xref: /netbsd-src/sys/kern/exec_subr.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: exec_subr.c,v 1.46 2005/12/11 12:24:29 christos 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 1.46 2005/12/11 12:24:29 christos 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 lwp * l, 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 lwp *l, struct exec_vmcmd *cmd)
150 {
151 	struct uvm_object *uobj;
152 	struct proc *p = l->l_proc;
153 	int error;
154 
155 	KASSERT(cmd->ev_vp->v_flag & VTEXT);
156 
157 	/*
158 	 * map the vnode in using uvm_map.
159 	 */
160 
161         if (cmd->ev_len == 0)
162                 return(0);
163         if (cmd->ev_offset & PAGE_MASK)
164                 return(EINVAL);
165 	if (cmd->ev_addr & PAGE_MASK)
166 		return(EINVAL);
167 	if (cmd->ev_len & PAGE_MASK)
168 		return(EINVAL);
169 
170 	/*
171 	 * first, attach to the object
172 	 */
173 
174         uobj = uvn_attach(cmd->ev_vp, VM_PROT_READ|VM_PROT_EXECUTE);
175         if (uobj == NULL)
176                 return(ENOMEM);
177 	VREF(cmd->ev_vp);
178 
179 	/*
180 	 * do the map
181 	 */
182 
183 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len,
184 		uobj, cmd->ev_offset, 0,
185 		UVM_MAPFLAG(cmd->ev_prot, VM_PROT_ALL, UVM_INH_COPY,
186 			UVM_ADV_NORMAL, UVM_FLAG_COPYONW|UVM_FLAG_FIXED));
187 	if (error) {
188 		uobj->pgops->pgo_detach(uobj);
189 	}
190 	return error;
191 }
192 
193 /*
194  * vmcmd_map_readvn():
195  *	handle vmcmd which specifies that a vnode should be read from.
196  *	appropriate for non-demand-paged text/data segments, i.e. impure
197  *	objects (a la OMAGIC and NMAGIC).
198  */
199 int
200 vmcmd_map_readvn(struct lwp *l, struct exec_vmcmd *cmd)
201 {
202 	struct proc *p = l->l_proc;
203 	int error;
204 	long diff;
205 
206 	if (cmd->ev_len == 0)
207 		return 0;
208 
209 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
210 	cmd->ev_addr -= diff;			/* required by uvm_map */
211 	cmd->ev_offset -= diff;
212 	cmd->ev_len += diff;
213 
214 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
215 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
216 			UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_COPY,
217 			UVM_ADV_NORMAL,
218 			UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW));
219 
220 	if (error)
221 		return error;
222 
223 	return vmcmd_readvn(l, cmd);
224 }
225 
226 int
227 vmcmd_readvn(struct lwp *l, struct exec_vmcmd *cmd)
228 {
229 	struct proc *p = l->l_proc;
230 	int error;
231 
232 	error = vn_rdwr(UIO_READ, cmd->ev_vp, (caddr_t)cmd->ev_addr,
233 	    cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT,
234 	    p->p_ucred, NULL, l);
235 	if (error)
236 		return error;
237 
238 #ifdef PMAP_NEED_PROCWR
239 	/*
240 	 * we had to write the process, make sure the pages are synched
241 	 * with the instruction cache.
242 	 */
243 	if (cmd->ev_prot & VM_PROT_EXECUTE)
244 		pmap_procwr(p, cmd->ev_addr, cmd->ev_len);
245 #endif
246 
247 	if (cmd->ev_prot != (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)) {
248 
249 		/*
250 		 * we had to map in the area at PROT_ALL so that vn_rdwr()
251 		 * could write to it.   however, the caller seems to want
252 		 * it mapped read-only, so now we are going to have to call
253 		 * uvm_map_protect() to fix up the protection.  ICK.
254 		 */
255 
256 		return uvm_map_protect(&p->p_vmspace->vm_map,
257 				trunc_page(cmd->ev_addr),
258 				round_page(cmd->ev_addr + cmd->ev_len),
259 				cmd->ev_prot, FALSE);
260 	}
261 	return 0;
262 }
263 
264 /*
265  * vmcmd_map_zero():
266  *	handle vmcmd which specifies a zero-filled address space region.  The
267  *	address range must be first allocated, then protected appropriately.
268  */
269 
270 int
271 vmcmd_map_zero(struct lwp *l, struct exec_vmcmd *cmd)
272 {
273 	struct proc *p = l->l_proc;
274 	int error;
275 	long diff;
276 
277 	diff = cmd->ev_addr - trunc_page(cmd->ev_addr);
278 	cmd->ev_addr -= diff;			/* required by uvm_map */
279 	cmd->ev_len += diff;
280 
281 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
282 			round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
283 			UVM_MAPFLAG(cmd->ev_prot, UVM_PROT_ALL, UVM_INH_COPY,
284 			UVM_ADV_NORMAL,
285 			UVM_FLAG_FIXED|UVM_FLAG_COPYONW));
286 	return error;
287 }
288 
289 /*
290  * exec_read_from():
291  *
292  *	Read from vnode into buffer at offset.
293  */
294 int
295 exec_read_from(struct lwp *l, struct vnode *vp, u_long off, void *bf,
296     size_t size)
297 {
298 	int error;
299 	size_t resid;
300 
301 	if ((error = vn_rdwr(UIO_READ, vp, bf, size, off, UIO_SYSSPACE,
302 	    0, l->l_proc->p_ucred, &resid, NULL)) != 0)
303 		return error;
304 	/*
305 	 * See if we got all of it
306 	 */
307 	if (resid != 0)
308 		return ENOEXEC;
309 	return 0;
310 }
311 
312 /*
313  * exec_setup_stack(): Set up the stack segment for an elf
314  * executable.
315  *
316  * Note that the ep_ssize parameter must be set to be the current stack
317  * limit; this is adjusted in the body of execve() to yield the
318  * appropriate stack segment usage once the argument length is
319  * calculated.
320  *
321  * This function returns an int for uniformity with other (future) formats'
322  * stack setup functions.  They might have errors to return.
323  */
324 
325 int
326 exec_setup_stack(struct lwp *l, struct exec_package *epp)
327 {
328 	u_long max_stack_size;
329 	u_long access_linear_min, access_size;
330 	u_long noaccess_linear_min, noaccess_size;
331 
332 #ifndef	USRSTACK32
333 #define USRSTACK32	(0x00000000ffffffffL&~PGOFSET)
334 #endif
335 
336 	if (epp->ep_flags & EXEC_32) {
337 		epp->ep_minsaddr = USRSTACK32;
338 		max_stack_size = MAXSSIZ;
339 	} else {
340 		epp->ep_minsaddr = USRSTACK;
341 		max_stack_size = MAXSSIZ;
342 	}
343 	epp->ep_maxsaddr = (u_long)STACK_GROW(epp->ep_minsaddr,
344 		max_stack_size);
345 	epp->ep_ssize = l->l_proc->p_rlimit[RLIMIT_STACK].rlim_cur;
346 
347 	/*
348 	 * set up commands for stack.  note that this takes *two*, one to
349 	 * map the part of the stack which we can access, and one to map
350 	 * the part which we can't.
351 	 *
352 	 * arguably, it could be made into one, but that would require the
353 	 * addition of another mapping proc, which is unnecessary
354 	 */
355 	access_size = epp->ep_ssize;
356 	access_linear_min = (u_long)STACK_ALLOC(epp->ep_minsaddr, access_size);
357 	noaccess_size = max_stack_size - access_size;
358 	noaccess_linear_min = (u_long)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
359 	    access_size), noaccess_size);
360 	if (noaccess_size > 0) {
361 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, noaccess_size,
362 		    noaccess_linear_min, NULL, 0, VM_PROT_NONE);
363 	}
364 	KASSERT(access_size > 0);
365 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, access_size,
366 	    access_linear_min, NULL, 0, VM_PROT_READ | VM_PROT_WRITE);
367 
368 	return 0;
369 }
370