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>
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 if ((error = nlookup(nd)) != 0)
250 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
251 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
252 nd->nl_flags &= ~NLC_NCPISLOCKED;
253 cache_unlock(&nd->nl_nch);
258 * Check file permissions (also 'opens' file).
259 * Include also the top level mount in the check.
261 error = exec_check_permissions(imgp, nd->nl_nch.mount);
264 goto exec_fail_dealloc;
267 error = exec_map_first_page(imgp);
270 goto exec_fail_dealloc;
272 imgp->proc->p_osrel = 0;
274 if (debug_execve_args && imgp->interpreted) {
275 kprintf(" target is interpreted -- recursive pass\n");
276 kprintf(" interpreter: %s\n", imgp->interpreter_name);
277 print_execve_args(args);
281 * If the current process has a special image activator it
282 * wants to try first, call it. For example, emulating shell
283 * scripts differently.
286 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
287 error = img_first(imgp);
290 * If the vnode has a registered vmspace, exec the vmspace
292 if (error == -1 && imgp->vp->v_resident) {
293 error = exec_resident_imgact(imgp);
297 * Loop through the list of image activators, calling each one.
298 * An activator returns -1 if there is no match, 0 on success,
299 * and an error otherwise.
301 for (i = 0; error == -1 && execsw[i]; ++i) {
302 if (execsw[i]->ex_imgact == NULL ||
303 execsw[i]->ex_imgact == img_first) {
306 error = (*execsw[i]->ex_imgact)(imgp);
312 goto exec_fail_dealloc;
316 * Special interpreter operation, cleanup and loop up to try to
317 * activate the interpreter.
319 if (imgp->interpreted) {
320 exec_unmap_first_page(imgp);
324 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
332 * Do the best to calculate the full path to the image file
334 if (imgp->auxargs != NULL &&
335 ((args->fname != NULL && args->fname[0] == '/') ||
336 vn_fullpath(imgp->proc,
341 imgp->execpath = args->fname;
344 * Copy out strings (args and env) and initialize stack base
346 stack_base = exec_copyout_strings(imgp);
347 p->p_vmspace->vm_minsaddr = (char *)stack_base;
350 * If custom stack fixup routine present for this process
351 * let it do the stack setup. If we are running a resident
352 * image there is no auxinfo or other image activator context
353 * so don't try to add fixups to the stack.
355 * Else stuff argument count as first item on stack
357 if (p->p_sysent->sv_fixup && imgp->resident == 0)
358 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
360 suword(--stack_base, imgp->args->argc);
363 * For security and other reasons, the file descriptor table cannot
364 * be shared after an exec.
366 if (p->p_fd->fd_refcnt > 1) {
367 struct filedesc *tmp;
369 error = fdcopy(p, &tmp);
376 * For security and other reasons, signal handlers cannot
377 * be shared after an exec. The new proces gets a copy of the old
378 * handlers. In execsigs(), the new process will have its signals
382 if (ops->ps_refcnt > 1) {
383 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
384 bcopy(ops, nps, sizeof(*nps));
385 refcount_init(&nps->ps_refcnt, 1);
387 if (refcount_release(&ops->ps_refcnt)) {
388 kfree(ops, M_SUBPROC);
394 * For security and other reasons virtual kernels cannot be
395 * inherited by an exec. This also allows a virtual kernel
396 * to fork/exec unrelated applications.
404 /* close files on exec */
407 /* reset caught signals */
410 /* name this process - nameiexec(p, ndp) */
411 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
412 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
414 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
417 * mark as execed, wakeup the process that vforked (if any) and tell
418 * it that it now has its own resources back
420 p->p_flags |= P_EXEC;
421 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
422 p->p_flags &= ~P_PPWAIT;
423 wakeup((caddr_t)p->p_pptr);
427 * Implement image setuid/setgid.
429 * Don't honor setuid/setgid if the filesystem prohibits it or if
430 * the process is being traced.
432 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
433 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
434 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
435 (p->p_flags & P_TRACED) == 0) {
437 * Turn off syscall tracing for set-id programs, except for
438 * root. Record any set-id flags first to make sure that
439 * we do not regain any tracing during a possible block.
442 if (p->p_tracenode && ktrace_suid == 0 &&
443 priv_check(td, PRIV_ROOT) != 0) {
444 ktrdestroy(&p->p_tracenode);
447 /* Close any file descriptors 0..2 that reference procfs */
449 /* Make sure file descriptors 0..2 are in use. */
450 error = fdcheckstd(lp);
452 goto exec_fail_dealloc;
454 * Set the new credentials.
457 if (attr.va_mode & VSUID)
458 change_euid(attr.va_uid);
459 if (attr.va_mode & VSGID)
460 p->p_ucred->cr_gid = attr.va_gid;
463 * Clear local varsym variables
465 varsymset_clean(&p->p_varsymset);
467 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
468 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
469 p->p_flags &= ~P_SUGID;
473 * Implement correct POSIX saved-id behavior.
475 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
476 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
478 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
479 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
483 * Store the vp for use in procfs. Be sure to keep p_textvp
484 * consistent if we block during the switch-over.
487 vref(imgp->vp); /* ref new vp */
488 p->p_textvp = imgp->vp;
489 if (ovp) /* release old vp */
492 /* Release old namecache handle to text file */
493 if (p->p_textnch.ncp)
494 cache_drop(&p->p_textnch);
496 if (nd->nl_nch.mount)
497 cache_copy(&nd->nl_nch, &p->p_textnch);
500 * Notify others that we exec'd, and clear the P_INEXEC flag
501 * as we're now a bona fide freshly-execed process.
503 KNOTE(&p->p_klist, NOTE_EXEC);
504 p->p_flags &= ~P_INEXEC;
507 * If tracing the process, trap to debugger so breakpoints
508 * can be set before the program executes.
510 STOPEVENT(p, S_EXEC, 0);
512 if (p->p_flags & P_TRACED)
515 /* clear "fork but no exec" flag, as we _are_ execing */
516 p->p_acflag &= ~AFORK;
518 /* Set values passed into the program in registers. */
519 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
522 /* Set the access time on the vnode */
523 vn_mark_atime(imgp->vp, td);
526 * Free any previous argument cache
530 if (pa && refcount_release(&pa->ar_ref)) {
536 * Cache arguments if they fit inside our allowance
538 i = imgp->args->begin_envv - imgp->args->begin_argv;
539 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
540 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
541 refcount_init(&pa->ar_ref, 1);
543 bcopy(imgp->args->begin_argv, pa->ar_args, i);
544 KKASSERT(p->p_args == NULL);
551 * free various allocated resources
554 exec_unmap_first_page(imgp);
562 kfree(imgp->freepath, M_TEMP);
565 ++mycpu->gd_cnt.v_exec;
566 lwkt_reltoken(&p->p_token);
572 * we're done here, clear P_INEXEC if we were the ones that
573 * set it. Otherwise if vmspace_destroyed is still set we
574 * raced another thread and that thread is responsible for
577 if (imgp->vmspace_destroyed & 2)
578 p->p_flags &= ~P_INEXEC;
579 lwkt_reltoken(&p->p_token);
580 if (imgp->vmspace_destroyed) {
582 * Sorry, no more process anymore. exit gracefully.
583 * However we can't die right here, because our
584 * caller might have to clean up, so indicate a
585 * lethal error by returning -1.
594 * execve() system call.
597 sys_execve(struct execve_args *uap)
599 struct nlookupdata nd;
600 struct image_args args;
603 bzero(&args, sizeof(args));
605 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
607 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
608 uap->argv, uap->envv);
611 error = kern_execve(&nd, &args);
613 exec_free_args(&args);
616 /* We hit a lethal error condition. Let's die now. */
617 exit1(W_EXITCODE(0, SIGABRT));
622 * The syscall result is returned in registers to the new program.
623 * Linux will register %edx as an atexit function and we must be
624 * sure to set it to 0. XXX
627 uap->sysmsg_result64 = 0;
633 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
634 struct lwbuf **plwb, const char **pdata)
642 * The file has to be mappable.
644 if ((object = imgp->vp->v_object) == NULL)
647 if (pageno >= object->size)
650 vm_object_hold(object);
651 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
652 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
656 * get_pages unbusies all the requested pages except the
657 * primary page (at index 0 in this case). The primary
658 * page may have been wired during the pagein (e.g. by
659 * the buffer cache) so vnode_pager_freepage() must be
660 * used to properly release it.
662 rv = vm_pager_get_page(object, &ma, 1);
663 m = vm_page_lookup(object, pageno);
665 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
667 vm_page_protect(m, VM_PROT_NONE);
668 vnode_pager_freepage(m);
670 vm_object_drop(object);
675 vm_page_wakeup(m); /* unbusy the page */
676 vm_object_drop(object);
678 *plwb = lwbuf_alloc(m, *plwb);
679 *pdata = (void *)lwbuf_kva(*plwb);
685 * Map the first page of an executable image.
687 * NOTE: If the mapping fails we have to NULL-out firstpage which may
688 * still be pointing to our supplied lwp structure.
691 exec_map_first_page(struct image_params *imgp)
696 exec_unmap_first_page(imgp);
698 imgp->firstpage = &imgp->firstpage_cache;
699 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
702 imgp->firstpage = NULL;
710 exec_unmap_page(struct lwbuf *lwb)
724 exec_unmap_first_page(struct image_params *imgp)
726 exec_unmap_page(imgp->firstpage);
727 imgp->firstpage = NULL;
728 imgp->image_header = NULL;
732 * Destroy old address space, and allocate a new stack
733 * The new stack is only SGROWSIZ large because it is grown
734 * automatically in trap.c.
736 * This is the point of no return.
739 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
741 struct vmspace *vmspace = imgp->proc->p_vmspace;
742 vm_offset_t stack_addr = USRSTACK - maxssiz;
748 * Indicate that we cannot gracefully error out any more, kill
749 * any other threads present, and set P_INEXEC to indicate that
750 * we are now messing with the process structure proper.
752 * If killalllwps() races return an error which coupled with
753 * vmspace_destroyed will cause us to exit. This is what we
754 * want since another thread is patiently waiting for us to exit
758 imgp->vmspace_destroyed = 1;
760 if (curthread->td_proc->p_nthreads > 1) {
761 error = killalllwps(1);
765 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
766 p->p_flags |= P_INEXEC;
769 * After setting P_INEXEC wait for any remaining references to
770 * the process (p) to go away.
772 * In particular, a vfork/exec sequence will replace p->p_vmspace
773 * and we must interlock anyone trying to access the space (aka
774 * procfs or sys_process.c calling procfs_domem()).
776 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
778 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
781 * Blow away entire process VM, if address space not shared,
782 * otherwise, create a new VM space so that other threads are
783 * not disrupted. If we are execing a resident vmspace we
784 * create a duplicate of it and remap the stack.
786 map = &vmspace->vm_map;
788 vmspace_exec(imgp->proc, vmcopy);
789 vmspace = imgp->proc->p_vmspace;
790 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
791 map = &vmspace->vm_map;
792 } else if (vmspace->vm_sysref.refcnt == 1) {
794 if (vmspace->vm_upcalls)
795 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
796 pmap_remove_pages(vmspace_pmap(vmspace),
797 0, VM_MAX_USER_ADDRESS);
798 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
800 vmspace_exec(imgp->proc, NULL);
801 vmspace = imgp->proc->p_vmspace;
802 map = &vmspace->vm_map;
805 /* Allocate a new stack */
806 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
807 0, VM_PROT_ALL, VM_PROT_ALL, 0);
811 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
812 * VM_STACK case, but they are still used to monitor the size of the
813 * process stack so we can check the stack rlimit.
815 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
816 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
822 * Copy out argument and environment strings from the old process
823 * address space into the temporary string buffer.
826 exec_copyin_args(struct image_args *args, char *fname,
827 enum exec_path_segflg segflg, char **argv, char **envv)
833 args->buf = objcache_get(exec_objcache, M_WAITOK);
834 if (args->buf == NULL)
836 args->begin_argv = args->buf;
837 args->endp = args->begin_argv;
838 args->space = ARG_MAX;
840 args->fname = args->buf + ARG_MAX;
843 * Copy the file name.
845 if (segflg == PATH_SYSSPACE) {
846 error = copystr(fname, args->fname, PATH_MAX, &length);
847 } else if (segflg == PATH_USERSPACE) {
848 error = copyinstr(fname, args->fname, PATH_MAX, &length);
852 * Extract argument strings. argv may not be NULL. The argv
853 * array is terminated by a NULL entry. We special-case the
854 * situation where argv[0] is NULL by passing { filename, NULL }
855 * to the new program to guarentee that the interpreter knows what
856 * file to open in case we exec an interpreted file. Note that
857 * a NULL argv[0] terminates the argv[] array.
859 * XXX the special-casing of argv[0] is historical and needs to be
865 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
866 if (argp == (caddr_t)-1) {
870 error = copyinstr(argp, args->endp,
871 args->space, &length);
873 if (error == ENAMETOOLONG)
877 args->space -= length;
878 args->endp += length;
881 if (args->argc == 0 && error == 0) {
882 length = strlen(args->fname) + 1;
883 if (length > args->space) {
886 bcopy(args->fname, args->endp, length);
887 args->space -= length;
888 args->endp += length;
894 args->begin_envv = args->endp;
897 * extract environment strings. envv may be NULL.
899 if (envv && error == 0) {
900 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
901 if (envp == (caddr_t) -1) {
905 error = copyinstr(envp, args->endp,
906 args->space, &length);
908 if (error == ENAMETOOLONG)
912 args->space -= length;
913 args->endp += length;
921 exec_free_args(struct image_args *args)
924 objcache_put(exec_objcache, args->buf);
930 * Copy strings out to the new process address space, constructing
931 * new arg and env vector tables. Return a pointer to the base
932 * so that it can be used as the initial stack pointer.
934 * The format is, roughly:
940 * [args & env] <-- destp
945 * [ps_strings] top of user stack
949 exec_copyout_strings(struct image_params *imgp)
951 int argc, envc, sgap;
955 char *stringp, *destp;
956 register_t *stack_base;
957 struct ps_strings *arginfo;
962 * Calculate string base and vector table pointers.
963 * Also deal with signal trampoline code for this exec type.
965 if (imgp->execpath != NULL && imgp->auxargs != NULL)
966 execpath_len = strlen(imgp->execpath) + 1;
969 arginfo = (struct ps_strings *)PS_STRINGS;
970 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
972 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
973 gap = stackgap_random;
979 sgap = ALIGN(karc4random() & (gap - 1));
985 * Calculate destp, which points to [args & env] and above.
987 destp = (caddr_t)arginfo -
989 roundup(execpath_len, sizeof(char *)) -
998 copyout(imgp->proc->p_sysent->sv_sigcode,
999 ((caddr_t)arginfo - szsigcode), szsigcode);
1003 * Copy the image path for the rtld
1006 imgp->execpathp = (uintptr_t)arginfo
1008 - roundup(execpath_len, sizeof(char *));
1009 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1013 * Calculate base for argv[], envp[], and ELF_Auxargs.
1015 vectp = (char **)destp - (AT_COUNT * 2);
1016 vectp -= imgp->args->argc + imgp->args->envc + 2;
1018 stack_base = (register_t *)vectp;
1020 stringp = imgp->args->begin_argv;
1021 argc = imgp->args->argc;
1022 envc = imgp->args->envc;
1025 * Copy out strings - arguments and environment (at destp)
1027 copyout(stringp, destp, ARG_MAX - imgp->args->space);
1030 * Fill in "ps_strings" struct for ps, w, etc.
1032 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1033 suword32(&arginfo->ps_nargvstr, argc);
1036 * Fill in argument portion of vector table.
1038 for (; argc > 0; --argc) {
1039 suword(vectp++, (long)(intptr_t)destp);
1040 while (*stringp++ != 0)
1045 /* a null vector table pointer separates the argp's from the envp's */
1048 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1049 suword32(&arginfo->ps_nenvstr, envc);
1052 * Fill in environment portion of vector table.
1054 for (; envc > 0; --envc) {
1055 suword(vectp++, (long)(intptr_t)destp);
1056 while (*stringp++ != 0)
1061 /* end of vector table is a null pointer */
1064 return (stack_base);
1068 * Check permissions of file to execute.
1069 * Return 0 for success or error code on failure.
1072 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1074 struct proc *p = imgp->proc;
1075 struct vnode *vp = imgp->vp;
1076 struct vattr *attr = imgp->attr;
1079 /* Get file attributes */
1080 error = VOP_GETATTR(vp, attr);
1085 * 1) Check if file execution is disabled for the filesystem that this
1087 * 2) Insure that at least one execute bit is on - otherwise root
1088 * will always succeed, and we don't want to happen unless the
1089 * file really is executable.
1090 * 3) Insure that the file is a regular file.
1092 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1093 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1094 ((attr->va_mode & 0111) == 0) ||
1095 (attr->va_type != VREG)) {
1100 * Zero length files can't be exec'd
1102 if (attr->va_size == 0)
1106 * Check for execute permission to file based on current credentials.
1108 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1113 * Check number of open-for-writes on the file and deny execution
1116 if (vp->v_writecount)
1120 * Call filesystem specific open routine, which allows us to read,
1121 * write, and mmap the file. Without the VOP_OPEN we can only
1124 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1132 * Exec handler registration
1135 exec_register(const struct execsw *execsw_arg)
1137 const struct execsw **es, **xs, **newexecsw;
1138 int count = 2; /* New slot and trailing NULL */
1141 for (es = execsw; *es; es++)
1143 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1146 for (es = execsw; *es; es++)
1151 kfree(execsw, M_TEMP);
1157 exec_unregister(const struct execsw *execsw_arg)
1159 const struct execsw **es, **xs, **newexecsw;
1163 panic("unregister with no handlers left?");
1165 for (es = execsw; *es; es++) {
1166 if (*es == execsw_arg)
1171 for (es = execsw; *es; es++)
1172 if (*es != execsw_arg)
1174 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1176 for (es = execsw; *es; es++)
1177 if (*es != execsw_arg)
1181 kfree(execsw, M_TEMP);