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.
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.
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]
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #pragma ident "@(#)space_map.c 1.7 08/03/11 SMI"
28 #include <sys/zfs_context.h>
32 #include <sys/space_map.h>
36 * NOTE: caller is responsible for all locking.
39 space_map_seg_compare(const void *x1, const void *x2)
41 const space_seg_t *s1 = x1;
42 const space_seg_t *s2 = x2;
44 if (s1->ss_start < s2->ss_start) {
45 if (s1->ss_end > s2->ss_start)
49 if (s1->ss_start > s2->ss_start) {
50 if (s1->ss_start < s2->ss_end)
58 space_map_create(space_map_t *sm, uint64_t start, uint64_t size, uint8_t shift,
61 bzero(sm, sizeof (*sm));
63 avl_create(&sm->sm_root, space_map_seg_compare,
64 sizeof (space_seg_t), offsetof(struct space_seg, ss_node));
73 space_map_destroy(space_map_t *sm)
75 ASSERT(!sm->sm_loaded && !sm->sm_loading);
76 VERIFY3U(sm->sm_space, ==, 0);
77 avl_destroy(&sm->sm_root);
81 space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
84 space_seg_t ssearch, *ss_before, *ss_after, *ss;
85 uint64_t end = start + size;
86 int merge_before, merge_after;
88 ASSERT(MUTEX_HELD(sm->sm_lock));
90 VERIFY3U(start, >=, sm->sm_start);
91 VERIFY3U(end, <=, sm->sm_start + sm->sm_size);
92 VERIFY(sm->sm_space + size <= sm->sm_size);
93 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
94 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
96 ssearch.ss_start = start;
98 ss = avl_find(&sm->sm_root, &ssearch, &where);
100 if (ss != NULL && ss->ss_start <= start && ss->ss_end >= end) {
101 zfs_panic_recover("zfs: allocating allocated segment"
102 "(offset=%llu size=%llu)\n",
103 (longlong_t)start, (longlong_t)size);
107 /* Make sure we don't overlap with either of our neighbors */
110 ss_before = avl_nearest(&sm->sm_root, where, AVL_BEFORE);
111 ss_after = avl_nearest(&sm->sm_root, where, AVL_AFTER);
113 merge_before = (ss_before != NULL && ss_before->ss_end == start);
114 merge_after = (ss_after != NULL && ss_after->ss_start == end);
116 if (merge_before && merge_after) {
117 avl_remove(&sm->sm_root, ss_before);
118 ss_after->ss_start = ss_before->ss_start;
119 kmem_free(ss_before, sizeof (*ss_before));
120 } else if (merge_before) {
121 ss_before->ss_end = end;
122 } else if (merge_after) {
123 ss_after->ss_start = start;
125 ss = kmem_alloc(sizeof (*ss), KM_SLEEP);
126 ss->ss_start = start;
128 avl_insert(&sm->sm_root, ss, where);
131 sm->sm_space += size;
135 space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
138 space_seg_t ssearch, *ss, *newseg;
139 uint64_t end = start + size;
140 int left_over, right_over;
142 ASSERT(MUTEX_HELD(sm->sm_lock));
144 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
145 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
147 ssearch.ss_start = start;
148 ssearch.ss_end = end;
149 ss = avl_find(&sm->sm_root, &ssearch, &where);
151 /* Make sure we completely overlap with someone */
153 zfs_panic_recover("zfs: freeing free segment "
154 "(offset=%llu size=%llu)",
155 (longlong_t)start, (longlong_t)size);
158 VERIFY3U(ss->ss_start, <=, start);
159 VERIFY3U(ss->ss_end, >=, end);
160 VERIFY(sm->sm_space - size <= sm->sm_size);
162 left_over = (ss->ss_start != start);
163 right_over = (ss->ss_end != end);
165 if (left_over && right_over) {
166 newseg = kmem_alloc(sizeof (*newseg), KM_SLEEP);
167 newseg->ss_start = end;
168 newseg->ss_end = ss->ss_end;
170 avl_insert_here(&sm->sm_root, newseg, ss, AVL_AFTER);
171 } else if (left_over) {
173 } else if (right_over) {
176 avl_remove(&sm->sm_root, ss);
177 kmem_free(ss, sizeof (*ss));
180 sm->sm_space -= size;
184 space_map_contains(space_map_t *sm, uint64_t start, uint64_t size)
187 space_seg_t ssearch, *ss;
188 uint64_t end = start + size;
190 ASSERT(MUTEX_HELD(sm->sm_lock));
192 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
193 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
195 ssearch.ss_start = start;
196 ssearch.ss_end = end;
197 ss = avl_find(&sm->sm_root, &ssearch, &where);
199 return (ss != NULL && ss->ss_start <= start && ss->ss_end >= end);
203 space_map_vacate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
208 ASSERT(MUTEX_HELD(sm->sm_lock));
210 while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
212 func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
213 kmem_free(ss, sizeof (*ss));
219 space_map_walk(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
223 for (ss = avl_first(&sm->sm_root); ss; ss = AVL_NEXT(&sm->sm_root, ss))
224 func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
228 space_map_excise(space_map_t *sm, uint64_t start, uint64_t size)
230 avl_tree_t *t = &sm->sm_root;
232 space_seg_t *ss, search;
233 uint64_t end = start + size;
234 uint64_t rm_start, rm_end;
236 ASSERT(MUTEX_HELD(sm->sm_lock));
238 search.ss_start = start;
239 search.ss_end = start;
242 ss = avl_find(t, &search, &where);
245 ss = avl_nearest(t, where, AVL_AFTER);
247 if (ss == NULL || ss->ss_start >= end)
250 rm_start = MAX(ss->ss_start, start);
251 rm_end = MIN(ss->ss_end, end);
253 space_map_remove(sm, rm_start, rm_end - rm_start);
258 * Replace smd with the union of smd and sms.
261 space_map_union(space_map_t *smd, space_map_t *sms)
263 avl_tree_t *t = &sms->sm_root;
266 ASSERT(MUTEX_HELD(smd->sm_lock));
269 * For each source segment, remove any intersections with the
270 * destination, then add the source segment to the destination.
272 for (ss = avl_first(t); ss != NULL; ss = AVL_NEXT(t, ss)) {
273 space_map_excise(smd, ss->ss_start, ss->ss_end - ss->ss_start);
274 space_map_add(smd, ss->ss_start, ss->ss_end - ss->ss_start);
279 * Wait for any in-progress space_map_load() to complete.
282 space_map_load_wait(space_map_t *sm)
284 ASSERT(MUTEX_HELD(sm->sm_lock));
286 while (sm->sm_loading)
287 cv_wait(&sm->sm_load_cv, sm->sm_lock);
291 * Note: space_map_load() will drop sm_lock across dmu_read() calls.
292 * The caller must be OK with this.
295 space_map_load(space_map_t *sm, space_map_ops_t *ops, uint8_t maptype,
296 space_map_obj_t *smo, objset_t *os)
298 uint64_t *entry, *entry_map, *entry_map_end;
299 uint64_t bufsize, size, offset, end, space;
300 uint64_t mapstart = sm->sm_start;
303 ASSERT(MUTEX_HELD(sm->sm_lock));
305 space_map_load_wait(sm);
310 sm->sm_loading = B_TRUE;
311 end = smo->smo_objsize;
312 space = smo->smo_alloc;
314 ASSERT(sm->sm_ops == NULL);
315 VERIFY3U(sm->sm_space, ==, 0);
317 if (maptype == SM_FREE) {
318 space_map_add(sm, sm->sm_start, sm->sm_size);
319 space = sm->sm_size - space;
322 bufsize = 1ULL << SPACE_MAP_BLOCKSHIFT;
323 entry_map = zio_buf_alloc(bufsize);
325 mutex_exit(sm->sm_lock);
327 dmu_prefetch(os, smo->smo_object, bufsize, end - bufsize);
328 mutex_enter(sm->sm_lock);
330 for (offset = 0; offset < end; offset += bufsize) {
331 size = MIN(end - offset, bufsize);
332 VERIFY(P2PHASE(size, sizeof (uint64_t)) == 0);
335 dprintf("object=%llu offset=%llx size=%llx\n",
336 smo->smo_object, offset, size);
338 mutex_exit(sm->sm_lock);
339 error = dmu_read(os, smo->smo_object, offset, size, entry_map);
340 mutex_enter(sm->sm_lock);
344 entry_map_end = entry_map + (size / sizeof (uint64_t));
345 for (entry = entry_map; entry < entry_map_end; entry++) {
348 if (SM_DEBUG_DECODE(e)) /* Skip debug entries */
351 (SM_TYPE_DECODE(e) == maptype ?
352 space_map_add : space_map_remove)(sm,
353 (SM_OFFSET_DECODE(e) << sm->sm_shift) + mapstart,
354 SM_RUN_DECODE(e) << sm->sm_shift);
359 VERIFY3U(sm->sm_space, ==, space);
361 sm->sm_loaded = B_TRUE;
366 space_map_vacate(sm, NULL, NULL);
369 zio_buf_free(entry_map, bufsize);
371 sm->sm_loading = B_FALSE;
373 cv_broadcast(&sm->sm_load_cv);
379 space_map_unload(space_map_t *sm)
381 ASSERT(MUTEX_HELD(sm->sm_lock));
383 if (sm->sm_loaded && sm->sm_ops != NULL)
384 sm->sm_ops->smop_unload(sm);
386 sm->sm_loaded = B_FALSE;
389 space_map_vacate(sm, NULL, NULL);
393 space_map_alloc(space_map_t *sm, uint64_t size)
397 start = sm->sm_ops->smop_alloc(sm, size);
399 space_map_remove(sm, start, size);
404 space_map_claim(space_map_t *sm, uint64_t start, uint64_t size)
406 sm->sm_ops->smop_claim(sm, start, size);
407 space_map_remove(sm, start, size);
411 space_map_free(space_map_t *sm, uint64_t start, uint64_t size)
413 space_map_add(sm, start, size);
414 sm->sm_ops->smop_free(sm, start, size);
418 * Note: space_map_sync() will drop sm_lock across dmu_write() calls.
421 space_map_sync(space_map_t *sm, uint8_t maptype,
422 space_map_obj_t *smo, objset_t *os, dmu_tx_t *tx)
424 spa_t *spa = dmu_objset_spa(os);
427 uint64_t bufsize, start, size, run_len;
428 uint64_t *entry, *entry_map, *entry_map_end;
430 ASSERT(MUTEX_HELD(sm->sm_lock));
432 if (sm->sm_space == 0)
435 dprintf("object %4llu, txg %llu, pass %d, %c, count %lu, space %llx\n",
436 smo->smo_object, dmu_tx_get_txg(tx), spa_sync_pass(spa),
437 maptype == SM_ALLOC ? 'A' : 'F', avl_numnodes(&sm->sm_root),
440 if (maptype == SM_ALLOC)
441 smo->smo_alloc += sm->sm_space;
443 smo->smo_alloc -= sm->sm_space;
445 bufsize = (8 + avl_numnodes(&sm->sm_root)) * sizeof (uint64_t);
446 bufsize = MIN(bufsize, 1ULL << SPACE_MAP_BLOCKSHIFT);
447 entry_map = zio_buf_alloc(bufsize);
448 entry_map_end = entry_map + (bufsize / sizeof (uint64_t));
451 *entry++ = SM_DEBUG_ENCODE(1) |
452 SM_DEBUG_ACTION_ENCODE(maptype) |
453 SM_DEBUG_SYNCPASS_ENCODE(spa_sync_pass(spa)) |
454 SM_DEBUG_TXG_ENCODE(dmu_tx_get_txg(tx));
456 while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
457 size = ss->ss_end - ss->ss_start;
458 start = (ss->ss_start - sm->sm_start) >> sm->sm_shift;
460 sm->sm_space -= size;
461 size >>= sm->sm_shift;
464 run_len = MIN(size, SM_RUN_MAX);
466 if (entry == entry_map_end) {
467 mutex_exit(sm->sm_lock);
468 dmu_write(os, smo->smo_object, smo->smo_objsize,
469 bufsize, entry_map, tx);
470 mutex_enter(sm->sm_lock);
471 smo->smo_objsize += bufsize;
475 *entry++ = SM_OFFSET_ENCODE(start) |
476 SM_TYPE_ENCODE(maptype) |
477 SM_RUN_ENCODE(run_len);
482 kmem_free(ss, sizeof (*ss));
485 if (entry != entry_map) {
486 size = (entry - entry_map) * sizeof (uint64_t);
487 mutex_exit(sm->sm_lock);
488 dmu_write(os, smo->smo_object, smo->smo_objsize,
489 size, entry_map, tx);
490 mutex_enter(sm->sm_lock);
491 smo->smo_objsize += size;
494 zio_buf_free(entry_map, bufsize);
496 VERIFY3U(sm->sm_space, ==, 0);
500 space_map_truncate(space_map_obj_t *smo, objset_t *os, dmu_tx_t *tx)
502 VERIFY(dmu_free_range(os, smo->smo_object, 0, -1ULL, tx) == 0);
504 smo->smo_objsize = 0;