xref: /openbsd-src/sys/uvm/uvm_unix.c (revision 46035553bfdd96e63c94e32da0210227ec2e3cf1)
1 /*	$OpenBSD: uvm_unix.c,v 1.71 2020/10/21 21:24:57 deraadt Exp $	*/
2 /*	$NetBSD: uvm_unix.c,v 1.18 2000/09/13 15:00:25 thorpej Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * Copyright (c) 1991, 1993 The Regents of the University of California.
7  * Copyright (c) 1988 University of Utah.
8  *
9  * All rights reserved.
10  *
11  * This code is derived from software contributed to Berkeley by
12  * the Systems Programming Group of the University of Utah Computer
13  * Science Department.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  * from: Utah $Hdr: vm_unix.c 1.1 89/11/07$
40  *      @(#)vm_unix.c   8.1 (Berkeley) 6/11/93
41  * from: Id: uvm_unix.c,v 1.1.2.2 1997/08/25 18:52:30 chuck Exp
42  */
43 
44 /*
45  * uvm_unix.c: traditional sbrk/grow interface to vm.
46  */
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/proc.h>
51 #include <sys/resourcevar.h>
52 #include <sys/vnode.h>
53 
54 #include <sys/mount.h>
55 #include <sys/syscallargs.h>
56 
57 #include <uvm/uvm.h>
58 
59 /*
60  * sys_obreak: set break
61  */
62 
63 int
64 sys_obreak(struct proc *p, void *v, register_t *retval)
65 {
66 	struct sys_obreak_args /* {
67 		syscallarg(char *) nsize;
68 	} */ *uap = v;
69 	struct vmspace *vm = p->p_vmspace;
70 	vaddr_t new, old, base;
71 	int error;
72 
73 	base = (vaddr_t)vm->vm_daddr;
74 	new = round_page((vaddr_t)SCARG(uap, nsize));
75 	if (new < base || (new - base) > lim_cur(RLIMIT_DATA))
76 		return (ENOMEM);
77 
78 	old = round_page(base + ptoa(vm->vm_dsize));
79 
80 	if (new == old)
81 		return (0);
82 
83 	/* grow or shrink? */
84 	if (new > old) {
85 		error = uvm_map(&vm->vm_map, &old, new - old, NULL,
86 		    UVM_UNKNOWN_OFFSET, 0,
87 		    UVM_MAPFLAG(PROT_READ | PROT_WRITE,
88 		    PROT_READ | PROT_WRITE | PROT_EXEC, MAP_INHERIT_COPY,
89 		    MADV_NORMAL, UVM_FLAG_FIXED|UVM_FLAG_COPYONW));
90 		if (error) {
91 			uprintf("sbrk: grow %ld failed, error = %d\n",
92 			    new - old, error);
93 			return (ENOMEM);
94 		}
95 		vm->vm_dsize += atop(new - old);
96 	} else {
97 		uvm_unmap(&vm->vm_map, new, old);
98 		vm->vm_dsize -= atop(old - new);
99 	}
100 
101 	return (0);
102 }
103 
104 /*
105  * uvm_grow: enlarge the "stack segment" to include sp.
106  */
107 void
108 uvm_grow(struct proc *p, vaddr_t sp)
109 {
110 	struct vmspace *vm = p->p_vmspace;
111 	vm_map_t map = &vm->vm_map;
112 	int si;
113 
114 	/* For user defined stacks (from sendsig). */
115 	if (sp < (vaddr_t)vm->vm_maxsaddr)
116 		return;
117 #ifdef MACHINE_STACK_GROWS_UP
118 	if (sp >= (vaddr_t)vm->vm_minsaddr)
119 		return;
120 #endif
121 
122 	vm_map_lock(map);
123 
124 	/* For common case of already allocated (from trap). */
125 #ifdef MACHINE_STACK_GROWS_UP
126 	if (sp < (vaddr_t)vm->vm_maxsaddr + ptoa(vm->vm_ssize))
127 #else
128 	if (sp >= (vaddr_t)vm->vm_minsaddr - ptoa(vm->vm_ssize))
129 #endif
130 		goto out;
131 
132 	/* Really need to check vs limit and increment stack size if ok. */
133 #ifdef MACHINE_STACK_GROWS_UP
134 	si = atop(sp - (vaddr_t)vm->vm_maxsaddr) - vm->vm_ssize + 1;
135 #else
136 	si = atop((vaddr_t)vm->vm_minsaddr - sp) - vm->vm_ssize;
137 #endif
138 	if (vm->vm_ssize + si <= atop(lim_cur(RLIMIT_STACK)))
139 		vm->vm_ssize += si;
140 out:
141 	vm_map_unlock(map);
142 }
143 
144 #ifndef SMALL_KERNEL
145 
146 #define WALK_CHUNK	32
147 /*
148  * Not all the pages in an amap may be present.  When dumping core,
149  * we don't want to force all the pages to be present: it's a waste
150  * of time and memory when we already know what they contain (zeros)
151  * and the ELF format at least can adequately represent them as a
152  * segment with memory size larger than its file size.
153  *
154  * So, we walk the amap with calls to amap_lookups() and scan the
155  * resulting pointers to find ranges of zero or more present pages
156  * followed by at least one absent page or the end of the amap.
157  * When then pass that range to the walk callback with 'start'
158  * pointing to the start of the present range, 'realend' pointing
159  * to the first absent page (or the end of the entry), and 'end'
160  * pointing to the page past the last absent page (or the end of
161  * the entry).
162  *
163  * Note that if the first page of the amap is empty then the callback
164  * must be invoked with 'start' == 'realend' so it can present that
165  * first range of absent pages.
166  */
167 int
168 uvm_coredump_walk_amap(struct vm_map_entry *entry, int *nsegmentp,
169     uvm_coredump_walk_cb *walk, void *cookie)
170 {
171 	struct vm_anon *anons[WALK_CHUNK];
172 	vaddr_t pos, start, realend, end, entry_end;
173 	vm_prot_t prot;
174 	int nsegment, absent, npages, i, error;
175 
176 	prot = entry->protection;
177 	nsegment = *nsegmentp;
178 	start = entry->start;
179 	entry_end = MIN(entry->end, VM_MAXUSER_ADDRESS);
180 
181 	absent = 0;
182 	for (pos = start; pos < entry_end; pos += npages << PAGE_SHIFT) {
183 		npages = (entry_end - pos) >> PAGE_SHIFT;
184 		if (npages > WALK_CHUNK)
185 			npages = WALK_CHUNK;
186 		amap_lookups(&entry->aref, pos - entry->start, anons, npages);
187 		for (i = 0; i < npages; i++) {
188 			if ((anons[i] == NULL) == absent)
189 				continue;
190 			if (!absent) {
191 				/* going from present to absent: set realend */
192 				realend = pos + (i << PAGE_SHIFT);
193 				absent = 1;
194 				continue;
195 			}
196 
197 			/* going from absent to present: invoke callback */
198 			end = pos + (i << PAGE_SHIFT);
199 			if (start != end) {
200 				error = (*walk)(start, realend, end, prot,
201 				    nsegment, cookie);
202 				if (error)
203 					return error;
204 				nsegment++;
205 			}
206 			start = realend = end;
207 			absent = 0;
208 		}
209 	}
210 
211 	if (!absent)
212 		realend = entry_end;
213 	error = (*walk)(start, realend, entry_end, prot, nsegment, cookie);
214 	*nsegmentp = nsegment + 1;
215 	return error;
216 }
217 
218 /*
219  * Common logic for whether a map entry should be included in a coredump
220  */
221 static inline int
222 uvm_should_coredump(struct proc *p, struct vm_map_entry *entry)
223 {
224 	if (!(entry->protection & PROT_WRITE) &&
225 	    entry->aref.ar_amap == NULL &&
226 	    entry->start != p->p_p->ps_sigcode &&
227 	    entry->start != p->p_p->ps_timekeep)
228 		return 0;
229 
230 	/*
231 	 * Skip ranges marked as unreadable, as uiomove(UIO_USERSPACE)
232 	 * will fail on them.  Maybe this really should be a test of
233 	 * entry->max_protection, but doing
234 	 *	uvm_map_extract(UVM_EXTRACT_FIXPROT)
235 	 * on each such page would suck.
236 	 */
237 	if ((entry->protection & PROT_READ) == 0)
238 		return 0;
239 
240 	/* Skip ranges excluded from coredumps. */
241 	if (UVM_ET_ISCONCEAL(entry))
242 		return 0;
243 
244 	/* Don't dump mmaped devices. */
245 	if (entry->object.uvm_obj != NULL &&
246 	    UVM_OBJ_IS_DEVICE(entry->object.uvm_obj))
247 		return 0;
248 
249 	if (entry->start >= VM_MAXUSER_ADDRESS)
250 		return 0;
251 
252 	return 1;
253 }
254 
255 
256 /* do nothing callback for uvm_coredump_walk_amap() */
257 static int
258 noop(vaddr_t start, vaddr_t realend, vaddr_t end, vm_prot_t prot,
259     int nsegment, void *cookie)
260 {
261 	return 0;
262 }
263 
264 /*
265  * Walk the VA space for a process to identify what to write to
266  * a coredump.  First the number of contiguous ranges is counted,
267  * then the 'setup' callback is invoked to prepare for actually
268  * recording the ranges, then the VA is walked again, invoking
269  * the 'walk' callback for each range.  The number of ranges walked
270  * is guaranteed to match the count seen by the 'setup' callback.
271  */
272 
273 int
274 uvm_coredump_walkmap(struct proc *p, uvm_coredump_setup_cb *setup,
275     uvm_coredump_walk_cb *walk, void *cookie)
276 {
277 	struct vmspace *vm = p->p_vmspace;
278 	struct vm_map *map = &vm->vm_map;
279 	struct vm_map_entry *entry;
280 	vaddr_t end;
281 	int refed_amaps = 0;
282 	int nsegment, error;
283 
284 	/*
285 	 * Walk the map once to count the segments.  If an amap is
286 	 * referenced more than once than take *another* reference
287 	 * and treat the amap as exactly one segment instead of
288 	 * checking page presence inside it.  On the second pass
289 	 * we'll recognize which amaps we did that for by the ref
290 	 * count being >1...and decrement it then.
291 	 */
292 	nsegment = 0;
293 	RBT_FOREACH(entry, uvm_map_addr, &map->addr) {
294 		/* should never happen for a user process */
295 		if (UVM_ET_ISSUBMAP(entry)) {
296 			panic("%s: user process with submap?", __func__);
297 		}
298 
299 		if (! uvm_should_coredump(p, entry))
300 			continue;
301 
302 		if (entry->aref.ar_amap != NULL) {
303 			if (entry->aref.ar_amap->am_ref == 1) {
304 				uvm_coredump_walk_amap(entry, &nsegment,
305 				    &noop, cookie);
306 				continue;
307 			}
308 
309 			/*
310 			 * Multiple refs currently, so take another and
311 			 * treat it as a single segment
312 			 */
313 			entry->aref.ar_amap->am_ref++;
314 			refed_amaps++;
315 		}
316 
317 		nsegment++;
318 	}
319 
320 	/*
321 	 * Okay, we have a count in nsegment.  Prepare to
322 	 * walk it again, then invoke the setup callback.
323 	 */
324 	entry = RBT_MIN(uvm_map_addr, &map->addr);
325 	error = (*setup)(nsegment, cookie);
326 	if (error)
327 		goto cleanup;
328 
329 	/*
330 	 * Setup went okay, so do the second walk, invoking the walk
331 	 * callback on the counted segments and cleaning up references
332 	 * as we go.
333 	 */
334 	nsegment = 0;
335 	for (; entry != NULL; entry = RBT_NEXT(uvm_map_addr, entry)) {
336 		if (! uvm_should_coredump(p, entry))
337 			continue;
338 
339 		if (entry->aref.ar_amap != NULL &&
340 		    entry->aref.ar_amap->am_ref == 1) {
341 			error = uvm_coredump_walk_amap(entry, &nsegment,
342 			    walk, cookie);
343 			if (error)
344 				break;
345 			continue;
346 		}
347 
348 		end = entry->end;
349 		if (end > VM_MAXUSER_ADDRESS)
350 			end = VM_MAXUSER_ADDRESS;
351 
352 		error = (*walk)(entry->start, end, end, entry->protection,
353 		    nsegment, cookie);
354 		if (error)
355 			break;
356 		nsegment++;
357 
358 		if (entry->aref.ar_amap != NULL &&
359 		    entry->aref.ar_amap->am_ref > 1) {
360 			/* multiple refs, so we need to drop one */
361 			entry->aref.ar_amap->am_ref--;
362 			refed_amaps--;
363 		}
364 	}
365 
366 	if (error) {
367 cleanup:
368 		/* clean up the extra references from where we left off */
369 		if (refed_amaps > 0) {
370 			for (; entry != NULL;
371 			    entry = RBT_NEXT(uvm_map_addr, entry)) {
372 				if (entry->aref.ar_amap == NULL ||
373 				    entry->aref.ar_amap->am_ref == 1)
374 					continue;
375 				if (! uvm_should_coredump(p, entry))
376 					continue;
377 				entry->aref.ar_amap->am_ref--;
378 				if (refed_amaps-- == 0)
379 					break;
380 			}
381 		}
382 	}
383 
384 	return error;
385 }
386 
387 #endif	/* !SMALL_KERNEL */
388