xref: /netbsd-src/sys/kern/exec_subr.c (revision 9c1da17e908379b8a470f1117a6395bd6a0ca559)
1 /*	$NetBSD: exec_subr.c,v 1.45 2005/07/06 23:08:57 thorpej 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 1.45 2005/07/06 23:08:57 thorpej Exp $");
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/malloc.h>
40 #include <sys/vnode.h>
41 #include <sys/filedesc.h>
42 #include <sys/exec.h>
43 #include <sys/mman.h>
44 #include <sys/resourcevar.h>
45 #include <sys/device.h>
46 
47 #include <uvm/uvm.h>
48 
49 #define	VMCMD_EVCNT_DECL(name)					\
50 static struct evcnt vmcmd_ev_##name =				\
51     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "vmcmd", #name);	\
52 EVCNT_ATTACH_STATIC(vmcmd_ev_##name)
53 
54 #define	VMCMD_EVCNT_INCR(name)					\
55     vmcmd_ev_##name.ev_count++
56 
57 VMCMD_EVCNT_DECL(calls);
58 VMCMD_EVCNT_DECL(extends);
59 VMCMD_EVCNT_DECL(kills);
60 
61 /*
62  * new_vmcmd():
63  *	create a new vmcmd structure and fill in its fields based
64  *	on function call arguments.  make sure objects ref'd by
65  *	the vmcmd are 'held'.
66  */
67 
68 void
69 new_vmcmd(struct exec_vmcmd_set *evsp,
70     int (*proc)(struct proc * p, struct exec_vmcmd *),
71     u_long len, u_long addr, struct vnode *vp, u_long offset,
72     u_int prot, int flags)
73 {
74 	struct exec_vmcmd    *vcp;
75 
76 	VMCMD_EVCNT_INCR(calls);
77 
78 	if (evsp->evs_used >= evsp->evs_cnt)
79 		vmcmdset_extend(evsp);
80 	vcp = &evsp->evs_cmds[evsp->evs_used++];
81 	vcp->ev_proc = proc;
82 	vcp->ev_len = len;
83 	vcp->ev_addr = addr;
84 	if ((vcp->ev_vp = vp) != NULL)
85 		vref(vp);
86 	vcp->ev_offset = offset;
87 	vcp->ev_prot = prot;
88 	vcp->ev_flags = flags;
89 }
90 
91 void
92 vmcmdset_extend(struct exec_vmcmd_set *evsp)
93 {
94 	struct exec_vmcmd *nvcp;
95 	u_int ocnt;
96 
97 #ifdef DIAGNOSTIC
98 	if (evsp->evs_used < evsp->evs_cnt)
99 		panic("vmcmdset_extend: not necessary");
100 #endif
101 
102 	/* figure out number of entries in new set */
103 	if ((ocnt = evsp->evs_cnt) != 0) {
104 		evsp->evs_cnt += ocnt;
105 		VMCMD_EVCNT_INCR(extends);
106 	} else
107 		evsp->evs_cnt = EXEC_DEFAULT_VMCMD_SETSIZE;
108 
109 	/* allocate it */
110 	nvcp = malloc(evsp->evs_cnt * sizeof(struct exec_vmcmd),
111 	    M_EXEC, M_WAITOK);
112 
113 	/* free the old struct, if there was one, and record the new one */
114 	if (ocnt) {
115 		memcpy(nvcp, evsp->evs_cmds,
116 		    (ocnt * sizeof(struct exec_vmcmd)));
117 		free(evsp->evs_cmds, M_EXEC);
118 	}
119 	evsp->evs_cmds = nvcp;
120 }
121 
122 void
123 kill_vmcmds(struct exec_vmcmd_set *evsp)
124 {
125 	struct exec_vmcmd *vcp;
126 	u_int i;
127 
128 	VMCMD_EVCNT_INCR(kills);
129 
130 	if (evsp->evs_cnt == 0)
131 		return;
132 
133 	for (i = 0; i < evsp->evs_used; i++) {
134 		vcp = &evsp->evs_cmds[i];
135 		if (vcp->ev_vp != NULL)
136 			vrele(vcp->ev_vp);
137 	}
138 	evsp->evs_used = evsp->evs_cnt = 0;
139 	free(evsp->evs_cmds, M_EXEC);
140 }
141 
142 /*
143  * vmcmd_map_pagedvn():
144  *	handle vmcmd which specifies that a vnode should be mmap'd.
145  *	appropriate for handling demand-paged text and data segments.
146  */
147 
148 int
149 vmcmd_map_pagedvn(struct proc *p, struct exec_vmcmd *cmd)
150 {
151 	struct uvm_object *uobj;
152 	int error;
153 
154 	KASSERT(cmd->ev_vp->v_flag & VTEXT);
155 
156 	/*
157 	 * map the vnode in using uvm_map.
158 	 */
159 
160         if (cmd->ev_len == 0)
161                 return(0);
162         if (cmd->ev_offset & PAGE_MASK)
163                 return(EINVAL);
164 	if (cmd->ev_addr & PAGE_MASK)
165 		return(EINVAL);
166 	if (cmd->ev_len & PAGE_MASK)
167 		return(EINVAL);
168 
169 	/*
170 	 * first, attach to the object
171 	 */
172 
173         uobj = uvn_attach(cmd->ev_vp, VM_PROT_READ|VM_PROT_EXECUTE);
174         if (uobj == NULL)
175                 return(ENOMEM);
176 	VREF(cmd->ev_vp);
177 
178 	/*
179 	 * do the map
180 	 */
181 
182 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len,
183 		uobj, cmd->ev_offset, 0,
184 		UVM_MAPFLAG(cmd->ev_prot, VM_PROT_ALL, UVM_INH_COPY,
185 			UVM_ADV_NORMAL, UVM_FLAG_COPYONW|UVM_FLAG_FIXED));
186 	if (error) {
187 		uobj->pgops->pgo_detach(uobj);
188 	}
189 	return error;
190 }
191 
192 /*
193  * vmcmd_map_readvn():
194  *	handle vmcmd which specifies that a vnode should be read from.
195  *	appropriate for non-demand-paged text/data segments, i.e. impure
196  *	objects (a la OMAGIC and NMAGIC).
197  */
198 int
199 vmcmd_map_readvn(struct proc *p, struct exec_vmcmd *cmd)
200 {
201 	int error;
202 	long diff;
203 
204 	if (cmd->ev_len == 0)
205 		return 0;
206 
207 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
208 	cmd->ev_addr -= diff;			/* required by uvm_map */
209 	cmd->ev_offset -= diff;
210 	cmd->ev_len += diff;
211 
212 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
213 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
214 			UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_COPY,
215 			UVM_ADV_NORMAL,
216 			UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW));
217 
218 	if (error)
219 		return error;
220 
221 	return vmcmd_readvn(p, cmd);
222 }
223 
224 int
225 vmcmd_readvn(struct proc *p, struct exec_vmcmd *cmd)
226 {
227 	int error;
228 
229 	error = vn_rdwr(UIO_READ, cmd->ev_vp, (caddr_t)cmd->ev_addr,
230 	    cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT,
231 	    p->p_ucred, NULL, p);
232 	if (error)
233 		return error;
234 
235 #ifdef PMAP_NEED_PROCWR
236 	/*
237 	 * we had to write the process, make sure the pages are synched
238 	 * with the instruction cache.
239 	 */
240 	if (cmd->ev_prot & VM_PROT_EXECUTE)
241 		pmap_procwr(p, cmd->ev_addr, cmd->ev_len);
242 #endif
243 
244 	if (cmd->ev_prot != (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)) {
245 
246 		/*
247 		 * we had to map in the area at PROT_ALL so that vn_rdwr()
248 		 * could write to it.   however, the caller seems to want
249 		 * it mapped read-only, so now we are going to have to call
250 		 * uvm_map_protect() to fix up the protection.  ICK.
251 		 */
252 
253 		return uvm_map_protect(&p->p_vmspace->vm_map,
254 				trunc_page(cmd->ev_addr),
255 				round_page(cmd->ev_addr + cmd->ev_len),
256 				cmd->ev_prot, FALSE);
257 	}
258 	return 0;
259 }
260 
261 /*
262  * vmcmd_map_zero():
263  *	handle vmcmd which specifies a zero-filled address space region.  The
264  *	address range must be first allocated, then protected appropriately.
265  */
266 
267 int
268 vmcmd_map_zero(struct proc *p, struct exec_vmcmd *cmd)
269 {
270 	int error;
271 	long diff;
272 
273 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
274 	cmd->ev_addr -= diff;			/* required by uvm_map */
275 	cmd->ev_len += diff;
276 
277 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
278 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
279 			UVM_MAPFLAG(cmd->ev_prot, UVM_PROT_ALL, UVM_INH_COPY,
280 			UVM_ADV_NORMAL,
281 			UVM_FLAG_FIXED|UVM_FLAG_COPYONW));
282 	return error;
283 }
284 
285 /*
286  * exec_read_from():
287  *
288  *	Read from vnode into buffer at offset.
289  */
290 int
291 exec_read_from(struct proc *p, struct vnode *vp, u_long off, void *bf,
292     size_t size)
293 {
294 	int error;
295 	size_t resid;
296 
297 	if ((error = vn_rdwr(UIO_READ, vp, bf, size, off, UIO_SYSSPACE,
298 	    0, p->p_ucred, &resid, NULL)) != 0)
299 		return error;
300 	/*
301 	 * See if we got all of it
302 	 */
303 	if (resid != 0)
304 		return ENOEXEC;
305 	return 0;
306 }
307 
308 /*
309  * exec_setup_stack(): Set up the stack segment for an elf
310  * executable.
311  *
312  * Note that the ep_ssize parameter must be set to be the current stack
313  * limit; this is adjusted in the body of execve() to yield the
314  * appropriate stack segment usage once the argument length is
315  * calculated.
316  *
317  * This function returns an int for uniformity with other (future) formats'
318  * stack setup functions.  They might have errors to return.
319  */
320 
321 int
322 exec_setup_stack(struct proc *p, struct exec_package *epp)
323 {
324 	u_long max_stack_size;
325 	u_long access_linear_min, access_size;
326 	u_long noaccess_linear_min, noaccess_size;
327 
328 #ifndef	USRSTACK32
329 #define USRSTACK32	(0x00000000ffffffffL&~PGOFSET)
330 #endif
331 
332 	if (epp->ep_flags & EXEC_32) {
333 		epp->ep_minsaddr = USRSTACK32;
334 		max_stack_size = MAXSSIZ;
335 	} else {
336 		epp->ep_minsaddr = USRSTACK;
337 		max_stack_size = MAXSSIZ;
338 	}
339 	epp->ep_maxsaddr = (u_long)STACK_GROW(epp->ep_minsaddr,
340 		max_stack_size);
341 	epp->ep_ssize = p->p_rlimit[RLIMIT_STACK].rlim_cur;
342 
343 	/*
344 	 * set up commands for stack.  note that this takes *two*, one to
345 	 * map the part of the stack which we can access, and one to map
346 	 * the part which we can't.
347 	 *
348 	 * arguably, it could be made into one, but that would require the
349 	 * addition of another mapping proc, which is unnecessary
350 	 */
351 	access_size = epp->ep_ssize;
352 	access_linear_min = (u_long)STACK_ALLOC(epp->ep_minsaddr, access_size);
353 	noaccess_size = max_stack_size - access_size;
354 	noaccess_linear_min = (u_long)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
355 	    access_size), noaccess_size);
356 	if (noaccess_size > 0) {
357 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, noaccess_size,
358 		    noaccess_linear_min, NULL, 0, VM_PROT_NONE);
359 	}
360 	KASSERT(access_size > 0);
361 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, access_size,
362 	    access_linear_min, NULL, 0, VM_PROT_READ | VM_PROT_WRITE);
363 
364 	return 0;
365 }
366