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
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_UINT,
158 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
161 print_execve_args(struct image_args *args)
166 cp = args->begin_argv;
167 for (ndx = 0; ndx < args->argc; ndx++) {
168 kprintf("\targv[%d]: %s\n", ndx, cp);
169 while (*cp++ != '\0');
171 for (ndx = 0; ndx < args->envc; ndx++) {
172 kprintf("\tenvv[%d]: %s\n", ndx, cp);
173 while (*cp++ != '\0');
178 * Each of the items is a pointer to a `const struct execsw', hence the
179 * double pointer here.
181 static const struct execsw **execsw;
184 * Replace current vmspace with a new binary.
185 * Returns 0 on success, > 0 on recoverable error (use as errno).
186 * Returns -1 on lethal error which demands killing of the current
190 kern_execve(struct nlookupdata *nd, struct image_args *args)
192 struct thread *td = curthread;
193 struct lwp *lp = td->td_lwp;
194 struct proc *p = td->td_proc;
195 register_t *stack_base;
200 struct image_params image_params, *imgp;
202 int (*img_first) (struct image_params *);
204 if (debug_execve_args) {
205 kprintf("%s()\n", __func__);
206 print_execve_args(args);
210 lwkt_gettoken(&p->p_token);
211 imgp = &image_params;
214 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
215 * no modifications to the process at all until we get there.
217 * Note that multiple threads may be trying to exec at the same
218 * time. exec_new_vmspace() handles that too.
222 * Initialize part of the common data
227 imgp->entry_addr = 0;
229 imgp->vmspace_destroyed = 0;
230 imgp->interpreted = 0;
231 imgp->interpreter_name[0] = 0;
232 imgp->auxargs = NULL;
234 imgp->firstpage = NULL;
235 imgp->ps_strings = 0;
236 imgp->execpath = imgp->freepath = NULL;
238 imgp->image_header = NULL;
243 * Translate the file name to a vnode. Unlock the cache entry to
244 * improve parallelism for programs exec'd in parallel.
246 if ((error = nlookup(nd)) != 0)
248 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
249 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
250 nd->nl_flags &= ~NLC_NCPISLOCKED;
251 cache_unlock(&nd->nl_nch);
256 * Check file permissions (also 'opens' file).
257 * Include also the top level mount in the check.
259 error = exec_check_permissions(imgp, nd->nl_nch.mount);
262 goto exec_fail_dealloc;
265 error = exec_map_first_page(imgp);
268 goto exec_fail_dealloc;
270 imgp->proc->p_osrel = 0;
272 if (debug_execve_args && imgp->interpreted) {
273 kprintf(" target is interpreted -- recursive pass\n");
274 kprintf(" interpreter: %s\n", imgp->interpreter_name);
275 print_execve_args(args);
279 * If the current process has a special image activator it
280 * wants to try first, call it. For example, emulating shell
281 * scripts differently.
284 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
285 error = img_first(imgp);
288 * If the vnode has a registered vmspace, exec the vmspace
290 if (error == -1 && imgp->vp->v_resident) {
291 error = exec_resident_imgact(imgp);
295 * Loop through the list of image activators, calling each one.
296 * An activator returns -1 if there is no match, 0 on success,
297 * and an error otherwise.
299 for (i = 0; error == -1 && execsw[i]; ++i) {
300 if (execsw[i]->ex_imgact == NULL ||
301 execsw[i]->ex_imgact == img_first) {
304 error = (*execsw[i]->ex_imgact)(imgp);
310 goto exec_fail_dealloc;
314 * Special interpreter operation, cleanup and loop up to try to
315 * activate the interpreter.
317 if (imgp->interpreted) {
318 exec_unmap_first_page(imgp);
322 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
330 * Do the best to calculate the full path to the image file
332 if (imgp->auxargs != NULL &&
333 ((args->fname != NULL && args->fname[0] == '/') ||
334 vn_fullpath(imgp->proc,
339 imgp->execpath = args->fname;
342 * Copy out strings (args and env) and initialize stack base
344 stack_base = exec_copyout_strings(imgp);
345 p->p_vmspace->vm_minsaddr = (char *)stack_base;
348 * If custom stack fixup routine present for this process
349 * let it do the stack setup. If we are running a resident
350 * image there is no auxinfo or other image activator context
351 * so don't try to add fixups to the stack.
353 * Else stuff argument count as first item on stack
355 if (p->p_sysent->sv_fixup && imgp->resident == 0)
356 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
358 suword(--stack_base, imgp->args->argc);
361 * For security and other reasons, the file descriptor table cannot
362 * be shared after an exec.
364 if (p->p_fd->fd_refcnt > 1) {
365 struct filedesc *tmp;
367 error = fdcopy(p, &tmp);
374 * For security and other reasons, signal handlers cannot
375 * be shared after an exec. The new proces gets a copy of the old
376 * handlers. In execsigs(), the new process will have its signals
380 if (ops->ps_refcnt > 1) {
381 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
382 bcopy(ops, nps, sizeof(*nps));
383 refcount_init(&nps->ps_refcnt, 1);
385 if (refcount_release(&ops->ps_refcnt)) {
386 kfree(ops, M_SUBPROC);
392 * For security and other reasons virtual kernels cannot be
393 * inherited by an exec. This also allows a virtual kernel
394 * to fork/exec unrelated applications.
402 /* close files on exec */
405 /* reset caught signals */
408 /* name this process - nameiexec(p, ndp) */
409 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
410 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
412 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
415 * mark as execed, wakeup the process that vforked (if any) and tell
416 * it that it now has its own resources back
419 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
420 p->p_flag &= ~P_PPWAIT;
421 wakeup((caddr_t)p->p_pptr);
425 * Implement image setuid/setgid.
427 * Don't honor setuid/setgid if the filesystem prohibits it or if
428 * the process is being traced.
430 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
431 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
432 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
433 (p->p_flag & P_TRACED) == 0) {
435 * Turn off syscall tracing for set-id programs, except for
436 * root. Record any set-id flags first to make sure that
437 * we do not regain any tracing during a possible block.
440 if (p->p_tracenode && ktrace_suid == 0 &&
441 priv_check(td, PRIV_ROOT) != 0) {
442 ktrdestroy(&p->p_tracenode);
445 /* Close any file descriptors 0..2 that reference procfs */
447 /* Make sure file descriptors 0..2 are in use. */
448 error = fdcheckstd(lp);
450 goto exec_fail_dealloc;
452 * Set the new credentials.
455 if (attr.va_mode & VSUID)
456 change_euid(attr.va_uid);
457 if (attr.va_mode & VSGID)
458 p->p_ucred->cr_gid = attr.va_gid;
461 * Clear local varsym variables
463 varsymset_clean(&p->p_varsymset);
465 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
466 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
467 p->p_flag &= ~P_SUGID;
471 * Implement correct POSIX saved-id behavior.
473 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
474 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
476 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
477 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
481 * Store the vp for use in procfs
483 if (p->p_textvp) /* release old reference */
485 p->p_textvp = imgp->vp;
488 /* Release old namecache handle to text file */
489 if (p->p_textnch.ncp)
490 cache_drop(&p->p_textnch);
492 if (nd->nl_nch.mount)
493 cache_copy(&nd->nl_nch, &p->p_textnch);
496 * Notify others that we exec'd, and clear the P_INEXEC flag
497 * as we're now a bona fide freshly-execed process.
499 KNOTE(&p->p_klist, NOTE_EXEC);
500 p->p_flag &= ~P_INEXEC;
503 * If tracing the process, trap to debugger so breakpoints
504 * can be set before the program executes.
506 STOPEVENT(p, S_EXEC, 0);
508 if (p->p_flag & P_TRACED)
511 /* clear "fork but no exec" flag, as we _are_ execing */
512 p->p_acflag &= ~AFORK;
514 /* Set values passed into the program in registers. */
515 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
518 /* Set the access time on the vnode */
519 vn_mark_atime(imgp->vp, td);
522 * Free any previous argument cache
526 if (pa && refcount_release(&pa->ar_ref)) {
532 * Cache arguments if they fit inside our allowance
534 i = imgp->args->begin_envv - imgp->args->begin_argv;
535 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
536 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
537 refcount_init(&pa->ar_ref, 1);
539 bcopy(imgp->args->begin_argv, pa->ar_args, i);
540 KKASSERT(p->p_args == NULL);
547 * free various allocated resources
550 exec_unmap_first_page(imgp);
558 kfree(imgp->freepath, M_TEMP);
561 ++mycpu->gd_cnt.v_exec;
562 lwkt_reltoken(&p->p_token);
568 * we're done here, clear P_INEXEC if we were the ones that
569 * set it. Otherwise if vmspace_destroyed is still set we
570 * raced another thread and that thread is responsible for
573 if (imgp->vmspace_destroyed & 2)
574 p->p_flag &= ~P_INEXEC;
575 lwkt_reltoken(&p->p_token);
576 if (imgp->vmspace_destroyed) {
578 * Sorry, no more process anymore. exit gracefully.
579 * However we can't die right here, because our
580 * caller might have to clean up, so indicate a
581 * lethal error by returning -1.
590 * execve() system call.
593 sys_execve(struct execve_args *uap)
595 struct nlookupdata nd;
596 struct image_args args;
599 bzero(&args, sizeof(args));
601 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
603 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
604 uap->argv, uap->envv);
607 error = kern_execve(&nd, &args);
609 exec_free_args(&args);
612 /* We hit a lethal error condition. Let's die now. */
613 exit1(W_EXITCODE(0, SIGABRT));
618 * The syscall result is returned in registers to the new program.
619 * Linux will register %edx as an atexit function and we must be
620 * sure to set it to 0. XXX
623 uap->sysmsg_result64 = 0;
629 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
630 struct lwbuf **plwb, const char **pdata)
638 * The file has to be mappable.
640 if ((object = imgp->vp->v_object) == NULL)
643 if (pageno >= object->size)
646 vm_object_hold(object);
647 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
648 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
652 * get_pages unbusies all the requested pages except the
653 * primary page (at index 0 in this case). The primary
654 * page may have been wired during the pagein (e.g. by
655 * the buffer cache) so vnode_pager_freepage() must be
656 * used to properly release it.
658 rv = vm_pager_get_page(object, &ma, 1);
659 m = vm_page_lookup(object, pageno);
661 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
663 vm_page_protect(m, VM_PROT_NONE);
664 vnode_pager_freepage(m);
670 vm_page_wakeup(m); /* unbusy the page */
671 vm_object_drop(object);
673 *plwb = lwbuf_alloc(m, *plwb);
674 *pdata = (void *)lwbuf_kva(*plwb);
680 * Map the first page of an executable image.
682 * NOTE: If the mapping fails we have to NULL-out firstpage which may
683 * still be pointing to our supplied lwp structure.
686 exec_map_first_page(struct image_params *imgp)
691 exec_unmap_first_page(imgp);
693 imgp->firstpage = &imgp->firstpage_cache;
694 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
697 imgp->firstpage = NULL;
705 exec_unmap_page(struct lwbuf *lwb)
719 exec_unmap_first_page(struct image_params *imgp)
721 exec_unmap_page(imgp->firstpage);
722 imgp->firstpage = NULL;
723 imgp->image_header = NULL;
727 * Destroy old address space, and allocate a new stack
728 * The new stack is only SGROWSIZ large because it is grown
729 * automatically in trap.c.
731 * This is the point of no return.
734 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
736 struct vmspace *vmspace = imgp->proc->p_vmspace;
737 vm_offset_t stack_addr = USRSTACK - maxssiz;
743 * Indicate that we cannot gracefully error out any more, kill
744 * any other threads present, and set P_INEXEC to indicate that
745 * we are now messing with the process structure proper.
747 * If killalllwps() races return an error which coupled with
748 * vmspace_destroyed will cause us to exit. This is what we
749 * want since another thread is patiently waiting for us to exit
753 imgp->vmspace_destroyed = 1;
755 if (curthread->td_proc->p_nthreads > 1) {
756 error = killalllwps(1);
760 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
761 p->p_flag |= P_INEXEC;
764 * Blow away entire process VM, if address space not shared,
765 * otherwise, create a new VM space so that other threads are
766 * not disrupted. If we are execing a resident vmspace we
767 * create a duplicate of it and remap the stack.
769 * The exitingcnt test is not strictly necessary but has been
770 * included for code sanity (to make the code more deterministic).
772 map = &vmspace->vm_map;
774 vmspace_exec(imgp->proc, vmcopy);
775 vmspace = imgp->proc->p_vmspace;
776 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
777 map = &vmspace->vm_map;
778 } else if (vmspace->vm_sysref.refcnt == 1 &&
779 vmspace->vm_exitingcnt == 0) {
781 if (vmspace->vm_upcalls)
782 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
783 pmap_remove_pages(vmspace_pmap(vmspace),
784 0, VM_MAX_USER_ADDRESS);
785 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
787 vmspace_exec(imgp->proc, NULL);
788 vmspace = imgp->proc->p_vmspace;
789 map = &vmspace->vm_map;
792 /* Allocate a new stack */
793 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
794 0, VM_PROT_ALL, VM_PROT_ALL, 0);
798 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
799 * VM_STACK case, but they are still used to monitor the size of the
800 * process stack so we can check the stack rlimit.
802 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
803 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
809 * Copy out argument and environment strings from the old process
810 * address space into the temporary string buffer.
813 exec_copyin_args(struct image_args *args, char *fname,
814 enum exec_path_segflg segflg, char **argv, char **envv)
820 args->buf = objcache_get(exec_objcache, M_WAITOK);
821 if (args->buf == NULL)
823 args->begin_argv = args->buf;
824 args->endp = args->begin_argv;
825 args->space = ARG_MAX;
827 args->fname = args->buf + ARG_MAX;
830 * Copy the file name.
832 if (segflg == PATH_SYSSPACE) {
833 error = copystr(fname, args->fname, PATH_MAX, &length);
834 } else if (segflg == PATH_USERSPACE) {
835 error = copyinstr(fname, args->fname, PATH_MAX, &length);
839 * Extract argument strings. argv may not be NULL. The argv
840 * array is terminated by a NULL entry. We special-case the
841 * situation where argv[0] is NULL by passing { filename, NULL }
842 * to the new program to guarentee that the interpreter knows what
843 * file to open in case we exec an interpreted file. Note that
844 * a NULL argv[0] terminates the argv[] array.
846 * XXX the special-casing of argv[0] is historical and needs to be
852 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
853 if (argp == (caddr_t)-1) {
857 error = copyinstr(argp, args->endp,
858 args->space, &length);
860 if (error == ENAMETOOLONG)
864 args->space -= length;
865 args->endp += length;
868 if (args->argc == 0 && error == 0) {
869 length = strlen(args->fname) + 1;
870 if (length > args->space) {
873 bcopy(args->fname, args->endp, length);
874 args->space -= length;
875 args->endp += length;
881 args->begin_envv = args->endp;
884 * extract environment strings. envv may be NULL.
886 if (envv && error == 0) {
887 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
888 if (envp == (caddr_t) -1) {
892 error = copyinstr(envp, args->endp, args->space,
895 if (error == ENAMETOOLONG)
899 args->space -= length;
900 args->endp += length;
908 exec_free_args(struct image_args *args)
911 objcache_put(exec_objcache, args->buf);
917 * Copy strings out to the new process address space, constructing
918 * new arg and env vector tables. Return a pointer to the base
919 * so that it can be used as the initial stack pointer.
922 exec_copyout_strings(struct image_params *imgp)
924 int argc, envc, sgap;
926 char *stringp, *destp;
927 register_t *stack_base;
928 struct ps_strings *arginfo;
933 * Calculate string base and vector table pointers.
934 * Also deal with signal trampoline code for this exec type.
936 if (imgp->execpath != NULL && imgp->auxargs != NULL)
937 execpath_len = strlen(imgp->execpath) + 1;
940 arginfo = (struct ps_strings *)PS_STRINGS;
941 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
942 if (stackgap_random != 0)
943 sgap = ALIGN(karc4random() & (stackgap_random - 1));
946 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
947 roundup(execpath_len, sizeof(char *)) -
948 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
954 copyout(imgp->proc->p_sysent->sv_sigcode,
955 ((caddr_t)arginfo - szsigcode), szsigcode);
958 * Copy the image path for the rtld
960 if (execpath_len != 0) {
961 imgp->execpathp = (uintptr_t)arginfo
964 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
968 * If we have a valid auxargs ptr, prepare some room
971 * The '+ 2' is for the null pointers at the end of each of the
972 * arg and env vector sets, and 'AT_COUNT*2' is room for the
976 vectp = (char **)(destp - (imgp->args->argc +
977 imgp->args->envc + 2 + (AT_COUNT * 2) + execpath_len) *
980 vectp = (char **)(destp - (imgp->args->argc +
981 imgp->args->envc + 2) * sizeof(char*));
985 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
986 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
990 * vectp also becomes our initial stack base
992 stack_base = (register_t *)vectp;
994 stringp = imgp->args->begin_argv;
995 argc = imgp->args->argc;
996 envc = imgp->args->envc;
999 * Copy out strings - arguments and environment.
1001 copyout(stringp, destp, ARG_MAX - imgp->args->space);
1004 * Fill in "ps_strings" struct for ps, w, etc.
1006 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1007 suword(&arginfo->ps_nargvstr, argc);
1010 * Fill in argument portion of vector table.
1012 for (; argc > 0; --argc) {
1013 suword(vectp++, (long)(intptr_t)destp);
1014 while (*stringp++ != 0)
1019 /* a null vector table pointer separates the argp's from the envp's */
1022 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1023 suword(&arginfo->ps_nenvstr, envc);
1026 * Fill in environment portion of vector table.
1028 for (; envc > 0; --envc) {
1029 suword(vectp++, (long)(intptr_t)destp);
1030 while (*stringp++ != 0)
1035 /* end of vector table is a null pointer */
1038 return (stack_base);
1042 * Check permissions of file to execute.
1043 * Return 0 for success or error code on failure.
1046 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1048 struct proc *p = imgp->proc;
1049 struct vnode *vp = imgp->vp;
1050 struct vattr *attr = imgp->attr;
1053 /* Get file attributes */
1054 error = VOP_GETATTR(vp, attr);
1059 * 1) Check if file execution is disabled for the filesystem that this
1061 * 2) Insure that at least one execute bit is on - otherwise root
1062 * will always succeed, and we don't want to happen unless the
1063 * file really is executable.
1064 * 3) Insure that the file is a regular file.
1066 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1067 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1068 ((attr->va_mode & 0111) == 0) ||
1069 (attr->va_type != VREG)) {
1074 * Zero length files can't be exec'd
1076 if (attr->va_size == 0)
1080 * Check for execute permission to file based on current credentials.
1082 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1087 * Check number of open-for-writes on the file and deny execution
1090 if (vp->v_writecount)
1094 * Call filesystem specific open routine, which allows us to read,
1095 * write, and mmap the file. Without the VOP_OPEN we can only
1098 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1106 * Exec handler registration
1109 exec_register(const struct execsw *execsw_arg)
1111 const struct execsw **es, **xs, **newexecsw;
1112 int count = 2; /* New slot and trailing NULL */
1115 for (es = execsw; *es; es++)
1117 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1120 for (es = execsw; *es; es++)
1125 kfree(execsw, M_TEMP);
1131 exec_unregister(const struct execsw *execsw_arg)
1133 const struct execsw **es, **xs, **newexecsw;
1137 panic("unregister with no handlers left?");
1139 for (es = execsw; *es; es++) {
1140 if (*es == execsw_arg)
1145 for (es = execsw; *es; es++)
1146 if (*es != execsw_arg)
1148 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1150 for (es = execsw; *es; es++)
1151 if (*es != execsw_arg)
1155 kfree(execsw, M_TEMP);