1 /* 2 * Copyright (c) 1993, David Greenman 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $ 27 * $DragonFly: src/sys/kern/kern_exec.c,v 1.61 2007/07/30 17:41:23 pavalos Exp $ 28 */ 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/sysproto.h> 33 #include <sys/kernel.h> 34 #include <sys/mount.h> 35 #include <sys/filedesc.h> 36 #include <sys/fcntl.h> 37 #include <sys/acct.h> 38 #include <sys/exec.h> 39 #include <sys/imgact.h> 40 #include <sys/imgact_elf.h> 41 #include <sys/kern_syscall.h> 42 #include <sys/wait.h> 43 #include <sys/malloc.h> 44 #include <sys/proc.h> 45 #include <sys/ktrace.h> 46 #include <sys/signalvar.h> 47 #include <sys/pioctl.h> 48 #include <sys/nlookup.h> 49 #include <sys/sfbuf.h> 50 #include <sys/sysent.h> 51 #include <sys/shm.h> 52 #include <sys/sysctl.h> 53 #include <sys/vnode.h> 54 #include <sys/vmmeter.h> 55 #include <sys/aio.h> 56 #include <sys/libkern.h> 57 58 #include <vm/vm.h> 59 #include <vm/vm_param.h> 60 #include <sys/lock.h> 61 #include <vm/pmap.h> 62 #include <vm/vm_page.h> 63 #include <vm/vm_map.h> 64 #include <vm/vm_kern.h> 65 #include <vm/vm_extern.h> 66 #include <vm/vm_object.h> 67 #include <vm/vm_pager.h> 68 69 #include <sys/user.h> 70 #include <sys/reg.h> 71 72 #include <sys/thread2.h> 73 74 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 75 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 76 77 static register_t *exec_copyout_strings (struct image_params *); 78 79 /* XXX This should be vm_size_t. */ 80 static u_long ps_strings = PS_STRINGS; 81 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 82 83 /* XXX This should be vm_size_t. */ 84 static u_long usrstack = USRSTACK; 85 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 86 87 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 88 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 89 &ps_arg_cache_limit, 0, ""); 90 91 int ps_argsopen = 1; 92 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 93 94 void print_execve_args(struct image_args *args); 95 int debug_execve_args = 0; 96 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 97 0, ""); 98 99 /* 100 * Exec arguments object cache 101 */ 102 static struct objcache *exec_objcache; 103 104 static 105 void 106 exec_objcache_init(void *arg __unused) 107 { 108 exec_objcache = objcache_create_mbacked( 109 M_EXECARGS, PATH_MAX + ARG_MAX, 110 16, /* up to this many objects */ 111 2, /* minimal magazine capacity */ 112 NULL, NULL, NULL); 113 } 114 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 115 116 /* 117 * stackgap_random specifies if the stackgap should have a random size added 118 * to it. It must be a power of 2. If non-zero, the stack gap will be 119 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 120 */ 121 static int stackgap_random = 1024; 122 static int 123 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 124 { 125 int error, new_val; 126 new_val = stackgap_random; 127 error = sysctl_handle_int(oidp, &new_val, 0, req); 128 if (error != 0 || req->newptr == NULL) 129 return (error); 130 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 131 return (EINVAL); 132 stackgap_random = new_val; 133 134 return(0); 135 } 136 137 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 138 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 139 140 void 141 print_execve_args(struct image_args *args) 142 { 143 char *cp; 144 int ndx; 145 146 cp = args->begin_argv; 147 for (ndx = 0; ndx < args->argc; ndx++) { 148 kprintf("\targv[%d]: %s\n", ndx, cp); 149 while (*cp++ != '\0'); 150 } 151 for (ndx = 0; ndx < args->envc; ndx++) { 152 kprintf("\tenvv[%d]: %s\n", ndx, cp); 153 while (*cp++ != '\0'); 154 } 155 } 156 157 /* 158 * Each of the items is a pointer to a `const struct execsw', hence the 159 * double pointer here. 160 */ 161 static const struct execsw **execsw; 162 163 /* 164 * Replace current vmspace with a new binary. 165 * Returns 0 on success, > 0 on recoverable error (use as errno). 166 * Returns -1 on lethal error which demands killing of the current 167 * process! 168 */ 169 int 170 kern_execve(struct nlookupdata *nd, struct image_args *args) 171 { 172 struct thread *td = curthread; 173 struct lwp *lp = td->td_lwp; 174 struct proc *p = td->td_proc; 175 register_t *stack_base; 176 int error, len, i; 177 struct image_params image_params, *imgp; 178 struct vattr attr; 179 int (*img_first) (struct image_params *); 180 181 if (debug_execve_args) { 182 kprintf("%s()\n", __func__); 183 print_execve_args(args); 184 } 185 186 KKASSERT(p); 187 imgp = &image_params; 188 189 /* 190 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 191 * no modifications to the process at all until we get there. 192 * 193 * Note that multiple threads may be trying to exec at the same 194 * time. exec_new_vmspace() handles that too. 195 */ 196 197 /* 198 * Initialize part of the common data 199 */ 200 imgp->proc = p; 201 imgp->args = args; 202 imgp->attr = &attr; 203 imgp->entry_addr = 0; 204 imgp->resident = 0; 205 imgp->vmspace_destroyed = 0; 206 imgp->interpreted = 0; 207 imgp->interpreter_name[0] = 0; 208 imgp->auxargs = NULL; 209 imgp->vp = NULL; 210 imgp->firstpage = NULL; 211 imgp->ps_strings = 0; 212 imgp->image_header = NULL; 213 214 interpret: 215 216 /* 217 * Translate the file name to a vnode. Unlock the cache entry to 218 * improve parallelism for programs exec'd in parallel. 219 */ 220 if ((error = nlookup(nd)) != 0) 221 goto exec_fail; 222 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 223 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 224 nd->nl_flags &= ~NLC_NCPISLOCKED; 225 cache_unlock(&nd->nl_nch); 226 if (error) 227 goto exec_fail; 228 229 /* 230 * Check file permissions (also 'opens' file) 231 */ 232 error = exec_check_permissions(imgp); 233 if (error) { 234 vn_unlock(imgp->vp); 235 goto exec_fail_dealloc; 236 } 237 238 error = exec_map_first_page(imgp); 239 vn_unlock(imgp->vp); 240 if (error) 241 goto exec_fail_dealloc; 242 243 if (debug_execve_args && imgp->interpreted) { 244 kprintf(" target is interpreted -- recursive pass\n"); 245 kprintf(" interpreter: %s\n", imgp->interpreter_name); 246 print_execve_args(args); 247 } 248 249 /* 250 * If the current process has a special image activator it 251 * wants to try first, call it. For example, emulating shell 252 * scripts differently. 253 */ 254 error = -1; 255 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 256 error = img_first(imgp); 257 258 /* 259 * If the vnode has a registered vmspace, exec the vmspace 260 */ 261 if (error == -1 && imgp->vp->v_resident) { 262 error = exec_resident_imgact(imgp); 263 } 264 265 /* 266 * Loop through the list of image activators, calling each one. 267 * An activator returns -1 if there is no match, 0 on success, 268 * and an error otherwise. 269 */ 270 for (i = 0; error == -1 && execsw[i]; ++i) { 271 if (execsw[i]->ex_imgact == NULL || 272 execsw[i]->ex_imgact == img_first) { 273 continue; 274 } 275 error = (*execsw[i]->ex_imgact)(imgp); 276 } 277 278 if (error) { 279 if (error == -1) 280 error = ENOEXEC; 281 goto exec_fail_dealloc; 282 } 283 284 /* 285 * Special interpreter operation, cleanup and loop up to try to 286 * activate the interpreter. 287 */ 288 if (imgp->interpreted) { 289 exec_unmap_first_page(imgp); 290 nlookup_done(nd); 291 vrele(imgp->vp); 292 imgp->vp = NULL; 293 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 294 NLC_FOLLOW); 295 if (error) 296 goto exec_fail; 297 goto interpret; 298 } 299 300 /* 301 * Copy out strings (args and env) and initialize stack base 302 */ 303 stack_base = exec_copyout_strings(imgp); 304 p->p_vmspace->vm_minsaddr = (char *)stack_base; 305 306 /* 307 * If custom stack fixup routine present for this process 308 * let it do the stack setup. If we are running a resident 309 * image there is no auxinfo or other image activator context 310 * so don't try to add fixups to the stack. 311 * 312 * Else stuff argument count as first item on stack 313 */ 314 if (p->p_sysent->sv_fixup && imgp->resident == 0) 315 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 316 else 317 suword(--stack_base, imgp->args->argc); 318 319 /* 320 * For security and other reasons, the file descriptor table cannot 321 * be shared after an exec. 322 */ 323 if (p->p_fd->fd_refcnt > 1) { 324 struct filedesc *tmp; 325 326 tmp = fdcopy(p); 327 fdfree(p); 328 p->p_fd = tmp; 329 } 330 331 /* 332 * For security and other reasons, signal handlers cannot 333 * be shared after an exec. The new proces gets a copy of the old 334 * handlers. In execsigs(), the new process will have its signals 335 * reset. 336 */ 337 if (p->p_sigacts->ps_refcnt > 1) { 338 struct sigacts *newsigacts; 339 340 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts), 341 M_SUBPROC, M_WAITOK); 342 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts)); 343 p->p_sigacts->ps_refcnt--; 344 p->p_sigacts = newsigacts; 345 p->p_sigacts->ps_refcnt = 1; 346 } 347 348 /* 349 * For security and other reasons virtual kernels cannot be 350 * inherited by an exec. This also allows a virtual kernel 351 * to fork/exec unrelated applications. 352 */ 353 if (p->p_vkernel) 354 vkernel_exit(p); 355 356 /* Stop profiling */ 357 stopprofclock(p); 358 359 /* close files on exec */ 360 fdcloseexec(p); 361 362 /* reset caught signals */ 363 execsigs(p); 364 365 /* name this process - nameiexec(p, ndp) */ 366 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 367 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 368 p->p_comm[len] = 0; 369 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 370 371 /* 372 * mark as execed, wakeup the process that vforked (if any) and tell 373 * it that it now has its own resources back 374 */ 375 p->p_flag |= P_EXEC; 376 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 377 p->p_flag &= ~P_PPWAIT; 378 wakeup((caddr_t)p->p_pptr); 379 } 380 381 /* 382 * Implement image setuid/setgid. 383 * 384 * Don't honor setuid/setgid if the filesystem prohibits it or if 385 * the process is being traced. 386 */ 387 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 388 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 389 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 390 (p->p_flag & P_TRACED) == 0) { 391 /* 392 * Turn off syscall tracing for set-id programs, except for 393 * root. Record any set-id flags first to make sure that 394 * we do not regain any tracing during a possible block. 395 */ 396 setsugid(); 397 if (p->p_tracenode && suser(td) != 0) { 398 ktrdestroy(&p->p_tracenode); 399 p->p_traceflag = 0; 400 } 401 /* Close any file descriptors 0..2 that reference procfs */ 402 setugidsafety(p); 403 /* Make sure file descriptors 0..2 are in use. */ 404 error = fdcheckstd(p); 405 if (error != 0) 406 goto exec_fail_dealloc; 407 /* 408 * Set the new credentials. 409 */ 410 cratom(&p->p_ucred); 411 if (attr.va_mode & VSUID) 412 change_euid(attr.va_uid); 413 if (attr.va_mode & VSGID) 414 p->p_ucred->cr_gid = attr.va_gid; 415 416 /* 417 * Clear local varsym variables 418 */ 419 varsymset_clean(&p->p_varsymset); 420 } else { 421 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 422 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 423 p->p_flag &= ~P_SUGID; 424 } 425 426 /* 427 * Implement correct POSIX saved-id behavior. 428 */ 429 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 430 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 431 cratom(&p->p_ucred); 432 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 433 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 434 } 435 436 /* 437 * Store the vp for use in procfs 438 */ 439 if (p->p_textvp) /* release old reference */ 440 vrele(p->p_textvp); 441 p->p_textvp = imgp->vp; 442 vref(p->p_textvp); 443 444 /* 445 * Notify others that we exec'd, and clear the P_INEXEC flag 446 * as we're now a bona fide freshly-execed process. 447 */ 448 KNOTE(&p->p_klist, NOTE_EXEC); 449 p->p_flag &= ~P_INEXEC; 450 451 /* 452 * If tracing the process, trap to debugger so breakpoints 453 * can be set before the program executes. 454 */ 455 STOPEVENT(p, S_EXEC, 0); 456 457 if (p->p_flag & P_TRACED) 458 ksignal(p, SIGTRAP); 459 460 /* clear "fork but no exec" flag, as we _are_ execing */ 461 p->p_acflag &= ~AFORK; 462 463 /* Set values passed into the program in registers. */ 464 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 465 imgp->ps_strings); 466 467 /* Free any previous argument cache */ 468 if (p->p_args && --p->p_args->ar_ref == 0) 469 FREE(p->p_args, M_PARGS); 470 p->p_args = NULL; 471 472 /* Cache arguments if they fit inside our allowance */ 473 i = imgp->args->begin_envv - imgp->args->begin_argv; 474 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 475 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 476 M_PARGS, M_WAITOK); 477 p->p_args->ar_ref = 1; 478 p->p_args->ar_length = i; 479 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 480 } 481 482 exec_fail_dealloc: 483 484 /* 485 * free various allocated resources 486 */ 487 if (imgp->firstpage) 488 exec_unmap_first_page(imgp); 489 490 if (imgp->vp) { 491 vrele(imgp->vp); 492 imgp->vp = NULL; 493 } 494 495 if (error == 0) { 496 ++mycpu->gd_cnt.v_exec; 497 return (0); 498 } 499 500 exec_fail: 501 /* 502 * we're done here, clear P_INEXEC if we were the ones that 503 * set it. Otherwise if vmspace_destroyed is still set we 504 * raced another thread and that thread is responsible for 505 * clearing it. 506 */ 507 if (imgp->vmspace_destroyed & 2) 508 p->p_flag &= ~P_INEXEC; 509 if (imgp->vmspace_destroyed) { 510 /* 511 * Sorry, no more process anymore. exit gracefully. 512 * However we can't die right here, because our 513 * caller might have to clean up, so indicate a 514 * lethal error by returning -1. 515 */ 516 return(-1); 517 } else { 518 return(error); 519 } 520 } 521 522 /* 523 * execve() system call. 524 */ 525 int 526 sys_execve(struct execve_args *uap) 527 { 528 struct nlookupdata nd; 529 struct image_args args; 530 int error; 531 532 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 533 if (error == 0) { 534 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 535 uap->argv, uap->envv); 536 } 537 if (error == 0) 538 error = kern_execve(&nd, &args); 539 nlookup_done(&nd); 540 exec_free_args(&args); 541 542 if (error < 0) { 543 /* We hit a lethal error condition. Let's die now. */ 544 exit1(W_EXITCODE(0, SIGABRT)); 545 /* NOTREACHED */ 546 } 547 548 /* 549 * The syscall result is returned in registers to the new program. 550 * Linux will register %edx as an atexit function and we must be 551 * sure to set it to 0. XXX 552 */ 553 if (error == 0) 554 uap->sysmsg_result64 = 0; 555 556 return (error); 557 } 558 559 int 560 exec_map_first_page(struct image_params *imgp) 561 { 562 int rv, i; 563 int initial_pagein; 564 vm_page_t ma[VM_INITIAL_PAGEIN]; 565 vm_page_t m; 566 vm_object_t object; 567 568 if (imgp->firstpage) 569 exec_unmap_first_page(imgp); 570 571 /* 572 * The file has to be mappable. 573 */ 574 if ((object = imgp->vp->v_object) == NULL) 575 return (EIO); 576 577 /* 578 * We shouldn't need protection for vm_page_grab() but we certainly 579 * need it for the lookup loop below (lookup/busy race), since 580 * an interrupt can unbusy and free the page before our busy check. 581 */ 582 crit_enter(); 583 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 584 585 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 586 ma[0] = m; 587 initial_pagein = VM_INITIAL_PAGEIN; 588 if (initial_pagein > object->size) 589 initial_pagein = object->size; 590 for (i = 1; i < initial_pagein; i++) { 591 if ((m = vm_page_lookup(object, i)) != NULL) { 592 if ((m->flags & PG_BUSY) || m->busy) 593 break; 594 if (m->valid) 595 break; 596 vm_page_busy(m); 597 } else { 598 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 599 if (m == NULL) 600 break; 601 } 602 ma[i] = m; 603 } 604 initial_pagein = i; 605 606 /* 607 * get_pages unbusies all the requested pages except the 608 * primary page (at index 0 in this case). 609 */ 610 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 611 m = vm_page_lookup(object, 0); 612 613 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 614 if (m) { 615 vm_page_protect(m, VM_PROT_NONE); 616 vm_page_free(m); 617 } 618 crit_exit(); 619 return EIO; 620 } 621 } 622 vm_page_hold(m); 623 vm_page_wakeup(m); /* unbusy the page */ 624 crit_exit(); 625 626 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 627 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 628 629 return 0; 630 } 631 632 void 633 exec_unmap_first_page(struct image_params *imgp) 634 { 635 vm_page_t m; 636 637 crit_enter(); 638 if (imgp->firstpage != NULL) { 639 m = sf_buf_page(imgp->firstpage); 640 sf_buf_free(imgp->firstpage); 641 imgp->firstpage = NULL; 642 imgp->image_header = NULL; 643 vm_page_unhold(m); 644 } 645 crit_exit(); 646 } 647 648 /* 649 * Destroy old address space, and allocate a new stack 650 * The new stack is only SGROWSIZ large because it is grown 651 * automatically in trap.c. 652 * 653 * This is the point of no return. 654 */ 655 int 656 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 657 { 658 struct vmspace *vmspace = imgp->proc->p_vmspace; 659 vm_offset_t stack_addr = USRSTACK - maxssiz; 660 struct proc *p; 661 vm_map_t map; 662 int error; 663 664 /* 665 * Indicate that we cannot gracefully error out any more, kill 666 * any other threads present, and set P_INEXEC to indicate that 667 * we are now messing with the process structure proper. 668 * 669 * If killalllwps() races return an error which coupled with 670 * vmspace_destroyed will cause us to exit. This is what we 671 * want since another thread is patiently waiting for us to exit 672 * in that case. 673 */ 674 p = curproc; 675 imgp->vmspace_destroyed = 1; 676 677 if (curthread->td_proc->p_nthreads > 1) { 678 error = killalllwps(1); 679 if (error) 680 return (error); 681 } 682 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 683 p->p_flag |= P_INEXEC; 684 685 /* 686 * Prevent a pending AIO from modifying the new address space. 687 */ 688 aio_proc_rundown(imgp->proc); 689 690 /* 691 * Blow away entire process VM, if address space not shared, 692 * otherwise, create a new VM space so that other threads are 693 * not disrupted. If we are execing a resident vmspace we 694 * create a duplicate of it and remap the stack. 695 * 696 * The exitingcnt test is not strictly necessary but has been 697 * included for code sanity (to make the code more deterministic). 698 */ 699 map = &vmspace->vm_map; 700 if (vmcopy) { 701 vmspace_exec(imgp->proc, vmcopy); 702 vmspace = imgp->proc->p_vmspace; 703 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 704 map = &vmspace->vm_map; 705 } else if (vmspace->vm_sysref.refcnt == 1 && 706 vmspace->vm_exitingcnt == 0) { 707 shmexit(vmspace); 708 if (vmspace->vm_upcalls) 709 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 710 pmap_remove_pages(vmspace_pmap(vmspace), 711 0, VM_MAX_USER_ADDRESS); 712 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 713 } else { 714 vmspace_exec(imgp->proc, NULL); 715 vmspace = imgp->proc->p_vmspace; 716 map = &vmspace->vm_map; 717 } 718 719 /* Allocate a new stack */ 720 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 721 VM_PROT_ALL, VM_PROT_ALL, 0); 722 if (error) 723 return (error); 724 725 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 726 * VM_STACK case, but they are still used to monitor the size of the 727 * process stack so we can check the stack rlimit. 728 */ 729 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 730 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 731 732 return(0); 733 } 734 735 /* 736 * Copy out argument and environment strings from the old process 737 * address space into the temporary string buffer. 738 */ 739 int 740 exec_copyin_args(struct image_args *args, char *fname, 741 enum exec_path_segflg segflg, char **argv, char **envv) 742 { 743 char *argp, *envp; 744 int error = 0; 745 size_t length; 746 747 bzero(args, sizeof(*args)); 748 749 args->buf = objcache_get(exec_objcache, M_WAITOK); 750 if (args->buf == NULL) 751 return (ENOMEM); 752 args->begin_argv = args->buf; 753 args->endp = args->begin_argv; 754 args->space = ARG_MAX; 755 756 args->fname = args->buf + ARG_MAX; 757 758 /* 759 * Copy the file name. 760 */ 761 if (segflg == PATH_SYSSPACE) { 762 error = copystr(fname, args->fname, PATH_MAX, &length); 763 } else if (segflg == PATH_USERSPACE) { 764 error = copyinstr(fname, args->fname, PATH_MAX, &length); 765 } 766 767 /* 768 * Extract argument strings. argv may not be NULL. The argv 769 * array is terminated by a NULL entry. We special-case the 770 * situation where argv[0] is NULL by passing { filename, NULL } 771 * to the new program to guarentee that the interpreter knows what 772 * file to open in case we exec an interpreted file. Note that 773 * a NULL argv[0] terminates the argv[] array. 774 * 775 * XXX the special-casing of argv[0] is historical and needs to be 776 * revisited. 777 */ 778 if (argv == NULL) 779 error = EFAULT; 780 if (error == 0) { 781 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 782 if (argp == (caddr_t)-1) { 783 error = EFAULT; 784 break; 785 } 786 error = copyinstr(argp, args->endp, 787 args->space, &length); 788 if (error) { 789 if (error == ENAMETOOLONG) 790 error = E2BIG; 791 break; 792 } 793 args->space -= length; 794 args->endp += length; 795 args->argc++; 796 } 797 if (args->argc == 0 && error == 0) { 798 length = strlen(args->fname) + 1; 799 if (length > args->space) { 800 error = E2BIG; 801 } else { 802 bcopy(args->fname, args->endp, length); 803 args->space -= length; 804 args->endp += length; 805 args->argc++; 806 } 807 } 808 } 809 810 args->begin_envv = args->endp; 811 812 /* 813 * extract environment strings. envv may be NULL. 814 */ 815 if (envv && error == 0) { 816 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 817 if (envp == (caddr_t) -1) { 818 error = EFAULT; 819 break; 820 } 821 error = copyinstr(envp, args->endp, args->space, 822 &length); 823 if (error) { 824 if (error == ENAMETOOLONG) 825 error = E2BIG; 826 break; 827 } 828 args->space -= length; 829 args->endp += length; 830 args->envc++; 831 } 832 } 833 return (error); 834 } 835 836 void 837 exec_free_args(struct image_args *args) 838 { 839 if (args->buf) { 840 objcache_put(exec_objcache, args->buf); 841 args->buf = NULL; 842 } 843 } 844 845 /* 846 * Copy strings out to the new process address space, constructing 847 * new arg and env vector tables. Return a pointer to the base 848 * so that it can be used as the initial stack pointer. 849 */ 850 register_t * 851 exec_copyout_strings(struct image_params *imgp) 852 { 853 int argc, envc, sgap; 854 char **vectp; 855 char *stringp, *destp; 856 register_t *stack_base; 857 struct ps_strings *arginfo; 858 int szsigcode; 859 860 /* 861 * Calculate string base and vector table pointers. 862 * Also deal with signal trampoline code for this exec type. 863 */ 864 arginfo = (struct ps_strings *)PS_STRINGS; 865 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 866 if (stackgap_random != 0) 867 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 868 else 869 sgap = 0; 870 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 871 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 872 873 /* 874 * install sigcode 875 */ 876 if (szsigcode) 877 copyout(imgp->proc->p_sysent->sv_sigcode, 878 ((caddr_t)arginfo - szsigcode), szsigcode); 879 880 /* 881 * If we have a valid auxargs ptr, prepare some room 882 * on the stack. 883 * 884 * The '+ 2' is for the null pointers at the end of each of the 885 * arg and env vector sets, and 'AT_COUNT*2' is room for the 886 * ELF Auxargs data. 887 */ 888 if (imgp->auxargs) { 889 vectp = (char **)(destp - (imgp->args->argc + 890 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 891 } else { 892 vectp = (char **)(destp - (imgp->args->argc + 893 imgp->args->envc + 2) * sizeof(char*)); 894 } 895 896 /* 897 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 898 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 899 */ 900 901 /* 902 * vectp also becomes our initial stack base 903 */ 904 stack_base = (register_t *)vectp; 905 906 stringp = imgp->args->begin_argv; 907 argc = imgp->args->argc; 908 envc = imgp->args->envc; 909 910 /* 911 * Copy out strings - arguments and environment. 912 */ 913 copyout(stringp, destp, ARG_MAX - imgp->args->space); 914 915 /* 916 * Fill in "ps_strings" struct for ps, w, etc. 917 */ 918 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 919 suword(&arginfo->ps_nargvstr, argc); 920 921 /* 922 * Fill in argument portion of vector table. 923 */ 924 for (; argc > 0; --argc) { 925 suword(vectp++, (long)(intptr_t)destp); 926 while (*stringp++ != 0) 927 destp++; 928 destp++; 929 } 930 931 /* a null vector table pointer separates the argp's from the envp's */ 932 suword(vectp++, 0); 933 934 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 935 suword(&arginfo->ps_nenvstr, envc); 936 937 /* 938 * Fill in environment portion of vector table. 939 */ 940 for (; envc > 0; --envc) { 941 suword(vectp++, (long)(intptr_t)destp); 942 while (*stringp++ != 0) 943 destp++; 944 destp++; 945 } 946 947 /* end of vector table is a null pointer */ 948 suword(vectp, 0); 949 950 return (stack_base); 951 } 952 953 /* 954 * Check permissions of file to execute. 955 * Return 0 for success or error code on failure. 956 */ 957 int 958 exec_check_permissions(struct image_params *imgp) 959 { 960 struct proc *p = imgp->proc; 961 struct vnode *vp = imgp->vp; 962 struct vattr *attr = imgp->attr; 963 int error; 964 965 /* Get file attributes */ 966 error = VOP_GETATTR(vp, attr); 967 if (error) 968 return (error); 969 970 /* 971 * 1) Check if file execution is disabled for the filesystem that this 972 * file resides on. 973 * 2) Insure that at least one execute bit is on - otherwise root 974 * will always succeed, and we don't want to happen unless the 975 * file really is executable. 976 * 3) Insure that the file is a regular file. 977 */ 978 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 979 ((attr->va_mode & 0111) == 0) || 980 (attr->va_type != VREG)) { 981 return (EACCES); 982 } 983 984 /* 985 * Zero length files can't be exec'd 986 */ 987 if (attr->va_size == 0) 988 return (ENOEXEC); 989 990 /* 991 * Check for execute permission to file based on current credentials. 992 */ 993 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 994 if (error) 995 return (error); 996 997 /* 998 * Check number of open-for-writes on the file and deny execution 999 * if there are any. 1000 */ 1001 if (vp->v_writecount) 1002 return (ETXTBSY); 1003 1004 /* 1005 * Call filesystem specific open routine, which allows us to read, 1006 * write, and mmap the file. Without the VOP_OPEN we can only 1007 * stat the file. 1008 */ 1009 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1010 if (error) 1011 return (error); 1012 1013 return (0); 1014 } 1015 1016 /* 1017 * Exec handler registration 1018 */ 1019 int 1020 exec_register(const struct execsw *execsw_arg) 1021 { 1022 const struct execsw **es, **xs, **newexecsw; 1023 int count = 2; /* New slot and trailing NULL */ 1024 1025 if (execsw) 1026 for (es = execsw; *es; es++) 1027 count++; 1028 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1029 if (newexecsw == NULL) 1030 return ENOMEM; 1031 xs = newexecsw; 1032 if (execsw) 1033 for (es = execsw; *es; es++) 1034 *xs++ = *es; 1035 *xs++ = execsw_arg; 1036 *xs = NULL; 1037 if (execsw) 1038 kfree(execsw, M_TEMP); 1039 execsw = newexecsw; 1040 return 0; 1041 } 1042 1043 int 1044 exec_unregister(const struct execsw *execsw_arg) 1045 { 1046 const struct execsw **es, **xs, **newexecsw; 1047 int count = 1; 1048 1049 if (execsw == NULL) 1050 panic("unregister with no handlers left?"); 1051 1052 for (es = execsw; *es; es++) { 1053 if (*es == execsw_arg) 1054 break; 1055 } 1056 if (*es == NULL) 1057 return ENOENT; 1058 for (es = execsw; *es; es++) 1059 if (*es != execsw_arg) 1060 count++; 1061 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1062 if (newexecsw == NULL) 1063 return ENOMEM; 1064 xs = newexecsw; 1065 for (es = execsw; *es; es++) 1066 if (*es != execsw_arg) 1067 *xs++ = *es; 1068 *xs = NULL; 1069 if (execsw) 1070 kfree(execsw, M_TEMP); 1071 execsw = newexecsw; 1072 return 0; 1073 } 1074