xref: /netbsd-src/sys/kern/exec_subr.c (revision ce099b40997c43048fb78bd578195f81d2456523)
1 /*	$NetBSD: exec_subr.c,v 1.60 2008/01/28 20:09:06 ad Exp $	*/
2 
3 /*
4  * Copyright (c) 1993, 1994, 1996 Christopher G. Demetriou
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *      This product includes software developed by Christopher G. Demetriou.
18  * 4. The name of the author may not be used to endorse or promote products
19  *    derived from this software without specific prior written permission
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: exec_subr.c,v 1.60 2008/01/28 20:09:06 ad Exp $");
35 
36 #include "opt_pax.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/kmem.h>
42 #include <sys/vnode.h>
43 #include <sys/filedesc.h>
44 #include <sys/exec.h>
45 #include <sys/mman.h>
46 #include <sys/resourcevar.h>
47 #include <sys/device.h>
48 
49 #ifdef PAX_MPROTECT
50 #include <sys/pax.h>
51 #endif /* PAX_MPROTECT */
52 
53 #include <uvm/uvm.h>
54 
55 #define	VMCMD_EVCNT_DECL(name)					\
56 static struct evcnt vmcmd_ev_##name =				\
57     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "vmcmd", #name);	\
58 EVCNT_ATTACH_STATIC(vmcmd_ev_##name)
59 
60 #define	VMCMD_EVCNT_INCR(name)					\
61     vmcmd_ev_##name.ev_count++
62 
63 VMCMD_EVCNT_DECL(calls);
64 VMCMD_EVCNT_DECL(extends);
65 VMCMD_EVCNT_DECL(kills);
66 
67 /*
68  * new_vmcmd():
69  *	create a new vmcmd structure and fill in its fields based
70  *	on function call arguments.  make sure objects ref'd by
71  *	the vmcmd are 'held'.
72  */
73 
74 void
75 new_vmcmd(struct exec_vmcmd_set *evsp,
76     int (*proc)(struct lwp * l, struct exec_vmcmd *),
77     u_long len, u_long addr, struct vnode *vp, u_long offset,
78     u_int prot, int flags)
79 {
80 	struct exec_vmcmd    *vcp;
81 
82 	VMCMD_EVCNT_INCR(calls);
83 
84 	if (evsp->evs_used >= evsp->evs_cnt)
85 		vmcmdset_extend(evsp);
86 	vcp = &evsp->evs_cmds[evsp->evs_used++];
87 	vcp->ev_proc = proc;
88 	vcp->ev_len = len;
89 	vcp->ev_addr = addr;
90 	if ((vcp->ev_vp = vp) != NULL)
91 		vref(vp);
92 	vcp->ev_offset = offset;
93 	vcp->ev_prot = prot;
94 	vcp->ev_flags = flags;
95 }
96 
97 void
98 vmcmdset_extend(struct exec_vmcmd_set *evsp)
99 {
100 	struct exec_vmcmd *nvcp;
101 	u_int ocnt;
102 
103 #ifdef DIAGNOSTIC
104 	if (evsp->evs_used < evsp->evs_cnt)
105 		panic("vmcmdset_extend: not necessary");
106 #endif
107 
108 	/* figure out number of entries in new set */
109 	if ((ocnt = evsp->evs_cnt) != 0) {
110 		evsp->evs_cnt += ocnt;
111 		VMCMD_EVCNT_INCR(extends);
112 	} else
113 		evsp->evs_cnt = EXEC_DEFAULT_VMCMD_SETSIZE;
114 
115 	/* allocate it */
116 	nvcp = kmem_alloc(evsp->evs_cnt * sizeof(struct exec_vmcmd), KM_SLEEP);
117 
118 	/* free the old struct, if there was one, and record the new one */
119 	if (ocnt) {
120 		memcpy(nvcp, evsp->evs_cmds,
121 		    (ocnt * sizeof(struct exec_vmcmd)));
122 		kmem_free(evsp->evs_cmds, ocnt * sizeof(struct exec_vmcmd));
123 	}
124 	evsp->evs_cmds = nvcp;
125 }
126 
127 void
128 kill_vmcmds(struct exec_vmcmd_set *evsp)
129 {
130 	struct exec_vmcmd *vcp;
131 	u_int i;
132 
133 	VMCMD_EVCNT_INCR(kills);
134 
135 	if (evsp->evs_cnt == 0)
136 		return;
137 
138 	for (i = 0; i < evsp->evs_used; i++) {
139 		vcp = &evsp->evs_cmds[i];
140 		if (vcp->ev_vp != NULL)
141 			vrele(vcp->ev_vp);
142 	}
143 	kmem_free(evsp->evs_cmds, evsp->evs_cnt * sizeof(struct exec_vmcmd));
144 	evsp->evs_used = evsp->evs_cnt = 0;
145 }
146 
147 /*
148  * vmcmd_map_pagedvn():
149  *	handle vmcmd which specifies that a vnode should be mmap'd.
150  *	appropriate for handling demand-paged text and data segments.
151  */
152 
153 int
154 vmcmd_map_pagedvn(struct lwp *l, struct exec_vmcmd *cmd)
155 {
156 	struct uvm_object *uobj;
157 	struct vnode *vp = cmd->ev_vp;
158 	struct proc *p = l->l_proc;
159 	int error;
160 	vm_prot_t prot, maxprot;
161 
162 	KASSERT(vp->v_iflag & VI_TEXT);
163 
164 	/*
165 	 * map the vnode in using uvm_map.
166 	 */
167 
168         if (cmd->ev_len == 0)
169                 return(0);
170         if (cmd->ev_offset & PAGE_MASK)
171                 return(EINVAL);
172 	if (cmd->ev_addr & PAGE_MASK)
173 		return(EINVAL);
174 	if (cmd->ev_len & PAGE_MASK)
175 		return(EINVAL);
176 
177 	prot = cmd->ev_prot;
178 	maxprot = UVM_PROT_ALL;
179 #ifdef PAX_MPROTECT
180 	pax_mprotect(l, &prot, &maxprot);
181 #endif /* PAX_MPROTECT */
182 
183 	/*
184 	 * check the file system's opinion about mmapping the file
185 	 */
186 
187 	error = VOP_MMAP(vp, prot, l->l_cred);
188 	if (error)
189 		return error;
190 
191 	if ((vp->v_vflag & VV_MAPPED) == 0) {
192 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
193 		mutex_enter(&vp->v_interlock);
194 		vp->v_vflag |= VV_MAPPED;
195 		vp->v_iflag |= VI_MAPPED;
196 		mutex_exit(&vp->v_interlock);
197 		VOP_UNLOCK(vp, 0);
198 	}
199 
200 	/*
201 	 * do the map, reference the object for this map entry
202 	 */
203 	uobj = &vp->v_uobj;
204 	vref(vp);
205 
206 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len,
207 		uobj, cmd->ev_offset, 0,
208 		UVM_MAPFLAG(prot, maxprot, UVM_INH_COPY,
209 			UVM_ADV_NORMAL, UVM_FLAG_COPYONW|UVM_FLAG_FIXED));
210 	if (error) {
211 		uobj->pgops->pgo_detach(uobj);
212 	}
213 	return error;
214 }
215 
216 /*
217  * vmcmd_map_readvn():
218  *	handle vmcmd which specifies that a vnode should be read from.
219  *	appropriate for non-demand-paged text/data segments, i.e. impure
220  *	objects (a la OMAGIC and NMAGIC).
221  */
222 int
223 vmcmd_map_readvn(struct lwp *l, struct exec_vmcmd *cmd)
224 {
225 	struct proc *p = l->l_proc;
226 	int error;
227 	long diff;
228 
229 	if (cmd->ev_len == 0)
230 		return 0;
231 
232 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
233 	cmd->ev_addr -= diff;			/* required by uvm_map */
234 	cmd->ev_offset -= diff;
235 	cmd->ev_len += diff;
236 
237 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
238 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
239 			UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_COPY,
240 			UVM_ADV_NORMAL,
241 			UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW));
242 
243 	if (error)
244 		return error;
245 
246 	return vmcmd_readvn(l, cmd);
247 }
248 
249 int
250 vmcmd_readvn(struct lwp *l, struct exec_vmcmd *cmd)
251 {
252 	struct proc *p = l->l_proc;
253 	int error;
254 	vm_prot_t prot, maxprot;
255 
256 	error = vn_rdwr(UIO_READ, cmd->ev_vp, (void *)cmd->ev_addr,
257 	    cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT,
258 	    l->l_cred, NULL, l);
259 	if (error)
260 		return error;
261 
262 	prot = cmd->ev_prot;
263 	maxprot = VM_PROT_ALL;
264 #ifdef PAX_MPROTECT
265 	pax_mprotect(l, &prot, &maxprot);
266 #endif /* PAX_MPROTECT */
267 
268 #ifdef PMAP_NEED_PROCWR
269 	/*
270 	 * we had to write the process, make sure the pages are synched
271 	 * with the instruction cache.
272 	 */
273 	if (prot & VM_PROT_EXECUTE)
274 		pmap_procwr(p, cmd->ev_addr, cmd->ev_len);
275 #endif
276 
277 	/*
278 	 * we had to map in the area at PROT_ALL so that vn_rdwr()
279 	 * could write to it.   however, the caller seems to want
280 	 * it mapped read-only, so now we are going to have to call
281 	 * uvm_map_protect() to fix up the protection.  ICK.
282 	 */
283 	if (maxprot != VM_PROT_ALL) {
284 		error = uvm_map_protect(&p->p_vmspace->vm_map,
285 				trunc_page(cmd->ev_addr),
286 				round_page(cmd->ev_addr + cmd->ev_len),
287 				maxprot, true);
288 		if (error)
289 			return (error);
290 	}
291 
292 	if (prot != maxprot) {
293 		error = uvm_map_protect(&p->p_vmspace->vm_map,
294 				trunc_page(cmd->ev_addr),
295 				round_page(cmd->ev_addr + cmd->ev_len),
296 				prot, false);
297 		if (error)
298 			return (error);
299 	}
300 
301 	return 0;
302 }
303 
304 /*
305  * vmcmd_map_zero():
306  *	handle vmcmd which specifies a zero-filled address space region.  The
307  *	address range must be first allocated, then protected appropriately.
308  */
309 
310 int
311 vmcmd_map_zero(struct lwp *l, struct exec_vmcmd *cmd)
312 {
313 	struct proc *p = l->l_proc;
314 	int error;
315 	long diff;
316 	vm_prot_t prot, maxprot;
317 
318 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
319 	cmd->ev_addr -= diff;			/* required by uvm_map */
320 	cmd->ev_len += diff;
321 
322 	prot = cmd->ev_prot;
323 	maxprot = UVM_PROT_ALL;
324 #ifdef PAX_MPROTECT
325 	pax_mprotect(l, &prot, &maxprot);
326 #endif /* PAX_MPROTECT */
327 
328 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
329 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
330 			UVM_MAPFLAG(prot, maxprot, UVM_INH_COPY,
331 			UVM_ADV_NORMAL,
332 			UVM_FLAG_FIXED|UVM_FLAG_COPYONW));
333 	return error;
334 }
335 
336 /*
337  * exec_read_from():
338  *
339  *	Read from vnode into buffer at offset.
340  */
341 int
342 exec_read_from(struct lwp *l, struct vnode *vp, u_long off, void *bf,
343     size_t size)
344 {
345 	int error;
346 	size_t resid;
347 
348 	if ((error = vn_rdwr(UIO_READ, vp, bf, size, off, UIO_SYSSPACE,
349 	    0, l->l_cred, &resid, NULL)) != 0)
350 		return error;
351 	/*
352 	 * See if we got all of it
353 	 */
354 	if (resid != 0)
355 		return ENOEXEC;
356 	return 0;
357 }
358 
359 /*
360  * exec_setup_stack(): Set up the stack segment for an elf
361  * executable.
362  *
363  * Note that the ep_ssize parameter must be set to be the current stack
364  * limit; this is adjusted in the body of execve() to yield the
365  * appropriate stack segment usage once the argument length is
366  * calculated.
367  *
368  * This function returns an int for uniformity with other (future) formats'
369  * stack setup functions.  They might have errors to return.
370  */
371 
372 int
373 exec_setup_stack(struct lwp *l, struct exec_package *epp)
374 {
375 	u_long max_stack_size;
376 	u_long access_linear_min, access_size;
377 	u_long noaccess_linear_min, noaccess_size;
378 
379 #ifndef	USRSTACK32
380 #define USRSTACK32	(0x00000000ffffffffL&~PGOFSET)
381 #endif
382 
383 	if (epp->ep_flags & EXEC_32) {
384 		epp->ep_minsaddr = USRSTACK32;
385 		max_stack_size = MAXSSIZ;
386 	} else {
387 		epp->ep_minsaddr = USRSTACK;
388 		max_stack_size = MAXSSIZ;
389 	}
390 
391 #ifdef PAX_ASLR
392 	pax_aslr_stack(l, epp, &max_stack_size);
393 #endif /* PAX_ASLR */
394 
395 	l->l_proc->p_stackbase = epp->ep_minsaddr;
396 
397 	epp->ep_maxsaddr = (u_long)STACK_GROW(epp->ep_minsaddr,
398 		max_stack_size);
399 	epp->ep_ssize = l->l_proc->p_rlimit[RLIMIT_STACK].rlim_cur;
400 
401 	/*
402 	 * set up commands for stack.  note that this takes *two*, one to
403 	 * map the part of the stack which we can access, and one to map
404 	 * the part which we can't.
405 	 *
406 	 * arguably, it could be made into one, but that would require the
407 	 * addition of another mapping proc, which is unnecessary
408 	 */
409 	access_size = epp->ep_ssize;
410 	access_linear_min = (u_long)STACK_ALLOC(epp->ep_minsaddr, access_size);
411 	noaccess_size = max_stack_size - access_size;
412 	noaccess_linear_min = (u_long)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
413 	    access_size), noaccess_size);
414 	if (noaccess_size > 0) {
415 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, noaccess_size,
416 		    noaccess_linear_min, NULL, 0, VM_PROT_NONE);
417 	}
418 	KASSERT(access_size > 0);
419 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, access_size,
420 	    access_linear_min, NULL, 0, VM_PROT_READ | VM_PROT_WRITE);
421 
422 	return 0;
423 }
424