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 $
27 * $DragonFly: src/sys/kern/kern_exec.c,v 1.14 2003/11/12 01:00:33 daver Exp $
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysproto.h>
33 #include <sys/kernel.h>
34 #include <sys/mount.h>
35 #include <sys/filedesc.h>
36 #include <sys/fcntl.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kern_syscall.h>
43 #include <sys/malloc.h>
45 #include <sys/signalvar.h>
46 #include <sys/pioctl.h>
47 #include <sys/namei.h>
48 #include <sys/sysent.h>
50 #include <sys/sysctl.h>
51 #include <sys/vnode.h>
55 #include <vm/vm_param.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_map.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vm_extern.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_pager.h>
66 #include <machine/reg.h>
68 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
70 static register_t *exec_copyout_strings (struct image_params *);
72 /* XXX This should be vm_size_t. */
73 static u_long ps_strings = PS_STRINGS;
74 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
76 /* XXX This should be vm_size_t. */
77 static u_long usrstack = USRSTACK;
78 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
80 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
81 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
82 &ps_arg_cache_limit, 0, "");
85 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
87 void print_execve_args(struct image_args *args);
88 int debug_execve_args = 0;
89 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
93 print_execve_args(struct image_args *args)
98 cp = args->begin_argv;
99 for (ndx = 0; ndx < args->argc; ndx++) {
100 printf("\targv[%d]: %s\n", ndx, cp);
101 while (*cp++ != '\0');
103 for (ndx = 0; ndx < args->envc; ndx++) {
104 printf("\tenvv[%d]: %s\n", ndx, cp);
105 while (*cp++ != '\0');
110 * Each of the items is a pointer to a `const struct execsw', hence the
111 * double pointer here.
113 static const struct execsw **execsw;
116 kern_execve(struct nameidata *ndp, struct image_args *args)
118 struct thread *td = curthread;
119 struct proc *p = td->td_proc;
120 register_t *stack_base;
122 struct image_params image_params, *imgp;
124 int (*img_first) (struct image_params *);
126 if (debug_execve_args) {
127 printf("%s()\n", __func__);
128 print_execve_args(args);
132 imgp = &image_params;
135 * Lock the process and set the P_INEXEC flag to indicate that
136 * it should be left alone until we're done here. This is
137 * necessary to avoid race conditions - e.g. in ptrace() -
138 * that might allow a local user to illicitly obtain elevated
141 p->p_flag |= P_INEXEC;
144 * Initialize part of the common data
149 imgp->entry_addr = 0;
150 imgp->vmspace_destroyed = 0;
151 imgp->interpreted = 0;
152 imgp->auxargs = NULL;
154 imgp->firstpage = NULL;
155 imgp->ps_strings = 0;
158 * Allocate temporary demand zeroed space for argument and
159 * environment strings
161 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
162 if (imgp->image_header == NULL) {
170 * Translate the file name. namei() returns a vnode pointer
171 * in ni_vp amoung other things.
175 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
180 imgp->vp = ndp->ni_vp;
183 * Check file permissions (also 'opens' file)
185 error = exec_check_permissions(imgp);
187 VOP_UNLOCK(imgp->vp, 0, td);
188 goto exec_fail_dealloc;
191 error = exec_map_first_page(imgp);
192 VOP_UNLOCK(imgp->vp, 0, td);
194 goto exec_fail_dealloc;
196 if (debug_execve_args && imgp->interpreted) {
197 printf(" target is interpreted -- recursive pass\n");
198 printf(" interpreter: %s\n", imgp->interpreter_name);
199 print_execve_args(args);
203 * If the current process has a special image activator it
204 * wants to try first, call it. For example, emulating shell
205 * scripts differently.
208 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
209 error = img_first(imgp);
212 * Loop through the list of image activators, calling each one.
213 * An activator returns -1 if there is no match, 0 on success,
214 * and an error otherwise.
216 for (i = 0; error == -1 && execsw[i]; ++i) {
217 if (execsw[i]->ex_imgact == NULL ||
218 execsw[i]->ex_imgact == img_first) {
221 error = (*execsw[i]->ex_imgact)(imgp);
227 goto exec_fail_dealloc;
231 * Special interpreter operation, cleanup and loop up to try to
232 * activate the interpreter.
234 if (imgp->interpreted) {
235 exec_unmap_first_page(imgp);
236 /* free name buffer and old vnode */
237 NDFREE(ndp, NDF_ONLY_PNBUF);
239 /* set new name to that of the interpreter */
240 NDINIT(ndp, NAMEI_LOOKUP,
241 CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
242 UIO_SYSSPACE, imgp->interpreter_name, td);
247 * Copy out strings (args and env) and initialize stack base
249 stack_base = exec_copyout_strings(imgp);
250 p->p_vmspace->vm_minsaddr = (char *)stack_base;
253 * If custom stack fixup routine present for this process
254 * let it do the stack setup.
255 * Else stuff argument count as first item on stack
257 if (p->p_sysent->sv_fixup)
258 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
260 suword(--stack_base, imgp->args->argc);
263 * For security and other reasons, the file descriptor table cannot
264 * be shared after an exec.
266 if (p->p_fd->fd_refcnt > 1) {
267 struct filedesc *tmp;
275 * For security and other reasons, signal handlers cannot
276 * be shared after an exec. The new proces gets a copy of the old
277 * handlers. In execsigs(), the new process will have its signals
280 if (p->p_procsig->ps_refcnt > 1) {
281 struct procsig *newprocsig;
283 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
284 M_SUBPROC, M_WAITOK);
285 bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
286 p->p_procsig->ps_refcnt--;
287 p->p_procsig = newprocsig;
288 p->p_procsig->ps_refcnt = 1;
289 if (p->p_sigacts == &p->p_addr->u_sigacts)
290 panic("shared procsig but private sigacts?");
292 p->p_addr->u_sigacts = *p->p_sigacts;
293 p->p_sigacts = &p->p_addr->u_sigacts;
299 /* close files on exec */
302 /* reset caught signals */
305 /* name this process - nameiexec(p, ndp) */
306 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
307 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
311 * mark as execed, wakeup the process that vforked (if any) and tell
312 * it that it now has its own resources back
315 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
316 p->p_flag &= ~P_PPWAIT;
317 wakeup((caddr_t)p->p_pptr);
321 * Implement image setuid/setgid.
323 * Don't honor setuid/setgid if the filesystem prohibits it or if
324 * the process is being traced.
326 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
327 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
328 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
329 (p->p_flag & P_TRACED) == 0) {
331 * Turn off syscall tracing for set-id programs, except for
332 * root. Record any set-id flags first to make sure that
333 * we do not regain any tracing during a possible block.
336 if (p->p_tracep && suser(td)) {
339 if ((vtmp = p->p_tracep) != NULL) {
345 /* Close any file descriptors 0..2 that reference procfs */
347 /* Make sure file descriptors 0..2 are in use. */
348 error = fdcheckstd(p);
350 goto exec_fail_dealloc;
352 * Set the new credentials.
355 if (attr.va_mode & VSUID)
356 change_euid(attr.va_uid);
357 if (attr.va_mode & VSGID)
358 p->p_ucred->cr_gid = attr.va_gid;
361 * Clear local varsym variables
363 varsymset_clean(&p->p_varsymset);
365 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
366 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
367 p->p_flag &= ~P_SUGID;
371 * Implement correct POSIX saved-id behavior.
373 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
374 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
376 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
377 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
381 * Store the vp for use in procfs
383 if (p->p_textvp) /* release old reference */
386 p->p_textvp = ndp->ni_vp;
389 * Notify others that we exec'd, and clear the P_INEXEC flag
390 * as we're now a bona fide freshly-execed process.
392 KNOTE(&p->p_klist, NOTE_EXEC);
393 p->p_flag &= ~P_INEXEC;
396 * If tracing the process, trap to debugger so breakpoints
397 * can be set before the program executes.
399 STOPEVENT(p, S_EXEC, 0);
401 if (p->p_flag & P_TRACED)
404 /* clear "fork but no exec" flag, as we _are_ execing */
405 p->p_acflag &= ~AFORK;
407 /* Set values passed into the program in registers. */
408 setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
411 /* Free any previous argument cache */
412 if (p->p_args && --p->p_args->ar_ref == 0)
413 FREE(p->p_args, M_PARGS);
416 /* Cache arguments if they fit inside our allowance */
417 i = imgp->args->begin_envv - imgp->args->begin_argv;
418 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
419 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
421 p->p_args->ar_ref = 1;
422 p->p_args->ar_length = i;
423 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
429 * free various allocated resources
432 exec_unmap_first_page(imgp);
434 if (imgp->image_header != NULL)
435 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
439 NDFREE(ndp, NDF_ONLY_PNBUF);
447 /* we're done here, clear P_INEXEC */
448 p->p_flag &= ~P_INEXEC;
449 if (imgp->vmspace_destroyed) {
450 /* sorry, no more process anymore. exit gracefully */
451 exit1(W_EXITCODE(0, SIGABRT));
460 * execve() system call.
463 execve(struct execve_args *uap)
465 struct thread *td = curthread;
467 struct image_args args;
470 NDINIT(&nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
471 UIO_USERSPACE, uap->fname, td);
473 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
474 uap->argv, uap->envv);
478 error = kern_execve(&nd, &args);
480 exec_free_args(&args);
483 * The syscall result is returned in registers to the new program.
484 * Linux will register %edx as an atexit function and we must be
485 * sure to set it to 0. XXX
488 uap->sysmsg_result64 = 0;
494 exec_map_first_page(struct image_params *imgp)
498 vm_page_t ma[VM_INITIAL_PAGEIN];
502 if (imgp->firstpage) {
503 exec_unmap_first_page(imgp);
506 VOP_GETVOBJECT(imgp->vp, &object);
509 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
511 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
512 initial_pagein = VM_INITIAL_PAGEIN;
513 if (initial_pagein > object->size)
514 initial_pagein = object->size;
515 for (i = 1; i < initial_pagein; i++) {
516 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
517 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
523 ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
530 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
531 ma[0] = vm_page_lookup(object, 0);
533 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
535 vm_page_protect(ma[0], VM_PROT_NONE);
544 vm_page_wakeup(ma[0]);
547 pmap_kenter((vm_offset_t) imgp->image_header, VM_PAGE_TO_PHYS(ma[0]));
548 imgp->firstpage = ma[0];
554 exec_unmap_first_page(imgp)
555 struct image_params *imgp;
557 if (imgp->firstpage) {
558 pmap_kremove((vm_offset_t) imgp->image_header);
559 vm_page_unwire(imgp->firstpage, 1);
560 imgp->firstpage = NULL;
565 * Destroy old address space, and allocate a new stack
566 * The new stack is only SGROWSIZ large because it is grown
567 * automatically in trap.c.
570 exec_new_vmspace(imgp)
571 struct image_params *imgp;
574 struct vmspace *vmspace = imgp->proc->p_vmspace;
575 vm_offset_t stack_addr = USRSTACK - maxssiz;
576 vm_map_t map = &vmspace->vm_map;
578 imgp->vmspace_destroyed = 1;
581 * Prevent a pending AIO from modifying the new address space.
583 aio_proc_rundown(imgp->proc);
586 * Blow away entire process VM, if address space not shared,
587 * otherwise, create a new VM space so that other threads are
590 if (vmspace->vm_refcnt == 1) {
592 pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
593 vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
595 vmspace_exec(imgp->proc);
596 vmspace = imgp->proc->p_vmspace;
597 map = &vmspace->vm_map;
600 /* Allocate a new stack */
601 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
602 VM_PROT_ALL, VM_PROT_ALL, 0);
606 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
607 * VM_STACK case, but they are still used to monitor the size of the
608 * process stack so we can check the stack rlimit.
610 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
611 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
617 * Copy out argument and environment strings from the old process
618 * address space into the temporary string buffer.
621 exec_copyin_args(struct image_args *args, char *fname,
622 enum exec_path_segflg segflg, char **argv, char **envv)
628 args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
629 if (args->buf == NULL)
631 args->begin_argv = args->buf;
632 args->endp = args->begin_argv;
633 args->space = ARG_MAX;
637 args->fname = args->buf + ARG_MAX;
640 * Copy the file name.
642 if (segflg == PATH_SYSSPACE) {
643 error = copystr(fname, args->fname, PATH_MAX, &length);
644 } else if (segflg == PATH_USERSPACE) {
645 error = copyinstr(fname, args->fname, PATH_MAX, &length);
649 * extract argument strings
653 argp = (caddr_t) (intptr_t) fuword(argv);
655 * First argument is a special case. If it is a null
656 * string, we must copy the file name into the string
657 * buffer as the first argument. This guarantees that
658 * the interpreter knows what file to open in the case
659 * that we exec an interpreted file.
661 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
662 if (argp == (caddr_t) -1) {
666 error = copyinstr(argp, args->endp,
667 args->space, &length);
668 if (error == ENAMETOOLONG)
672 args->space -= length;
673 args->endp += length;
678 args->begin_envv = args->endp;
681 * extract environment strings
685 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
686 if (envp == (caddr_t) -1) {
690 error = copyinstr(envp, args->endp, args->space,
692 if (error == ENAMETOOLONG)
696 args->space -= length;
697 args->endp += length;
704 exec_free_args(args);
709 exec_free_args(struct image_args *args)
711 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, PATH_MAX + ARG_MAX);
715 * Copy strings out to the new process address space, constructing
716 * new arg and env vector tables. Return a pointer to the base
717 * so that it can be used as the initial stack pointer.
720 exec_copyout_strings(struct image_params *imgp)
724 char *stringp, *destp;
725 register_t *stack_base;
726 struct ps_strings *arginfo;
730 * Calculate string base and vector table pointers.
731 * Also deal with signal trampoline code for this exec type.
733 arginfo = (struct ps_strings *)PS_STRINGS;
734 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
735 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
736 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
742 copyout(imgp->proc->p_sysent->sv_sigcode,
743 ((caddr_t)arginfo - szsigcode), szsigcode);
746 * If we have a valid auxargs ptr, prepare some room
751 * The '+ 2' is for the null pointers at the end of each of the
752 * arg and env vector sets, and 'AT_COUNT*2' is room for the
755 vectp = (char **)(destp - (imgp->args->argc +
756 imgp->args->envc + 2 + AT_COUNT*2) * sizeof(char*));
759 * The '+ 2' is for the null pointers at the end of each of the
760 * arg and env vector sets
763 (destp - (imgp->args->argc + imgp->args->envc + 2) *
767 * vectp also becomes our initial stack base
769 stack_base = (register_t *)vectp;
771 stringp = imgp->args->begin_argv;
772 argc = imgp->args->argc;
773 envc = imgp->args->envc;
776 * Copy out strings - arguments and environment.
778 copyout(stringp, destp, ARG_MAX - imgp->args->space);
781 * Fill in "ps_strings" struct for ps, w, etc.
783 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
784 suword(&arginfo->ps_nargvstr, argc);
787 * Fill in argument portion of vector table.
789 for (; argc > 0; --argc) {
790 suword(vectp++, (long)(intptr_t)destp);
791 while (*stringp++ != 0)
796 /* a null vector table pointer seperates the argp's from the envp's */
799 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
800 suword(&arginfo->ps_nenvstr, envc);
803 * Fill in environment portion of vector table.
805 for (; envc > 0; --envc) {
806 suword(vectp++, (long)(intptr_t)destp);
807 while (*stringp++ != 0)
812 /* end of vector table is a null pointer */
819 * Check permissions of file to execute.
820 * Return 0 for success or error code on failure.
823 exec_check_permissions(imgp)
824 struct image_params *imgp;
826 struct proc *p = imgp->proc;
827 struct vnode *vp = imgp->vp;
828 struct vattr *attr = imgp->attr;
829 struct thread *td = p->p_thread;
832 /* Get file attributes */
833 error = VOP_GETATTR(vp, attr, td);
838 * 1) Check if file execution is disabled for the filesystem that this
840 * 2) Insure that at least one execute bit is on - otherwise root
841 * will always succeed, and we don't want to happen unless the
842 * file really is executable.
843 * 3) Insure that the file is a regular file.
845 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
846 ((attr->va_mode & 0111) == 0) ||
847 (attr->va_type != VREG)) {
852 * Zero length files can't be exec'd
854 if (attr->va_size == 0)
858 * Check for execute permission to file based on current credentials.
860 error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
865 * Check number of open-for-writes on the file and deny execution
868 if (vp->v_writecount)
872 * Call filesystem specific open routine (which does nothing in the
875 error = VOP_OPEN(vp, FREAD, p->p_ucred, td);
883 * Exec handler registration
886 exec_register(execsw_arg)
887 const struct execsw *execsw_arg;
889 const struct execsw **es, **xs, **newexecsw;
890 int count = 2; /* New slot and trailing NULL */
893 for (es = execsw; *es; es++)
895 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
896 if (newexecsw == NULL)
900 for (es = execsw; *es; es++)
905 free(execsw, M_TEMP);
911 exec_unregister(execsw_arg)
912 const struct execsw *execsw_arg;
914 const struct execsw **es, **xs, **newexecsw;
918 panic("unregister with no handlers left?\n");
920 for (es = execsw; *es; es++) {
921 if (*es == execsw_arg)
926 for (es = execsw; *es; es++)
927 if (*es != execsw_arg)
929 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
930 if (newexecsw == NULL)
933 for (es = execsw; *es; es++)
934 if (*es != execsw_arg)
938 free(execsw, M_TEMP);