xref: /openbsd/sys/uvm/uvm_map.h (revision 713a145f)
1 /*	$OpenBSD: uvm_map.h,v 1.93 2024/10/31 05:00:00 dlg 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_GUARDPAGES	r/o; must be specified at map
218  *					initialization time.
219  *					If set, guards will appear between
220  *					automatic allocations.
221  *					No locking required.
222  *
223  *		VM_MAP_ISVMSPACE	r/o; set by uvmspace_alloc.
224  *					Signifies that this map is a vmspace.
225  *					(The implementation treats all maps
226  *					without this bit as kernel maps.)
227  *					No locking required.
228  *
229  *
230  * All automatic allocations (uvm_map without MAP_FIXED) will allocate
231  * from vm_map.free.
232  * If that allocation fails:
233  * - vmspace maps will spill over into vm_map.bfree,
234  * - all other maps will call uvm_map_kmem_grow() to increase the arena.
235  *
236  * vmspace maps have their data, brk() and stack arenas automatically
237  * updated when uvm_map() is invoked without MAP_FIXED.
238  * The spill over arena (vm_map.bfree) will contain the space in the brk()
239  * and stack ranges.
240  * Kernel maps never have a bfree arena and this tree will always be empty.
241  *
242  *
243  * read_locks and write_locks are used in lock debugging code.
244  *
245  *  Locks used to protect struct members in this file:
246  *	a	atomic operations
247  *	I	immutable after creation or exec(2)
248  *	v	`vm_map_lock' (this map `lock' or `mtx')
249  *	f	flags_lock
250  */
251 struct vm_map {
252 	struct pmap		*pmap;		/* [I] Physical map */
253 	u_long			sserial;	/* [v] # stack changes */
254 
255 	struct uvm_map_addr	addr;		/* [v] Entry tree, by addr */
256 
257 	vsize_t			size;		/* virtual size */
258 	int			ref_count;	/* [a] Reference count */
259 	int			flags;		/* [f] flags */
260 	unsigned int		timestamp;	/* Version number */
261 	struct proc		*busy;		/* [f] thread holding map busy*/
262 	unsigned int		nbusy;		/* [f] waiters for busy */
263 
264 	vaddr_t			min_offset;	/* [I] First address in map. */
265 	vaddr_t			max_offset;	/* [I] Last address in map. */
266 
267 	/*
268 	 * Allocation overflow regions.
269 	 */
270 	vaddr_t			b_start;	/* [v] Start for brk() alloc. */
271 	vaddr_t			b_end;		/* [v] End for brk() alloc. */
272 	vaddr_t			s_start;	/* [v] Start for stack alloc. */
273 	vaddr_t			s_end;		/* [v] End for stack alloc. */
274 
275 	/*
276 	 * Special address selectors.
277 	 *
278 	 * The uaddr_exe mapping is used if:
279 	 * - protX is selected
280 	 * - the pointer is not NULL
281 	 *
282 	 * If uaddr_exe is not used, the other mappings are checked in
283 	 * order of appearance.
284 	 * If a hint is given, the selection will only be used if the hint
285 	 * falls in the range described by the mapping.
286 	 *
287 	 * The states are pointers because:
288 	 * - they may not all be in use
289 	 * - the struct size for different schemes is variable
290 	 *
291 	 * The uaddr_brk_stack selector will select addresses that are in
292 	 * the brk/stack area of the map.
293 	 */
294 	struct uvm_addr_state	*uaddr_exe;	/* Executable selector. */
295 	struct uvm_addr_state	*uaddr_any[4];	/* More selectors. */
296 	struct uvm_addr_state	*uaddr_brk_stack; /* Brk/stack selector. */
297 
298 #define UVM_MAP_CHECK_COPYIN_MAX 4	/* main, sigtramp, ld.so, libc.so */
299 	struct uvm_check_copyin {
300 		vaddr_t		start, end;
301 	}			check_copyin[UVM_MAP_CHECK_COPYIN_MAX];
302 	int			check_copyin_count;
303 
304 	/*
305 	 * XXX struct mutex changes size because of compile options, so
306 	 * place after fields which are inspected by libkvm / procmap(8)
307 	 */
308 	struct rwlock		lock;		/* Non-intrsafe lock */
309 	struct mutex		mtx;		/* Intrsafe lock */
310 	struct mutex		flags_lock;	/* flags lock */
311 };
312 
313 /* vm_map flags */
314 #define	VM_MAP_PAGEABLE		0x01		/* ro: entries are pageable */
315 #define	VM_MAP_INTRSAFE		0x02		/* ro: interrupt safe map */
316 #define	VM_MAP_WIREFUTURE	0x04		/* rw: wire future mappings */
317 #define	VM_MAP_GUARDPAGES	0x20		/* rw: add guard pgs to map */
318 #define	VM_MAP_ISVMSPACE	0x40		/* ro: map is a vmspace */
319 #define	VM_MAP_PINSYSCALL_ONCE	0x100		/* rw: pinsyscall done */
320 
321 /* Number of kernel maps and entries to statically allocate */
322 #define	MAX_KMAPENT	1024	/* Sufficient to make it to the scheduler. */
323 
324 #ifdef _KERNEL
325 /*
326  * globals:
327  */
328 
329 extern vaddr_t	uvm_maxkaddr;
330 
331 /*
332  * protos: the following prototypes define the interface to vm_map
333  */
334 
335 void		uvm_map_deallocate(struct vm_map *);
336 
337 int		uvm_map_clean(struct vm_map *, vaddr_t, vaddr_t, int);
338 void		uvm_map_clip_start(struct vm_map *, struct vm_map_entry *,
339 		    vaddr_t);
340 void		uvm_map_clip_end(struct vm_map *, struct vm_map_entry *,
341 		    vaddr_t);
342 int		uvm_map_extract(struct vm_map *, vaddr_t, vsize_t,
343 		    vaddr_t *, int);
344 struct vm_map *	uvm_map_create(pmap_t, vaddr_t, vaddr_t, int);
345 vaddr_t		uvm_map_pie(vaddr_t);
346 vaddr_t		uvm_map_hint(struct vmspace *, vm_prot_t, vaddr_t, vaddr_t);
347 int		uvm_map_check_copyin_add(struct vm_map *, vaddr_t, vaddr_t);
348 int		uvm_map_immutable(struct vm_map *, vaddr_t, vaddr_t, int);
349 int		uvm_map_inherit(struct vm_map *, vaddr_t, vaddr_t, vm_inherit_t);
350 int		uvm_map_advice(struct vm_map *, vaddr_t, vaddr_t, int);
351 void		uvm_map_init(void);
352 boolean_t	uvm_map_lookup_entry(struct vm_map *, vaddr_t, vm_map_entry_t *);
353 boolean_t	uvm_map_is_stack_remappable(struct vm_map *, vaddr_t, vsize_t, int);
354 int		uvm_map_remap_as_stack(struct proc *, vaddr_t, vsize_t);
355 void		uvm_map_setup(struct vm_map *, pmap_t, vaddr_t, vaddr_t, int);
356 int		uvm_map_submap(struct vm_map *, vaddr_t, vaddr_t,
357 		    struct vm_map *);
358 void		uvm_unmap(struct vm_map *, vaddr_t, vaddr_t);
359 void		uvm_unmap_detach(struct uvm_map_deadq *, int);
360 int		uvm_unmap_remove(struct vm_map*, vaddr_t, vaddr_t,
361 		    struct uvm_map_deadq *, boolean_t, boolean_t, boolean_t);
362 void		uvm_map_set_uaddr(struct vm_map*, struct uvm_addr_state**,
363 		    struct uvm_addr_state*);
364 int		uvm_map_mquery(struct vm_map*, vaddr_t*, vsize_t, voff_t, int);
365 
366 
367 struct p_inentry;
368 
369 int		uvm_map_inentry_sp(vm_map_entry_t);
370 boolean_t	uvm_map_inentry(struct proc *, struct p_inentry *, vaddr_t addr,
371 		    const char *fmt, int (*fn)(vm_map_entry_t), u_long serial);
372 
373 struct kinfo_vmentry;
374 
375 int		uvm_map_fill_vmmap(struct vm_map *, struct kinfo_vmentry *,
376 		    size_t *);
377 
378 /*
379  * VM map locking operations:
380  *
381  *	These operations perform locking on the data portion of the
382  *	map.
383  *
384  *	vm_map_lock_try: try to lock a map, failing if it is already locked.
385  *
386  *	vm_map_lock: acquire an exclusive (write) lock on a map.
387  *
388  *	vm_map_lock_read: acquire a shared (read) lock on a map.
389  *
390  *	vm_map_unlock: release an exclusive lock on a map.
391  *
392  *	vm_map_unlock_read: release a shared lock on a map.
393  *
394  *	vm_map_busy: mark a map as busy.
395  *
396  *	vm_map_unbusy: clear busy status on a map.
397  *
398  */
399 
400 boolean_t	vm_map_lock_try_ln(struct vm_map*, char*, int);
401 void		vm_map_lock_ln(struct vm_map*, char*, int);
402 void		vm_map_lock_read_ln(struct vm_map*, char*, int);
403 void		vm_map_unlock_ln(struct vm_map*, char*, int);
404 void		vm_map_unlock_read_ln(struct vm_map*, char*, int);
405 void		vm_map_busy_ln(struct vm_map*, char*, int);
406 void		vm_map_unbusy_ln(struct vm_map*, char*, int);
407 void		vm_map_assert_anylock_ln(struct vm_map*, char*, int);
408 void		vm_map_assert_wrlock_ln(struct vm_map*, char*, int);
409 
410 #ifdef DIAGNOSTIC
411 #define vm_map_lock_try(map)	vm_map_lock_try_ln(map, __FILE__, __LINE__)
412 #define vm_map_lock(map)	vm_map_lock_ln(map, __FILE__, __LINE__)
413 #define vm_map_lock_read(map)	vm_map_lock_read_ln(map, __FILE__, __LINE__)
414 #define vm_map_unlock(map)	vm_map_unlock_ln(map, __FILE__, __LINE__)
415 #define vm_map_unlock_read(map)	vm_map_unlock_read_ln(map, __FILE__, __LINE__)
416 #define vm_map_busy(map)	vm_map_busy_ln(map, __FILE__, __LINE__)
417 #define vm_map_unbusy(map)	vm_map_unbusy_ln(map, __FILE__, __LINE__)
418 #define vm_map_assert_anylock(map)	\
419 		vm_map_assert_anylock_ln(map, __FILE__, __LINE__)
420 #define vm_map_assert_wrlock(map)	\
421 		vm_map_assert_wrlock_ln(map, __FILE__, __LINE__)
422 #else
423 #define vm_map_lock_try(map)	vm_map_lock_try_ln(map, NULL, 0)
424 #define vm_map_lock(map)	vm_map_lock_ln(map, NULL, 0)
425 #define vm_map_lock_read(map)	vm_map_lock_read_ln(map, NULL, 0)
426 #define vm_map_unlock(map)	vm_map_unlock_ln(map, NULL, 0)
427 #define vm_map_unlock_read(map)	vm_map_unlock_read_ln(map, NULL, 0)
428 #define vm_map_busy(map)	vm_map_busy_ln(map, NULL, 0)
429 #define vm_map_unbusy(map)	vm_map_unbusy_ln(map, NULL, 0)
430 #define vm_map_assert_anylock(map)	vm_map_assert_anylock_ln(map, NULL, 0)
431 #define vm_map_assert_wrlock(map)	vm_map_assert_wrlock_ln(map, NULL, 0)
432 #endif
433 
434 void		uvm_map_lock_entry(struct vm_map_entry *);
435 void		uvm_map_unlock_entry(struct vm_map_entry *);
436 
437 #endif /* _KERNEL */
438 
439 /*
440  *	Functions implemented as macros
441  */
442 #define		vm_map_min(map)		((map)->min_offset)
443 #define		vm_map_max(map)		((map)->max_offset)
444 #define		vm_map_pmap(map)	((map)->pmap)
445 
446 #endif /* _UVM_UVM_MAP_H_ */
447