space_map.c revision ea8dc4b6d2251b437950c0056bc626b311c73c27
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#include <sys/zfs_context.h>
29#include <sys/spa.h>
30#include <sys/dmu.h>
31#include <sys/space_map.h>
32
33/*
34 * Space map routines.
35 * NOTE: caller is responsible for all locking.
36 */
37static int
38space_map_seg_compare(const void *x1, const void *x2)
39{
40	const space_seg_t *s1 = x1;
41	const space_seg_t *s2 = x2;
42
43	if (s1->ss_start < s2->ss_start) {
44		if (s1->ss_end > s2->ss_start)
45			return (0);
46		return (-1);
47	}
48	if (s1->ss_start > s2->ss_start) {
49		if (s1->ss_start < s2->ss_end)
50			return (0);
51		return (1);
52	}
53	return (0);
54}
55
56void
57space_map_create(space_map_t *sm, uint64_t start, uint64_t size, uint64_t shift,
58	kmutex_t *lp)
59{
60	avl_create(&sm->sm_root, space_map_seg_compare,
61	    sizeof (space_seg_t), offsetof(struct space_seg, ss_node));
62	sm->sm_start = start;
63	sm->sm_end = start + size;
64	sm->sm_size = size;
65	sm->sm_shift = shift;
66	sm->sm_space = 0;
67	sm->sm_lock = lp;
68}
69
70void
71space_map_destroy(space_map_t *sm)
72{
73	VERIFY3U(sm->sm_space, ==, 0);
74	avl_destroy(&sm->sm_root);
75}
76
77void
78space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
79{
80	avl_index_t where;
81	space_seg_t ssearch, *ss_before, *ss_after, *ss;
82	uint64_t end = start + size;
83	int merge_before, merge_after;
84
85	ASSERT(MUTEX_HELD(sm->sm_lock));
86	VERIFY(size != 0);
87	VERIFY3U(start, >=, sm->sm_start);
88	VERIFY3U(end, <=, sm->sm_end);
89	VERIFY(sm->sm_space + size <= sm->sm_size);
90	VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
91	VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
92
93	ssearch.ss_start = start;
94	ssearch.ss_end = end;
95	ss = avl_find(&sm->sm_root, &ssearch, &where);
96
97	/* Make sure we don't overlap with either of our neighbors */
98	VERIFY(ss == NULL);
99
100	ss_before = avl_nearest(&sm->sm_root, where, AVL_BEFORE);
101	ss_after = avl_nearest(&sm->sm_root, where, AVL_AFTER);
102
103	merge_before = (ss_before != NULL && ss_before->ss_end == start);
104	merge_after = (ss_after != NULL && ss_after->ss_start == end);
105
106	if (merge_before && merge_after) {
107		avl_remove(&sm->sm_root, ss_before);
108		ss_after->ss_start = ss_before->ss_start;
109		kmem_free(ss_before, sizeof (*ss_before));
110	} else if (merge_before) {
111		ss_before->ss_end = end;
112	} else if (merge_after) {
113		ss_after->ss_start = start;
114	} else {
115		ss = kmem_alloc(sizeof (*ss), KM_SLEEP);
116		ss->ss_start = start;
117		ss->ss_end = end;
118		avl_insert(&sm->sm_root, ss, where);
119	}
120
121	sm->sm_space += size;
122}
123
124void
125space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
126{
127	avl_index_t where;
128	space_seg_t ssearch, *ss, *newseg;
129	uint64_t end = start + size;
130	int left_over, right_over;
131
132	ASSERT(MUTEX_HELD(sm->sm_lock));
133	VERIFY(size != 0);
134	VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
135	VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
136
137	ssearch.ss_start = start;
138	ssearch.ss_end = end;
139	ss = avl_find(&sm->sm_root, &ssearch, &where);
140
141	/* Make sure we completely overlap with someone */
142	VERIFY(ss != NULL);
143	VERIFY3U(ss->ss_start, <=, start);
144	VERIFY3U(ss->ss_end, >=, end);
145	VERIFY(sm->sm_space - size <= sm->sm_size);
146
147	left_over = (ss->ss_start != start);
148	right_over = (ss->ss_end != end);
149
150	if (left_over && right_over) {
151		newseg = kmem_alloc(sizeof (*newseg), KM_SLEEP);
152		newseg->ss_start = end;
153		newseg->ss_end = ss->ss_end;
154		ss->ss_end = start;
155		avl_insert_here(&sm->sm_root, newseg, ss, AVL_AFTER);
156	} else if (left_over) {
157		ss->ss_end = start;
158	} else if (right_over) {
159		ss->ss_start = end;
160	} else {
161		avl_remove(&sm->sm_root, ss);
162		kmem_free(ss, sizeof (*ss));
163	}
164
165	sm->sm_space -= size;
166}
167
168int
169space_map_contains(space_map_t *sm, uint64_t start, uint64_t size)
170{
171	avl_index_t where;
172	space_seg_t ssearch, *ss;
173	uint64_t end = start + size;
174
175	ASSERT(MUTEX_HELD(sm->sm_lock));
176	VERIFY(size != 0);
177	VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
178	VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
179
180	ssearch.ss_start = start;
181	ssearch.ss_end = end;
182	ss = avl_find(&sm->sm_root, &ssearch, &where);
183
184	return (ss != NULL && ss->ss_start <= start && ss->ss_end >= end);
185}
186
187void
188space_map_vacate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
189{
190	space_seg_t *ss;
191	void *cookie = NULL;
192
193	ASSERT(MUTEX_HELD(sm->sm_lock));
194
195	while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
196		if (func != NULL)
197			func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
198		kmem_free(ss, sizeof (*ss));
199	}
200	sm->sm_space = 0;
201}
202
203void
204space_map_iterate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
205{
206	space_seg_t *ss;
207
208	for (ss = avl_first(&sm->sm_root); ss; ss = AVL_NEXT(&sm->sm_root, ss))
209		func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
210}
211
212void
213space_map_merge(space_map_t *src, space_map_t *dest)
214{
215	space_map_vacate(src, space_map_add, dest);
216}
217
218void
219space_map_excise(space_map_t *sm, uint64_t start, uint64_t size)
220{
221	avl_tree_t *t = &sm->sm_root;
222	avl_index_t where;
223	space_seg_t *ss, search;
224	uint64_t end = start + size;
225	uint64_t rm_start, rm_end;
226
227	ASSERT(MUTEX_HELD(sm->sm_lock));
228
229	search.ss_start = start;
230	search.ss_end = start;
231
232	for (;;) {
233		ss = avl_find(t, &search, &where);
234
235		if (ss == NULL)
236			ss = avl_nearest(t, where, AVL_AFTER);
237
238		if (ss == NULL || ss->ss_start >= end)
239			break;
240
241		rm_start = MAX(ss->ss_start, start);
242		rm_end = MIN(ss->ss_end, end);
243
244		space_map_remove(sm, rm_start, rm_end - rm_start);
245	}
246}
247
248/*
249 * Replace smd with the union of smd and sms.
250 */
251void
252space_map_union(space_map_t *smd, space_map_t *sms)
253{
254	avl_tree_t *t = &sms->sm_root;
255	space_seg_t *ss;
256
257	ASSERT(MUTEX_HELD(smd->sm_lock));
258
259	/*
260	 * For each source segment, remove any intersections with the
261	 * destination, then add the source segment to the destination.
262	 */
263	for (ss = avl_first(t); ss != NULL; ss = AVL_NEXT(t, ss)) {
264		space_map_excise(smd, ss->ss_start, ss->ss_end - ss->ss_start);
265		space_map_add(smd, ss->ss_start, ss->ss_end - ss->ss_start);
266	}
267}
268
269int
270space_map_load(space_map_t *sm, space_map_obj_t *smo, uint8_t maptype,
271	objset_t *os, uint64_t end, uint64_t space)
272{
273	uint64_t *entry, *entry_map, *entry_map_end;
274	uint64_t bufsize, size, offset;
275	uint64_t mapstart = sm->sm_start;
276
277	ASSERT(MUTEX_HELD(sm->sm_lock));
278	VERIFY3U(sm->sm_space, ==, 0);
279
280	bufsize = MIN(end, SPACE_MAP_CHUNKSIZE);
281	entry_map = kmem_alloc(bufsize, KM_SLEEP);
282
283	if (maptype == SM_FREE) {
284		space_map_add(sm, sm->sm_start, sm->sm_size);
285		space = sm->sm_size - space;
286	}
287
288	for (offset = 0; offset < end; offset += bufsize) {
289		size = MIN(end - offset, bufsize);
290		VERIFY(P2PHASE(size, sizeof (uint64_t)) == 0);
291		VERIFY(size != 0);
292
293		dprintf("object=%llu  offset=%llx  size=%llx\n",
294		    smo->smo_object, offset, size);
295		VERIFY(0 == dmu_read(os, smo->smo_object, offset, size,
296		    entry_map));
297
298		entry_map_end = entry_map + (size / sizeof (uint64_t));
299		for (entry = entry_map; entry < entry_map_end; entry++) {
300			uint64_t e = *entry;
301
302			if (SM_DEBUG_DECODE(e))		/* Skip debug entries */
303				continue;
304
305			(SM_TYPE_DECODE(e) == maptype ?
306			    space_map_add : space_map_remove)(sm,
307			    (SM_OFFSET_DECODE(e) << sm->sm_shift) + mapstart,
308			    SM_RUN_DECODE(e) << sm->sm_shift);
309		}
310	}
311	VERIFY3U(sm->sm_space, ==, space);
312
313	kmem_free(entry_map, bufsize);
314
315	return (0);
316}
317
318void
319space_map_sync(space_map_t *sm, space_map_t *dest, space_map_obj_t *smo,
320    uint8_t maptype, objset_t *os, dmu_tx_t *tx)
321{
322	spa_t *spa = dmu_objset_spa(os);
323	void *cookie = NULL;
324	space_seg_t *ss;
325	uint64_t bufsize, start, size, run_len;
326	uint64_t *entry, *entry_map, *entry_map_end;
327
328	ASSERT(MUTEX_HELD(sm->sm_lock));
329
330	if (sm->sm_space == 0)
331		return;
332
333	dprintf("object %4llu, txg %llu, pass %d, %c, count %lu, space %llx\n",
334	    smo->smo_object, dmu_tx_get_txg(tx), spa_sync_pass(spa),
335	    maptype == SM_ALLOC ? 'A' : 'F', avl_numnodes(&sm->sm_root),
336	    sm->sm_space);
337
338	bufsize = (8 + avl_numnodes(&sm->sm_root)) * sizeof (uint64_t);
339	bufsize = MIN(bufsize, SPACE_MAP_CHUNKSIZE);
340	entry_map = kmem_alloc(bufsize, KM_SLEEP);
341	entry_map_end = entry_map + (bufsize / sizeof (uint64_t));
342	entry = entry_map;
343
344	*entry++ = SM_DEBUG_ENCODE(1) |
345	    SM_DEBUG_ACTION_ENCODE(maptype) |
346	    SM_DEBUG_SYNCPASS_ENCODE(spa_sync_pass(spa)) |
347	    SM_DEBUG_TXG_ENCODE(dmu_tx_get_txg(tx));
348
349	while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
350		size = ss->ss_end - ss->ss_start;
351		start = (ss->ss_start - sm->sm_start) >> sm->sm_shift;
352
353		if (dest)
354			space_map_add(dest, ss->ss_start, size);
355
356		sm->sm_space -= size;
357		size >>= sm->sm_shift;
358
359		while (size) {
360			run_len = MIN(size, SM_RUN_MAX);
361
362			if (entry == entry_map_end) {
363				dmu_write(os, smo->smo_object, smo->smo_objsize,
364				    bufsize, entry_map, tx);
365				smo->smo_objsize += bufsize;
366				entry = entry_map;
367			}
368
369			*entry++ = SM_OFFSET_ENCODE(start) |
370			    SM_TYPE_ENCODE(maptype) |
371			    SM_RUN_ENCODE(run_len);
372
373			start += run_len;
374			size -= run_len;
375		}
376		kmem_free(ss, sizeof (*ss));
377	}
378
379	if (entry != entry_map) {
380		size = (entry - entry_map) * sizeof (uint64_t);
381		dmu_write(os, smo->smo_object, smo->smo_objsize,
382		    size, entry_map, tx);
383		smo->smo_objsize += size;
384	}
385
386	kmem_free(entry_map, bufsize);
387
388	VERIFY3U(sm->sm_space, ==, 0);
389}
390
391void
392space_map_write(space_map_t *sm, space_map_obj_t *smo, objset_t *os,
393    dmu_tx_t *tx)
394{
395	uint64_t oldsize = smo->smo_objsize;
396
397	VERIFY(0 == dmu_free_range(os, smo->smo_object, 0,
398	    smo->smo_objsize, tx));
399
400	smo->smo_objsize = 0;
401
402	VERIFY3U(sm->sm_space, ==, smo->smo_alloc);
403	space_map_sync(sm, NULL, smo, SM_ALLOC, os, tx);
404
405	dprintf("write sm object %llu from %llu to %llu bytes in txg %llu\n",
406	    smo->smo_object, oldsize, smo->smo_objsize, dmu_tx_get_txg(tx));
407}
408