xref: /openbsd/sys/uvm/uvm_map.h (revision 30d20579)
1 /*	$OpenBSD: uvm_map.h,v 1.89 2024/04/02 08:39:17 deraadt Exp $	*/
2 /*	$NetBSD: uvm_map.h,v 1.24 2001/02/18 21:19:08 chs Exp $	*/
3 
4 /*
5  * Copyright (c) 2011 Ariane van der Steldt <ariane@openbsd.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  *
19  *
20  * Copyright (c) 1997 Charles D. Cranor and Washington University.
21  * Copyright (c) 1991, 1993, The Regents of the University of California.
22  *
23  * All rights reserved.
24  *
25  * This code is derived from software contributed to Berkeley by
26  * The Mach Operating System project at Carnegie-Mellon University.
27  *
28  * Redistribution and use in source and binary forms, with or without
29  * modification, are permitted provided that the following conditions
30  * are met:
31  * 1. Redistributions of source code must retain the above copyright
32  *    notice, this list of conditions and the following disclaimer.
33  * 2. Redistributions in binary form must reproduce the above copyright
34  *    notice, this list of conditions and the following disclaimer in the
35  *    documentation and/or other materials provided with the distribution.
36  * 3. Neither the name of the University nor the names of its contributors
37  *    may be used to endorse or promote products derived from this software
38  *    without specific prior written permission.
39  *
40  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
41  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
42  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
43  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
44  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
45  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
46  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
47  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
48  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
49  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
50  * SUCH DAMAGE.
51  *
52  *	@(#)vm_map.h    8.3 (Berkeley) 3/15/94
53  * from: Id: uvm_map.h,v 1.1.2.3 1998/02/07 01:16:55 chs Exp
54  *
55  *
56  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
57  * All rights reserved.
58  *
59  * Permission to use, copy, modify and distribute this software and
60  * its documentation is hereby granted, provided that both the copyright
61  * notice and this permission notice appear in all copies of the
62  * software, derivative works or modified versions, and any portions
63  * thereof, and that both notices appear in supporting documentation.
64  *
65  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
66  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
67  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
68  *
69  * Carnegie Mellon requests users of this software to return to
70  *
71  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
72  *  School of Computer Science
73  *  Carnegie Mellon University
74  *  Pittsburgh PA 15213-3890
75  *
76  * any improvements or extensions that they make and grant Carnegie the
77  * rights to redistribute these changes.
78  */
79 
80 #ifndef _UVM_UVM_MAP_H_
81 #define _UVM_UVM_MAP_H_
82 
83 #include <sys/mutex.h>
84 #include <sys/rwlock.h>
85 
86 #ifdef _KERNEL
87 
88 /*
89  * UVM_MAP_CLIP_START: ensure that the entry begins at or after
90  * the starting address, if it doesn't we split the entry.
91  *
92  * => map must be locked by caller
93  */
94 
95 #define UVM_MAP_CLIP_START(_map, _entry, _addr)				\
96 	do {								\
97 		KASSERT((_entry)->end + (_entry)->fspace > (_addr));	\
98 		if ((_entry)->start < (_addr))				\
99 			uvm_map_clip_start((_map), (_entry), (_addr));	\
100 	} while (0)
101 
102 /*
103  * UVM_MAP_CLIP_END: ensure that the entry ends at or before
104  *      the ending address, if it doesn't we split the entry.
105  *
106  * => map must be locked by caller
107  */
108 
109 #define UVM_MAP_CLIP_END(_map, _entry, _addr)				\
110 	do {								\
111 		KASSERT((_entry)->start < (_addr));			\
112 		if ((_entry)->end > (_addr))				\
113 			uvm_map_clip_end((_map), (_entry), (_addr));	\
114 	} while (0)
115 
116 /*
117  * extract flags
118  */
119 #define UVM_EXTRACT_FIXPROT	0x8	/* set prot to maxprot as we go */
120 
121 #endif /* _KERNEL */
122 
123 #include <uvm/uvm_anon.h>
124 
125 /*
126  * Address map entries consist of start and end addresses,
127  * a VM object (or sharing map) and offset into that object,
128  * and user-exported inheritance and protection information.
129  * Also included is control information for virtual copy operations.
130  */
131 struct vm_map_entry {
132 	union {
133 		RBT_ENTRY(vm_map_entry)	addr_entry; /* address tree */
134 		SLIST_ENTRY(vm_map_entry) addr_kentry;
135 	} daddrs;
136 
137 	union {
138 		RBT_ENTRY(vm_map_entry)	rbtree;	/* Link freespace tree. */
139 		TAILQ_ENTRY(vm_map_entry) tailq;/* Link freespace queue. */
140 		TAILQ_ENTRY(vm_map_entry) deadq;/* dead entry queue */
141 	} dfree;
142 
143 #define uvm_map_entry_start_copy start
144 	vaddr_t			start;		/* start address */
145 	vaddr_t			end;		/* end address */
146 
147 	vsize_t			guard;		/* bytes in guard */
148 	vsize_t			fspace;		/* free space */
149 
150 	union {
151 		struct uvm_object *uvm_obj;	/* uvm object */
152 		struct vm_map	*sub_map;	/* belongs to another map */
153 	} object;				/* object I point to */
154 	voff_t			offset;		/* offset into object */
155 	struct vm_aref		aref;		/* anonymous overlay */
156 	int			etype;		/* entry type */
157 	vm_prot_t		protection;	/* protection code */
158 	vm_prot_t		max_protection;	/* maximum protection */
159 	vm_inherit_t		inheritance;	/* inheritance */
160 	int			wired_count;	/* can be paged if == 0 */
161 	int			advice;		/* madvise advice */
162 #define uvm_map_entry_stop_copy flags
163 	u_int8_t		flags;		/* flags */
164 
165 #define	UVM_MAP_STATIC		0x01		/* static map entry */
166 #define	UVM_MAP_KMEM		0x02		/* from kmem entry pool */
167 
168 	vsize_t			fspace_augment;	/* max(fspace) in subtree */
169 };
170 
171 #define	VM_MAPENT_ISWIRED(entry)	((entry)->wired_count != 0)
172 
173 TAILQ_HEAD(uvm_map_deadq, vm_map_entry);	/* dead entry queue */
174 RBT_HEAD(uvm_map_addr, vm_map_entry);
175 #ifdef _KERNEL
176 RBT_PROTOTYPE(uvm_map_addr, vm_map_entry, daddrs.addr_entry,
177     uvm_mapentry_addrcmp);
178 #endif
179 
180 /*
181  *	A Map is a rbtree of map entries, kept sorted by address.
182  *	In addition, free space entries are also kept in a rbtree,
183  *	indexed by free size.
184  *
185  *
186  *
187  *	LOCKING PROTOCOL NOTES:
188  *	-----------------------
189  *
190  *	VM map locking is a little complicated.  There are both shared
191  *	and exclusive locks on maps.  However, it is sometimes required
192  *	to unlock a VM map (to prevent lock ordering issues) without
193  *	allowing any other thread to modify it.
194  *
195  *	In order to prevent this scenario, we introduce the notion of
196  *	a `busy' map.  A `busy' map is unlocked, but other threads
197  *	attempting to write-lock wait for this flag to clear before
198  *	entering the lock manager.  A map may only be marked busy
199  *	when the map is write-locked and may only be marked unbusy by
200  *	the thread which marked it busy.
201  *
202  *	Access to the map `flags' member is controlled by the `flags_lock'
203  *	simple lock.  Note that some flags are static (set once at map
204  *	creation time, and never changed), and thus require no locking
205  *	to check those flags.  All flags which are r/w must be set or
206  *	cleared while the `flags_lock' is asserted.  Additional locking
207  *	requirements are:
208  *
209  *		VM_MAP_PAGEABLE		r/o static flag; no locking required
210  *
211  *		VM_MAP_INTRSAFE		r/o static flag; no locking required
212  *
213  *		VM_MAP_WIREFUTURE	r/w; may only be set or cleared when
214  *					map is write-locked.  may be tested
215  *					without asserting `flags_lock'.
216  *
217  *		VM_MAP_BUSY		r/w; may only be set when map is
218  *					write-locked, may only be cleared by
219  *					thread which set it, map read-locked
220  *					or write-locked.  must be tested
221  *					while `flags_lock' is asserted.
222  *
223  *		VM_MAP_WANTLOCK		r/w; may only be set when the map
224  *					is busy, and thread is attempting
225  *					to write-lock.  must be tested
226  *					while `flags_lock' is asserted.
227  *
228  *		VM_MAP_GUARDPAGES	r/o; must be specified at map
229  *					initialization time.
230  *					If set, guards will appear between
231  *					automatic allocations.
232  *					No locking required.
233  *
234  *		VM_MAP_ISVMSPACE	r/o; set by uvmspace_alloc.
235  *					Signifies that this map is a vmspace.
236  *					(The implementation treats all maps
237  *					without this bit as kernel maps.)
238  *					No locking required.
239  *
240  *
241  * All automatic allocations (uvm_map without MAP_FIXED) will allocate
242  * from vm_map.free.
243  * If that allocation fails:
244  * - vmspace maps will spill over into vm_map.bfree,
245  * - all other maps will call uvm_map_kmem_grow() to increase the arena.
246  *
247  * vmspace maps have their data, brk() and stack arenas automatically
248  * updated when uvm_map() is invoked without MAP_FIXED.
249  * The spill over arena (vm_map.bfree) will contain the space in the brk()
250  * and stack ranges.
251  * Kernel maps never have a bfree arena and this tree will always be empty.
252  *
253  *
254  * read_locks and write_locks are used in lock debugging code.
255  *
256  *  Locks used to protect struct members in this file:
257  *	a	atomic operations
258  *	I	immutable after creation or exec(2)
259  *	v	`vm_map_lock' (this map `lock' or `mtx')
260  */
261 struct vm_map {
262 	struct pmap		*pmap;		/* [I] Physical map */
263 	u_long			sserial;	/* [v] # stack changes */
264 
265 	struct uvm_map_addr	addr;		/* [v] Entry tree, by addr */
266 
267 	vsize_t			size;		/* virtual size */
268 	int			ref_count;	/* [a] Reference count */
269 	int			flags;		/* flags */
270 	unsigned int		timestamp;	/* Version number */
271 	struct proc		*busy;		/* [v] thread holding map busy*/
272 
273 	vaddr_t			min_offset;	/* [I] First address in map. */
274 	vaddr_t			max_offset;	/* [I] Last address in map. */
275 
276 	/*
277 	 * Allocation overflow regions.
278 	 */
279 	vaddr_t			b_start;	/* [v] Start for brk() alloc. */
280 	vaddr_t			b_end;		/* [v] End for brk() alloc. */
281 	vaddr_t			s_start;	/* [v] Start for stack alloc. */
282 	vaddr_t			s_end;		/* [v] End for stack alloc. */
283 
284 	/*
285 	 * Special address selectors.
286 	 *
287 	 * The uaddr_exe mapping is used if:
288 	 * - protX is selected
289 	 * - the pointer is not NULL
290 	 *
291 	 * If uaddr_exe is not used, the other mappings are checked in
292 	 * order of appearance.
293 	 * If a hint is given, the selection will only be used if the hint
294 	 * falls in the range described by the mapping.
295 	 *
296 	 * The states are pointers because:
297 	 * - they may not all be in use
298 	 * - the struct size for different schemes is variable
299 	 *
300 	 * The uaddr_brk_stack selector will select addresses that are in
301 	 * the brk/stack area of the map.
302 	 */
303 	struct uvm_addr_state	*uaddr_exe;	/* Executable selector. */
304 	struct uvm_addr_state	*uaddr_any[4];	/* More selectors. */
305 	struct uvm_addr_state	*uaddr_brk_stack; /* Brk/stack selector. */
306 
307 #define UVM_MAP_CHECK_COPYIN_MAX 4	/* main, sigtramp, ld.so, libc.so */
308 	struct uvm_check_copyin {
309 		vaddr_t		start, end;
310 	}			check_copyin[UVM_MAP_CHECK_COPYIN_MAX];
311 	int			check_copyin_count;
312 
313 	/*
314 	 * XXX struct mutex changes size because of compile options, so
315 	 * place after fields which are inspected by libkvm / procmap(8)
316 	 */
317 	struct rwlock		lock;		/* Non-intrsafe lock */
318 	struct mutex		mtx;		/* Intrsafe lock */
319 	struct mutex		flags_lock;	/* flags lock */
320 };
321 
322 /* vm_map flags */
323 #define	VM_MAP_PAGEABLE		0x01		/* ro: entries are pageable */
324 #define	VM_MAP_INTRSAFE		0x02		/* ro: interrupt safe map */
325 #define	VM_MAP_WIREFUTURE	0x04		/* rw: wire future mappings */
326 #define	VM_MAP_BUSY		0x08		/* rw: map is busy */
327 #define	VM_MAP_WANTLOCK		0x10		/* rw: want to write-lock */
328 #define	VM_MAP_GUARDPAGES	0x20		/* rw: add guard pgs to map */
329 #define	VM_MAP_ISVMSPACE	0x40		/* ro: map is a vmspace */
330 #define	VM_MAP_PINSYSCALL_ONCE	0x100		/* rw: pinsyscall done */
331 
332 /* Number of kernel maps and entries to statically allocate */
333 #define	MAX_KMAPENT	1024	/* Sufficient to make it to the scheduler. */
334 
335 #ifdef _KERNEL
336 /*
337  * globals:
338  */
339 
340 extern vaddr_t	uvm_maxkaddr;
341 
342 /*
343  * protos: the following prototypes define the interface to vm_map
344  */
345 
346 void		uvm_map_deallocate(struct vm_map *);
347 
348 int		uvm_map_clean(struct vm_map *, vaddr_t, vaddr_t, int);
349 void		uvm_map_clip_start(struct vm_map *, struct vm_map_entry *,
350 		    vaddr_t);
351 void		uvm_map_clip_end(struct vm_map *, struct vm_map_entry *,
352 		    vaddr_t);
353 int		uvm_map_extract(struct vm_map *, vaddr_t, vsize_t,
354 		    vaddr_t *, int);
355 struct vm_map *	uvm_map_create(pmap_t, vaddr_t, vaddr_t, int);
356 vaddr_t		uvm_map_pie(vaddr_t);
357 vaddr_t		uvm_map_hint(struct vmspace *, vm_prot_t, vaddr_t, vaddr_t);
358 int		uvm_map_check_copyin_add(struct vm_map *, vaddr_t, vaddr_t);
359 int		uvm_map_immutable(struct vm_map *, vaddr_t, vaddr_t, int);
360 int		uvm_map_inherit(struct vm_map *, vaddr_t, vaddr_t, vm_inherit_t);
361 int		uvm_map_advice(struct vm_map *, vaddr_t, vaddr_t, int);
362 void		uvm_map_init(void);
363 boolean_t	uvm_map_lookup_entry(struct vm_map *, vaddr_t, vm_map_entry_t *);
364 boolean_t	uvm_map_is_stack_remappable(struct vm_map *, vaddr_t, vsize_t, int);
365 int		uvm_map_remap_as_stack(struct proc *, vaddr_t, vsize_t);
366 int		uvm_map_replace(struct vm_map *, vaddr_t, vaddr_t,
367 		    vm_map_entry_t, int);
368 int		uvm_map_reserve(struct vm_map *, vsize_t, vaddr_t, vsize_t,
369 		    vaddr_t *);
370 void		uvm_map_setup(struct vm_map *, pmap_t, vaddr_t, vaddr_t, int);
371 int		uvm_map_submap(struct vm_map *, vaddr_t, vaddr_t,
372 		    struct vm_map *);
373 void		uvm_unmap(struct vm_map *, vaddr_t, vaddr_t);
374 void		uvm_unmap_detach(struct uvm_map_deadq *, int);
375 int		uvm_unmap_remove(struct vm_map*, vaddr_t, vaddr_t,
376 		    struct uvm_map_deadq *, boolean_t, boolean_t, boolean_t);
377 void		uvm_map_set_uaddr(struct vm_map*, struct uvm_addr_state**,
378 		    struct uvm_addr_state*);
379 int		uvm_map_mquery(struct vm_map*, vaddr_t*, vsize_t, voff_t, int);
380 
381 
382 struct p_inentry;
383 
384 int		uvm_map_inentry_sp(vm_map_entry_t);
385 boolean_t	uvm_map_inentry(struct proc *, struct p_inentry *, vaddr_t addr,
386 		    const char *fmt, int (*fn)(vm_map_entry_t), u_long serial);
387 
388 struct kinfo_vmentry;
389 
390 int		uvm_map_fill_vmmap(struct vm_map *, struct kinfo_vmentry *,
391 		    size_t *);
392 
393 /*
394  * VM map locking operations:
395  *
396  *	These operations perform locking on the data portion of the
397  *	map.
398  *
399  *	vm_map_lock_try: try to lock a map, failing if it is already locked.
400  *
401  *	vm_map_lock: acquire an exclusive (write) lock on a map.
402  *
403  *	vm_map_lock_read: acquire a shared (read) lock on a map.
404  *
405  *	vm_map_unlock: release an exclusive lock on a map.
406  *
407  *	vm_map_unlock_read: release a shared lock on a map.
408  *
409  *	vm_map_busy: mark a map as busy.
410  *
411  *	vm_map_unbusy: clear busy status on a map.
412  *
413  */
414 
415 boolean_t	vm_map_lock_try_ln(struct vm_map*, char*, int);
416 void		vm_map_lock_ln(struct vm_map*, char*, int);
417 void		vm_map_lock_read_ln(struct vm_map*, char*, int);
418 void		vm_map_unlock_ln(struct vm_map*, char*, int);
419 void		vm_map_unlock_read_ln(struct vm_map*, char*, int);
420 void		vm_map_busy_ln(struct vm_map*, char*, int);
421 void		vm_map_unbusy_ln(struct vm_map*, char*, int);
422 void		vm_map_assert_anylock_ln(struct vm_map*, char*, int);
423 void		vm_map_assert_wrlock_ln(struct vm_map*, char*, int);
424 
425 #ifdef DIAGNOSTIC
426 #define vm_map_lock_try(map)	vm_map_lock_try_ln(map, __FILE__, __LINE__)
427 #define vm_map_lock(map)	vm_map_lock_ln(map, __FILE__, __LINE__)
428 #define vm_map_lock_read(map)	vm_map_lock_read_ln(map, __FILE__, __LINE__)
429 #define vm_map_unlock(map)	vm_map_unlock_ln(map, __FILE__, __LINE__)
430 #define vm_map_unlock_read(map)	vm_map_unlock_read_ln(map, __FILE__, __LINE__)
431 #define vm_map_busy(map)	vm_map_busy_ln(map, __FILE__, __LINE__)
432 #define vm_map_unbusy(map)	vm_map_unbusy_ln(map, __FILE__, __LINE__)
433 #define vm_map_assert_anylock(map)	\
434 		vm_map_assert_anylock_ln(map, __FILE__, __LINE__)
435 #define vm_map_assert_wrlock(map)	\
436 		vm_map_assert_wrlock_ln(map, __FILE__, __LINE__)
437 #else
438 #define vm_map_lock_try(map)	vm_map_lock_try_ln(map, NULL, 0)
439 #define vm_map_lock(map)	vm_map_lock_ln(map, NULL, 0)
440 #define vm_map_lock_read(map)	vm_map_lock_read_ln(map, NULL, 0)
441 #define vm_map_unlock(map)	vm_map_unlock_ln(map, NULL, 0)
442 #define vm_map_unlock_read(map)	vm_map_unlock_read_ln(map, NULL, 0)
443 #define vm_map_busy(map)	vm_map_busy_ln(map, NULL, 0)
444 #define vm_map_unbusy(map)	vm_map_unbusy_ln(map, NULL, 0)
445 #define vm_map_assert_anylock(map)	vm_map_assert_anylock_ln(map, NULL, 0)
446 #define vm_map_assert_wrlock(map)	vm_map_assert_wrlock_ln(map, NULL, 0)
447 #endif
448 
449 void		uvm_map_lock_entry(struct vm_map_entry *);
450 void		uvm_map_unlock_entry(struct vm_map_entry *);
451 
452 #endif /* _KERNEL */
453 
454 /*
455  *	Functions implemented as macros
456  */
457 #define		vm_map_min(map)		((map)->min_offset)
458 #define		vm_map_max(map)		((map)->max_offset)
459 #define		vm_map_pmap(map)	((map)->pmap)
460 
461 #endif /* _UVM_UVM_MAP_H_ */
462