2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
26 * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysproto.h>
32 #include <sys/kernel.h>
33 #include <sys/mount.h>
34 #include <sys/filedesc.h>
35 #include <sys/fcntl.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kern_syscall.h>
42 #include <sys/malloc.h>
45 #include <sys/ktrace.h>
46 #include <sys/signalvar.h>
47 #include <sys/pioctl.h>
48 #include <sys/nlookup.h>
49 #include <sys/sysent.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
53 #include <sys/vmmeter.h>
54 #include <sys/libkern.h>
56 #include <cpu/lwbuf.h>
59 #include <vm/vm_param.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/vnode_pager.h>
68 #include <vm/vm_pager.h>
73 #include <sys/refcount.h>
74 #include <sys/thread2.h>
75 #include <vm/vm_page2.h>
77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
78 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
80 static register_t *exec_copyout_strings (struct image_params *);
82 /* XXX This should be vm_size_t. */
83 static u_long ps_strings = PS_STRINGS;
84 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
86 /* XXX This should be vm_size_t. */
87 static u_long usrstack = USRSTACK;
88 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
90 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
91 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
92 &ps_arg_cache_limit, 0, "");
95 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
97 static int ktrace_suid = 0;
98 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
100 void print_execve_args(struct image_args *args);
101 int debug_execve_args = 0;
102 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
106 * Exec arguments object cache
108 static struct objcache *exec_objcache;
112 exec_objcache_init(void *arg __unused)
118 * Maximum number of concurrent execs. This can be limiting on
119 * systems with a lot of cpu cores but it also eats a significant
122 cluster_limit = (ncpus < 16) ? 16 : ncpus;
123 limsize = kmem_lim_size();
124 if (limsize > 7 * 1024)
126 if (limsize > 15 * 1024)
129 exec_objcache = objcache_create_mbacked(
130 M_EXECARGS, PATH_MAX + ARG_MAX,
134 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
137 * stackgap_random specifies if the stackgap should have a random size added
138 * to it. It must be a power of 2. If non-zero, the stack gap will be
139 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
141 static int stackgap_random = 1024;
143 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
146 new_val = stackgap_random;
147 error = sysctl_handle_int(oidp, &new_val, 0, req);
148 if (error != 0 || req->newptr == NULL)
150 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val)))
152 stackgap_random = new_val;
157 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT,
158 0, 0, sysctl_kern_stackgap, "I",
159 "Max random stack gap (power of 2), static gap if negative");
162 print_execve_args(struct image_args *args)
167 cp = args->begin_argv;
168 for (ndx = 0; ndx < args->argc; ndx++) {
169 kprintf("\targv[%d]: %s\n", ndx, cp);
170 while (*cp++ != '\0');
172 for (ndx = 0; ndx < args->envc; ndx++) {
173 kprintf("\tenvv[%d]: %s\n", ndx, cp);
174 while (*cp++ != '\0');
179 * Each of the items is a pointer to a `const struct execsw', hence the
180 * double pointer here.
182 static const struct execsw **execsw;
185 * Replace current vmspace with a new binary.
186 * Returns 0 on success, > 0 on recoverable error (use as errno).
187 * Returns -1 on lethal error which demands killing of the current
191 kern_execve(struct nlookupdata *nd, struct image_args *args)
193 struct thread *td = curthread;
194 struct lwp *lp = td->td_lwp;
195 struct proc *p = td->td_proc;
197 register_t *stack_base;
202 struct image_params image_params, *imgp;
204 int (*img_first) (struct image_params *);
206 if (debug_execve_args) {
207 kprintf("%s()\n", __func__);
208 print_execve_args(args);
212 lwkt_gettoken(&p->p_token);
213 imgp = &image_params;
216 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
217 * no modifications to the process at all until we get there.
219 * Note that multiple threads may be trying to exec at the same
220 * time. exec_new_vmspace() handles that too.
224 * Initialize part of the common data
229 imgp->entry_addr = 0;
231 imgp->vmspace_destroyed = 0;
232 imgp->interpreted = 0;
233 imgp->interpreter_name[0] = 0;
234 imgp->auxargs = NULL;
236 imgp->firstpage = NULL;
237 imgp->ps_strings = 0;
238 imgp->execpath = imgp->freepath = NULL;
240 imgp->image_header = NULL;
245 * Translate the file name to a vnode. Unlock the cache entry to
246 * improve parallelism for programs exec'd in parallel.
248 nd->nl_flags |= NLC_SHAREDLOCK;
249 if ((error = nlookup(nd)) != 0)
251 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &imgp->vp);
252 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
253 nd->nl_flags &= ~NLC_NCPISLOCKED;
254 cache_unlock(&nd->nl_nch);
259 * Check file permissions (also 'opens' file).
260 * Include also the top level mount in the check.
262 error = exec_check_permissions(imgp, nd->nl_nch.mount);
265 goto exec_fail_dealloc;
268 error = exec_map_first_page(imgp);
271 goto exec_fail_dealloc;
273 imgp->proc->p_osrel = 0;
275 if (debug_execve_args && imgp->interpreted) {
276 kprintf(" target is interpreted -- recursive pass\n");
277 kprintf(" interpreter: %s\n", imgp->interpreter_name);
278 print_execve_args(args);
282 * If the current process has a special image activator it
283 * wants to try first, call it. For example, emulating shell
284 * scripts differently.
287 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
288 error = img_first(imgp);
291 * If the vnode has a registered vmspace, exec the vmspace
293 if (error == -1 && imgp->vp->v_resident) {
294 error = exec_resident_imgact(imgp);
298 * Loop through the list of image activators, calling each one.
299 * An activator returns -1 if there is no match, 0 on success,
300 * and an error otherwise.
302 for (i = 0; error == -1 && execsw[i]; ++i) {
303 if (execsw[i]->ex_imgact == NULL ||
304 execsw[i]->ex_imgact == img_first) {
307 error = (*execsw[i]->ex_imgact)(imgp);
313 goto exec_fail_dealloc;
317 * Special interpreter operation, cleanup and loop up to try to
318 * activate the interpreter.
320 if (imgp->interpreted) {
321 exec_unmap_first_page(imgp);
325 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
333 * Do the best to calculate the full path to the image file
335 if (imgp->auxargs != NULL &&
336 ((args->fname != NULL && args->fname[0] == '/') ||
337 vn_fullpath(imgp->proc,
342 imgp->execpath = args->fname;
345 * Copy out strings (args and env) and initialize stack base
347 stack_base = exec_copyout_strings(imgp);
348 p->p_vmspace->vm_minsaddr = (char *)stack_base;
351 * If custom stack fixup routine present for this process
352 * let it do the stack setup. If we are running a resident
353 * image there is no auxinfo or other image activator context
354 * so don't try to add fixups to the stack.
356 * Else stuff argument count as first item on stack
358 if (p->p_sysent->sv_fixup && imgp->resident == 0)
359 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
361 suword(--stack_base, imgp->args->argc);
364 * For security and other reasons, the file descriptor table cannot
365 * be shared after an exec.
367 if (p->p_fd->fd_refcnt > 1) {
368 struct filedesc *tmp;
370 error = fdcopy(p, &tmp);
377 * For security and other reasons, signal handlers cannot
378 * be shared after an exec. The new proces gets a copy of the old
379 * handlers. In execsigs(), the new process will have its signals
383 if (ops->ps_refcnt > 1) {
384 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
385 bcopy(ops, nps, sizeof(*nps));
386 refcount_init(&nps->ps_refcnt, 1);
388 if (refcount_release(&ops->ps_refcnt)) {
389 kfree(ops, M_SUBPROC);
395 * For security and other reasons virtual kernels cannot be
396 * inherited by an exec. This also allows a virtual kernel
397 * to fork/exec unrelated applications.
405 /* close files on exec */
408 /* reset caught signals */
411 /* name this process - nameiexec(p, ndp) */
412 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
413 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
415 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
418 * mark as execed, wakeup the process that vforked (if any) and tell
419 * it that it now has its own resources back
421 * We are using the P_PPWAIT as an interlock so an atomic op is
422 * necessary to synchronize with the parent's cpu.
424 p->p_flags |= P_EXEC;
425 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
426 if (p->p_pptr->p_upmap)
427 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
428 atomic_clear_int(&p->p_flags, P_PPWAIT);
433 * Implement image setuid/setgid.
435 * Don't honor setuid/setgid if the filesystem prohibits it or if
436 * the process is being traced.
438 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
439 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
440 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
441 (p->p_flags & P_TRACED) == 0) {
443 * Turn off syscall tracing for set-id programs, except for
444 * root. Record any set-id flags first to make sure that
445 * we do not regain any tracing during a possible block.
448 if (p->p_tracenode && ktrace_suid == 0 &&
449 priv_check(td, PRIV_ROOT) != 0) {
450 ktrdestroy(&p->p_tracenode);
453 /* Close any file descriptors 0..2 that reference procfs */
455 /* Make sure file descriptors 0..2 are in use. */
456 error = fdcheckstd(lp);
458 goto exec_fail_dealloc;
460 * Set the new credentials.
463 if (attr.va_mode & VSUID)
464 change_euid(attr.va_uid);
465 if (attr.va_mode & VSGID)
466 p->p_ucred->cr_gid = attr.va_gid;
469 * Clear local varsym variables
471 varsymset_clean(&p->p_varsymset);
473 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
474 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
475 p->p_flags &= ~P_SUGID;
479 * Implement correct POSIX saved-id behavior.
481 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
482 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
484 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
485 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
489 * Store the vp for use in procfs. Be sure to keep p_textvp
490 * consistent if we block during the switch-over.
493 vref(imgp->vp); /* ref new vp */
494 p->p_textvp = imgp->vp;
495 if (ovp) /* release old vp */
498 /* Release old namecache handle to text file */
499 if (p->p_textnch.ncp)
500 cache_drop(&p->p_textnch);
502 if (nd->nl_nch.mount)
503 cache_copy(&nd->nl_nch, &p->p_textnch);
506 * Notify others that we exec'd, and clear the P_INEXEC flag
507 * as we're now a bona fide freshly-execed process.
509 KNOTE(&p->p_klist, NOTE_EXEC);
510 p->p_flags &= ~P_INEXEC;
515 * If tracing the process, trap to debugger so breakpoints
516 * can be set before the program executes.
518 STOPEVENT(p, S_EXEC, 0);
520 if (p->p_flags & P_TRACED)
523 /* clear "fork but no exec" flag, as we _are_ execing */
524 p->p_acflag &= ~AFORK;
526 /* Set values passed into the program in registers. */
527 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
530 /* Set the access time on the vnode */
531 vn_mark_atime(imgp->vp, td);
534 * Free any previous argument cache
538 if (pa && refcount_release(&pa->ar_ref)) {
544 * Cache arguments if they fit inside our allowance
546 i = imgp->args->begin_envv - imgp->args->begin_argv;
547 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
548 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
549 refcount_init(&pa->ar_ref, 1);
551 bcopy(imgp->args->begin_argv, pa->ar_args, i);
552 KKASSERT(p->p_args == NULL);
559 * free various allocated resources
562 exec_unmap_first_page(imgp);
570 kfree(imgp->freepath, M_TEMP);
573 ++mycpu->gd_cnt.v_exec;
574 lwkt_reltoken(&p->p_token);
580 * we're done here, clear P_INEXEC if we were the ones that
581 * set it. Otherwise if vmspace_destroyed is still set we
582 * raced another thread and that thread is responsible for
585 if (imgp->vmspace_destroyed & 2) {
586 p->p_flags &= ~P_INEXEC;
590 lwkt_reltoken(&p->p_token);
591 if (imgp->vmspace_destroyed) {
593 * Sorry, no more process anymore. exit gracefully.
594 * However we can't die right here, because our
595 * caller might have to clean up, so indicate a
596 * lethal error by returning -1.
605 * execve() system call.
608 sys_execve(struct execve_args *uap)
610 struct nlookupdata nd;
611 struct image_args args;
614 bzero(&args, sizeof(args));
616 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
618 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
619 uap->argv, uap->envv);
622 error = kern_execve(&nd, &args);
624 exec_free_args(&args);
627 /* We hit a lethal error condition. Let's die now. */
628 exit1(W_EXITCODE(0, SIGABRT));
633 * The syscall result is returned in registers to the new program.
634 * Linux will register %edx as an atexit function and we must be
635 * sure to set it to 0. XXX
638 uap->sysmsg_result64 = 0;
644 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
645 struct lwbuf **plwb, const char **pdata)
653 * The file has to be mappable.
655 if ((object = imgp->vp->v_object) == NULL)
658 if (pageno >= object->size)
662 * Shortcut using shared locks, improve concurrent execs.
664 vm_object_hold_shared(object);
665 m = vm_page_lookup(object, pageno);
667 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) {
669 vm_page_sleep_busy(m, FALSE, "execpg");
670 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL &&
671 m->object == object && m->pindex == pageno) {
672 vm_object_drop(object);
678 vm_object_drop(object);
683 vm_object_hold(object);
684 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
685 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
689 * get_pages unbusies all the requested pages except the
690 * primary page (at index 0 in this case). The primary
691 * page may have been wired during the pagein (e.g. by
692 * the buffer cache) so vnode_pager_freepage() must be
693 * used to properly release it.
695 rv = vm_pager_get_page(object, &ma, 1);
696 m = vm_page_lookup(object, pageno);
698 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
700 vm_page_protect(m, VM_PROT_NONE);
701 vnode_pager_freepage(m);
703 vm_object_drop(object);
708 vm_page_wakeup(m); /* unbusy the page */
709 vm_object_drop(object);
712 *plwb = lwbuf_alloc(m, *plwb);
713 *pdata = (void *)lwbuf_kva(*plwb);
719 * Map the first page of an executable image.
721 * NOTE: If the mapping fails we have to NULL-out firstpage which may
722 * still be pointing to our supplied lwp structure.
725 exec_map_first_page(struct image_params *imgp)
730 exec_unmap_first_page(imgp);
732 imgp->firstpage = &imgp->firstpage_cache;
733 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
736 imgp->firstpage = NULL;
744 exec_unmap_page(struct lwbuf *lwb)
758 exec_unmap_first_page(struct image_params *imgp)
760 exec_unmap_page(imgp->firstpage);
761 imgp->firstpage = NULL;
762 imgp->image_header = NULL;
766 * Destroy old address space, and allocate a new stack
767 * The new stack is only SGROWSIZ large because it is grown
768 * automatically in trap.c.
770 * This is the point of no return.
773 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
775 struct vmspace *vmspace = imgp->proc->p_vmspace;
776 vm_offset_t stack_addr = USRSTACK - maxssiz;
782 * Indicate that we cannot gracefully error out any more, kill
783 * any other threads present, and set P_INEXEC to indicate that
784 * we are now messing with the process structure proper.
786 * If killalllwps() races return an error which coupled with
787 * vmspace_destroyed will cause us to exit. This is what we
788 * want since another thread is patiently waiting for us to exit
792 imgp->vmspace_destroyed = 1;
794 if (curthread->td_proc->p_nthreads > 1) {
795 error = killalllwps(1);
799 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
800 p->p_flags |= P_INEXEC;
803 * Tell procfs to release its hold on the process. It
804 * will return EAGAIN.
810 * After setting P_INEXEC wait for any remaining references to
811 * the process (p) to go away.
813 * In particular, a vfork/exec sequence will replace p->p_vmspace
814 * and we must interlock anyone trying to access the space (aka
815 * procfs or sys_process.c calling procfs_domem()).
817 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
819 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
822 * Blow away entire process VM, if address space not shared,
823 * otherwise, create a new VM space so that other threads are
824 * not disrupted. If we are execing a resident vmspace we
825 * create a duplicate of it and remap the stack.
827 map = &vmspace->vm_map;
829 vmspace_exec(imgp->proc, vmcopy);
830 vmspace = imgp->proc->p_vmspace;
831 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
832 map = &vmspace->vm_map;
833 } else if (vmspace_getrefs(vmspace) == 1) {
835 pmap_remove_pages(vmspace_pmap(vmspace),
836 0, VM_MAX_USER_ADDRESS);
837 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
839 vmspace_exec(imgp->proc, NULL);
840 vmspace = imgp->proc->p_vmspace;
841 map = &vmspace->vm_map;
844 /* Allocate a new stack */
845 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
846 0, VM_PROT_ALL, VM_PROT_ALL, 0);
850 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
851 * VM_STACK case, but they are still used to monitor the size of the
852 * process stack so we can check the stack rlimit.
854 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
855 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
861 * Copy out argument and environment strings from the old process
862 * address space into the temporary string buffer.
865 exec_copyin_args(struct image_args *args, char *fname,
866 enum exec_path_segflg segflg, char **argv, char **envv)
872 args->buf = objcache_get(exec_objcache, M_WAITOK);
873 if (args->buf == NULL)
875 args->begin_argv = args->buf;
876 args->endp = args->begin_argv;
877 args->space = ARG_MAX;
879 args->fname = args->buf + ARG_MAX;
882 * Copy the file name.
884 if (segflg == PATH_SYSSPACE) {
885 error = copystr(fname, args->fname, PATH_MAX, &length);
886 } else if (segflg == PATH_USERSPACE) {
887 error = copyinstr(fname, args->fname, PATH_MAX, &length);
891 * Extract argument strings. argv may not be NULL. The argv
892 * array is terminated by a NULL entry. We special-case the
893 * situation where argv[0] is NULL by passing { filename, NULL }
894 * to the new program to guarentee that the interpreter knows what
895 * file to open in case we exec an interpreted file. Note that
896 * a NULL argv[0] terminates the argv[] array.
898 * XXX the special-casing of argv[0] is historical and needs to be
904 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
905 if (argp == (caddr_t)-1) {
909 error = copyinstr(argp, args->endp,
910 args->space, &length);
912 if (error == ENAMETOOLONG)
916 args->space -= length;
917 args->endp += length;
920 if (args->argc == 0 && error == 0) {
921 length = strlen(args->fname) + 1;
922 if (length > args->space) {
925 bcopy(args->fname, args->endp, length);
926 args->space -= length;
927 args->endp += length;
933 args->begin_envv = args->endp;
936 * extract environment strings. envv may be NULL.
938 if (envv && error == 0) {
939 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
940 if (envp == (caddr_t) -1) {
944 error = copyinstr(envp, args->endp,
945 args->space, &length);
947 if (error == ENAMETOOLONG)
951 args->space -= length;
952 args->endp += length;
960 exec_free_args(struct image_args *args)
963 objcache_put(exec_objcache, args->buf);
969 * Copy strings out to the new process address space, constructing
970 * new arg and env vector tables. Return a pointer to the base
971 * so that it can be used as the initial stack pointer.
973 * The format is, roughly:
979 * [args & env] <-- destp
984 * [ps_strings] top of user stack
988 exec_copyout_strings(struct image_params *imgp)
990 int argc, envc, sgap;
994 char *stringp, *destp;
995 register_t *stack_base;
996 struct ps_strings *arginfo;
1001 * Calculate string base and vector table pointers.
1002 * Also deal with signal trampoline code for this exec type.
1004 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1005 execpath_len = strlen(imgp->execpath) + 1;
1008 arginfo = (struct ps_strings *)PS_STRINGS;
1009 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
1011 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
1012 gap = stackgap_random;
1018 sgap = ALIGN(karc4random() & (gap - 1));
1024 * Calculate destp, which points to [args & env] and above.
1026 destp = (caddr_t)arginfo -
1028 roundup(execpath_len, sizeof(char *)) -
1037 copyout(imgp->proc->p_sysent->sv_sigcode,
1038 ((caddr_t)arginfo - szsigcode), szsigcode);
1042 * Copy the image path for the rtld
1045 imgp->execpathp = (uintptr_t)arginfo
1047 - roundup(execpath_len, sizeof(char *));
1048 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1052 * Calculate base for argv[], envp[], and ELF_Auxargs.
1054 vectp = (char **)destp - (AT_COUNT * 2);
1055 vectp -= imgp->args->argc + imgp->args->envc + 2;
1057 stack_base = (register_t *)vectp;
1059 stringp = imgp->args->begin_argv;
1060 argc = imgp->args->argc;
1061 envc = imgp->args->envc;
1064 * Copy out strings - arguments and environment (at destp)
1066 copyout(stringp, destp, ARG_MAX - imgp->args->space);
1069 * Fill in "ps_strings" struct for ps, w, etc.
1071 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1072 suword32(&arginfo->ps_nargvstr, argc);
1075 * Fill in argument portion of vector table.
1077 for (; argc > 0; --argc) {
1078 suword(vectp++, (long)(intptr_t)destp);
1079 while (*stringp++ != 0)
1084 /* a null vector table pointer separates the argp's from the envp's */
1087 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1088 suword32(&arginfo->ps_nenvstr, envc);
1091 * Fill in environment portion of vector table.
1093 for (; envc > 0; --envc) {
1094 suword(vectp++, (long)(intptr_t)destp);
1095 while (*stringp++ != 0)
1100 /* end of vector table is a null pointer */
1103 return (stack_base);
1107 * Check permissions of file to execute.
1108 * Return 0 for success or error code on failure.
1111 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1113 struct proc *p = imgp->proc;
1114 struct vnode *vp = imgp->vp;
1115 struct vattr *attr = imgp->attr;
1118 /* Get file attributes */
1119 error = VOP_GETATTR(vp, attr);
1124 * 1) Check if file execution is disabled for the filesystem that this
1126 * 2) Insure that at least one execute bit is on - otherwise root
1127 * will always succeed, and we don't want to happen unless the
1128 * file really is executable.
1129 * 3) Insure that the file is a regular file.
1131 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1132 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1133 ((attr->va_mode & 0111) == 0) ||
1134 (attr->va_type != VREG)) {
1139 * Zero length files can't be exec'd
1141 if (attr->va_size == 0)
1145 * Check for execute permission to file based on current credentials.
1147 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1152 * Check number of open-for-writes on the file and deny execution
1155 if (vp->v_writecount)
1159 * Call filesystem specific open routine, which allows us to read,
1160 * write, and mmap the file. Without the VOP_OPEN we can only
1163 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1171 * Exec handler registration
1174 exec_register(const struct execsw *execsw_arg)
1176 const struct execsw **es, **xs, **newexecsw;
1177 int count = 2; /* New slot and trailing NULL */
1180 for (es = execsw; *es; es++)
1182 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1185 for (es = execsw; *es; es++)
1190 kfree(execsw, M_TEMP);
1196 exec_unregister(const struct execsw *execsw_arg)
1198 const struct execsw **es, **xs, **newexecsw;
1202 panic("unregister with no handlers left?");
1204 for (es = execsw; *es; es++) {
1205 if (*es == execsw_arg)
1210 for (es = execsw; *es; es++)
1211 if (*es != execsw_arg)
1213 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1215 for (es = execsw; *es; es++)
1216 if (*es != execsw_arg)
1220 kfree(execsw, M_TEMP);