sys/kern/kern_exec.c

   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
  27 #include <sys/cdefs.h>
  28 __FBSDID("$FreeBSD$");
  29
  30 #include "opt_hwpmc_hooks.h"
  31 #include "opt_kdtrace.h"
  32 #include "opt_ktrace.h"
  33 #include "opt_mac.h"
  34 #include "opt_vm.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/eventhandler.h>
  39 #include <sys/lock.h>
  40 #include <sys/mutex.h>
  41 #include <sys/sysproto.h>
  42 #include <sys/signalvar.h>
  43 #include <sys/kernel.h>
  44 #include <sys/mount.h>
  45 #include <sys/filedesc.h>
  46 #include <sys/fcntl.h>
  47 #include <sys/acct.h>
  48 #include <sys/exec.h>
  49 #include <sys/imgact.h>
  50 #include <sys/imgact_elf.h>
  51 #include <sys/wait.h>
  52 #include <sys/malloc.h>
  53 #include <sys/priv.h>
  54 #include <sys/proc.h>
  55 #include <sys/pioctl.h>
  56 #include <sys/namei.h>
  57 #include <sys/resourcevar.h>
  58 #include <sys/sdt.h>
  59 #include <sys/sf_buf.h>
  60 #include <sys/syscallsubr.h>
  61 #include <sys/sysent.h>
  62 #include <sys/shm.h>
  63 #include <sys/sysctl.h>
  64 #include <sys/vnode.h>
  65 #include <sys/stat.h>
  66 #ifdef KTRACE
  67 #include <sys/ktrace.h>
  68 #endif
  69
  70 #include <vm/vm.h>
  71 #include <vm/vm_param.h>
  72 #include <vm/pmap.h>
  73 #include <vm/vm_page.h>
  74 #include <vm/vm_map.h>
  75 #include <vm/vm_kern.h>
  76 #include <vm/vm_extern.h>
  77 #include <vm/vm_object.h>
  78 #include <vm/vm_pager.h>
  79
  80 #ifdef  HWPMC_HOOKS
  81 #include <sys/pmckern.h>
  82 #endif
  83
  84 #include <machine/reg.h>
  85
  86 #include <security/audit/audit.h>
  87 #include <security/mac/mac_framework.h>
  88
  89 #ifdef KDTRACE_HOOKS
  90 #include <sys/dtrace_bsd.h>
  91 dtrace_execexit_func_t  dtrace_fasttrap_exec;
  92 #endif
  93
  94 SDT_PROVIDER_DECLARE(proc);
  95 SDT_PROBE_DEFINE(proc, kernel, , exec);
  96 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
  97 SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
  98 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
  99 SDT_PROBE_DEFINE(proc, kernel, , exec_success);
 100 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
 101
 102 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
 103
 104 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
 105 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
 106 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
 107 static int do_execve(struct thread *td, struct image_args *args,
 108     struct mac *mac_p);
 109 static void exec_free_args(struct image_args *);
 110
 111 /* XXX This should be vm_size_t. */
 112 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
 113     NULL, 0, sysctl_kern_ps_strings, "LU", "");
 114
 115 /* XXX This should be vm_size_t. */
 116 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
 117     NULL, 0, sysctl_kern_usrstack, "LU", "");
 118
 119 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
 120     NULL, 0, sysctl_kern_stackprot, "I", "");
 121
 122 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
 123 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
 124     &ps_arg_cache_limit, 0, "");
 125
 126 static int
 127 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
 128 {
 129         struct proc *p;
 130         int error;
 131
 132         p = curproc;
 133 #ifdef SCTL_MASK32
 134         if (req->flags & SCTL_MASK32) {
 135                 unsigned int val;
 136                 val = (unsigned int)p->p_sysent->sv_psstrings;
 137                 error = SYSCTL_OUT(req, &val, sizeof(val));
 138         } else
 139 #endif
 140                 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
 141                    sizeof(p->p_sysent->sv_psstrings));
 142         return error;
 143 }
 144
 145 static int
 146 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
 147 {
 148         struct proc *p;
 149         int error;
 150
 151         p = curproc;
 152 #ifdef SCTL_MASK32
 153         if (req->flags & SCTL_MASK32) {
 154                 unsigned int val;
 155                 val = (unsigned int)p->p_sysent->sv_usrstack;
 156                 error = SYSCTL_OUT(req, &val, sizeof(val));
 157         } else
 158 #endif
 159                 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
 160                     sizeof(p->p_sysent->sv_usrstack));
 161         return error;
 162 }
 163
 164 static int
 165 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
 166 {
 167         struct proc *p;
 168
 169         p = curproc;
 170         return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
 171             sizeof(p->p_sysent->sv_stackprot)));
 172 }
 173
 174 /*
 175  * Each of the items is a pointer to a `const struct execsw', hence the
 176  * double pointer here.
 177  */
 178 static const struct execsw **execsw;
 179
 180 #ifndef _SYS_SYSPROTO_H_
 181 struct execve_args {
 182         char    *fname;
 183         char    **argv;
 184         char    **envv;
 185 };
 186 #endif
 187
 188 int
 189 execve(td, uap)
 190         struct thread *td;
 191         struct execve_args /* {
 192                 char *fname;
 193                 char **argv;
 194                 char **envv;
 195         } */ *uap;
 196 {
 197         int error;
 198         struct image_args args;
 199
 200         error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
 201             uap->argv, uap->envv);
 202         if (error == 0)
 203                 error = kern_execve(td, &args, NULL);
 204         return (error);
 205 }
 206
 207 #ifndef _SYS_SYSPROTO_H_
 208 struct fexecve_args {
 209         int     fd;
 210         char    **argv;
 211         char    **envv;
 212 }
 213 #endif
 214 int
 215 fexecve(struct thread *td, struct fexecve_args *uap)
 216 {
 217         int error;
 218         struct image_args args;
 219
 220         error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
 221             uap->argv, uap->envv);
 222         if (error == 0) {
 223                 args.fd = uap->fd;
 224                 error = kern_execve(td, &args, NULL);
 225         }
 226         return (error);
 227 }
 228
 229 #ifndef _SYS_SYSPROTO_H_
 230 struct __mac_execve_args {
 231         char    *fname;
 232         char    **argv;
 233         char    **envv;
 234         struct mac      *mac_p;
 235 };
 236 #endif
 237
 238 int
 239 __mac_execve(td, uap)
 240         struct thread *td;
 241         struct __mac_execve_args /* {
 242                 char *fname;
 243                 char **argv;
 244                 char **envv;
 245                 struct mac *mac_p;
 246         } */ *uap;
 247 {
 248 #ifdef MAC
 249         int error;
 250         struct image_args args;
 251
 252         error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
 253             uap->argv, uap->envv);
 254         if (error == 0)
 255                 error = kern_execve(td, &args, uap->mac_p);
 256         return (error);
 257 #else
 258         return (ENOSYS);
 259 #endif
 260 }
 261
 262 /*
 263  * XXX: kern_execve has the astonishing property of not always returning to
 264  * the caller.  If sufficiently bad things happen during the call to
 265  * do_execve(), it can end up calling exit1(); as a result, callers must
 266  * avoid doing anything which they might need to undo (e.g., allocating
 267  * memory).
 268  */
 269 int
 270 kern_execve(td, args, mac_p)
 271         struct thread *td;
 272         struct image_args *args;
 273         struct mac *mac_p;
 274 {
 275         struct proc *p = td->td_proc;
 276         int error;
 277
 278         AUDIT_ARG(argv, args->begin_argv, args->argc,
 279             args->begin_envv - args->begin_argv);
 280         AUDIT_ARG(envv, args->begin_envv, args->envc,
 281             args->endp - args->begin_envv);
 282         if (p->p_flag & P_HADTHREADS) {
 283                 PROC_LOCK(p);
 284                 if (thread_single(SINGLE_BOUNDARY)) {
 285                         PROC_UNLOCK(p);
 286                         exec_free_args(args);
 287                         return (ERESTART);      /* Try again later. */
 288                 }
 289                 PROC_UNLOCK(p);
 290         }
 291
 292         error = do_execve(td, args, mac_p);
 293
 294         if (p->p_flag & P_HADTHREADS) {
 295                 PROC_LOCK(p);
 296                 /*
 297                  * If success, we upgrade to SINGLE_EXIT state to
 298                  * force other threads to suicide.
 299                  */
 300                 if (error == 0)
 301                         thread_single(SINGLE_EXIT);
 302                 else
 303                         thread_single_end();
 304                 PROC_UNLOCK(p);
 305         }
 306
 307         return (error);
 308 }
 309
 310 /*
 311  * In-kernel implementation of execve().  All arguments are assumed to be
 312  * userspace pointers from the passed thread.
 313  */
 314 static int
 315 do_execve(td, args, mac_p)
 316         struct thread *td;
 317         struct image_args *args;
 318         struct mac *mac_p;
 319 {
 320         struct proc *p = td->td_proc;
 321         struct nameidata nd, *ndp;
 322         struct ucred *newcred = NULL, *oldcred;
 323         struct uidinfo *euip;
 324         register_t *stack_base;
 325         int error, len = 0, i;
 326         struct image_params image_params, *imgp;
 327         struct vattr attr;
 328         int (*img_first)(struct image_params *);
 329         struct pargs *oldargs = NULL, *newargs = NULL;
 330         struct sigacts *oldsigacts, *newsigacts;
 331 #ifdef KTRACE
 332         struct vnode *tracevp = NULL;
 333         struct ucred *tracecred = NULL;
 334 #endif
 335         struct vnode *textvp = NULL, *binvp = NULL;
 336         int credential_changing;
 337         int vfslocked;
 338         int textset;
 339 #ifdef MAC
 340         struct label *interpvplabel = NULL;
 341         int will_transition;
 342 #endif
 343 #ifdef HWPMC_HOOKS
 344         struct pmckern_procexec pe;
 345 #endif
 346         static const char fexecv_proc_title[] = "(fexecv)";
 347
 348         vfslocked = 0;
 349         imgp = &image_params;
 350
 351         /*
 352          * Lock the process and set the P_INEXEC flag to indicate that
 353          * it should be left alone until we're done here.  This is
 354          * necessary to avoid race conditions - e.g. in ptrace() -
 355          * that might allow a local user to illicitly obtain elevated
 356          * privileges.
 357          */
 358         PROC_LOCK(p);
 359         KASSERT((p->p_flag & P_INEXEC) == 0,
 360             ("%s(): process already has P_INEXEC flag", __func__));
 361         p->p_flag |= P_INEXEC;
 362         PROC_UNLOCK(p);
 363
 364         /*
 365          * Initialize part of the common data
 366          */
 367         imgp->proc = p;
 368         imgp->execlabel = NULL;
 369         imgp->attr = &attr;
 370         imgp->entry_addr = 0;
 371         imgp->vmspace_destroyed = 0;
 372         imgp->interpreted = 0;
 373         imgp->opened = 0;
 374         imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
 375         imgp->auxargs = NULL;
 376         imgp->vp = NULL;
 377         imgp->object = NULL;
 378         imgp->firstpage = NULL;
 379         imgp->ps_strings = 0;
 380         imgp->auxarg_size = 0;
 381         imgp->args = args;
 382
 383 #ifdef MAC
 384         error = mac_execve_enter(imgp, mac_p);
 385         if (error)
 386                 goto exec_fail;
 387 #endif
 388
 389         imgp->image_header = NULL;
 390
 391         SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
 392
 393         /*
 394          * Translate the file name. namei() returns a vnode pointer
 395          *      in ni_vp amoung other things.
 396          *
 397          * XXXAUDIT: It would be desirable to also audit the name of the
 398          * interpreter if this is an interpreted binary.
 399          */
 400         if (args->fname != NULL) {
 401                 ndp = &nd;
 402                 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
 403                     | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
 404         }
 405
 406 interpret:
 407         if (args->fname != NULL) {
 408                 error = namei(ndp);
 409                 if (error)
 410                         goto exec_fail;
 411
 412                 vfslocked = NDHASGIANT(ndp);
 413                 binvp  = ndp->ni_vp;
 414                 imgp->vp = binvp;
 415         } else {
 416                 AUDIT_ARG(fd, args->fd);
 417                 error = fgetvp(td, args->fd, &binvp);
 418                 if (error)
 419                         goto exec_fail;
 420                 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
 421                 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
 422                 AUDIT_ARG(vnode, binvp, ARG_VNODE1);
 423                 imgp->vp = binvp;
 424         }
 425
 426         /*
 427          * Check file permissions (also 'opens' file)
 428          */
 429         error = exec_check_permissions(imgp);
 430         if (error)
 431                 goto exec_fail_dealloc;
 432
 433         imgp->object = imgp->vp->v_object;
 434         if (imgp->object != NULL)
 435                 vm_object_reference(imgp->object);
 436
 437         /*
 438          * Set VV_TEXT now so no one can write to the executable while we're
 439          * activating it.
 440          *
 441          * Remember if this was set before and unset it in case this is not
 442          * actually an executable image.
 443          */
 444         textset = imgp->vp->v_vflag & VV_TEXT;
 445         imgp->vp->v_vflag |= VV_TEXT;
 446
 447         error = exec_map_first_page(imgp);
 448         if (error)
 449                 goto exec_fail_dealloc;
 450
 451         imgp->proc->p_osrel = 0;
 452         /*
 453          *      If the current process has a special image activator it
 454          *      wants to try first, call it.   For example, emulating shell
 455          *      scripts differently.
 456          */
 457         error = -1;
 458         if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
 459                 error = img_first(imgp);
 460
 461         /*
 462          *      Loop through the list of image activators, calling each one.
 463          *      An activator returns -1 if there is no match, 0 on success,
 464          *      and an error otherwise.
 465          */
 466         for (i = 0; error == -1 && execsw[i]; ++i) {
 467                 if (execsw[i]->ex_imgact == NULL ||
 468                     execsw[i]->ex_imgact == img_first) {
 469                         continue;
 470                 }
 471                 error = (*execsw[i]->ex_imgact)(imgp);
 472         }
 473
 474         if (error) {
 475                 if (error == -1) {
 476                         if (textset == 0)
 477                                 imgp->vp->v_vflag &= ~VV_TEXT;
 478                         error = ENOEXEC;
 479                 }
 480                 goto exec_fail_dealloc;
 481         }
 482
 483         /*
 484          * Special interpreter operation, cleanup and loop up to try to
 485          * activate the interpreter.
 486          */
 487         if (imgp->interpreted) {
 488                 exec_unmap_first_page(imgp);
 489                 /*
 490                  * VV_TEXT needs to be unset for scripts.  There is a short
 491                  * period before we determine that something is a script where
 492                  * VV_TEXT will be set. The vnode lock is held over this
 493                  * entire period so nothing should illegitimately be blocked.
 494                  */
 495                 imgp->vp->v_vflag &= ~VV_TEXT;
 496                 /* free name buffer and old vnode */
 497                 if (args->fname != NULL)
 498                         NDFREE(ndp, NDF_ONLY_PNBUF);
 499 #ifdef MAC
 500                 mac_execve_interpreter_enter(binvp, &interpvplabel);
 501 #endif
 502                 if (imgp->opened) {
 503                         VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
 504                         imgp->opened = 0;
 505                 }
 506                 vput(binvp);
 507                 vm_object_deallocate(imgp->object);
 508                 imgp->object = NULL;
 509                 VFS_UNLOCK_GIANT(vfslocked);
 510                 vfslocked = 0;
 511                 /* set new name to that of the interpreter */
 512                 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
 513                     UIO_SYSSPACE, imgp->interpreter_name, td);
 514                 args->fname = imgp->interpreter_name;
 515                 goto interpret;
 516         }
 517
 518         /*
 519          * NB: We unlock the vnode here because it is believed that none
 520          * of the sv_copyout_strings/sv_fixup operations require the vnode.
 521          */
 522         VOP_UNLOCK(imgp->vp, 0);
 523         /*
 524          * Copy out strings (args and env) and initialize stack base
 525          */
 526         if (p->p_sysent->sv_copyout_strings)
 527                 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
 528         else
 529                 stack_base = exec_copyout_strings(imgp);
 530
 531         /*
 532          * If custom stack fixup routine present for this process
 533          * let it do the stack setup.
 534          * Else stuff argument count as first item on stack
 535          */
 536         if (p->p_sysent->sv_fixup != NULL)
 537                 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
 538         else
 539                 suword(--stack_base, imgp->args->argc);
 540
 541         /*
 542          * For security and other reasons, the file descriptor table cannot
 543          * be shared after an exec.
 544          */
 545         fdunshare(p, td);
 546
 547         /*
 548          * Malloc things before we need locks.
 549          */
 550         newcred = crget();
 551         euip = uifind(attr.va_uid);
 552         i = imgp->args->begin_envv - imgp->args->begin_argv;
 553         /* Cache arguments if they fit inside our allowance */
 554         if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
 555                 newargs = pargs_alloc(i);
 556                 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
 557         }
 558
 559         /* close files on exec */
 560         fdcloseexec(td);
 561         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 562
 563         /* Get a reference to the vnode prior to locking the proc */
 564         VREF(binvp);
 565
 566         /*
 567          * For security and other reasons, signal handlers cannot
 568          * be shared after an exec. The new process gets a copy of the old
 569          * handlers. In execsigs(), the new process will have its signals
 570          * reset.
 571          */
 572         PROC_LOCK(p);
 573         if (sigacts_shared(p->p_sigacts)) {
 574                 oldsigacts = p->p_sigacts;
 575                 PROC_UNLOCK(p);
 576                 newsigacts = sigacts_alloc();
 577                 sigacts_copy(newsigacts, oldsigacts);
 578                 PROC_LOCK(p);
 579                 p->p_sigacts = newsigacts;
 580         } else
 581                 oldsigacts = NULL;
 582
 583         /* Stop profiling */
 584         stopprofclock(p);
 585
 586         /* reset caught signals */
 587         execsigs(p);
 588
 589         /* name this process - nameiexec(p, ndp) */
 590         if (args->fname) {
 591                 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
 592                 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
 593         } else {
 594                 len = MAXCOMLEN;
 595                 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0)
 596                         len = MAXCOMLEN;
 597                 else {
 598                         len = sizeof(fexecv_proc_title);
 599                         bcopy(fexecv_proc_title, p->p_comm, len);
 600                 }
 601         }
 602         p->p_comm[len] = 0;
 603         bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
 604
 605         /*
 606          * mark as execed, wakeup the process that vforked (if any) and tell
 607          * it that it now has its own resources back
 608          */
 609         p->p_flag |= P_EXEC;
 610         if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
 611                 p->p_flag &= ~P_PPWAIT;
 612                 wakeup(p->p_pptr);
 613         }
 614
 615         /*
 616          * Implement image setuid/setgid.
 617          *
 618          * Don't honor setuid/setgid if the filesystem prohibits it or if
 619          * the process is being traced.
 620          *
 621          * XXXMAC: For the time being, use NOSUID to also prohibit
 622          * transitions on the file system.
 623          */
 624         oldcred = p->p_ucred;
 625         credential_changing = 0;
 626         credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
 627             attr.va_uid;
 628         credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
 629             attr.va_gid;
 630 #ifdef MAC
 631         will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
 632             interpvplabel, imgp);
 633         credential_changing |= will_transition;
 634 #endif
 635
 636         if (credential_changing &&
 637             (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
 638             (p->p_flag & P_TRACED) == 0) {
 639                 /*
 640                  * Turn off syscall tracing for set-id programs, except for
 641                  * root.  Record any set-id flags first to make sure that
 642                  * we do not regain any tracing during a possible block.
 643                  */
 644                 setsugid(p);
 645
 646 #ifdef KTRACE
 647                 if (p->p_tracevp != NULL &&
 648                     priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
 649                         mtx_lock(&ktrace_mtx);
 650                         p->p_traceflag = 0;
 651                         tracevp = p->p_tracevp;
 652                         p->p_tracevp = NULL;
 653                         tracecred = p->p_tracecred;
 654                         p->p_tracecred = NULL;
 655                         mtx_unlock(&ktrace_mtx);
 656                 }
 657 #endif
 658                 /*
 659                  * Close any file descriptors 0..2 that reference procfs,
 660                  * then make sure file descriptors 0..2 are in use.
 661                  *
 662                  * setugidsafety() may call closef() and then pfind()
 663                  * which may grab the process lock.
 664                  * fdcheckstd() may call falloc() which may block to
 665                  * allocate memory, so temporarily drop the process lock.
 666                  */
 667                 PROC_UNLOCK(p);
 668                 setugidsafety(td);
 669                 VOP_UNLOCK(imgp->vp, 0);
 670                 error = fdcheckstd(td);
 671                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 672                 if (error != 0)
 673                         goto done1;
 674                 PROC_LOCK(p);
 675                 /*
 676                  * Set the new credentials.
 677                  */
 678                 crcopy(newcred, oldcred);
 679                 if (attr.va_mode & S_ISUID)
 680                         change_euid(newcred, euip);
 681                 if (attr.va_mode & S_ISGID)
 682                         change_egid(newcred, attr.va_gid);
 683 #ifdef MAC
 684                 if (will_transition) {
 685                         mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
 686                             interpvplabel, imgp);
 687                 }
 688 #endif
 689                 /*
 690                  * Implement correct POSIX saved-id behavior.
 691                  *
 692                  * XXXMAC: Note that the current logic will save the
 693                  * uid and gid if a MAC domain transition occurs, even
 694                  * though maybe it shouldn't.
 695                  */
 696                 change_svuid(newcred, newcred->cr_uid);
 697                 change_svgid(newcred, newcred->cr_gid);
 698                 p->p_ucred = newcred;
 699                 newcred = NULL;
 700         } else {
 701                 if (oldcred->cr_uid == oldcred->cr_ruid &&
 702                     oldcred->cr_gid == oldcred->cr_rgid)
 703                         p->p_flag &= ~P_SUGID;
 704                 /*
 705                  * Implement correct POSIX saved-id behavior.
 706                  *
 707                  * XXX: It's not clear that the existing behavior is
 708                  * POSIX-compliant.  A number of sources indicate that the
 709                  * saved uid/gid should only be updated if the new ruid is
 710                  * not equal to the old ruid, or the new euid is not equal
 711                  * to the old euid and the new euid is not equal to the old
 712                  * ruid.  The FreeBSD code always updates the saved uid/gid.
 713                  * Also, this code uses the new (replaced) euid and egid as
 714                  * the source, which may or may not be the right ones to use.
 715                  */
 716                 if (oldcred->cr_svuid != oldcred->cr_uid ||
 717                     oldcred->cr_svgid != oldcred->cr_gid) {
 718                         crcopy(newcred, oldcred);
 719                         change_svuid(newcred, newcred->cr_uid);
 720                         change_svgid(newcred, newcred->cr_gid);
 721                         p->p_ucred = newcred;
 722                         newcred = NULL;
 723                 }
 724         }
 725
 726         /*
 727          * Store the vp for use in procfs.  This vnode was referenced prior
 728          * to locking the proc lock.
 729          */
 730         textvp = p->p_textvp;
 731         p->p_textvp = binvp;
 732
 733 #ifdef KDTRACE_HOOKS
 734         /*
 735          * Tell the DTrace fasttrap provider about the exec if it
 736          * has declared an interest.
 737          */
 738         if (dtrace_fasttrap_exec)
 739                 dtrace_fasttrap_exec(p);
 740 #endif
 741
 742         /*
 743          * Notify others that we exec'd, and clear the P_INEXEC flag
 744          * as we're now a bona fide freshly-execed process.
 745          */
 746         KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
 747         p->p_flag &= ~P_INEXEC;
 748
 749         /*
 750          * If tracing the process, trap to debugger so breakpoints
 751          * can be set before the program executes.
 752          * Use tdsignal to deliver signal to current thread, use
 753          * psignal may cause the signal to be delivered to wrong thread
 754          * because that thread will exit, remember we are going to enter
 755          * single thread mode.
 756          */
 757         if (p->p_flag & P_TRACED)
 758                 tdsignal(p, td, SIGTRAP, NULL);
 759
 760         /* clear "fork but no exec" flag, as we _are_ execing */
 761         p->p_acflag &= ~AFORK;
 762
 763         /*
 764          * Free any previous argument cache and replace it with
 765          * the new argument cache, if any.
 766          */
 767         oldargs = p->p_args;
 768         p->p_args = newargs;
 769         newargs = NULL;
 770
 771 #ifdef  HWPMC_HOOKS
 772         /*
 773          * Check if system-wide sampling is in effect or if the
 774          * current process is using PMCs.  If so, do exec() time
 775          * processing.  This processing needs to happen AFTER the
 776          * P_INEXEC flag is cleared.
 777          *
 778          * The proc lock needs to be released before taking the PMC
 779          * SX.
 780          */
 781         if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
 782                 PROC_UNLOCK(p);
 783                 pe.pm_credentialschanged = credential_changing;
 784                 pe.pm_entryaddr = imgp->entry_addr;
 785
 786                 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
 787         } else
 788                 PROC_UNLOCK(p);
 789 #else  /* !HWPMC_HOOKS */
 790         PROC_UNLOCK(p);
 791 #endif
 792
 793         /* Set values passed into the program in registers. */
 794         if (p->p_sysent->sv_setregs)
 795                 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
 796                     (u_long)(uintptr_t)stack_base, imgp->ps_strings);
 797         else
 798                 exec_setregs(td, imgp->entry_addr,
 799                     (u_long)(uintptr_t)stack_base, imgp->ps_strings);
 800
 801         vfs_mark_atime(imgp->vp, td->td_ucred);
 802
 803 done1:
 804
 805         /*
 806          * Free any resources malloc'd earlier that we didn't use.
 807          */
 808         uifree(euip);
 809         if (newcred == NULL)
 810                 crfree(oldcred);
 811         else
 812                 crfree(newcred);
 813         VOP_UNLOCK(imgp->vp, 0);
 814
 815         SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
 816
 817         /*
 818          * Handle deferred decrement of ref counts.
 819          */
 820         if (textvp != NULL) {
 821                 int tvfslocked;
 822
 823                 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
 824                 vrele(textvp);
 825                 VFS_UNLOCK_GIANT(tvfslocked);
 826         }
 827         if (binvp && error != 0)
 828                 vrele(binvp);
 829 #ifdef KTRACE
 830         if (tracevp != NULL) {
 831                 int tvfslocked;
 832
 833                 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
 834                 vrele(tracevp);
 835                 VFS_UNLOCK_GIANT(tvfslocked);
 836         }
 837         if (tracecred != NULL)
 838                 crfree(tracecred);
 839 #endif
 840         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 841         pargs_drop(oldargs);
 842         pargs_drop(newargs);
 843         if (oldsigacts != NULL)
 844                 sigacts_free(oldsigacts);
 845
 846 exec_fail_dealloc:
 847
 848         /*
 849          * free various allocated resources
 850          */
 851         if (imgp->firstpage != NULL)
 852                 exec_unmap_first_page(imgp);
 853
 854         if (imgp->vp != NULL) {
 855                 if (args->fname)
 856                         NDFREE(ndp, NDF_ONLY_PNBUF);
 857                 if (imgp->opened)
 858                         VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
 859                 vput(imgp->vp);
 860         }
 861
 862         if (imgp->object != NULL)
 863                 vm_object_deallocate(imgp->object);
 864
 865         if (error == 0) {
 866                 /*
 867                  * Stop the process here if its stop event mask has
 868                  * the S_EXEC bit set.
 869                  */
 870                 STOPEVENT(p, S_EXEC, 0);
 871                 goto done2;
 872         }
 873
 874 exec_fail:
 875         /* we're done here, clear P_INEXEC */
 876         PROC_LOCK(p);
 877         p->p_flag &= ~P_INEXEC;
 878         PROC_UNLOCK(p);
 879
 880         SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
 881
 882 done2:
 883 #ifdef MAC
 884         mac_execve_exit(imgp);
 885         mac_execve_interpreter_exit(interpvplabel);
 886 #endif
 887         VFS_UNLOCK_GIANT(vfslocked);
 888         exec_free_args(args);
 889
 890         if (error && imgp->vmspace_destroyed) {
 891                 /* sorry, no more process anymore. exit gracefully */
 892                 exit1(td, W_EXITCODE(0, SIGABRT));
 893                 /* NOT REACHED */
 894         }
 895         return (error);
 896 }
 897
 898 int
 899 exec_map_first_page(imgp)
 900         struct image_params *imgp;
 901 {
 902         int rv, i;
 903         int initial_pagein;
 904         vm_page_t ma[VM_INITIAL_PAGEIN];
 905         vm_object_t object;
 906
 907         if (imgp->firstpage != NULL)
 908                 exec_unmap_first_page(imgp);
 909
 910         object = imgp->vp->v_object;
 911         if (object == NULL)
 912                 return (EACCES);
 913         VM_OBJECT_LOCK(object);
 914 #if VM_NRESERVLEVEL > 0
 915         if ((object->flags & OBJ_COLORED) == 0) {
 916                 object->flags |= OBJ_COLORED;
 917                 object->pg_color = 0;
 918         }
 919 #endif
 920         ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
 921         if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
 922                 initial_pagein = VM_INITIAL_PAGEIN;
 923                 if (initial_pagein > object->size)
 924                         initial_pagein = object->size;
 925                 for (i = 1; i < initial_pagein; i++) {
 926                         if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
 927                                 if (ma[i]->valid)
 928                                         break;
 929                                 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
 930                                         break;
 931                                 vm_page_busy(ma[i]);
 932                         } else {
 933                                 ma[i] = vm_page_alloc(object, i,
 934                                     VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
 935                                 if (ma[i] == NULL)
 936                                         break;
 937                         }
 938                 }
 939                 initial_pagein = i;
 940                 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
 941                 ma[0] = vm_page_lookup(object, 0);
 942                 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
 943                     (ma[0]->valid == 0)) {
 944                         if (ma[0]) {
 945                                 vm_page_lock_queues();
 946                                 vm_page_free(ma[0]);
 947                                 vm_page_unlock_queues();
 948                         }
 949                         VM_OBJECT_UNLOCK(object);
 950                         return (EIO);
 951                 }
 952         }
 953         vm_page_lock_queues();
 954         vm_page_hold(ma[0]);
 955         vm_page_unlock_queues();
 956         vm_page_wakeup(ma[0]);
 957         VM_OBJECT_UNLOCK(object);
 958
 959         imgp->firstpage = sf_buf_alloc(ma[0], 0);
 960         imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
 961
 962         return (0);
 963 }
 964
 965 void
 966 exec_unmap_first_page(imgp)
 967         struct image_params *imgp;
 968 {
 969         vm_page_t m;
 970
 971         if (imgp->firstpage != NULL) {
 972                 m = sf_buf_page(imgp->firstpage);
 973                 sf_buf_free(imgp->firstpage);
 974                 imgp->firstpage = NULL;
 975                 vm_page_lock_queues();
 976                 vm_page_unhold(m);
 977                 vm_page_unlock_queues();
 978         }
 979 }
 980
 981 /*
 982  * Destroy old address space, and allocate a new stack
 983  *      The new stack is only SGROWSIZ large because it is grown
 984  *      automatically in trap.c.
 985  */
 986 int
 987 exec_new_vmspace(imgp, sv)
 988         struct image_params *imgp;
 989         struct sysentvec *sv;
 990 {
 991         int error;
 992         struct proc *p = imgp->proc;
 993         struct vmspace *vmspace = p->p_vmspace;
 994         vm_offset_t stack_addr;
 995         vm_map_t map;
 996         u_long ssiz;
 997
 998         imgp->vmspace_destroyed = 1;
 999         imgp->sysent = sv;
1000
1001         /* May be called with Giant held */
1002         EVENTHANDLER_INVOKE(process_exec, p, imgp);
1003
1004         /*
1005          * Blow away entire process VM, if address space not shared,
1006          * otherwise, create a new VM space so that other threads are
1007          * not disrupted
1008          */
1009         map = &vmspace->vm_map;
1010         if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
1011             vm_map_max(map) == sv->sv_maxuser) {
1012                 shmexit(vmspace);
1013                 pmap_remove_pages(vmspace_pmap(vmspace));
1014                 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1015         } else {
1016                 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
1017                 if (error)
1018                         return (error);
1019                 vmspace = p->p_vmspace;
1020                 map = &vmspace->vm_map;
1021         }
1022
1023         /* Allocate a new stack */
1024         if (sv->sv_maxssiz != NULL)
1025                 ssiz = *sv->sv_maxssiz;
1026         else
1027                 ssiz = maxssiz;
1028         stack_addr = sv->sv_usrstack - ssiz;
1029         error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1030             sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1031         if (error)
1032                 return (error);
1033
1034 #ifdef __ia64__
1035         /* Allocate a new register stack */
1036         stack_addr = IA64_BACKINGSTORE;
1037         error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1038             sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1039         if (error)
1040                 return (error);
1041 #endif
1042
1043         /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1044          * VM_STACK case, but they are still used to monitor the size of the
1045          * process stack so we can check the stack rlimit.
1046          */
1047         vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1048         vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1049
1050         return (0);
1051 }
1052
1053 /*
1054  * Copy out argument and environment strings from the old process address
1055  * space into the temporary string buffer.
1056  */
1057 int
1058 exec_copyin_args(struct image_args *args, char *fname,
1059     enum uio_seg segflg, char **argv, char **envv)
1060 {
1061         char *argp, *envp;
1062         int error;
1063         size_t length;
1064
1065         error = 0;
1066
1067         bzero(args, sizeof(*args));
1068         if (argv == NULL)
1069                 return (EFAULT);
1070         /*
1071          * Allocate temporary demand zeroed space for argument and
1072          *      environment strings:
1073          *
1074          * o ARG_MAX for argument and environment;
1075          * o MAXSHELLCMDLEN for the name of interpreters.
1076          */
1077         args->buf = (char *) kmem_alloc_wait(exec_map,
1078             PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1079         if (args->buf == NULL)
1080                 return (ENOMEM);
1081         args->begin_argv = args->buf;
1082         args->endp = args->begin_argv;
1083         args->stringspace = ARG_MAX;
1084         /*
1085          * Copy the file name.
1086          */
1087         if (fname != NULL) {
1088                 args->fname = args->buf + ARG_MAX;
1089                 error = (segflg == UIO_SYSSPACE) ?
1090                     copystr(fname, args->fname, PATH_MAX, &length) :
1091                     copyinstr(fname, args->fname, PATH_MAX, &length);
1092                 if (error != 0)
1093                         goto err_exit;
1094         } else
1095                 args->fname = NULL;
1096
1097         /*
1098          * extract arguments first
1099          */
1100         while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1101                 if (argp == (caddr_t) -1) {
1102                         error = EFAULT;
1103                         goto err_exit;
1104                 }
1105                 if ((error = copyinstr(argp, args->endp,
1106                     args->stringspace, &length))) {
1107                         if (error == ENAMETOOLONG)
1108                                 error = E2BIG;
1109                         goto err_exit;
1110                 }
1111                 args->stringspace -= length;
1112                 args->endp += length;
1113                 args->argc++;
1114         }
1115
1116         args->begin_envv = args->endp;
1117
1118         /*
1119          * extract environment strings
1120          */
1121         if (envv) {
1122                 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1123                         if (envp == (caddr_t)-1) {
1124                                 error = EFAULT;
1125                                 goto err_exit;
1126                         }
1127                         if ((error = copyinstr(envp, args->endp,
1128                             args->stringspace, &length))) {
1129                                 if (error == ENAMETOOLONG)
1130                                         error = E2BIG;
1131                                 goto err_exit;
1132                         }
1133                         args->stringspace -= length;
1134                         args->endp += length;
1135                         args->envc++;
1136                 }
1137         }
1138
1139         return (0);
1140
1141 err_exit:
1142         exec_free_args(args);
1143         return (error);
1144 }
1145
1146 static void
1147 exec_free_args(struct image_args *args)
1148 {
1149
1150         if (args->buf) {
1151                 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1152                     PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1153                 args->buf = NULL;
1154         }
1155 }
1156
1157 /*
1158  * Copy strings out to the new process address space, constructing new arg
1159  * and env vector tables. Return a pointer to the base so that it can be used
1160  * as the initial stack pointer.
1161  */
1162 register_t *
1163 exec_copyout_strings(imgp)
1164         struct image_params *imgp;
1165 {
1166         int argc, envc;
1167         char **vectp;
1168         char *stringp, *destp;
1169         register_t *stack_base;
1170         struct ps_strings *arginfo;
1171         struct proc *p;
1172         int szsigcode;
1173
1174         /*
1175          * Calculate string base and vector table pointers.
1176          * Also deal with signal trampoline code for this exec type.
1177          */
1178         p = imgp->proc;
1179         szsigcode = 0;
1180         arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1181         if (p->p_sysent->sv_szsigcode != NULL)
1182                 szsigcode = *(p->p_sysent->sv_szsigcode);
1183         destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1184             roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1185
1186         /*
1187          * install sigcode
1188          */
1189         if (szsigcode)
1190                 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1191                     szsigcode), szsigcode);
1192
1193         /*
1194          * If we have a valid auxargs ptr, prepare some room
1195          * on the stack.
1196          */
1197         if (imgp->auxargs) {
1198                 /*
1199                  * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1200                  * lower compatibility.
1201                  */
1202                 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1203                     (AT_COUNT * 2);
1204                 /*
1205                  * The '+ 2' is for the null pointers at the end of each of
1206                  * the arg and env vector sets,and imgp->auxarg_size is room
1207                  * for argument of Runtime loader.
1208                  */
1209                 vectp = (char **)(destp - (imgp->args->argc +
1210                     imgp->args->envc + 2 + imgp->auxarg_size) *
1211                     sizeof(char *));
1212
1213         } else {
1214                 /*
1215                  * The '+ 2' is for the null pointers at the end of each of
1216                  * the arg and env vector sets
1217                  */
1218                 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1219                     sizeof(char *));
1220         }
1221
1222         /*
1223          * vectp also becomes our initial stack base
1224          */
1225         stack_base = (register_t *)vectp;
1226
1227         stringp = imgp->args->begin_argv;
1228         argc = imgp->args->argc;
1229         envc = imgp->args->envc;
1230
1231         /*
1232          * Copy out strings - arguments and environment.
1233          */
1234         copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1235
1236         /*
1237          * Fill in "ps_strings" struct for ps, w, etc.
1238          */
1239         suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1240         suword(&arginfo->ps_nargvstr, argc);
1241
1242         /*
1243          * Fill in argument portion of vector table.
1244          */
1245         for (; argc > 0; --argc) {
1246                 suword(vectp++, (long)(intptr_t)destp);
1247                 while (*stringp++ != 0)
1248                         destp++;
1249                 destp++;
1250         }
1251
1252         /* a null vector table pointer separates the argp's from the envp's */
1253         suword(vectp++, 0);
1254
1255         suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1256         suword(&arginfo->ps_nenvstr, envc);
1257
1258         /*
1259          * Fill in environment portion of vector table.
1260          */
1261         for (; envc > 0; --envc) {
1262                 suword(vectp++, (long)(intptr_t)destp);
1263                 while (*stringp++ != 0)
1264                         destp++;
1265                 destp++;
1266         }
1267
1268         /* end of vector table is a null pointer */
1269         suword(vectp, 0);
1270
1271         return (stack_base);
1272 }
1273
1274 /*
1275  * Check permissions of file to execute.
1276  *      Called with imgp->vp locked.
1277  *      Return 0 for success or error code on failure.
1278  */
1279 int
1280 exec_check_permissions(imgp)
1281         struct image_params *imgp;
1282 {
1283         struct vnode *vp = imgp->vp;
1284         struct vattr *attr = imgp->attr;
1285         struct thread *td;
1286         int error;
1287
1288         td = curthread;
1289
1290         /* Get file attributes */
1291         error = VOP_GETATTR(vp, attr, td->td_ucred);
1292         if (error)
1293                 return (error);
1294
1295 #ifdef MAC
1296         error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1297         if (error)
1298                 return (error);
1299 #endif
1300
1301         /*
1302          * 1) Check if file execution is disabled for the filesystem that this
1303          *      file resides on.
1304          * 2) Insure that at least one execute bit is on - otherwise root
1305          *      will always succeed, and we don't want to happen unless the
1306          *      file really is executable.
1307          * 3) Insure that the file is a regular file.
1308          */
1309         if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1310             ((attr->va_mode & 0111) == 0) ||
1311             (attr->va_type != VREG))
1312                 return (EACCES);
1313
1314         /*
1315          * Zero length files can't be exec'd
1316          */
1317         if (attr->va_size == 0)
1318                 return (ENOEXEC);
1319
1320         /*
1321          *  Check for execute permission to file based on current credentials.
1322          */
1323         error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1324         if (error)
1325                 return (error);
1326
1327         /*
1328          * Check number of open-for-writes on the file and deny execution
1329          * if there are any.
1330          */
1331         if (vp->v_writecount)
1332                 return (ETXTBSY);
1333
1334         /*
1335          * Call filesystem specific open routine (which does nothing in the
1336          * general case).
1337          */
1338         error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1339         if (error == 0)
1340                 imgp->opened = 1;
1341         return (error);
1342 }
1343
1344 /*
1345  * Exec handler registration
1346  */
1347 int
1348 exec_register(execsw_arg)
1349         const struct execsw *execsw_arg;
1350 {
1351         const struct execsw **es, **xs, **newexecsw;
1352         int count = 2;  /* New slot and trailing NULL */
1353
1354         if (execsw)
1355                 for (es = execsw; *es; es++)
1356                         count++;
1357         newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1358         if (newexecsw == NULL)
1359                 return (ENOMEM);
1360         xs = newexecsw;
1361         if (execsw)
1362                 for (es = execsw; *es; es++)
1363                         *xs++ = *es;
1364         *xs++ = execsw_arg;
1365         *xs = NULL;
1366         if (execsw)
1367                 free(execsw, M_TEMP);
1368         execsw = newexecsw;
1369         return (0);
1370 }
1371
1372 int
1373 exec_unregister(execsw_arg)
1374         const struct execsw *execsw_arg;
1375 {
1376         const struct execsw **es, **xs, **newexecsw;
1377         int count = 1;
1378
1379         if (execsw == NULL)
1380                 panic("unregister with no handlers left?\n");
1381
1382         for (es = execsw; *es; es++) {
1383                 if (*es == execsw_arg)
1384                         break;
1385         }
1386         if (*es == NULL)
1387                 return (ENOENT);
1388         for (es = execsw; *es; es++)
1389                 if (*es != execsw_arg)
1390                         count++;
1391         newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1392         if (newexecsw == NULL)
1393                 return (ENOMEM);
1394         xs = newexecsw;
1395         for (es = execsw; *es; es++)
1396                 if (*es != execsw_arg)
1397                         *xs++ = *es;
1398         *xs = NULL;
1399         if (execsw)
1400                 free(execsw, M_TEMP);
1401         execsw = newexecsw;
1402         return (0);
1403 }