xref: /openbsd/sys/kern/exec_subr.c (revision 73471bf0)
1 /*	$OpenBSD: exec_subr.c,v 1.57 2019/11/29 06:34:45 deraadt Exp $	*/
2 /*	$NetBSD: exec_subr.c,v 1.9 1994/12/04 03:10:42 mycroft Exp $	*/
3 
4 /*
5  * Copyright (c) 1993, 1994 Christopher G. Demetriou
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *      This product includes software developed by Christopher G. Demetriou.
19  * 4. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/proc.h>
37 #include <sys/malloc.h>
38 #include <sys/vnode.h>
39 #include <sys/exec.h>
40 #include <sys/mman.h>
41 #include <sys/resourcevar.h>
42 
43 #include <uvm/uvm_extern.h>
44 
45 #ifdef DEBUG
46 /*
47  * new_vmcmd():
48  *	create a new vmcmd structure and fill in its fields based
49  *	on function call arguments.  make sure objects ref'd by
50  *	the vmcmd are 'held'.
51  *
52  * If not debugging, this is a macro, so it's expanded inline.
53  */
54 
55 void
56 new_vmcmd(struct exec_vmcmd_set *evsp,
57     int (*proc)(struct proc *, struct exec_vmcmd *), u_long len, u_long addr,
58     struct vnode *vp, u_long offset, u_int prot, int flags)
59 {
60 	struct exec_vmcmd    *vcp;
61 
62 	if (evsp->evs_used >= evsp->evs_cnt)
63 		vmcmdset_extend(evsp);
64 	vcp = &evsp->evs_cmds[evsp->evs_used++];
65 	vcp->ev_proc = proc;
66 	vcp->ev_len = len;
67 	vcp->ev_addr = addr;
68 	if ((vcp->ev_vp = vp) != NULL)
69 		vref(vp);
70 	vcp->ev_offset = offset;
71 	vcp->ev_prot = prot;
72 	vcp->ev_flags = flags;
73 }
74 #endif /* DEBUG */
75 
76 void
77 vmcmdset_extend(struct exec_vmcmd_set *evsp)
78 {
79 	struct exec_vmcmd *nvcp;
80 	u_int ocnt;
81 
82 #ifdef DIAGNOSTIC
83 	if (evsp->evs_used < evsp->evs_cnt)
84 		panic("vmcmdset_extend: not necessary");
85 #endif
86 
87 	ocnt = evsp->evs_cnt;
88 	KASSERT(ocnt > 0);
89 	/* figure out number of entries in new set */
90 	evsp->evs_cnt += ocnt;
91 
92 	/* reallocate the command set */
93 	nvcp = mallocarray(evsp->evs_cnt, sizeof(*nvcp), M_EXEC,
94 	    M_WAITOK);
95 	memcpy(nvcp, evsp->evs_cmds, ocnt * sizeof(*nvcp));
96 	if (evsp->evs_cmds != evsp->evs_start)
97 		free(evsp->evs_cmds, M_EXEC, ocnt * sizeof(*nvcp));
98 	evsp->evs_cmds = nvcp;
99 }
100 
101 void
102 kill_vmcmds(struct exec_vmcmd_set *evsp)
103 {
104 	struct exec_vmcmd *vcp;
105 	int i;
106 
107 	for (i = 0; i < evsp->evs_used; i++) {
108 		vcp = &evsp->evs_cmds[i];
109 		if (vcp->ev_vp != NULLVP)
110 			vrele(vcp->ev_vp);
111 	}
112 
113 	/*
114 	 * Free old vmcmds and reset the array.
115 	 */
116 	evsp->evs_used = 0;
117 	if (evsp->evs_cmds != evsp->evs_start)
118 		free(evsp->evs_cmds, M_EXEC,
119 		    evsp->evs_cnt * sizeof(struct exec_vmcmd));
120 	evsp->evs_cmds = evsp->evs_start;
121 	evsp->evs_cnt = EXEC_DEFAULT_VMCMD_SETSIZE;
122 }
123 
124 int
125 exec_process_vmcmds(struct proc *p, struct exec_package *epp)
126 {
127 	struct exec_vmcmd *base_vc = NULL;
128 	int error = 0;
129 	int i;
130 
131 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
132 		struct exec_vmcmd *vcp;
133 
134 		vcp = &epp->ep_vmcmds.evs_cmds[i];
135 
136 		if (vcp->ev_flags & VMCMD_RELATIVE) {
137 #ifdef DIAGNOSTIC
138 			if (base_vc == NULL)
139 				panic("exec_process_vmcmds: RELATIVE no base");
140 #endif
141 			vcp->ev_addr += base_vc->ev_addr;
142 		}
143 		error = (*vcp->ev_proc)(p, vcp);
144 		if (vcp->ev_flags & VMCMD_BASE) {
145 			base_vc = vcp;
146 		}
147 	}
148 
149 	kill_vmcmds(&epp->ep_vmcmds);
150 
151 	return (error);
152 }
153 
154 /*
155  * vmcmd_map_pagedvn():
156  *	handle vmcmd which specifies that a vnode should be mmap'd.
157  *	appropriate for handling demand-paged text and data segments.
158  */
159 
160 int
161 vmcmd_map_pagedvn(struct proc *p, struct exec_vmcmd *cmd)
162 {
163 	/*
164 	 * note that if you're going to map part of a process as being
165 	 * paged from a vnode, that vnode had damn well better be marked as
166 	 * VTEXT.  that's handled in the routine which sets up the vmcmd to
167 	 * call this routine.
168 	 */
169 	struct uvm_object *uobj;
170 	unsigned int syscalls = 0;
171 	int error;
172 
173 	/*
174 	 * map the vnode in using uvm_map.
175 	 */
176 
177 	if (cmd->ev_len == 0)
178 		return (0);
179 	if (cmd->ev_offset & PAGE_MASK)
180 		return (EINVAL);
181 	if (cmd->ev_addr & PAGE_MASK)
182 		return (EINVAL);
183 	if (cmd->ev_len & PAGE_MASK)
184 		return (EINVAL);
185 
186 	/*
187 	 * first, attach to the object
188 	 */
189 
190 	uobj = uvn_attach(cmd->ev_vp, PROT_READ | PROT_EXEC);
191 	if (uobj == NULL)
192 		return (ENOMEM);
193 
194 	/*
195 	 * do the map
196 	 */
197 	if ((cmd->ev_flags & VMCMD_SYSCALL) && (cmd->ev_prot & PROT_EXEC))
198 		syscalls |= UVM_FLAG_SYSCALL;
199 
200 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len,
201 	    uobj, cmd->ev_offset, 0,
202 	    UVM_MAPFLAG(cmd->ev_prot, PROT_MASK, MAP_INHERIT_COPY,
203 	    MADV_NORMAL, UVM_FLAG_COPYONW | UVM_FLAG_FIXED | syscalls));
204 
205 	/*
206 	 * check for error
207 	 */
208 
209 	if (error) {
210 		/*
211 		 * error: detach from object
212 		 */
213 		uobj->pgops->pgo_detach(uobj);
214 	}
215 
216 	return (error);
217 }
218 
219 /*
220  * vmcmd_map_readvn():
221  *	handle vmcmd which specifies that a vnode should be read from.
222  *	appropriate for non-demand-paged text/data segments, i.e. impure
223  *	objects (a la OMAGIC and NMAGIC).
224  */
225 
226 int
227 vmcmd_map_readvn(struct proc *p, struct exec_vmcmd *cmd)
228 {
229 	int error;
230 	vm_prot_t prot;
231 
232 	if (cmd->ev_len == 0)
233 		return (0);
234 
235 	prot = cmd->ev_prot;
236 
237 	cmd->ev_addr = trunc_page(cmd->ev_addr); /* required by uvm_map */
238 	error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
239 	    round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
240 	    UVM_MAPFLAG(prot | PROT_WRITE, PROT_MASK, MAP_INHERIT_COPY,
241 	    MADV_NORMAL, UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW));
242 
243 	if (error)
244 		return (error);
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_ucred, NULL, p);
249 	if (error)
250 		return (error);
251 
252 	if ((prot & PROT_WRITE) == 0) {
253 		/*
254 		 * we had to map in the area at PROT_WRITE so that vn_rdwr()
255 		 * could write to it.   however, the caller seems to want
256 		 * it mapped read-only, so now we are going to have to call
257 		 * uvm_map_protect() to fix up the protection.  ICK.
258 		 */
259 		return (uvm_map_protect(&p->p_vmspace->vm_map,
260 		    trunc_page(cmd->ev_addr),
261 		    round_page(cmd->ev_addr + cmd->ev_len),
262 		    prot, FALSE));
263 	}
264 	return (0);
265 }
266 
267 /*
268  * vmcmd_map_zero():
269  *	handle vmcmd which specifies a zero-filled address space region.
270  */
271 
272 int
273 vmcmd_map_zero(struct proc *p, struct exec_vmcmd *cmd)
274 {
275 	if (cmd->ev_len == 0)
276 		return (0);
277 
278 	cmd->ev_addr = trunc_page(cmd->ev_addr); /* required by uvm_map */
279 	return (uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr,
280 	    round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0,
281 	    UVM_MAPFLAG(cmd->ev_prot, PROT_MASK, MAP_INHERIT_COPY,
282 	    MADV_NORMAL, UVM_FLAG_FIXED|UVM_FLAG_COPYONW |
283 	    (cmd->ev_flags & VMCMD_STACK ? UVM_FLAG_STACK : 0))));
284 }
285 
286 /*
287  * vmcmd_randomize():
288  *	handle vmcmd which specifies a randomized address space region.
289  */
290 #define RANDOMIZE_CTX_THRESHOLD 512
291 int
292 vmcmd_randomize(struct proc *p, struct exec_vmcmd *cmd)
293 {
294 	int error;
295 	struct arc4random_ctx *ctx;
296 	char *buf;
297 	size_t sublen, off = 0;
298 	size_t len = cmd->ev_len;
299 
300 	if (len == 0)
301 		return (0);
302 	if (len > ELF_RANDOMIZE_LIMIT)
303 		return (EINVAL);
304 
305 	buf = malloc(PAGE_SIZE, M_TEMP, M_WAITOK);
306 	if (len < RANDOMIZE_CTX_THRESHOLD) {
307 		arc4random_buf(buf, len);
308 		error = copyout(buf, (void *)cmd->ev_addr, len);
309 		explicit_bzero(buf, len);
310 	} else {
311 		ctx = arc4random_ctx_new();
312 		do {
313 			sublen = MIN(len, PAGE_SIZE);
314 			arc4random_ctx_buf(ctx, buf, sublen);
315 			error = copyout(buf, (void *)cmd->ev_addr + off, sublen);
316 			if (error)
317 				break;
318 			off += sublen;
319 			len -= sublen;
320 			sched_pause(yield);
321 		} while (len);
322 		arc4random_ctx_free(ctx);
323 		explicit_bzero(buf, PAGE_SIZE);
324 	}
325 	free(buf, M_TEMP, PAGE_SIZE);
326 	return (error);
327 }
328 
329 #ifndef MAXSSIZ_GUARD
330 #define MAXSSIZ_GUARD	(1024 * 1024)
331 #endif
332 
333 /*
334  * exec_setup_stack(): Set up the stack segment for an executable.
335  *
336  * Note that the ep_ssize parameter must be set to be the current stack
337  * limit; this is adjusted in the body of execve() to yield the
338  * appropriate stack segment usage once the argument length is
339  * calculated.
340  *
341  * This function returns an int for uniformity with other (future) formats'
342  * stack setup functions.  They might have errors to return.
343  */
344 
345 int
346 exec_setup_stack(struct proc *p, struct exec_package *epp)
347 {
348 	vaddr_t sgap;
349 
350 #ifdef MACHINE_STACK_GROWS_UP
351 	epp->ep_maxsaddr = USRSTACK;
352 	epp->ep_minsaddr = USRSTACK + MAXSSIZ;
353 #else
354 	epp->ep_maxsaddr = USRSTACK - MAXSSIZ - MAXSSIZ_GUARD;
355 	epp->ep_minsaddr = USRSTACK;
356 #endif
357 	epp->ep_ssize = round_page(lim_cur(RLIMIT_STACK));
358 
359 	if (stackgap_random != 0) {
360 		sgap = arc4random() & (stackgap_random - 1);
361 		sgap = trunc_page(sgap);
362 
363 #ifdef MACHINE_STACK_GROWS_UP
364 		epp->ep_maxsaddr += sgap;
365 		epp->ep_minsaddr += sgap;
366 #else
367 		epp->ep_maxsaddr -= sgap;
368 		epp->ep_minsaddr -= sgap;
369 #endif
370 	}
371 
372 	/*
373 	 * set up commands for stack.  note that this takes *two*, one to
374 	 * map the part of the stack which we can access, and one to map
375 	 * the part which we can't.
376 	 *
377 	 * arguably, it could be made into one, but that would require the
378 	 * addition of another mapping proc, which is unnecessary
379 	 *
380 	 * note that in memory, things assumed to be: 0 ....... ep_maxsaddr
381 	 * <stack> ep_minsaddr
382 	 */
383 #ifdef MACHINE_STACK_GROWS_UP
384 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero,
385 	    ((epp->ep_minsaddr - epp->ep_ssize) - epp->ep_maxsaddr),
386 	    epp->ep_maxsaddr + epp->ep_ssize, NULLVP, 0,
387 	    PROT_NONE);
388 	NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, epp->ep_ssize,
389 	    epp->ep_maxsaddr, NULLVP, 0,
390 	    PROT_READ | PROT_WRITE, VMCMD_STACK);
391 #else
392 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero,
393 	    ((epp->ep_minsaddr - epp->ep_ssize) - epp->ep_maxsaddr),
394 	    epp->ep_maxsaddr, NULLVP, 0,
395 	    PROT_NONE);
396 	NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, epp->ep_ssize,
397 	    (epp->ep_minsaddr - epp->ep_ssize), NULLVP, 0,
398 	    PROT_READ | PROT_WRITE, VMCMD_STACK);
399 #endif
400 
401 	return (0);
402 }
403