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 1.11 2003/08/26 21:09:02 rob 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>
42 #include <sys/malloc.h>
44 #include <sys/signalvar.h>
45 #include <sys/pioctl.h>
46 #include <sys/namei.h>
47 #include <sys/sysent.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
54 #include <vm/vm_param.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_pager.h>
65 #include <machine/reg.h>
67 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
69 static register_t *exec_copyout_strings (struct image_params *);
71 /* XXX This should be vm_size_t. */
72 static u_long ps_strings = PS_STRINGS;
73 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
75 /* XXX This should be vm_size_t. */
76 static u_long usrstack = USRSTACK;
77 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
79 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
80 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
81 &ps_arg_cache_limit, 0, "");
84 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
87 * Each of the items is a pointer to a `const struct execsw', hence the
88 * double pointer here.
90 static const struct execsw **execsw;
93 * execve() system call.
96 execve(struct execve_args *uap)
98 struct thread *td = curthread;
99 struct proc *p = td->td_proc;
100 struct nameidata nd, *ndp;
101 register_t *stack_base;
103 struct image_params image_params, *imgp;
105 int (*img_first) (struct image_params *);
108 imgp = &image_params;
111 * Lock the process and set the P_INEXEC flag to indicate that
112 * it should be left alone until we're done here. This is
113 * necessary to avoid race conditions - e.g. in ptrace() -
114 * that might allow a local user to illicitly obtain elevated
117 p->p_flag |= P_INEXEC;
120 * Initialize part of the common data
125 imgp->argc = imgp->envc = 0;
127 imgp->entry_addr = 0;
128 imgp->vmspace_destroyed = 0;
129 imgp->interpreted = 0;
130 imgp->interpreter_name[0] = '\0';
131 imgp->auxargs = NULL;
133 imgp->firstpage = NULL;
134 imgp->ps_strings = 0;
137 * Allocate temporary demand zeroed space for argument and
138 * environment strings
140 imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX + PAGE_SIZE);
141 if (imgp->stringbase == NULL) {
145 imgp->stringp = imgp->stringbase;
146 imgp->stringspace = ARG_MAX;
147 imgp->image_header = imgp->stringbase + ARG_MAX;
150 * Translate the file name. namei() returns a vnode pointer
151 * in ni_vp amoung other things.
154 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
155 UIO_USERSPACE, uap->fname, td);
161 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
162 ARG_MAX + PAGE_SIZE);
166 imgp->vp = ndp->ni_vp;
167 imgp->fname = uap->fname;
170 * Check file permissions (also 'opens' file)
172 error = exec_check_permissions(imgp);
174 VOP_UNLOCK(imgp->vp, 0, td);
175 goto exec_fail_dealloc;
178 error = exec_map_first_page(imgp);
179 VOP_UNLOCK(imgp->vp, 0, td);
181 goto exec_fail_dealloc;
184 * If the current process has a special image activator it
185 * wants to try first, call it. For example, emulating shell
186 * scripts differently.
189 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
190 error = img_first(imgp);
193 * Loop through the list of image activators, calling each one.
194 * An activator returns -1 if there is no match, 0 on success,
195 * and an error otherwise.
197 for (i = 0; error == -1 && execsw[i]; ++i) {
198 if (execsw[i]->ex_imgact == NULL ||
199 execsw[i]->ex_imgact == img_first) {
202 error = (*execsw[i]->ex_imgact)(imgp);
208 goto exec_fail_dealloc;
212 * Special interpreter operation, cleanup and loop up to try to
213 * activate the interpreter.
215 if (imgp->interpreted) {
216 exec_unmap_first_page(imgp);
217 /* free name buffer and old vnode */
218 NDFREE(ndp, NDF_ONLY_PNBUF);
220 /* set new name to that of the interpreter */
221 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
222 UIO_SYSSPACE, imgp->interpreter_name, td);
227 * Copy out strings (args and env) and initialize stack base
229 stack_base = exec_copyout_strings(imgp);
230 p->p_vmspace->vm_minsaddr = (char *)stack_base;
233 * If custom stack fixup routine present for this process
234 * let it do the stack setup.
235 * Else stuff argument count as first item on stack
237 if (p->p_sysent->sv_fixup)
238 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
240 suword(--stack_base, imgp->argc);
243 * For security and other reasons, the file descriptor table cannot
244 * be shared after an exec.
246 if (p->p_fd->fd_refcnt > 1) {
247 struct filedesc *tmp;
255 * For security and other reasons, signal handlers cannot
256 * be shared after an exec. The new proces gets a copy of the old
257 * handlers. In execsigs(), the new process will have its signals
260 if (p->p_procsig->ps_refcnt > 1) {
261 struct procsig *newprocsig;
263 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
264 M_SUBPROC, M_WAITOK);
265 bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
266 p->p_procsig->ps_refcnt--;
267 p->p_procsig = newprocsig;
268 p->p_procsig->ps_refcnt = 1;
269 if (p->p_sigacts == &p->p_addr->u_sigacts)
270 panic("shared procsig but private sigacts?");
272 p->p_addr->u_sigacts = *p->p_sigacts;
273 p->p_sigacts = &p->p_addr->u_sigacts;
279 /* close files on exec */
282 /* reset caught signals */
285 /* name this process - nameiexec(p, ndp) */
286 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
287 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
291 * mark as execed, wakeup the process that vforked (if any) and tell
292 * it that it now has its own resources back
295 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
296 p->p_flag &= ~P_PPWAIT;
297 wakeup((caddr_t)p->p_pptr);
301 * Implement image setuid/setgid.
303 * Don't honor setuid/setgid if the filesystem prohibits it or if
304 * the process is being traced.
306 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
307 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
308 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
309 (p->p_flag & P_TRACED) == 0) {
311 * Turn off syscall tracing for set-id programs, except for
312 * root. Record any set-id flags first to make sure that
313 * we do not regain any tracing during a possible block.
316 if (p->p_tracep && suser(td)) {
319 if ((vtmp = p->p_tracep) != NULL) {
325 /* Close any file descriptors 0..2 that reference procfs */
327 /* Make sure file descriptors 0..2 are in use. */
328 error = fdcheckstd(p);
330 goto exec_fail_dealloc;
332 * Set the new credentials.
335 if (attr.va_mode & VSUID)
336 change_euid(attr.va_uid);
337 if (attr.va_mode & VSGID)
338 p->p_ucred->cr_gid = attr.va_gid;
340 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
341 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
342 p->p_flag &= ~P_SUGID;
346 * Implement correct POSIX saved-id behavior.
348 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
349 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
351 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
352 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
356 * Store the vp for use in procfs
358 if (p->p_textvp) /* release old reference */
361 p->p_textvp = ndp->ni_vp;
364 * Notify others that we exec'd, and clear the P_INEXEC flag
365 * as we're now a bona fide freshly-execed process.
367 KNOTE(&p->p_klist, NOTE_EXEC);
368 p->p_flag &= ~P_INEXEC;
371 * If tracing the process, trap to debugger so breakpoints
372 * can be set before the program executes.
374 STOPEVENT(p, S_EXEC, 0);
376 if (p->p_flag & P_TRACED)
379 /* clear "fork but no exec" flag, as we _are_ execing */
380 p->p_acflag &= ~AFORK;
382 /* Set values passed into the program in registers. */
383 setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
387 * The syscall result is returned in registers to the new program.
388 * Linux will register %edx as an atexit function and we must be
389 * sure to set it to 0. XXX
391 uap->sysmsg_result64 = 0;
393 /* Free any previous argument cache */
394 if (p->p_args && --p->p_args->ar_ref == 0)
395 FREE(p->p_args, M_PARGS);
398 /* Cache arguments if they fit inside our allowance */
399 i = imgp->endargs - imgp->stringbase;
400 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
401 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
403 p->p_args->ar_ref = 1;
404 p->p_args->ar_length = i;
405 bcopy(imgp->stringbase, p->p_args->ar_args, i);
411 * free various allocated resources
414 exec_unmap_first_page(imgp);
416 if (imgp->stringbase != NULL)
417 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
418 ARG_MAX + PAGE_SIZE);
421 NDFREE(ndp, NDF_ONLY_PNBUF);
429 /* we're done here, clear P_INEXEC */
430 p->p_flag &= ~P_INEXEC;
431 if (imgp->vmspace_destroyed) {
432 /* sorry, no more process anymore. exit gracefully */
433 exit1(W_EXITCODE(0, SIGABRT));
442 exec_map_first_page(struct image_params *imgp)
446 vm_page_t ma[VM_INITIAL_PAGEIN];
450 if (imgp->firstpage) {
451 exec_unmap_first_page(imgp);
454 VOP_GETVOBJECT(imgp->vp, &object);
457 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
459 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
460 initial_pagein = VM_INITIAL_PAGEIN;
461 if (initial_pagein > object->size)
462 initial_pagein = object->size;
463 for (i = 1; i < initial_pagein; i++) {
464 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
465 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
471 ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
478 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
479 ma[0] = vm_page_lookup(object, 0);
481 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
483 vm_page_protect(ma[0], VM_PROT_NONE);
492 vm_page_wakeup(ma[0]);
495 pmap_kenter((vm_offset_t) imgp->image_header, VM_PAGE_TO_PHYS(ma[0]));
496 imgp->firstpage = ma[0];
502 exec_unmap_first_page(imgp)
503 struct image_params *imgp;
505 if (imgp->firstpage) {
506 pmap_kremove((vm_offset_t) imgp->image_header);
507 vm_page_unwire(imgp->firstpage, 1);
508 imgp->firstpage = NULL;
513 * Destroy old address space, and allocate a new stack
514 * The new stack is only SGROWSIZ large because it is grown
515 * automatically in trap.c.
518 exec_new_vmspace(imgp)
519 struct image_params *imgp;
522 struct vmspace *vmspace = imgp->proc->p_vmspace;
523 vm_offset_t stack_addr = USRSTACK - maxssiz;
524 vm_map_t map = &vmspace->vm_map;
526 imgp->vmspace_destroyed = 1;
529 * Prevent a pending AIO from modifying the new address space.
531 aio_proc_rundown(imgp->proc);
534 * Blow away entire process VM, if address space not shared,
535 * otherwise, create a new VM space so that other threads are
538 if (vmspace->vm_refcnt == 1) {
540 pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
541 vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
543 vmspace_exec(imgp->proc);
544 vmspace = imgp->proc->p_vmspace;
545 map = &vmspace->vm_map;
548 /* Allocate a new stack */
549 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
550 VM_PROT_ALL, VM_PROT_ALL, 0);
554 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
555 * VM_STACK case, but they are still used to monitor the size of the
556 * process stack so we can check the stack rlimit.
558 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
559 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
565 * Copy out argument and environment strings from the old process
566 * address space into the temporary string buffer.
569 exec_extract_strings(imgp)
570 struct image_params *imgp;
578 * extract arguments first
581 argv = imgp->uap->argv;
584 argp = (caddr_t) (intptr_t) fuword(argv);
585 if (argp == (caddr_t) -1)
593 if (argp == (caddr_t) -1)
595 if ((error = copyinstr(argp, imgp->stringp,
596 imgp->stringspace, &length))) {
597 if (error == ENAMETOOLONG)
601 imgp->stringspace -= length;
602 imgp->stringp += length;
604 } while ((argp = (caddr_t) (intptr_t) fuword(argv++)));
608 imgp->endargs = imgp->stringp;
611 * extract environment strings
614 envv = imgp->uap->envv;
617 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
618 if (envp == (caddr_t) -1)
620 if ((error = copyinstr(envp, imgp->stringp,
621 imgp->stringspace, &length))) {
622 if (error == ENAMETOOLONG)
626 imgp->stringspace -= length;
627 imgp->stringp += length;
636 * Copy strings out to the new process address space, constructing
637 * new arg and env vector tables. Return a pointer to the base
638 * so that it can be used as the initial stack pointer.
641 exec_copyout_strings(imgp)
642 struct image_params *imgp;
646 char *stringp, *destp;
647 register_t *stack_base;
648 struct ps_strings *arginfo;
652 * Calculate string base and vector table pointers.
653 * Also deal with signal trampoline code for this exec type.
655 arginfo = (struct ps_strings *)PS_STRINGS;
656 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
657 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
658 roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
664 copyout(imgp->proc->p_sysent->sv_sigcode,
665 ((caddr_t)arginfo - szsigcode), szsigcode);
668 * If we have a valid auxargs ptr, prepare some room
673 * The '+ 2' is for the null pointers at the end of each of the
674 * arg and env vector sets, and 'AT_COUNT*2' is room for the
677 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
678 AT_COUNT*2) * sizeof(char*));
681 * The '+ 2' is for the null pointers at the end of each of the
682 * arg and env vector sets
685 (destp - (imgp->argc + imgp->envc + 2) * sizeof(char*));
688 * vectp also becomes our initial stack base
690 stack_base = (register_t *)vectp;
692 stringp = imgp->stringbase;
697 * Copy out strings - arguments and environment.
699 copyout(stringp, destp, ARG_MAX - imgp->stringspace);
702 * Fill in "ps_strings" struct for ps, w, etc.
704 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
705 suword(&arginfo->ps_nargvstr, argc);
708 * Fill in argument portion of vector table.
710 for (; argc > 0; --argc) {
711 suword(vectp++, (long)(intptr_t)destp);
712 while (*stringp++ != 0)
717 /* a null vector table pointer seperates the argp's from the envp's */
720 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
721 suword(&arginfo->ps_nenvstr, envc);
724 * Fill in environment portion of vector table.
726 for (; envc > 0; --envc) {
727 suword(vectp++, (long)(intptr_t)destp);
728 while (*stringp++ != 0)
733 /* end of vector table is a null pointer */
740 * Check permissions of file to execute.
741 * Return 0 for success or error code on failure.
744 exec_check_permissions(imgp)
745 struct image_params *imgp;
747 struct proc *p = imgp->proc;
748 struct vnode *vp = imgp->vp;
749 struct vattr *attr = imgp->attr;
750 struct thread *td = p->p_thread;
753 /* Get file attributes */
754 error = VOP_GETATTR(vp, attr, td);
759 * 1) Check if file execution is disabled for the filesystem that this
761 * 2) Insure that at least one execute bit is on - otherwise root
762 * will always succeed, and we don't want to happen unless the
763 * file really is executable.
764 * 3) Insure that the file is a regular file.
766 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
767 ((attr->va_mode & 0111) == 0) ||
768 (attr->va_type != VREG)) {
773 * Zero length files can't be exec'd
775 if (attr->va_size == 0)
779 * Check for execute permission to file based on current credentials.
781 error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
786 * Check number of open-for-writes on the file and deny execution
789 if (vp->v_writecount)
793 * Call filesystem specific open routine (which does nothing in the
796 error = VOP_OPEN(vp, FREAD, p->p_ucred, td);
804 * Exec handler registration
807 exec_register(execsw_arg)
808 const struct execsw *execsw_arg;
810 const struct execsw **es, **xs, **newexecsw;
811 int count = 2; /* New slot and trailing NULL */
814 for (es = execsw; *es; es++)
816 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
817 if (newexecsw == NULL)
821 for (es = execsw; *es; es++)
826 free(execsw, M_TEMP);
832 exec_unregister(execsw_arg)
833 const struct execsw *execsw_arg;
835 const struct execsw **es, **xs, **newexecsw;
839 panic("unregister with no handlers left?\n");
841 for (es = execsw; *es; es++) {
842 if (*es == execsw_arg)
847 for (es = execsw; *es; es++)
848 if (*es != execsw_arg)
850 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
851 if (newexecsw == NULL)
854 for (es = execsw; *es; es++)
855 if (*es != execsw_arg)
859 free(execsw, M_TEMP);