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 <sys/mplock2.h>
76 #include <vm/vm_page2.h>
78 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
79 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
81 static register_t *exec_copyout_strings (struct image_params *);
83 /* XXX This should be vm_size_t. */
84 static u_long ps_strings = PS_STRINGS;
85 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
87 /* XXX This should be vm_size_t. */
88 static u_long usrstack = USRSTACK;
89 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
91 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
92 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
93 &ps_arg_cache_limit, 0, "");
96 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
98 static int ktrace_suid = 0;
99 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
101 void print_execve_args(struct image_args *args);
102 int debug_execve_args = 0;
103 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
107 * Exec arguments object cache
109 static struct objcache *exec_objcache;
113 exec_objcache_init(void *arg __unused)
119 * Maximum number of concurrent execs. This can be limiting on
120 * systems with a lot of cpu cores but it also eats a significant
123 cluster_limit = (ncpus < 16) ? 16 : ncpus;
124 limsize = kmem_lim_size();
125 if (limsize > 7 * 1024)
127 if (limsize > 15 * 1024)
130 exec_objcache = objcache_create_mbacked(
131 M_EXECARGS, PATH_MAX + ARG_MAX,
135 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
138 * stackgap_random specifies if the stackgap should have a random size added
139 * to it. It must be a power of 2. If non-zero, the stack gap will be
140 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
142 static int stackgap_random = 1024;
144 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
147 new_val = stackgap_random;
148 error = sysctl_handle_int(oidp, &new_val, 0, req);
149 if (error != 0 || req->newptr == NULL)
151 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val)))
153 stackgap_random = new_val;
158 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT,
159 0, 0, sysctl_kern_stackgap, "I",
160 "Max random stack gap (power of 2), static gap if negative");
163 print_execve_args(struct image_args *args)
168 cp = args->begin_argv;
169 for (ndx = 0; ndx < args->argc; ndx++) {
170 kprintf("\targv[%d]: %s\n", ndx, cp);
171 while (*cp++ != '\0');
173 for (ndx = 0; ndx < args->envc; ndx++) {
174 kprintf("\tenvv[%d]: %s\n", ndx, cp);
175 while (*cp++ != '\0');
180 * Each of the items is a pointer to a `const struct execsw', hence the
181 * double pointer here.
183 static const struct execsw **execsw;
186 * Replace current vmspace with a new binary.
187 * Returns 0 on success, > 0 on recoverable error (use as errno).
188 * Returns -1 on lethal error which demands killing of the current
192 kern_execve(struct nlookupdata *nd, struct image_args *args)
194 struct thread *td = curthread;
195 struct lwp *lp = td->td_lwp;
196 struct proc *p = td->td_proc;
198 register_t *stack_base;
203 struct image_params image_params, *imgp;
205 int (*img_first) (struct image_params *);
207 if (debug_execve_args) {
208 kprintf("%s()\n", __func__);
209 print_execve_args(args);
213 lwkt_gettoken(&p->p_token);
214 imgp = &image_params;
217 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
218 * no modifications to the process at all until we get there.
220 * Note that multiple threads may be trying to exec at the same
221 * time. exec_new_vmspace() handles that too.
225 * Initialize part of the common data
230 imgp->entry_addr = 0;
232 imgp->vmspace_destroyed = 0;
233 imgp->interpreted = 0;
234 imgp->interpreter_name[0] = 0;
235 imgp->auxargs = NULL;
237 imgp->firstpage = NULL;
238 imgp->ps_strings = 0;
239 imgp->execpath = imgp->freepath = NULL;
241 imgp->image_header = NULL;
246 * Translate the file name to a vnode. Unlock the cache entry to
247 * improve parallelism for programs exec'd in parallel.
249 if ((error = nlookup(nd)) != 0)
251 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &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 p->p_flags |= P_EXEC;
422 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
423 p->p_flags &= ~P_PPWAIT;
424 wakeup((caddr_t)p->p_pptr);
428 * Implement image setuid/setgid.
430 * Don't honor setuid/setgid if the filesystem prohibits it or if
431 * the process is being traced.
433 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
434 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
435 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
436 (p->p_flags & P_TRACED) == 0) {
438 * Turn off syscall tracing for set-id programs, except for
439 * root. Record any set-id flags first to make sure that
440 * we do not regain any tracing during a possible block.
443 if (p->p_tracenode && ktrace_suid == 0 &&
444 priv_check(td, PRIV_ROOT) != 0) {
445 ktrdestroy(&p->p_tracenode);
448 /* Close any file descriptors 0..2 that reference procfs */
450 /* Make sure file descriptors 0..2 are in use. */
451 error = fdcheckstd(lp);
453 goto exec_fail_dealloc;
455 * Set the new credentials.
458 if (attr.va_mode & VSUID)
459 change_euid(attr.va_uid);
460 if (attr.va_mode & VSGID)
461 p->p_ucred->cr_gid = attr.va_gid;
464 * Clear local varsym variables
466 varsymset_clean(&p->p_varsymset);
468 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
469 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
470 p->p_flags &= ~P_SUGID;
474 * Implement correct POSIX saved-id behavior.
476 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
477 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
479 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
480 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
484 * Store the vp for use in procfs. Be sure to keep p_textvp
485 * consistent if we block during the switch-over.
488 vref(imgp->vp); /* ref new vp */
489 p->p_textvp = imgp->vp;
490 if (ovp) /* release old vp */
493 /* Release old namecache handle to text file */
494 if (p->p_textnch.ncp)
495 cache_drop(&p->p_textnch);
497 if (nd->nl_nch.mount)
498 cache_copy(&nd->nl_nch, &p->p_textnch);
501 * Notify others that we exec'd, and clear the P_INEXEC flag
502 * as we're now a bona fide freshly-execed process.
504 KNOTE(&p->p_klist, NOTE_EXEC);
505 p->p_flags &= ~P_INEXEC;
510 * If tracing the process, trap to debugger so breakpoints
511 * can be set before the program executes.
513 STOPEVENT(p, S_EXEC, 0);
515 if (p->p_flags & P_TRACED)
518 /* clear "fork but no exec" flag, as we _are_ execing */
519 p->p_acflag &= ~AFORK;
521 /* Set values passed into the program in registers. */
522 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
525 /* Set the access time on the vnode */
526 vn_mark_atime(imgp->vp, td);
529 * Free any previous argument cache
533 if (pa && refcount_release(&pa->ar_ref)) {
539 * Cache arguments if they fit inside our allowance
541 i = imgp->args->begin_envv - imgp->args->begin_argv;
542 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
543 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
544 refcount_init(&pa->ar_ref, 1);
546 bcopy(imgp->args->begin_argv, pa->ar_args, i);
547 KKASSERT(p->p_args == NULL);
554 * free various allocated resources
557 exec_unmap_first_page(imgp);
565 kfree(imgp->freepath, M_TEMP);
568 ++mycpu->gd_cnt.v_exec;
569 lwkt_reltoken(&p->p_token);
575 * we're done here, clear P_INEXEC if we were the ones that
576 * set it. Otherwise if vmspace_destroyed is still set we
577 * raced another thread and that thread is responsible for
580 if (imgp->vmspace_destroyed & 2) {
581 p->p_flags &= ~P_INEXEC;
585 lwkt_reltoken(&p->p_token);
586 if (imgp->vmspace_destroyed) {
588 * Sorry, no more process anymore. exit gracefully.
589 * However we can't die right here, because our
590 * caller might have to clean up, so indicate a
591 * lethal error by returning -1.
600 * execve() system call.
603 sys_execve(struct execve_args *uap)
605 struct nlookupdata nd;
606 struct image_args args;
609 bzero(&args, sizeof(args));
611 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
613 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
614 uap->argv, uap->envv);
617 error = kern_execve(&nd, &args);
619 exec_free_args(&args);
622 /* We hit a lethal error condition. Let's die now. */
623 exit1(W_EXITCODE(0, SIGABRT));
628 * The syscall result is returned in registers to the new program.
629 * Linux will register %edx as an atexit function and we must be
630 * sure to set it to 0. XXX
633 uap->sysmsg_result64 = 0;
639 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
640 struct lwbuf **plwb, const char **pdata)
648 * The file has to be mappable.
650 if ((object = imgp->vp->v_object) == NULL)
653 if (pageno >= object->size)
657 * Shortcut using shared locks, improve concurrent execs.
659 vm_object_hold_shared(object);
660 m = vm_page_lookup(object, pageno);
662 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) {
664 vm_page_sleep_busy(m, FALSE, "execpg");
665 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) {
666 vm_object_drop(object);
672 vm_object_drop(object);
677 vm_object_hold(object);
678 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
679 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
683 * get_pages unbusies all the requested pages except the
684 * primary page (at index 0 in this case). The primary
685 * page may have been wired during the pagein (e.g. by
686 * the buffer cache) so vnode_pager_freepage() must be
687 * used to properly release it.
689 rv = vm_pager_get_page(object, &ma, 1);
690 m = vm_page_lookup(object, pageno);
692 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
694 vm_page_protect(m, VM_PROT_NONE);
695 vnode_pager_freepage(m);
697 vm_object_drop(object);
702 vm_page_wakeup(m); /* unbusy the page */
703 vm_object_drop(object);
706 *plwb = lwbuf_alloc(m, *plwb);
707 *pdata = (void *)lwbuf_kva(*plwb);
713 * Map the first page of an executable image.
715 * NOTE: If the mapping fails we have to NULL-out firstpage which may
716 * still be pointing to our supplied lwp structure.
719 exec_map_first_page(struct image_params *imgp)
724 exec_unmap_first_page(imgp);
726 imgp->firstpage = &imgp->firstpage_cache;
727 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
730 imgp->firstpage = NULL;
738 exec_unmap_page(struct lwbuf *lwb)
752 exec_unmap_first_page(struct image_params *imgp)
754 exec_unmap_page(imgp->firstpage);
755 imgp->firstpage = NULL;
756 imgp->image_header = NULL;
760 * Destroy old address space, and allocate a new stack
761 * The new stack is only SGROWSIZ large because it is grown
762 * automatically in trap.c.
764 * This is the point of no return.
767 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
769 struct vmspace *vmspace = imgp->proc->p_vmspace;
770 vm_offset_t stack_addr = USRSTACK - maxssiz;
776 * Indicate that we cannot gracefully error out any more, kill
777 * any other threads present, and set P_INEXEC to indicate that
778 * we are now messing with the process structure proper.
780 * If killalllwps() races return an error which coupled with
781 * vmspace_destroyed will cause us to exit. This is what we
782 * want since another thread is patiently waiting for us to exit
786 imgp->vmspace_destroyed = 1;
788 if (curthread->td_proc->p_nthreads > 1) {
789 error = killalllwps(1);
793 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
794 p->p_flags |= P_INEXEC;
797 * Tell procfs to release its hold on the process. It
798 * will return EAGAIN.
804 * After setting P_INEXEC wait for any remaining references to
805 * the process (p) to go away.
807 * In particular, a vfork/exec sequence will replace p->p_vmspace
808 * and we must interlock anyone trying to access the space (aka
809 * procfs or sys_process.c calling procfs_domem()).
811 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
813 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
816 * Blow away entire process VM, if address space not shared,
817 * otherwise, create a new VM space so that other threads are
818 * not disrupted. If we are execing a resident vmspace we
819 * create a duplicate of it and remap the stack.
821 map = &vmspace->vm_map;
823 vmspace_exec(imgp->proc, vmcopy);
824 vmspace = imgp->proc->p_vmspace;
825 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
826 map = &vmspace->vm_map;
827 } else if (vmspace->vm_sysref.refcnt == 1) {
829 pmap_remove_pages(vmspace_pmap(vmspace),
830 0, VM_MAX_USER_ADDRESS);
831 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
833 vmspace_exec(imgp->proc, NULL);
834 vmspace = imgp->proc->p_vmspace;
835 map = &vmspace->vm_map;
838 /* Allocate a new stack */
839 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
840 0, VM_PROT_ALL, VM_PROT_ALL, 0);
844 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
845 * VM_STACK case, but they are still used to monitor the size of the
846 * process stack so we can check the stack rlimit.
848 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
849 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
855 * Copy out argument and environment strings from the old process
856 * address space into the temporary string buffer.
859 exec_copyin_args(struct image_args *args, char *fname,
860 enum exec_path_segflg segflg, char **argv, char **envv)
866 args->buf = objcache_get(exec_objcache, M_WAITOK);
867 if (args->buf == NULL)
869 args->begin_argv = args->buf;
870 args->endp = args->begin_argv;
871 args->space = ARG_MAX;
873 args->fname = args->buf + ARG_MAX;
876 * Copy the file name.
878 if (segflg == PATH_SYSSPACE) {
879 error = copystr(fname, args->fname, PATH_MAX, &length);
880 } else if (segflg == PATH_USERSPACE) {
881 error = copyinstr(fname, args->fname, PATH_MAX, &length);
885 * Extract argument strings. argv may not be NULL. The argv
886 * array is terminated by a NULL entry. We special-case the
887 * situation where argv[0] is NULL by passing { filename, NULL }
888 * to the new program to guarentee that the interpreter knows what
889 * file to open in case we exec an interpreted file. Note that
890 * a NULL argv[0] terminates the argv[] array.
892 * XXX the special-casing of argv[0] is historical and needs to be
898 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
899 if (argp == (caddr_t)-1) {
903 error = copyinstr(argp, args->endp,
904 args->space, &length);
906 if (error == ENAMETOOLONG)
910 args->space -= length;
911 args->endp += length;
914 if (args->argc == 0 && error == 0) {
915 length = strlen(args->fname) + 1;
916 if (length > args->space) {
919 bcopy(args->fname, args->endp, length);
920 args->space -= length;
921 args->endp += length;
927 args->begin_envv = args->endp;
930 * extract environment strings. envv may be NULL.
932 if (envv && error == 0) {
933 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
934 if (envp == (caddr_t) -1) {
938 error = copyinstr(envp, args->endp,
939 args->space, &length);
941 if (error == ENAMETOOLONG)
945 args->space -= length;
946 args->endp += length;
954 exec_free_args(struct image_args *args)
957 objcache_put(exec_objcache, args->buf);
963 * Copy strings out to the new process address space, constructing
964 * new arg and env vector tables. Return a pointer to the base
965 * so that it can be used as the initial stack pointer.
967 * The format is, roughly:
973 * [args & env] <-- destp
978 * [ps_strings] top of user stack
982 exec_copyout_strings(struct image_params *imgp)
984 int argc, envc, sgap;
988 char *stringp, *destp;
989 register_t *stack_base;
990 struct ps_strings *arginfo;
995 * Calculate string base and vector table pointers.
996 * Also deal with signal trampoline code for this exec type.
998 if (imgp->execpath != NULL && imgp->auxargs != NULL)
999 execpath_len = strlen(imgp->execpath) + 1;
1002 arginfo = (struct ps_strings *)PS_STRINGS;
1003 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
1005 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
1006 gap = stackgap_random;
1012 sgap = ALIGN(karc4random() & (gap - 1));
1018 * Calculate destp, which points to [args & env] and above.
1020 destp = (caddr_t)arginfo -
1022 roundup(execpath_len, sizeof(char *)) -
1031 copyout(imgp->proc->p_sysent->sv_sigcode,
1032 ((caddr_t)arginfo - szsigcode), szsigcode);
1036 * Copy the image path for the rtld
1039 imgp->execpathp = (uintptr_t)arginfo
1041 - roundup(execpath_len, sizeof(char *));
1042 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1046 * Calculate base for argv[], envp[], and ELF_Auxargs.
1048 vectp = (char **)destp - (AT_COUNT * 2);
1049 vectp -= imgp->args->argc + imgp->args->envc + 2;
1051 stack_base = (register_t *)vectp;
1053 stringp = imgp->args->begin_argv;
1054 argc = imgp->args->argc;
1055 envc = imgp->args->envc;
1058 * Copy out strings - arguments and environment (at destp)
1060 copyout(stringp, destp, ARG_MAX - imgp->args->space);
1063 * Fill in "ps_strings" struct for ps, w, etc.
1065 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1066 suword32(&arginfo->ps_nargvstr, argc);
1069 * Fill in argument portion of vector table.
1071 for (; argc > 0; --argc) {
1072 suword(vectp++, (long)(intptr_t)destp);
1073 while (*stringp++ != 0)
1078 /* a null vector table pointer separates the argp's from the envp's */
1081 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1082 suword32(&arginfo->ps_nenvstr, envc);
1085 * Fill in environment portion of vector table.
1087 for (; envc > 0; --envc) {
1088 suword(vectp++, (long)(intptr_t)destp);
1089 while (*stringp++ != 0)
1094 /* end of vector table is a null pointer */
1097 return (stack_base);
1101 * Check permissions of file to execute.
1102 * Return 0 for success or error code on failure.
1105 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1107 struct proc *p = imgp->proc;
1108 struct vnode *vp = imgp->vp;
1109 struct vattr *attr = imgp->attr;
1112 /* Get file attributes */
1113 error = VOP_GETATTR(vp, attr);
1118 * 1) Check if file execution is disabled for the filesystem that this
1120 * 2) Insure that at least one execute bit is on - otherwise root
1121 * will always succeed, and we don't want to happen unless the
1122 * file really is executable.
1123 * 3) Insure that the file is a regular file.
1125 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1126 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1127 ((attr->va_mode & 0111) == 0) ||
1128 (attr->va_type != VREG)) {
1133 * Zero length files can't be exec'd
1135 if (attr->va_size == 0)
1139 * Check for execute permission to file based on current credentials.
1141 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1146 * Check number of open-for-writes on the file and deny execution
1149 if (vp->v_writecount)
1153 * Call filesystem specific open routine, which allows us to read,
1154 * write, and mmap the file. Without the VOP_OPEN we can only
1157 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1165 * Exec handler registration
1168 exec_register(const struct execsw *execsw_arg)
1170 const struct execsw **es, **xs, **newexecsw;
1171 int count = 2; /* New slot and trailing NULL */
1174 for (es = execsw; *es; es++)
1176 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1179 for (es = execsw; *es; es++)
1184 kfree(execsw, M_TEMP);
1190 exec_unregister(const struct execsw *execsw_arg)
1192 const struct execsw **es, **xs, **newexecsw;
1196 panic("unregister with no handlers left?");
1198 for (es = execsw; *es; es++) {
1199 if (*es == execsw_arg)
1204 for (es = execsw; *es; es++)
1205 if (*es != execsw_arg)
1207 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1209 for (es = execsw; *es; es++)
1210 if (*es != execsw_arg)
1214 kfree(execsw, M_TEMP);