1 /* $OpenBSD: kern_exec.c,v 1.241 2023/01/05 21:39:57 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/mount.h> 41 #include <sys/malloc.h> 42 #include <sys/pool.h> 43 #include <sys/namei.h> 44 #include <sys/vnode.h> 45 #include <sys/fcntl.h> 46 #include <sys/file.h> 47 #include <sys/acct.h> 48 #include <sys/exec.h> 49 #include <sys/exec_elf.h> 50 #include <sys/ktrace.h> 51 #include <sys/resourcevar.h> 52 #include <sys/mman.h> 53 #include <sys/signalvar.h> 54 #include <sys/stat.h> 55 #include <sys/conf.h> 56 #include <sys/pledge.h> 57 #ifdef SYSVSHM 58 #include <sys/shm.h> 59 #endif 60 61 #include <sys/syscallargs.h> 62 63 #include <uvm/uvm_extern.h> 64 #include <machine/tcb.h> 65 66 #include <sys/timetc.h> 67 68 struct uvm_object *sigobject; /* shared sigcode object */ 69 struct uvm_object *timekeep_object; 70 struct timekeep *timekeep; 71 72 void unveil_destroy(struct process *ps); 73 74 const struct kmem_va_mode kv_exec = { 75 .kv_wait = 1, 76 .kv_map = &exec_map 77 }; 78 79 /* 80 * Map the shared signal code. 81 */ 82 int exec_sigcode_map(struct process *); 83 84 /* 85 * Map the shared timekeep page. 86 */ 87 int exec_timekeep_map(struct process *); 88 89 /* 90 * If non-zero, stackgap_random specifies the upper limit of the random gap size 91 * added to the fixed stack position. Must be n^2. 92 */ 93 int stackgap_random = STACKGAP_RANDOM; 94 95 /* 96 * check exec: 97 * given an "executable" described in the exec package's namei info, 98 * see what we can do with it. 99 * 100 * ON ENTRY: 101 * exec package with appropriate namei info 102 * proc pointer of exec'ing proc 103 * NO SELF-LOCKED VNODES 104 * 105 * ON EXIT: 106 * error: nothing held, etc. exec header still allocated. 107 * ok: filled exec package, one locked vnode. 108 * 109 * EXEC SWITCH ENTRY: 110 * Locked vnode to check, exec package, proc. 111 * 112 * EXEC SWITCH EXIT: 113 * ok: return 0, filled exec package, one locked vnode. 114 * error: destructive: 115 * everything deallocated except exec header. 116 * non-destructive: 117 * error code, locked vnode, exec header unmodified 118 */ 119 int 120 check_exec(struct proc *p, struct exec_package *epp) 121 { 122 int error, i; 123 struct vnode *vp; 124 struct nameidata *ndp; 125 size_t resid; 126 127 ndp = epp->ep_ndp; 128 ndp->ni_cnd.cn_nameiop = LOOKUP; 129 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 130 if (epp->ep_flags & EXEC_INDIR) 131 ndp->ni_cnd.cn_flags |= BYPASSUNVEIL; 132 /* first get the vnode */ 133 if ((error = namei(ndp)) != 0) 134 return (error); 135 epp->ep_vp = vp = ndp->ni_vp; 136 137 /* check for regular file */ 138 if (vp->v_type != VREG) { 139 error = EACCES; 140 goto bad1; 141 } 142 143 /* get attributes */ 144 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 145 goto bad1; 146 147 /* Check mount point */ 148 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 149 error = EACCES; 150 goto bad1; 151 } 152 153 /* SUID programs may not be started with execpromises */ 154 if ((epp->ep_vap->va_mode & (VSUID | VSGID)) && 155 (p->p_p->ps_flags & PS_EXECPLEDGE)) { 156 error = EACCES; 157 goto bad1; 158 } 159 160 if ((vp->v_mount->mnt_flag & MNT_NOSUID)) 161 epp->ep_vap->va_mode &= ~(VSUID | VSGID); 162 163 /* check access. for root we have to see if any exec bit on */ 164 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 165 goto bad1; 166 if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 167 error = EACCES; 168 goto bad1; 169 } 170 171 /* try to open it */ 172 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 173 goto bad1; 174 175 /* unlock vp, we need it unlocked from here */ 176 VOP_UNLOCK(vp); 177 178 /* now we have the file, get the exec header */ 179 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 180 UIO_SYSSPACE, 0, p->p_ucred, &resid, p); 181 if (error) 182 goto bad2; 183 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 184 185 /* 186 * set up the vmcmds for creation of the process 187 * address space 188 */ 189 error = ENOEXEC; 190 for (i = 0; i < nexecs && error != 0; i++) { 191 int newerror; 192 193 if (execsw[i].es_check == NULL) 194 continue; 195 newerror = (*execsw[i].es_check)(p, epp); 196 /* make sure the first "interesting" error code is saved. */ 197 if (!newerror || error == ENOEXEC) 198 error = newerror; 199 if (epp->ep_flags & EXEC_DESTR && error != 0) 200 return (error); 201 } 202 if (!error) { 203 /* check that entry point is sane */ 204 if (epp->ep_entry > VM_MAXUSER_ADDRESS) { 205 error = ENOEXEC; 206 } 207 208 /* check limits */ 209 if ((epp->ep_tsize > MAXTSIZ) || 210 (epp->ep_dsize > lim_cur(RLIMIT_DATA))) 211 error = ENOMEM; 212 213 if (!error) 214 return (0); 215 } 216 217 /* 218 * free any vmspace-creation commands, 219 * and release their references 220 */ 221 kill_vmcmds(&epp->ep_vmcmds); 222 223 bad2: 224 /* 225 * close the vnode, free the pathname buf, and punt. 226 */ 227 vn_close(vp, FREAD, p->p_ucred, p); 228 pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf); 229 return (error); 230 231 bad1: 232 /* 233 * free the namei pathname buffer, and put the vnode 234 * (which we don't yet have open). 235 */ 236 pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf); 237 vput(vp); 238 return (error); 239 } 240 241 /* 242 * exec system call 243 */ 244 int 245 sys_execve(struct proc *p, void *v, register_t *retval) 246 { 247 struct sys_execve_args /* { 248 syscallarg(const char *) path; 249 syscallarg(char *const *) argp; 250 syscallarg(char *const *) envp; 251 } */ *uap = v; 252 int error; 253 struct exec_package pack; 254 struct nameidata nid; 255 struct vattr attr; 256 struct ucred *cred = p->p_ucred; 257 char *argp; 258 char * const *cpp, *dp, *sp; 259 #ifdef KTRACE 260 char *env_start; 261 #endif 262 struct process *pr = p->p_p; 263 long argc, envc; 264 size_t len, sgap, dstsize; 265 #ifdef MACHINE_STACK_GROWS_UP 266 size_t slen; 267 #endif 268 char *stack; 269 struct ps_strings arginfo; 270 struct vmspace *vm; 271 struct vnode *otvp; 272 273 /* get other threads to stop */ 274 if ((error = single_thread_set(p, SINGLE_UNWIND, 1))) 275 return (error); 276 277 /* 278 * Cheap solution to complicated problems. 279 * Mark this process as "leave me alone, I'm execing". 280 */ 281 atomic_setbits_int(&pr->ps_flags, PS_INEXEC); 282 283 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 284 nid.ni_pledge = PLEDGE_EXEC; 285 nid.ni_unveil = UNVEIL_EXEC; 286 287 /* 288 * initialize the fields of the exec package. 289 */ 290 pack.ep_name = (char *)SCARG(uap, path); 291 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 292 pack.ep_hdrlen = exec_maxhdrsz; 293 pack.ep_hdrvalid = 0; 294 pack.ep_ndp = &nid; 295 pack.ep_interp = NULL; 296 pack.ep_args = NULL; 297 pack.ep_auxinfo = NULL; 298 VMCMDSET_INIT(&pack.ep_vmcmds); 299 pack.ep_vap = &attr; 300 pack.ep_flags = 0; 301 302 /* see if we can run it. */ 303 if ((error = check_exec(p, &pack)) != 0) { 304 goto freehdr; 305 } 306 307 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 308 309 /* allocate an argument buffer */ 310 argp = km_alloc(NCARGS, &kv_exec, &kp_pageable, &kd_waitok); 311 #ifdef DIAGNOSTIC 312 if (argp == NULL) 313 panic("execve: argp == NULL"); 314 #endif 315 dp = argp; 316 argc = 0; 317 318 /* 319 * Copy the fake args list, if there's one, freeing it as we go. 320 * exec_script_makecmds() allocates either 2 or 3 fake args bounded 321 * by MAXINTERP + MAXPATHLEN < NCARGS so no overflow can happen. 322 */ 323 if (pack.ep_flags & EXEC_HASARGL) { 324 dstsize = NCARGS; 325 for(; pack.ep_fa[argc] != NULL; argc++) { 326 len = strlcpy(dp, pack.ep_fa[argc], dstsize); 327 len++; 328 dp += len; dstsize -= len; 329 if (pack.ep_fa[argc+1] != NULL) 330 free(pack.ep_fa[argc], M_EXEC, len); 331 else 332 free(pack.ep_fa[argc], M_EXEC, MAXPATHLEN); 333 } 334 free(pack.ep_fa, M_EXEC, 4 * sizeof(char *)); 335 pack.ep_flags &= ~EXEC_HASARGL; 336 } 337 338 /* Now get argv & environment */ 339 if (!(cpp = SCARG(uap, argp))) { 340 error = EFAULT; 341 goto bad; 342 } 343 344 if (pack.ep_flags & EXEC_SKIPARG) 345 cpp++; 346 347 while (1) { 348 len = argp + ARG_MAX - dp; 349 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 350 goto bad; 351 if (!sp) 352 break; 353 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 354 if (error == ENAMETOOLONG) 355 error = E2BIG; 356 goto bad; 357 } 358 dp += len; 359 cpp++; 360 argc++; 361 } 362 363 /* must have at least one argument */ 364 if (argc == 0) { 365 error = EINVAL; 366 goto bad; 367 } 368 369 #ifdef KTRACE 370 if (KTRPOINT(p, KTR_EXECARGS)) 371 ktrexec(p, KTR_EXECARGS, argp, dp - argp); 372 #endif 373 374 envc = 0; 375 /* environment does not need to be there */ 376 if ((cpp = SCARG(uap, envp)) != NULL ) { 377 #ifdef KTRACE 378 env_start = dp; 379 #endif 380 while (1) { 381 len = argp + ARG_MAX - dp; 382 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 383 goto bad; 384 if (!sp) 385 break; 386 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 387 if (error == ENAMETOOLONG) 388 error = E2BIG; 389 goto bad; 390 } 391 dp += len; 392 cpp++; 393 envc++; 394 } 395 396 #ifdef KTRACE 397 if (KTRPOINT(p, KTR_EXECENV)) 398 ktrexec(p, KTR_EXECENV, env_start, dp - env_start); 399 #endif 400 } 401 402 dp = (char *)(((long)dp + _STACKALIGNBYTES) & ~_STACKALIGNBYTES); 403 404 /* 405 * If we have enabled random stackgap, the stack itself has already 406 * been moved from a random location, but is still aligned to a page 407 * boundary. Provide the lower bits of random placement now. 408 */ 409 if (stackgap_random == 0) { 410 sgap = 0; 411 } else { 412 sgap = arc4random() & PAGE_MASK; 413 sgap = (sgap + _STACKALIGNBYTES) & ~_STACKALIGNBYTES; 414 } 415 416 /* Now check if args & environ fit into new stack */ 417 len = ((argc + envc + 2 + ELF_AUX_WORDS) * sizeof(char *) + 418 sizeof(long) + dp + sgap + sizeof(struct ps_strings)) - argp; 419 420 len = (len + _STACKALIGNBYTES) &~ _STACKALIGNBYTES; 421 422 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 423 error = ENOMEM; 424 goto bad; 425 } 426 427 /* adjust "active stack depth" for process VSZ */ 428 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 429 430 /* 431 * we're committed: any further errors will kill the process, so 432 * kill the other threads now. 433 */ 434 single_thread_set(p, SINGLE_EXIT, 1); 435 436 /* 437 * Prepare vmspace for remapping. Note that uvmspace_exec can replace 438 * ps_vmspace! 439 */ 440 uvmspace_exec(p, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 441 442 vm = pr->ps_vmspace; 443 /* Now map address space */ 444 vm->vm_taddr = (char *)trunc_page(pack.ep_taddr); 445 vm->vm_tsize = atop(round_page(pack.ep_taddr + pack.ep_tsize) - 446 trunc_page(pack.ep_taddr)); 447 vm->vm_daddr = (char *)trunc_page(pack.ep_daddr); 448 vm->vm_dsize = atop(round_page(pack.ep_daddr + pack.ep_dsize) - 449 trunc_page(pack.ep_daddr)); 450 vm->vm_dused = 0; 451 vm->vm_ssize = atop(round_page(pack.ep_ssize)); 452 vm->vm_maxsaddr = (char *)pack.ep_maxsaddr; 453 vm->vm_minsaddr = (char *)pack.ep_minsaddr; 454 455 /* create the new process's VM space by running the vmcmds */ 456 #ifdef DIAGNOSTIC 457 if (pack.ep_vmcmds.evs_used == 0) 458 panic("execve: no vmcmds"); 459 #endif 460 error = exec_process_vmcmds(p, &pack); 461 462 /* if an error happened, deallocate and punt */ 463 if (error) 464 goto exec_abort; 465 466 #ifdef MACHINE_STACK_GROWS_UP 467 pr->ps_strings = (vaddr_t)vm->vm_maxsaddr + sgap; 468 if (uvm_map_protect(&vm->vm_map, (vaddr_t)vm->vm_maxsaddr, 469 trunc_page(pr->ps_strings), PROT_NONE, 0, TRUE, FALSE)) 470 goto exec_abort; 471 #else 472 pr->ps_strings = (vaddr_t)vm->vm_minsaddr - sizeof(arginfo) - sgap; 473 if (uvm_map_protect(&vm->vm_map, 474 round_page(pr->ps_strings + sizeof(arginfo)), 475 (vaddr_t)vm->vm_minsaddr, PROT_NONE, 0, TRUE, FALSE)) 476 goto exec_abort; 477 #endif 478 479 memset(&arginfo, 0, sizeof(arginfo)); 480 481 /* remember information about the process */ 482 arginfo.ps_nargvstr = argc; 483 arginfo.ps_nenvstr = envc; 484 485 #ifdef MACHINE_STACK_GROWS_UP 486 stack = (char *)vm->vm_maxsaddr + sizeof(arginfo) + sgap; 487 slen = len - sizeof(arginfo) - sgap; 488 #else 489 stack = (char *)(vm->vm_minsaddr - len); 490 #endif 491 /* Now copy argc, args & environ to new stack */ 492 if (!copyargs(&pack, &arginfo, stack, argp)) 493 goto exec_abort; 494 495 pr->ps_auxinfo = (vaddr_t)pack.ep_auxinfo; 496 497 /* copy out the process's ps_strings structure */ 498 if (copyout(&arginfo, (char *)pr->ps_strings, sizeof(arginfo))) 499 goto exec_abort; 500 501 stopprofclock(pr); /* stop profiling */ 502 fdcloseexec(p); /* handle close on exec */ 503 execsigs(p); /* reset caught signals */ 504 TCB_SET(p, NULL); /* reset the TCB address */ 505 pr->ps_kbind_addr = 0; /* reset the kbind bits */ 506 pr->ps_kbind_cookie = 0; 507 arc4random_buf(&pr->ps_sigcookie, sizeof pr->ps_sigcookie); 508 509 /* set command name & other accounting info */ 510 memset(pr->ps_comm, 0, sizeof(pr->ps_comm)); 511 strlcpy(pr->ps_comm, nid.ni_cnd.cn_nameptr, sizeof(pr->ps_comm)); 512 pr->ps_acflag &= ~AFORK; 513 514 /* record proc's vnode, for use by sysctl */ 515 otvp = pr->ps_textvp; 516 vref(pack.ep_vp); 517 pr->ps_textvp = pack.ep_vp; 518 if (otvp) 519 vrele(otvp); 520 521 atomic_setbits_int(&pr->ps_flags, PS_EXEC); 522 if (pr->ps_flags & PS_PPWAIT) { 523 atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT); 524 atomic_clearbits_int(&pr->ps_pptr->ps_flags, PS_ISPWAIT); 525 wakeup(pr->ps_pptr); 526 } 527 528 /* 529 * If process does execve() while it has a mismatched real, 530 * effective, or saved uid/gid, we set PS_SUGIDEXEC. 531 */ 532 if (cred->cr_uid != cred->cr_ruid || 533 cred->cr_uid != cred->cr_svuid || 534 cred->cr_gid != cred->cr_rgid || 535 cred->cr_gid != cred->cr_svgid) 536 atomic_setbits_int(&pr->ps_flags, PS_SUGIDEXEC); 537 else 538 atomic_clearbits_int(&pr->ps_flags, PS_SUGIDEXEC); 539 540 if (pr->ps_flags & PS_EXECPLEDGE) { 541 pr->ps_pledge = pr->ps_execpledge; 542 atomic_setbits_int(&pr->ps_flags, PS_PLEDGE); 543 } else { 544 atomic_clearbits_int(&pr->ps_flags, PS_PLEDGE); 545 pr->ps_pledge = 0; 546 /* XXX XXX XXX XXX */ 547 /* Clear our unveil paths out so the child 548 * starts afresh 549 */ 550 unveil_destroy(pr); 551 pr->ps_uvdone = 0; 552 } 553 554 /* 555 * deal with set[ug]id. 556 * MNT_NOEXEC has already been used to disable s[ug]id. 557 */ 558 if ((attr.va_mode & (VSUID | VSGID)) && proc_cansugid(p)) { 559 int i; 560 561 atomic_setbits_int(&pr->ps_flags, PS_SUGID|PS_SUGIDEXEC); 562 563 #ifdef KTRACE 564 /* 565 * If process is being ktraced, turn off - unless 566 * root set it. 567 */ 568 if (pr->ps_tracevp && !(pr->ps_traceflag & KTRFAC_ROOT)) 569 ktrcleartrace(pr); 570 #endif 571 p->p_ucred = cred = crcopy(cred); 572 if (attr.va_mode & VSUID) 573 cred->cr_uid = attr.va_uid; 574 if (attr.va_mode & VSGID) 575 cred->cr_gid = attr.va_gid; 576 577 /* 578 * For set[ug]id processes, a few caveats apply to 579 * stdin, stdout, and stderr. 580 */ 581 error = 0; 582 fdplock(p->p_fd); 583 for (i = 0; i < 3; i++) { 584 struct file *fp = NULL; 585 586 /* 587 * NOTE - This will never return NULL because of 588 * immature fds. The file descriptor table is not 589 * shared because we're suid. 590 */ 591 fp = fd_getfile(p->p_fd, i); 592 593 /* 594 * Ensure that stdin, stdout, and stderr are already 595 * allocated. We do not want userland to accidentally 596 * allocate descriptors in this range which has implied 597 * meaning to libc. 598 */ 599 if (fp == NULL) { 600 short flags = FREAD | (i == 0 ? 0 : FWRITE); 601 struct vnode *vp; 602 int indx; 603 604 if ((error = falloc(p, &fp, &indx)) != 0) 605 break; 606 #ifdef DIAGNOSTIC 607 if (indx != i) 608 panic("sys_execve: falloc indx != i"); 609 #endif 610 if ((error = cdevvp(getnulldev(), &vp)) != 0) { 611 fdremove(p->p_fd, indx); 612 closef(fp, p); 613 break; 614 } 615 if ((error = VOP_OPEN(vp, flags, cred, p)) != 0) { 616 fdremove(p->p_fd, indx); 617 closef(fp, p); 618 vrele(vp); 619 break; 620 } 621 if (flags & FWRITE) 622 vp->v_writecount++; 623 fp->f_flag = flags; 624 fp->f_type = DTYPE_VNODE; 625 fp->f_ops = &vnops; 626 fp->f_data = (caddr_t)vp; 627 fdinsert(p->p_fd, indx, 0, fp); 628 } 629 FRELE(fp, p); 630 } 631 fdpunlock(p->p_fd); 632 if (error) 633 goto exec_abort; 634 } else 635 atomic_clearbits_int(&pr->ps_flags, PS_SUGID); 636 637 /* 638 * Reset the saved ugids and update the process's copy of the 639 * creds if the creds have been changed 640 */ 641 if (cred->cr_uid != cred->cr_svuid || 642 cred->cr_gid != cred->cr_svgid) { 643 /* make sure we have unshared ucreds */ 644 p->p_ucred = cred = crcopy(cred); 645 cred->cr_svuid = cred->cr_uid; 646 cred->cr_svgid = cred->cr_gid; 647 } 648 649 if (pr->ps_ucred != cred) { 650 struct ucred *ocred; 651 652 ocred = pr->ps_ucred; 653 crhold(cred); 654 pr->ps_ucred = cred; 655 crfree(ocred); 656 } 657 658 if (pr->ps_flags & PS_SUGIDEXEC) { 659 cancel_all_itimers(); 660 } 661 662 /* reset CPU time usage for the thread, but not the process */ 663 timespecclear(&p->p_tu.tu_runtime); 664 p->p_tu.tu_uticks = p->p_tu.tu_sticks = p->p_tu.tu_iticks = 0; 665 666 km_free(argp, NCARGS, &kv_exec, &kp_pageable); 667 668 pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf); 669 vn_close(pack.ep_vp, FREAD, cred, p); 670 671 /* 672 * notify others that we exec'd 673 */ 674 KNOTE(&pr->ps_klist, NOTE_EXEC); 675 676 /* map the process's timekeep page, needs to be before exec_elf_fixup */ 677 if (exec_timekeep_map(pr)) 678 goto free_pack_abort; 679 680 /* setup new registers and do misc. setup. */ 681 if (exec_elf_fixup(p, &pack) != 0) 682 goto free_pack_abort; 683 #ifdef MACHINE_STACK_GROWS_UP 684 setregs(p, &pack, (u_long)stack + slen, &arginfo); 685 #else 686 setregs(p, &pack, (u_long)stack, &arginfo); 687 #endif 688 689 /* map the process's signal trampoline code */ 690 if (exec_sigcode_map(pr)) 691 goto free_pack_abort; 692 693 #ifdef __HAVE_EXEC_MD_MAP 694 /* perform md specific mappings that process might need */ 695 if (exec_md_map(p, &pack)) 696 goto free_pack_abort; 697 #endif 698 699 if (pr->ps_flags & PS_TRACED) 700 psignal(p, SIGTRAP); 701 702 free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen); 703 704 p->p_descfd = 255; 705 if ((pack.ep_flags & EXEC_HASFD) && pack.ep_fd < 255) 706 p->p_descfd = pack.ep_fd; 707 708 if (pack.ep_flags & EXEC_WXNEEDED) 709 atomic_setbits_int(&p->p_p->ps_flags, PS_WXNEEDED); 710 else 711 atomic_clearbits_int(&p->p_p->ps_flags, PS_WXNEEDED); 712 713 atomic_clearbits_int(&pr->ps_flags, PS_INEXEC); 714 single_thread_clear(p, P_SUSPSIG); 715 716 /* setregs() sets up all the registers, so just 'return' */ 717 return EJUSTRETURN; 718 719 bad: 720 /* free the vmspace-creation commands, and release their references */ 721 kill_vmcmds(&pack.ep_vmcmds); 722 /* kill any opened file descriptor, if necessary */ 723 if (pack.ep_flags & EXEC_HASFD) { 724 pack.ep_flags &= ~EXEC_HASFD; 725 fdplock(p->p_fd); 726 /* fdrelease unlocks p->p_fd. */ 727 (void) fdrelease(p, pack.ep_fd); 728 } 729 if (pack.ep_interp != NULL) 730 pool_put(&namei_pool, pack.ep_interp); 731 free(pack.ep_args, M_TEMP, sizeof *pack.ep_args); 732 /* close and put the exec'd file */ 733 vn_close(pack.ep_vp, FREAD, cred, p); 734 pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf); 735 km_free(argp, NCARGS, &kv_exec, &kp_pageable); 736 737 freehdr: 738 free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen); 739 atomic_clearbits_int(&pr->ps_flags, PS_INEXEC); 740 single_thread_clear(p, P_SUSPSIG); 741 742 return (error); 743 744 exec_abort: 745 /* 746 * the old process doesn't exist anymore. exit gracefully. 747 * get rid of the (new) address space we have created, if any, get rid 748 * of our namei data and vnode, and exit noting failure 749 */ 750 uvm_unmap(&vm->vm_map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 751 if (pack.ep_interp != NULL) 752 pool_put(&namei_pool, pack.ep_interp); 753 free(pack.ep_args, M_TEMP, sizeof *pack.ep_args); 754 pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf); 755 vn_close(pack.ep_vp, FREAD, cred, p); 756 km_free(argp, NCARGS, &kv_exec, &kp_pageable); 757 758 free_pack_abort: 759 free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen); 760 exit1(p, 0, SIGABRT, EXIT_NORMAL); 761 762 /* NOTREACHED */ 763 atomic_clearbits_int(&pr->ps_flags, PS_INEXEC); 764 765 return (0); 766 } 767 768 769 int 770 copyargs(struct exec_package *pack, struct ps_strings *arginfo, void *stack, 771 void *argp) 772 { 773 char **cpp = stack; 774 char *dp, *sp; 775 size_t len; 776 void *nullp = NULL; 777 long argc = arginfo->ps_nargvstr; 778 int envc = arginfo->ps_nenvstr; 779 780 if (copyout(&argc, cpp++, sizeof(argc))) 781 return (0); 782 783 dp = (char *) (cpp + argc + envc + 2 + ELF_AUX_WORDS); 784 sp = argp; 785 786 /* XXX don't copy them out, remap them! */ 787 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 788 789 for (; --argc >= 0; sp += len, dp += len) 790 if (copyout(&dp, cpp++, sizeof(dp)) || 791 copyoutstr(sp, dp, ARG_MAX, &len)) 792 return (0); 793 794 if (copyout(&nullp, cpp++, sizeof(nullp))) 795 return (0); 796 797 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 798 799 for (; --envc >= 0; sp += len, dp += len) 800 if (copyout(&dp, cpp++, sizeof(dp)) || 801 copyoutstr(sp, dp, ARG_MAX, &len)) 802 return (0); 803 804 if (copyout(&nullp, cpp++, sizeof(nullp))) 805 return (0); 806 807 /* if this process needs auxinfo, note where to place it */ 808 if (pack->ep_args != NULL) 809 pack->ep_auxinfo = cpp; 810 811 return (1); 812 } 813 814 int 815 exec_sigcode_map(struct process *pr) 816 { 817 extern char sigcode[], esigcode[], sigcoderet[]; 818 vsize_t sz; 819 820 sz = (vaddr_t)esigcode - (vaddr_t)sigcode; 821 822 /* 823 * If we don't have a sigobject yet, create one. 824 * 825 * sigobject is an anonymous memory object (just like SYSV shared 826 * memory) that we keep a permanent reference to and that we map 827 * in all processes that need this sigcode. The creation is simple, 828 * we create an object, add a permanent reference to it, map it in 829 * kernel space, copy out the sigcode to it and unmap it. Then we map 830 * it with PROT_EXEC into the process just the way sys_mmap would map it. 831 */ 832 if (sigobject == NULL) { 833 extern int sigfillsiz; 834 extern u_char sigfill[]; 835 size_t off, left; 836 vaddr_t va; 837 int r; 838 839 sigobject = uao_create(sz, 0); 840 uao_reference(sigobject); /* permanent reference */ 841 842 if ((r = uvm_map(kernel_map, &va, round_page(sz), sigobject, 843 0, 0, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE, 844 MAP_INHERIT_SHARE, MADV_RANDOM, 0)))) { 845 uao_detach(sigobject); 846 return (ENOMEM); 847 } 848 849 for (off = 0, left = round_page(sz); left != 0; 850 off += sigfillsiz) { 851 size_t chunk = ulmin(left, sigfillsiz); 852 memcpy((caddr_t)va + off, sigfill, chunk); 853 left -= chunk; 854 } 855 memcpy((caddr_t)va, sigcode, sz); 856 uvm_unmap(kernel_map, va, va + round_page(sz)); 857 } 858 859 pr->ps_sigcode = 0; /* no hint */ 860 uao_reference(sigobject); 861 if (uvm_map(&pr->ps_vmspace->vm_map, &pr->ps_sigcode, round_page(sz), 862 sigobject, 0, 0, UVM_MAPFLAG(PROT_EXEC, 863 PROT_READ | PROT_WRITE | PROT_EXEC, MAP_INHERIT_COPY, 864 MADV_RANDOM, UVM_FLAG_COPYONW | UVM_FLAG_SYSCALL))) { 865 uao_detach(sigobject); 866 return (ENOMEM); 867 } 868 uvm_map_immutable(&pr->ps_vmspace->vm_map, pr->ps_sigcode, 869 pr->ps_sigcode + round_page(sz), 1); 870 871 /* Calculate PC at point of sigreturn entry */ 872 pr->ps_sigcoderet = pr->ps_sigcode + (sigcoderet - sigcode); 873 874 return (0); 875 } 876 877 int 878 exec_timekeep_map(struct process *pr) 879 { 880 size_t timekeep_sz = round_page(sizeof(struct timekeep)); 881 882 /* 883 * Similar to the sigcode object 884 */ 885 if (timekeep_object == NULL) { 886 vaddr_t va = 0; 887 888 timekeep_object = uao_create(timekeep_sz, 0); 889 uao_reference(timekeep_object); 890 891 if (uvm_map(kernel_map, &va, timekeep_sz, timekeep_object, 892 0, 0, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE, 893 MAP_INHERIT_SHARE, MADV_RANDOM, 0))) { 894 uao_detach(timekeep_object); 895 timekeep_object = NULL; 896 return (ENOMEM); 897 } 898 if (uvm_fault_wire(kernel_map, va, va + timekeep_sz, 899 PROT_READ | PROT_WRITE)) { 900 uvm_unmap(kernel_map, va, va + timekeep_sz); 901 uao_detach(timekeep_object); 902 timekeep_object = NULL; 903 return (ENOMEM); 904 } 905 906 timekeep = (struct timekeep *)va; 907 timekeep->tk_version = TK_VERSION; 908 } 909 910 pr->ps_timekeep = 0; /* no hint */ 911 uao_reference(timekeep_object); 912 if (uvm_map(&pr->ps_vmspace->vm_map, &pr->ps_timekeep, timekeep_sz, 913 timekeep_object, 0, 0, UVM_MAPFLAG(PROT_READ, PROT_READ, 914 MAP_INHERIT_COPY, MADV_RANDOM, 0))) { 915 uao_detach(timekeep_object); 916 return (ENOMEM); 917 } 918 uvm_map_immutable(&pr->ps_vmspace->vm_map, pr->ps_timekeep, 919 pr->ps_timekeep + timekeep_sz, 1); 920 921 return (0); 922 } 923