xref: /netbsd-src/sys/kern/exec_subr.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: exec_subr.c,v 1.48 2006/05/20 15:45:37 elad 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 1.48 2006/05/20 15:45:37 elad 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/malloc.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 = malloc(evsp->evs_cnt * sizeof(struct exec_vmcmd),
117 	    M_EXEC, M_WAITOK);
118 
119 	/* free the old struct, if there was one, and record the new one */
120 	if (ocnt) {
121 		memcpy(nvcp, evsp->evs_cmds,
122 		    (ocnt * sizeof(struct exec_vmcmd)));
123 		free(evsp->evs_cmds, M_EXEC);
124 	}
125 	evsp->evs_cmds = nvcp;
126 }
127 
128 void
129 kill_vmcmds(struct exec_vmcmd_set *evsp)
130 {
131 	struct exec_vmcmd *vcp;
132 	u_int i;
133 
134 	VMCMD_EVCNT_INCR(kills);
135 
136 	if (evsp->evs_cnt == 0)
137 		return;
138 
139 	for (i = 0; i < evsp->evs_used; i++) {
140 		vcp = &evsp->evs_cmds[i];
141 		if (vcp->ev_vp != NULL)
142 			vrele(vcp->ev_vp);
143 	}
144 	evsp->evs_used = evsp->evs_cnt = 0;
145 	free(evsp->evs_cmds, M_EXEC);
146 }
147 
148 /*
149  * vmcmd_map_pagedvn():
150  *	handle vmcmd which specifies that a vnode should be mmap'd.
151  *	appropriate for handling demand-paged text and data segments.
152  */
153 
154 int
155 vmcmd_map_pagedvn(struct lwp *l, struct exec_vmcmd *cmd)
156 {
157 	struct uvm_object *uobj;
158 	struct proc *p = l->l_proc;
159 	int error;
160 	vm_prot_t prot, maxprot;
161 
162 	KASSERT(cmd->ev_vp->v_flag & VTEXT);
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 	/*
178 	 * first, attach to the object
179 	 */
180 
181         uobj = uvn_attach(cmd->ev_vp, VM_PROT_READ|VM_PROT_EXECUTE);
182         if (uobj == NULL)
183                 return(ENOMEM);
184 	VREF(cmd->ev_vp);
185 
186 	prot = cmd->ev_prot;
187 	maxprot = UVM_PROT_ALL;
188 #ifdef PAX_MPROTECT
189 	pax_mprotect(l, &prot, &maxprot);
190 #endif /* PAX_MPROTECT */
191 
192 	/*
193 	 * do the map
194 	 */
195 
196 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len,
197 		uobj, cmd->ev_offset, 0,
198 		UVM_MAPFLAG(prot, maxprot, UVM_INH_COPY,
199 			UVM_ADV_NORMAL, UVM_FLAG_COPYONW|UVM_FLAG_FIXED));
200 	if (error) {
201 		uobj->pgops->pgo_detach(uobj);
202 	}
203 	return error;
204 }
205 
206 /*
207  * vmcmd_map_readvn():
208  *	handle vmcmd which specifies that a vnode should be read from.
209  *	appropriate for non-demand-paged text/data segments, i.e. impure
210  *	objects (a la OMAGIC and NMAGIC).
211  */
212 int
213 vmcmd_map_readvn(struct lwp *l, struct exec_vmcmd *cmd)
214 {
215 	struct proc *p = l->l_proc;
216 	int error;
217 	long diff;
218 
219 	if (cmd->ev_len == 0)
220 		return 0;
221 
222 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
223 	cmd->ev_addr -= diff;			/* required by uvm_map */
224 	cmd->ev_offset -= diff;
225 	cmd->ev_len += diff;
226 
227 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
228 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
229 			UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_COPY,
230 			UVM_ADV_NORMAL,
231 			UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW));
232 
233 	if (error)
234 		return error;
235 
236 	return vmcmd_readvn(l, cmd);
237 }
238 
239 int
240 vmcmd_readvn(struct lwp *l, struct exec_vmcmd *cmd)
241 {
242 	struct proc *p = l->l_proc;
243 	int error;
244 	vm_prot_t prot, maxprot;
245 
246 	error = vn_rdwr(UIO_READ, cmd->ev_vp, (caddr_t)cmd->ev_addr,
247 	    cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT,
248 	    p->p_cred, NULL, l);
249 	if (error)
250 		return error;
251 
252 	prot = cmd->ev_prot;
253 	maxprot = VM_PROT_ALL;
254 #ifdef PAX_MPROTECT
255 	pax_mprotect(l, &prot, &maxprot);
256 #endif /* PAX_MPROTECT */
257 
258 #ifdef PMAP_NEED_PROCWR
259 	/*
260 	 * we had to write the process, make sure the pages are synched
261 	 * with the instruction cache.
262 	 */
263 	if (prot & VM_PROT_EXECUTE)
264 		pmap_procwr(p, cmd->ev_addr, cmd->ev_len);
265 #endif
266 
267 	/*
268 	 * we had to map in the area at PROT_ALL so that vn_rdwr()
269 	 * could write to it.   however, the caller seems to want
270 	 * it mapped read-only, so now we are going to have to call
271 	 * uvm_map_protect() to fix up the protection.  ICK.
272 	 */
273 	if (maxprot != VM_PROT_ALL) {
274 		error = uvm_map_protect(&p->p_vmspace->vm_map,
275 				trunc_page(cmd->ev_addr),
276 				round_page(cmd->ev_addr + cmd->ev_len),
277 				maxprot, TRUE);
278 		if (error)
279 			return (error);
280 	}
281 
282 	if (prot != maxprot) {
283 		error = uvm_map_protect(&p->p_vmspace->vm_map,
284 				trunc_page(cmd->ev_addr),
285 				round_page(cmd->ev_addr + cmd->ev_len),
286 				prot, FALSE);
287 		if (error)
288 			return (error);
289 	}
290 
291 	return 0;
292 }
293 
294 /*
295  * vmcmd_map_zero():
296  *	handle vmcmd which specifies a zero-filled address space region.  The
297  *	address range must be first allocated, then protected appropriately.
298  */
299 
300 int
301 vmcmd_map_zero(struct lwp *l, struct exec_vmcmd *cmd)
302 {
303 	struct proc *p = l->l_proc;
304 	int error;
305 	long diff;
306 	vm_prot_t prot, maxprot;
307 
308 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
309 	cmd->ev_addr -= diff;			/* required by uvm_map */
310 	cmd->ev_len += diff;
311 
312 	prot = cmd->ev_prot;
313 	maxprot = UVM_PROT_ALL;
314 #ifdef PAX_MPROTECT
315 	pax_mprotect(l, &prot, &maxprot);
316 #endif /* PAX_MPROTECT */
317 
318 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
319 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
320 			UVM_MAPFLAG(prot, maxprot, UVM_INH_COPY,
321 			UVM_ADV_NORMAL,
322 			UVM_FLAG_FIXED|UVM_FLAG_COPYONW));
323 	return error;
324 }
325 
326 /*
327  * exec_read_from():
328  *
329  *	Read from vnode into buffer at offset.
330  */
331 int
332 exec_read_from(struct lwp *l, struct vnode *vp, u_long off, void *bf,
333     size_t size)
334 {
335 	int error;
336 	size_t resid;
337 
338 	if ((error = vn_rdwr(UIO_READ, vp, bf, size, off, UIO_SYSSPACE,
339 	    0, l->l_proc->p_cred, &resid, NULL)) != 0)
340 		return error;
341 	/*
342 	 * See if we got all of it
343 	 */
344 	if (resid != 0)
345 		return ENOEXEC;
346 	return 0;
347 }
348 
349 /*
350  * exec_setup_stack(): Set up the stack segment for an elf
351  * executable.
352  *
353  * Note that the ep_ssize parameter must be set to be the current stack
354  * limit; this is adjusted in the body of execve() to yield the
355  * appropriate stack segment usage once the argument length is
356  * calculated.
357  *
358  * This function returns an int for uniformity with other (future) formats'
359  * stack setup functions.  They might have errors to return.
360  */
361 
362 int
363 exec_setup_stack(struct lwp *l, struct exec_package *epp)
364 {
365 	u_long max_stack_size;
366 	u_long access_linear_min, access_size;
367 	u_long noaccess_linear_min, noaccess_size;
368 
369 #ifndef	USRSTACK32
370 #define USRSTACK32	(0x00000000ffffffffL&~PGOFSET)
371 #endif
372 
373 	if (epp->ep_flags & EXEC_32) {
374 		epp->ep_minsaddr = USRSTACK32;
375 		max_stack_size = MAXSSIZ;
376 	} else {
377 		epp->ep_minsaddr = USRSTACK;
378 		max_stack_size = MAXSSIZ;
379 	}
380 	epp->ep_maxsaddr = (u_long)STACK_GROW(epp->ep_minsaddr,
381 		max_stack_size);
382 	epp->ep_ssize = l->l_proc->p_rlimit[RLIMIT_STACK].rlim_cur;
383 
384 	/*
385 	 * set up commands for stack.  note that this takes *two*, one to
386 	 * map the part of the stack which we can access, and one to map
387 	 * the part which we can't.
388 	 *
389 	 * arguably, it could be made into one, but that would require the
390 	 * addition of another mapping proc, which is unnecessary
391 	 */
392 	access_size = epp->ep_ssize;
393 	access_linear_min = (u_long)STACK_ALLOC(epp->ep_minsaddr, access_size);
394 	noaccess_size = max_stack_size - access_size;
395 	noaccess_linear_min = (u_long)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
396 	    access_size), noaccess_size);
397 	if (noaccess_size > 0) {
398 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, noaccess_size,
399 		    noaccess_linear_min, NULL, 0, VM_PROT_NONE);
400 	}
401 	KASSERT(access_size > 0);
402 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, access_size,
403 	    access_linear_min, NULL, 0, VM_PROT_READ | VM_PROT_WRITE);
404 
405 	return 0;
406 }
407