1 /* $NetBSD: kern_exec.c,v 1.172 2003/08/29 13:29:32 enami Exp $ */ 2 3 /*- 4 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou 5 * Copyright (C) 1992 Wolfgang Solfrank. 6 * Copyright (C) 1992 TooLs GmbH. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by TooLs GmbH. 20 * 4. The name of TooLs GmbH may not be used to endorse or promote products 21 * derived from this software without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 26 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 28 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 29 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 30 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 31 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 32 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.172 2003/08/29 13:29:32 enami Exp $"); 37 38 #include "opt_ktrace.h" 39 #include "opt_syscall_debug.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/filedesc.h> 44 #include <sys/kernel.h> 45 #include <sys/proc.h> 46 #include <sys/mount.h> 47 #include <sys/malloc.h> 48 #include <sys/namei.h> 49 #include <sys/vnode.h> 50 #include <sys/file.h> 51 #include <sys/acct.h> 52 #include <sys/exec.h> 53 #include <sys/ktrace.h> 54 #include <sys/resourcevar.h> 55 #include <sys/wait.h> 56 #include <sys/mman.h> 57 #include <sys/ras.h> 58 #include <sys/signalvar.h> 59 #include <sys/stat.h> 60 #include <sys/syscall.h> 61 62 #include <sys/sa.h> 63 #include <sys/savar.h> 64 #include <sys/syscallargs.h> 65 66 #include <uvm/uvm_extern.h> 67 68 #include <machine/cpu.h> 69 #include <machine/reg.h> 70 71 static int exec_sigcode_map(struct proc *, const struct emul *); 72 73 #ifdef DEBUG_EXEC 74 #define DPRINTF(a) uprintf a 75 #else 76 #define DPRINTF(a) 77 #endif /* DEBUG_EXEC */ 78 79 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec"); 80 81 /* 82 * Exec function switch: 83 * 84 * Note that each makecmds function is responsible for loading the 85 * exec package with the necessary functions for any exec-type-specific 86 * handling. 87 * 88 * Functions for specific exec types should be defined in their own 89 * header file. 90 */ 91 extern const struct execsw execsw_builtin[]; 92 extern int nexecs_builtin; 93 static const struct execsw **execsw = NULL; 94 static int nexecs; 95 96 u_int exec_maxhdrsz; /* must not be static - netbsd32 needs it */ 97 98 #ifdef LKM 99 /* list of supported emulations */ 100 static 101 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head); 102 struct emul_entry { 103 LIST_ENTRY(emul_entry) el_list; 104 const struct emul *el_emul; 105 int ro_entry; 106 }; 107 108 /* list of dynamically loaded execsw entries */ 109 static 110 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head); 111 struct exec_entry { 112 LIST_ENTRY(exec_entry) ex_list; 113 const struct execsw *es; 114 }; 115 116 /* structure used for building execw[] */ 117 struct execsw_entry { 118 struct execsw_entry *next; 119 const struct execsw *es; 120 }; 121 #endif /* LKM */ 122 123 /* NetBSD emul struct */ 124 extern char sigcode[], esigcode[]; 125 #ifdef SYSCALL_DEBUG 126 extern const char * const syscallnames[]; 127 #endif 128 #ifdef __HAVE_SYSCALL_INTERN 129 void syscall_intern(struct proc *); 130 #else 131 void syscall(void); 132 #endif 133 134 struct uvm_object *emul_netbsd_object; 135 136 const struct emul emul_netbsd = { 137 "netbsd", 138 NULL, /* emulation path */ 139 #ifndef __HAVE_MINIMAL_EMUL 140 EMUL_HAS_SYS___syscall, 141 NULL, 142 SYS_syscall, 143 SYS_NSYSENT, 144 #endif 145 sysent, 146 #ifdef SYSCALL_DEBUG 147 syscallnames, 148 #else 149 NULL, 150 #endif 151 sendsig, 152 trapsignal, 153 sigcode, 154 esigcode, 155 &emul_netbsd_object, 156 setregs, 157 NULL, 158 NULL, 159 NULL, 160 #ifdef __HAVE_SYSCALL_INTERN 161 syscall_intern, 162 #else 163 syscall, 164 #endif 165 NULL, 166 NULL, 167 }; 168 169 #ifdef LKM 170 /* 171 * Exec lock. Used to control access to execsw[] structures. 172 * This must not be static so that netbsd32 can access it, too. 173 */ 174 struct lock exec_lock; 175 176 static void link_es(struct execsw_entry **, const struct execsw *); 177 #endif /* LKM */ 178 179 /* 180 * check exec: 181 * given an "executable" described in the exec package's namei info, 182 * see what we can do with it. 183 * 184 * ON ENTRY: 185 * exec package with appropriate namei info 186 * proc pointer of exec'ing proc 187 * iff verified exec enabled then flag indicating a direct exec or 188 * an indirect exec (i.e. for a shell script interpreter) 189 * NO SELF-LOCKED VNODES 190 * 191 * ON EXIT: 192 * error: nothing held, etc. exec header still allocated. 193 * ok: filled exec package, executable's vnode (unlocked). 194 * 195 * EXEC SWITCH ENTRY: 196 * Locked vnode to check, exec package, proc. 197 * 198 * EXEC SWITCH EXIT: 199 * ok: return 0, filled exec package, executable's vnode (unlocked). 200 * error: destructive: 201 * everything deallocated execept exec header. 202 * non-destructive: 203 * error code, executable's vnode (unlocked), 204 * exec header unmodified. 205 */ 206 int 207 #ifdef VERIFIED_EXEC 208 check_exec(struct proc *p, struct exec_package *epp, int direct_exec) 209 #else 210 check_exec(struct proc *p, struct exec_package *epp) 211 #endif 212 { 213 int error, i; 214 struct vnode *vp; 215 struct nameidata *ndp; 216 size_t resid; 217 218 ndp = epp->ep_ndp; 219 ndp->ni_cnd.cn_nameiop = LOOKUP; 220 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 221 /* first get the vnode */ 222 if ((error = namei(ndp)) != 0) 223 return error; 224 epp->ep_vp = vp = ndp->ni_vp; 225 226 /* check access and type */ 227 if (vp->v_type != VREG) { 228 error = EACCES; 229 goto bad1; 230 } 231 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 232 goto bad1; 233 234 /* get attributes */ 235 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 236 goto bad1; 237 238 /* Check mount point */ 239 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 240 error = EACCES; 241 goto bad1; 242 } 243 if (vp->v_mount->mnt_flag & MNT_NOSUID) 244 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 245 246 /* try to open it */ 247 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 248 goto bad1; 249 250 /* unlock vp, since we need it unlocked from here on out. */ 251 VOP_UNLOCK(vp, 0); 252 253 254 #ifdef VERIFIED_EXEC 255 /* Evaluate signature for file... */ 256 if ((error = check_veriexec(p, vp, epp, direct_exec)) != 0) 257 goto bad2; 258 #endif 259 260 /* now we have the file, get the exec header */ 261 uvn_attach(vp, VM_PROT_READ); 262 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 263 UIO_SYSSPACE, 0, p->p_ucred, &resid, p); 264 if (error) 265 goto bad2; 266 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 267 268 /* 269 * Set up default address space limits. Can be overridden 270 * by individual exec packages. 271 * 272 * XXX probably should be all done in the exec pakages. 273 */ 274 epp->ep_vm_minaddr = VM_MIN_ADDRESS; 275 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 276 /* 277 * set up the vmcmds for creation of the process 278 * address space 279 */ 280 error = ENOEXEC; 281 for (i = 0; i < nexecs && error != 0; i++) { 282 int newerror; 283 284 epp->ep_esch = execsw[i]; 285 newerror = (*execsw[i]->es_check)(p, epp); 286 /* make sure the first "interesting" error code is saved. */ 287 if (!newerror || error == ENOEXEC) 288 error = newerror; 289 290 /* if es_check call was successful, update epp->ep_es */ 291 if (!newerror && (epp->ep_flags & EXEC_HASES) == 0) 292 epp->ep_es = execsw[i]; 293 294 if (epp->ep_flags & EXEC_DESTR && error != 0) 295 return error; 296 } 297 if (!error) { 298 /* check that entry point is sane */ 299 if (epp->ep_entry > VM_MAXUSER_ADDRESS) 300 error = ENOEXEC; 301 302 /* check limits */ 303 if ((epp->ep_tsize > MAXTSIZ) || 304 (epp->ep_dsize > 305 (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur)) 306 error = ENOMEM; 307 308 if (!error) 309 return (0); 310 } 311 312 /* 313 * free any vmspace-creation commands, 314 * and release their references 315 */ 316 kill_vmcmds(&epp->ep_vmcmds); 317 318 bad2: 319 /* 320 * close and release the vnode, restore the old one, free the 321 * pathname buf, and punt. 322 */ 323 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 324 VOP_CLOSE(vp, FREAD, p->p_ucred, p); 325 vput(vp); 326 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 327 return error; 328 329 bad1: 330 /* 331 * free the namei pathname buffer, and put the vnode 332 * (which we don't yet have open). 333 */ 334 vput(vp); /* was still locked */ 335 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 336 return error; 337 } 338 339 /* 340 * exec system call 341 */ 342 /* ARGSUSED */ 343 int 344 sys_execve(struct lwp *l, void *v, register_t *retval) 345 { 346 struct sys_execve_args /* { 347 syscallarg(const char *) path; 348 syscallarg(char * const *) argp; 349 syscallarg(char * const *) envp; 350 } */ *uap = v; 351 int error; 352 u_int i; 353 struct exec_package pack; 354 struct nameidata nid; 355 struct vattr attr; 356 struct proc *p; 357 struct ucred *cred; 358 char *argp; 359 char * const *cpp; 360 char *dp, *sp; 361 long argc, envc; 362 size_t len; 363 char *stack; 364 struct ps_strings arginfo; 365 struct vmspace *vm; 366 char **tmpfap; 367 int szsigcode; 368 struct exec_vmcmd *base_vcp; 369 int oldlwpflags; 370 371 /* Disable scheduler activation upcalls. */ 372 oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL); 373 if (l->l_flag & L_SA) 374 l->l_flag &= ~(L_SA | L_SA_UPCALL); 375 376 p = l->l_proc; 377 /* 378 * Lock the process and set the P_INEXEC flag to indicate that 379 * it should be left alone until we're done here. This is 380 * necessary to avoid race conditions - e.g. in ptrace() - 381 * that might allow a local user to illicitly obtain elevated 382 * privileges. 383 */ 384 p->p_flag |= P_INEXEC; 385 386 cred = p->p_ucred; 387 base_vcp = NULL; 388 /* 389 * Init the namei data to point the file user's program name. 390 * This is done here rather than in check_exec(), so that it's 391 * possible to override this settings if any of makecmd/probe 392 * functions call check_exec() recursively - for example, 393 * see exec_script_makecmds(). 394 */ 395 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 396 397 /* 398 * initialize the fields of the exec package. 399 */ 400 pack.ep_name = SCARG(uap, path); 401 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 402 pack.ep_hdrlen = exec_maxhdrsz; 403 pack.ep_hdrvalid = 0; 404 pack.ep_ndp = &nid; 405 pack.ep_emul_arg = NULL; 406 pack.ep_vmcmds.evs_cnt = 0; 407 pack.ep_vmcmds.evs_used = 0; 408 pack.ep_vap = &attr; 409 pack.ep_flags = 0; 410 411 #ifdef LKM 412 lockmgr(&exec_lock, LK_SHARED, NULL); 413 #endif 414 415 /* see if we can run it. */ 416 #ifdef VERIFIED_EXEC 417 if ((error = check_exec(p, &pack, 1)) != 0) 418 /* if ((error = check_exec(p, &pack, 0)) != 0) */ 419 #else 420 if ((error = check_exec(p, &pack)) != 0) 421 #endif 422 goto freehdr; 423 424 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 425 426 /* allocate an argument buffer */ 427 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS); 428 #ifdef DIAGNOSTIC 429 if (argp == (vaddr_t) 0) 430 panic("execve: argp == NULL"); 431 #endif 432 dp = argp; 433 argc = 0; 434 435 /* copy the fake args list, if there's one, freeing it as we go */ 436 if (pack.ep_flags & EXEC_HASARGL) { 437 tmpfap = pack.ep_fa; 438 while (*tmpfap != NULL) { 439 char *cp; 440 441 cp = *tmpfap; 442 while (*cp) 443 *dp++ = *cp++; 444 dp++; 445 446 FREE(*tmpfap, M_EXEC); 447 tmpfap++; argc++; 448 } 449 FREE(pack.ep_fa, M_EXEC); 450 pack.ep_flags &= ~EXEC_HASARGL; 451 } 452 453 /* Now get argv & environment */ 454 if (!(cpp = SCARG(uap, argp))) { 455 error = EINVAL; 456 goto bad; 457 } 458 459 if (pack.ep_flags & EXEC_SKIPARG) 460 cpp++; 461 462 while (1) { 463 len = argp + ARG_MAX - dp; 464 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 465 goto bad; 466 if (!sp) 467 break; 468 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 469 if (error == ENAMETOOLONG) 470 error = E2BIG; 471 goto bad; 472 } 473 #ifdef KTRACE 474 if (KTRPOINT(p, KTR_EXEC_ARG)) 475 ktrkmem(p, KTR_EXEC_ARG, dp, len - 1); 476 #endif 477 dp += len; 478 cpp++; 479 argc++; 480 } 481 482 envc = 0; 483 /* environment need not be there */ 484 if ((cpp = SCARG(uap, envp)) != NULL ) { 485 while (1) { 486 len = argp + ARG_MAX - dp; 487 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 488 goto bad; 489 if (!sp) 490 break; 491 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 492 if (error == ENAMETOOLONG) 493 error = E2BIG; 494 goto bad; 495 } 496 #ifdef KTRACE 497 if (KTRPOINT(p, KTR_EXEC_ENV)) 498 ktrkmem(p, KTR_EXEC_ENV, dp, len - 1); 499 #endif 500 dp += len; 501 cpp++; 502 envc++; 503 } 504 } 505 506 dp = (char *) ALIGN(dp); 507 508 szsigcode = pack.ep_es->es_emul->e_esigcode - 509 pack.ep_es->es_emul->e_sigcode; 510 511 /* Now check if args & environ fit into new stack */ 512 if (pack.ep_flags & EXEC_32) 513 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 514 sizeof(int) + sizeof(int) + dp + STACKGAPLEN + 515 szsigcode + sizeof(struct ps_strings)) - argp; 516 else 517 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 518 sizeof(char *) + sizeof(int) + dp + STACKGAPLEN + 519 szsigcode + sizeof(struct ps_strings)) - argp; 520 521 len = ALIGN(len); /* make the stack "safely" aligned */ 522 523 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 524 error = ENOMEM; 525 goto bad; 526 } 527 528 /* Get rid of other LWPs/ */ 529 p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */ 530 exit_lwps(l); 531 p->p_flag &= ~P_WEXIT; 532 KDASSERT(p->p_nlwps == 1); 533 534 /* This is now LWP 1 */ 535 l->l_lid = 1; 536 p->p_nlwpid = 1; 537 538 /* Release any SA state. */ 539 if (p->p_sa) { 540 p->p_flag &= ~P_SA; 541 free(p->p_sa->sa_stacks, M_SA); 542 pool_put(&sadata_pool, p->p_sa); 543 p->p_sa = NULL; 544 } 545 546 /* Remove POSIX timers */ 547 timers_free(p, TIMERS_POSIX); 548 549 /* adjust "active stack depth" for process VSZ */ 550 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 551 552 /* 553 * Do whatever is necessary to prepare the address space 554 * for remapping. Note that this might replace the current 555 * vmspace with another! 556 */ 557 uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 558 559 /* Now map address space */ 560 vm = p->p_vmspace; 561 vm->vm_taddr = (caddr_t) pack.ep_taddr; 562 vm->vm_tsize = btoc(pack.ep_tsize); 563 vm->vm_daddr = (caddr_t) pack.ep_daddr; 564 vm->vm_dsize = btoc(pack.ep_dsize); 565 vm->vm_ssize = btoc(pack.ep_ssize); 566 vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr; 567 vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr; 568 569 /* create the new process's VM space by running the vmcmds */ 570 #ifdef DIAGNOSTIC 571 if (pack.ep_vmcmds.evs_used == 0) 572 panic("execve: no vmcmds"); 573 #endif 574 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 575 struct exec_vmcmd *vcp; 576 577 vcp = &pack.ep_vmcmds.evs_cmds[i]; 578 if (vcp->ev_flags & VMCMD_RELATIVE) { 579 #ifdef DIAGNOSTIC 580 if (base_vcp == NULL) 581 panic("execve: relative vmcmd with no base"); 582 if (vcp->ev_flags & VMCMD_BASE) 583 panic("execve: illegal base & relative vmcmd"); 584 #endif 585 vcp->ev_addr += base_vcp->ev_addr; 586 } 587 error = (*vcp->ev_proc)(p, vcp); 588 #ifdef DEBUG_EXEC 589 if (error) { 590 int j; 591 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 592 for (j = 0; j <= i; j++) 593 uprintf( 594 "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n", 595 j, vp[j].ev_addr, vp[j].ev_len, 596 vp[j].ev_offset, vp[j].ev_prot, 597 vp[j].ev_flags); 598 } 599 #endif /* DEBUG_EXEC */ 600 if (vcp->ev_flags & VMCMD_BASE) 601 base_vcp = vcp; 602 } 603 604 /* free the vmspace-creation commands, and release their references */ 605 kill_vmcmds(&pack.ep_vmcmds); 606 607 /* if an error happened, deallocate and punt */ 608 if (error) { 609 DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error)); 610 goto exec_abort; 611 } 612 613 /* remember information about the process */ 614 arginfo.ps_nargvstr = argc; 615 arginfo.ps_nenvstr = envc; 616 617 stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 618 sizeof(struct ps_strings) + szsigcode), 619 len - (sizeof(struct ps_strings) + szsigcode)); 620 #ifdef __MACHINE_STACK_GROWS_UP 621 /* 622 * The copyargs call always copies into lower addresses 623 * first, moving towards higher addresses, starting with 624 * the stack pointer that we give. When the stack grows 625 * down, this puts argc/argv/envp very shallow on the 626 * stack, right at the first user stack pointer, and puts 627 * STACKGAPLEN very deep in the stack. When the stack 628 * grows up, the situation is reversed. 629 * 630 * Normally, this is no big deal. But the ld_elf.so _rtld() 631 * function expects to be called with a single pointer to 632 * a region that has a few words it can stash values into, 633 * followed by argc/argv/envp. When the stack grows down, 634 * it's easy to decrement the stack pointer a little bit to 635 * allocate the space for these few words and pass the new 636 * stack pointer to _rtld. When the stack grows up, however, 637 * a few words before argc is part of the signal trampoline, XXX 638 * so we have a problem. 639 * 640 * Instead of changing how _rtld works, we take the easy way 641 * out and steal 32 bytes before we call copyargs. This 642 * space is effectively stolen from STACKGAPLEN. 643 */ 644 stack += 32; 645 #endif /* __MACHINE_STACK_GROWS_UP */ 646 647 /* Now copy argc, args & environ to new stack */ 648 error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp); 649 if (error) { 650 DPRINTF(("execve: copyargs failed %d\n", error)); 651 goto exec_abort; 652 } 653 /* Move the stack back to original point */ 654 stack = (char *)STACK_GROW(vm->vm_minsaddr, len); 655 656 /* fill process ps_strings info */ 657 p->p_psstr = (struct ps_strings *)STACK_ALLOC(vm->vm_minsaddr, 658 sizeof(struct ps_strings)); 659 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 660 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 661 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 662 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 663 664 /* copy out the process's ps_strings structure */ 665 if ((error = copyout(&arginfo, (char *)p->p_psstr, 666 sizeof(arginfo))) != 0) { 667 DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n", 668 &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo))); 669 goto exec_abort; 670 } 671 672 stopprofclock(p); /* stop profiling */ 673 fdcloseexec(p); /* handle close on exec */ 674 execsigs(p); /* reset catched signals */ 675 676 l->l_ctxlink = NULL; /* reset ucontext link */ 677 678 /* set command name & other accounting info */ 679 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 680 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 681 p->p_comm[len] = 0; 682 p->p_acflag &= ~AFORK; 683 684 /* record proc's vnode, for use by procfs and others */ 685 if (p->p_textvp) 686 vrele(p->p_textvp); 687 VREF(pack.ep_vp); 688 p->p_textvp = pack.ep_vp; 689 690 p->p_flag |= P_EXEC; 691 if (p->p_flag & P_PPWAIT) { 692 p->p_flag &= ~P_PPWAIT; 693 wakeup((caddr_t) p->p_pptr); 694 } 695 696 /* 697 * deal with set[ug]id. 698 * MNT_NOSUID has already been used to disable s[ug]id. 699 */ 700 if ((p->p_flag & P_TRACED) == 0 && 701 702 (((attr.va_mode & S_ISUID) != 0 && 703 p->p_ucred->cr_uid != attr.va_uid) || 704 705 ((attr.va_mode & S_ISGID) != 0 && 706 p->p_ucred->cr_gid != attr.va_gid))) { 707 /* 708 * Mark the process as SUGID before we do 709 * anything that might block. 710 */ 711 p_sugid(p); 712 713 /* Make sure file descriptors 0..2 are in use. */ 714 if ((error = fdcheckstd(p)) != 0) 715 goto exec_abort; 716 717 p->p_ucred = crcopy(cred); 718 #ifdef KTRACE 719 /* 720 * If process is being ktraced, turn off - unless 721 * root set it. 722 */ 723 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) 724 ktrderef(p); 725 #endif 726 if (attr.va_mode & S_ISUID) 727 p->p_ucred->cr_uid = attr.va_uid; 728 if (attr.va_mode & S_ISGID) 729 p->p_ucred->cr_gid = attr.va_gid; 730 } else 731 p->p_flag &= ~P_SUGID; 732 p->p_cred->p_svuid = p->p_ucred->cr_uid; 733 p->p_cred->p_svgid = p->p_ucred->cr_gid; 734 735 #if defined(__HAVE_RAS) 736 /* 737 * Remove all RASs from the address space. 738 */ 739 ras_purgeall(p); 740 #endif 741 742 doexechooks(p); 743 744 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 745 746 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 747 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 748 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 749 vput(pack.ep_vp); 750 751 /* notify others that we exec'd */ 752 KNOTE(&p->p_klist, NOTE_EXEC); 753 754 /* setup new registers and do misc. setup. */ 755 (*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack); 756 if (pack.ep_es->es_setregs) 757 (*pack.ep_es->es_setregs)(l, &pack, (u_long) stack); 758 759 /* map the process's signal trampoline code */ 760 if (exec_sigcode_map(p, pack.ep_es->es_emul)) 761 goto exec_abort; 762 763 if (p->p_flag & P_TRACED) 764 psignal(p, SIGTRAP); 765 766 free(pack.ep_hdr, M_EXEC); 767 768 /* 769 * Call emulation specific exec hook. This can setup setup per-process 770 * p->p_emuldata or do any other per-process stuff an emulation needs. 771 * 772 * If we are executing process of different emulation than the 773 * original forked process, call e_proc_exit() of the old emulation 774 * first, then e_proc_exec() of new emulation. If the emulation is 775 * same, the exec hook code should deallocate any old emulation 776 * resources held previously by this process. 777 */ 778 if (p->p_emul && p->p_emul->e_proc_exit 779 && p->p_emul != pack.ep_es->es_emul) 780 (*p->p_emul->e_proc_exit)(p); 781 782 /* 783 * Call exec hook. Emulation code may NOT store reference to anything 784 * from &pack. 785 */ 786 if (pack.ep_es->es_emul->e_proc_exec) 787 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack); 788 789 /* update p_emul, the old value is no longer needed */ 790 p->p_emul = pack.ep_es->es_emul; 791 792 /* ...and the same for p_execsw */ 793 p->p_execsw = pack.ep_es; 794 795 #ifdef __HAVE_SYSCALL_INTERN 796 (*p->p_emul->e_syscall_intern)(p); 797 #endif 798 #ifdef KTRACE 799 if (KTRPOINT(p, KTR_EMUL)) 800 ktremul(p); 801 #endif 802 803 #ifdef LKM 804 lockmgr(&exec_lock, LK_RELEASE, NULL); 805 #endif 806 p->p_flag &= ~P_INEXEC; 807 808 if (p->p_flag & P_STOPEXEC) { 809 int s; 810 811 sigminusset(&contsigmask, &p->p_sigctx.ps_siglist); 812 SCHED_LOCK(s); 813 p->p_stat = SSTOP; 814 l->l_stat = LSSTOP; 815 p->p_nrlwps--; 816 mi_switch(l, NULL); 817 SCHED_ASSERT_UNLOCKED(); 818 splx(s); 819 } 820 821 return (EJUSTRETURN); 822 823 bad: 824 p->p_flag &= ~P_INEXEC; 825 /* free the vmspace-creation commands, and release their references */ 826 kill_vmcmds(&pack.ep_vmcmds); 827 /* kill any opened file descriptor, if necessary */ 828 if (pack.ep_flags & EXEC_HASFD) { 829 pack.ep_flags &= ~EXEC_HASFD; 830 (void) fdrelease(p, pack.ep_fd); 831 } 832 /* close and put the exec'd file */ 833 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 834 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 835 vput(pack.ep_vp); 836 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 837 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 838 839 freehdr: 840 l->l_flag |= oldlwpflags; 841 p->p_flag &= ~P_INEXEC; 842 #ifdef LKM 843 lockmgr(&exec_lock, LK_RELEASE, NULL); 844 #endif 845 846 free(pack.ep_hdr, M_EXEC); 847 return error; 848 849 exec_abort: 850 p->p_flag &= ~P_INEXEC; 851 #ifdef LKM 852 lockmgr(&exec_lock, LK_RELEASE, NULL); 853 #endif 854 855 /* 856 * the old process doesn't exist anymore. exit gracefully. 857 * get rid of the (new) address space we have created, if any, get rid 858 * of our namei data and vnode, and exit noting failure 859 */ 860 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 861 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 862 if (pack.ep_emul_arg) 863 FREE(pack.ep_emul_arg, M_TEMP); 864 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 865 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 866 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 867 vput(pack.ep_vp); 868 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 869 free(pack.ep_hdr, M_EXEC); 870 exit1(l, W_EXITCODE(error, SIGABRT)); 871 872 /* NOTREACHED */ 873 return 0; 874 } 875 876 877 int 878 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo, 879 char **stackp, void *argp) 880 { 881 char **cpp, *dp, *sp; 882 size_t len; 883 void *nullp; 884 long argc, envc; 885 int error; 886 887 cpp = (char **)*stackp; 888 nullp = NULL; 889 argc = arginfo->ps_nargvstr; 890 envc = arginfo->ps_nenvstr; 891 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) 892 return error; 893 894 dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen); 895 sp = argp; 896 897 /* XXX don't copy them out, remap them! */ 898 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 899 900 for (; --argc >= 0; sp += len, dp += len) 901 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 902 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 903 return error; 904 905 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 906 return error; 907 908 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 909 910 for (; --envc >= 0; sp += len, dp += len) 911 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 912 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 913 return error; 914 915 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 916 return error; 917 918 *stackp = (char *)cpp; 919 return 0; 920 } 921 922 #ifdef LKM 923 /* 924 * Find an emulation of given name in list of emulations. 925 * Needs to be called with the exec_lock held. 926 */ 927 const struct emul * 928 emul_search(const char *name) 929 { 930 struct emul_entry *it; 931 932 LIST_FOREACH(it, &el_head, el_list) { 933 if (strcmp(name, it->el_emul->e_name) == 0) 934 return it->el_emul; 935 } 936 937 return NULL; 938 } 939 940 /* 941 * Add an emulation to list, if it's not there already. 942 */ 943 int 944 emul_register(const struct emul *emul, int ro_entry) 945 { 946 struct emul_entry *ee; 947 int error; 948 949 error = 0; 950 lockmgr(&exec_lock, LK_SHARED, NULL); 951 952 if (emul_search(emul->e_name)) { 953 error = EEXIST; 954 goto out; 955 } 956 957 MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry), 958 M_EXEC, M_WAITOK); 959 ee->el_emul = emul; 960 ee->ro_entry = ro_entry; 961 LIST_INSERT_HEAD(&el_head, ee, el_list); 962 963 out: 964 lockmgr(&exec_lock, LK_RELEASE, NULL); 965 return error; 966 } 967 968 /* 969 * Remove emulation with name 'name' from list of supported emulations. 970 */ 971 int 972 emul_unregister(const char *name) 973 { 974 const struct proclist_desc *pd; 975 struct emul_entry *it; 976 int i, error; 977 struct proc *ptmp; 978 979 error = 0; 980 lockmgr(&exec_lock, LK_SHARED, NULL); 981 982 LIST_FOREACH(it, &el_head, el_list) { 983 if (strcmp(it->el_emul->e_name, name) == 0) 984 break; 985 } 986 987 if (!it) { 988 error = ENOENT; 989 goto out; 990 } 991 992 if (it->ro_entry) { 993 error = EBUSY; 994 goto out; 995 } 996 997 /* test if any execw[] entry is still using this */ 998 for(i=0; i < nexecs; i++) { 999 if (execsw[i]->es_emul == it->el_emul) { 1000 error = EBUSY; 1001 goto out; 1002 } 1003 } 1004 1005 /* 1006 * Test if any process is running under this emulation - since 1007 * emul_unregister() is running quite sendomly, it's better 1008 * to do expensive check here than to use any locking. 1009 */ 1010 proclist_lock_read(); 1011 for (pd = proclists; pd->pd_list != NULL && !error; pd++) { 1012 LIST_FOREACH(ptmp, pd->pd_list, p_list) { 1013 if (ptmp->p_emul == it->el_emul) { 1014 error = EBUSY; 1015 break; 1016 } 1017 } 1018 } 1019 proclist_unlock_read(); 1020 1021 if (error) 1022 goto out; 1023 1024 1025 /* entry is not used, remove it */ 1026 LIST_REMOVE(it, el_list); 1027 FREE(it, M_EXEC); 1028 1029 out: 1030 lockmgr(&exec_lock, LK_RELEASE, NULL); 1031 return error; 1032 } 1033 1034 /* 1035 * Add execsw[] entry. 1036 */ 1037 int 1038 exec_add(struct execsw *esp, const char *e_name) 1039 { 1040 struct exec_entry *it; 1041 int error; 1042 1043 error = 0; 1044 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1045 1046 if (!esp->es_emul) { 1047 esp->es_emul = emul_search(e_name); 1048 if (!esp->es_emul) { 1049 error = ENOENT; 1050 goto out; 1051 } 1052 } 1053 1054 LIST_FOREACH(it, &ex_head, ex_list) { 1055 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1056 if (it->es->es_check == esp->es_check 1057 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1058 && it->es->es_emul == esp->es_emul) { 1059 error = EEXIST; 1060 goto out; 1061 } 1062 } 1063 1064 /* if we got here, the entry doesn't exist yet */ 1065 MALLOC(it, struct exec_entry *, sizeof(struct exec_entry), 1066 M_EXEC, M_WAITOK); 1067 it->es = esp; 1068 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1069 1070 /* update execsw[] */ 1071 exec_init(0); 1072 1073 out: 1074 lockmgr(&exec_lock, LK_RELEASE, NULL); 1075 return error; 1076 } 1077 1078 /* 1079 * Remove execsw[] entry. 1080 */ 1081 int 1082 exec_remove(const struct execsw *esp) 1083 { 1084 struct exec_entry *it; 1085 int error; 1086 1087 error = 0; 1088 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1089 1090 LIST_FOREACH(it, &ex_head, ex_list) { 1091 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1092 if (it->es->es_check == esp->es_check 1093 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1094 && it->es->es_emul == esp->es_emul) 1095 break; 1096 } 1097 if (!it) { 1098 error = ENOENT; 1099 goto out; 1100 } 1101 1102 /* remove item from list and free resources */ 1103 LIST_REMOVE(it, ex_list); 1104 FREE(it, M_EXEC); 1105 1106 /* update execsw[] */ 1107 exec_init(0); 1108 1109 out: 1110 lockmgr(&exec_lock, LK_RELEASE, NULL); 1111 return error; 1112 } 1113 1114 static void 1115 link_es(struct execsw_entry **listp, const struct execsw *esp) 1116 { 1117 struct execsw_entry *et, *e1; 1118 1119 MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry), 1120 M_TEMP, M_WAITOK); 1121 et->next = NULL; 1122 et->es = esp; 1123 if (*listp == NULL) { 1124 *listp = et; 1125 return; 1126 } 1127 1128 switch(et->es->es_prio) { 1129 case EXECSW_PRIO_FIRST: 1130 /* put new entry as the first */ 1131 et->next = *listp; 1132 *listp = et; 1133 break; 1134 case EXECSW_PRIO_ANY: 1135 /* put new entry after all *_FIRST and *_ANY entries */ 1136 for(e1 = *listp; e1->next 1137 && e1->next->es->es_prio != EXECSW_PRIO_LAST; 1138 e1 = e1->next); 1139 et->next = e1->next; 1140 e1->next = et; 1141 break; 1142 case EXECSW_PRIO_LAST: 1143 /* put new entry as the last one */ 1144 for(e1 = *listp; e1->next; e1 = e1->next); 1145 e1->next = et; 1146 break; 1147 default: 1148 #ifdef DIAGNOSTIC 1149 panic("execw[] entry with unknown priority %d found", 1150 et->es->es_prio); 1151 #endif 1152 break; 1153 } 1154 } 1155 1156 /* 1157 * Initialize exec structures. If init_boot is true, also does necessary 1158 * one-time initialization (it's called from main() that way). 1159 * Once system is multiuser, this should be called with exec_lock held, 1160 * i.e. via exec_{add|remove}(). 1161 */ 1162 int 1163 exec_init(int init_boot) 1164 { 1165 const struct execsw **new_es, * const *old_es; 1166 struct execsw_entry *list, *e1; 1167 struct exec_entry *e2; 1168 int i, es_sz; 1169 1170 if (init_boot) { 1171 /* do one-time initializations */ 1172 lockinit(&exec_lock, PWAIT, "execlck", 0, 0); 1173 1174 /* register compiled-in emulations */ 1175 for(i=0; i < nexecs_builtin; i++) { 1176 if (execsw_builtin[i].es_emul) 1177 emul_register(execsw_builtin[i].es_emul, 1); 1178 } 1179 #ifdef DIAGNOSTIC 1180 if (i == 0) 1181 panic("no emulations found in execsw_builtin[]"); 1182 #endif 1183 } 1184 1185 /* 1186 * Build execsw[] array from builtin entries and entries added 1187 * at runtime. 1188 */ 1189 list = NULL; 1190 for(i=0; i < nexecs_builtin; i++) 1191 link_es(&list, &execsw_builtin[i]); 1192 1193 /* Add dynamically loaded entries */ 1194 es_sz = nexecs_builtin; 1195 LIST_FOREACH(e2, &ex_head, ex_list) { 1196 link_es(&list, e2->es); 1197 es_sz++; 1198 } 1199 1200 /* 1201 * Now that we have sorted all execw entries, create new execsw[] 1202 * and free no longer needed memory in the process. 1203 */ 1204 new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK); 1205 for(i=0; list; i++) { 1206 new_es[i] = list->es; 1207 e1 = list->next; 1208 FREE(list, M_TEMP); 1209 list = e1; 1210 } 1211 1212 /* 1213 * New execsw[] array built, now replace old execsw[] and free 1214 * used memory. 1215 */ 1216 old_es = execsw; 1217 execsw = new_es; 1218 nexecs = es_sz; 1219 if (old_es) 1220 free((void *)old_es, M_EXEC); 1221 1222 /* 1223 * Figure out the maximum size of an exec header. 1224 */ 1225 exec_maxhdrsz = 0; 1226 for (i = 0; i < nexecs; i++) { 1227 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1228 exec_maxhdrsz = execsw[i]->es_hdrsz; 1229 } 1230 1231 return 0; 1232 } 1233 #endif 1234 1235 #ifndef LKM 1236 /* 1237 * Simplified exec_init() for kernels without LKMs. Only initialize 1238 * exec_maxhdrsz and execsw[]. 1239 */ 1240 int 1241 exec_init(int init_boot) 1242 { 1243 int i; 1244 1245 #ifdef DIAGNOSTIC 1246 if (!init_boot) 1247 panic("exec_init(): called with init_boot == 0"); 1248 #endif 1249 1250 /* do one-time initializations */ 1251 nexecs = nexecs_builtin; 1252 execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK); 1253 1254 /* 1255 * Fill in execsw[] and figure out the maximum size of an exec header. 1256 */ 1257 exec_maxhdrsz = 0; 1258 for(i=0; i < nexecs; i++) { 1259 execsw[i] = &execsw_builtin[i]; 1260 if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz) 1261 exec_maxhdrsz = execsw_builtin[i].es_hdrsz; 1262 } 1263 1264 return 0; 1265 1266 } 1267 #endif /* !LKM */ 1268 1269 static int 1270 exec_sigcode_map(struct proc *p, const struct emul *e) 1271 { 1272 vaddr_t va; 1273 vsize_t sz; 1274 int error; 1275 struct uvm_object *uobj; 1276 1277 if (e->e_sigobject == NULL) { 1278 return 0; 1279 } 1280 1281 /* 1282 * If we don't have a sigobject for this emulation, create one. 1283 * 1284 * sigobject is an anonymous memory object (just like SYSV shared 1285 * memory) that we keep a permanent reference to and that we map 1286 * in all processes that need this sigcode. The creation is simple, 1287 * we create an object, add a permanent reference to it, map it in 1288 * kernel space, copy out the sigcode to it and unmap it. 1289 * The we map it with PROT_READ|PROT_EXEC into the process just 1290 * the way sys_mmap would map it. 1291 */ 1292 1293 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1294 uobj = *e->e_sigobject; 1295 if (uobj == NULL) { 1296 uobj = uao_create(sz, 0); 1297 uao_reference(uobj); 1298 va = vm_map_min(kernel_map); 1299 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1300 uobj, 0, 0, 1301 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1302 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 1303 printf("kernel mapping failed %d\n", error); 1304 (*uobj->pgops->pgo_detach)(uobj); 1305 return (error); 1306 } 1307 memcpy((void *)va, e->e_sigcode, sz); 1308 #ifdef PMAP_NEED_PROCWR 1309 pmap_procwr(&proc0, va, sz); 1310 #endif 1311 uvm_unmap(kernel_map, va, va + round_page(sz)); 1312 *e->e_sigobject = uobj; 1313 } 1314 1315 /* Just a hint to uvm_map where to put it. */ 1316 va = VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, round_page(sz)); 1317 (*uobj->pgops->pgo_reference)(uobj); 1318 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 1319 uobj, 0, 0, 1320 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 1321 UVM_ADV_RANDOM, 0)); 1322 if (error) { 1323 (*uobj->pgops->pgo_detach)(uobj); 1324 return (error); 1325 } 1326 p->p_sigctx.ps_sigcode = (void *)va; 1327 return (0); 1328 } 1329