (0); 836 837 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 838 839 for (; --envc >= 0; sp += len, dp += len) 840 if (copyout(&dp, cpp++, sizeof(dp)) || 841 copyoutstr(sp, dp, ARG_MAX, &len)) 842 return (0); 843 844 if (copyout(&nullp, cpp++, sizeof(nullp))) 845 return (0); 846 847 /* if this process needs auxinfo, note where to place it */ 848 if (pack->ep_args != NULL) 849 pack->ep_auxinfo = cpp; 850 851 return (1); 852 } 853 854 int 855 exec_sigcode_map(struct process *pr) 856 { 857 extern char sigcode[], esigcode[], sigcoderet[]; 858 vsize_t sz; 859 860 sz = (vaddr_t)esigcode - (vaddr_t)sigcode; 861 862 /* 863 * If we don't have a sigobject yet, create one. 864 * 865 * sigobject is an anonymous memory object (just like SYSV shared 866 * memory) that we keep a permanent reference to and that we map 867 * in all processes that need this sigcode. The creation is simple, 868 * we create an object, add a permanent reference to it, map it in 869 * kernel space, copy out the sigcode to it and unmap it. Then we map 870 * it with PROT_EXEC into the process just the way sys_mmap would map it. 871 */ 872 if (sigobject == NULL) { 873 extern int sigfillsiz; 874 extern u_char sigfill[]; 875 size_t off, left; 876 vaddr_t va; 877 int r; 878 879 sigobject = uao_create(sz, 0); 880 uao_reference(sigobject); /* permanent reference */ 881 882 if ((r = uvm_map(kernel_map, &va, round_page(sz), sigobject, 883 0, 0, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE, 884 MAP_INHERIT_SHARE, MADV_RANDOM, 0)))) { 885 uao_detach(sigobject); 886 return (ENOMEM); 887 } 888 889 for (off = 0, left = round_page(sz); left != 0; 890 off += sigfillsiz) { 891 size_t chunk = ulmin(left, sigfillsiz); 892 memcpy((caddr_t)va + off, sigfill, chunk); 893 left -= chunk; 894 } 895 memcpy((caddr_t)va, sigcode, sz); 896 897 (void) uvm_map_protect(kernel_map, va, round_page(sz), 898 PROT_READ, 0, FALSE, FALSE); 899 sigcode_va = va; 900 sigcode_sz = round_page(sz); 901 } 902 903 pr->ps_sigcode = 0; /* no hint */ 904 uao_reference(sigobject); 905 if (uvm_map(&pr->ps_vmspace->vm_map, &pr->ps_sigcode, round_page(sz), 906 sigobject, 0, 0, UVM_MAPFLAG(PROT_EXEC, 907 PROT_READ | PROT_WRITE | PROT_EXEC, MAP_INHERIT_COPY, 908 MADV_RANDOM, UVM_FLAG_COPYONW | UVM_FLAG_SYSCALL))) { 909 uao_detach(sigobject); 910 return (ENOMEM); 911 } 912 uvm_map_immutable(&pr->ps_vmspace->vm_map, pr->ps_sigcode, 913 pr->ps_sigcode + round_page(sz), 1); 914 915 /* Calculate PC at point of sigreturn entry */ 916 pr->ps_sigcoderet = pr->ps_sigcode + (sigcoderet - sigcode); 917 918 return (0); 919 } 920 921 int 922 exec_timekeep_map(struct process *pr) 923 { 924 size_t timekeep_sz = round_page(sizeof(struct timekeep)); 925 926 /* 927 * Similar to the sigcode object 928 */ 929 if (timekeep_object == NULL) { 930 vaddr_t va = 0; 931 932 timekeep_object = uao_create(timekeep_sz, 0); 933 uao_reference(timekeep_object); 934 935 if (uvm_map(kernel_map, &va, timekeep_sz, timekeep_object, 936 0, 0, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE, 937 MAP_INHERIT_SHARE, MADV_RANDOM, 0))) { 938 uao_detach(timekeep_object); 939 timekeep_object = NULL; 940 return (ENOMEM); 941 } 942 if (uvm_fault_wire(kernel_map, va, va + timekeep_sz, 943 PROT_READ | PROT_WRITE)) { 944 uvm_unmap(kernel_map, va, va + timekeep_sz); 945 uao_detach(timekeep_object); 946 timekeep_object = NULL; 947 return (ENOMEM); 948 } 949 950 timekeep = (struct timekeep *)va; 951 timekeep->tk_version = TK_VERSION; 952 } 953 954 pr->ps_timekeep = 0; /* no hint */ 955 uao_reference(timekeep_object); 956 if (uvm_map(&pr->ps_vmspace->vm_map, &pr->ps_timekeep, timekeep_sz, 957 timekeep_object, 0, 0, UVM_MAPFLAG(PROT_READ, PROT_READ, 958 MAP_INHERIT_COPY, MADV_RANDOM, 0))) { 959 uao_detach(timekeep_object); 960 return (ENOMEM); 961 } 962 uvm_map_immutable(&pr->ps_vmspace->vm_map, pr->ps_timekeep, 963 pr->ps_timekeep + timekeep_sz, 1); 964 965 return (0); 966 } 967