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 #include <sys/zfs_context.h>
30 #include <sys/space_map.h>
34 * NOTE: caller is responsible for all locking.
37 space_map_seg_compare(const void *x1, const void *x2)
39 const space_seg_t *s1 = x1;
40 const space_seg_t *s2 = x2;
42 if (s1->ss_start < s2->ss_start) {
43 if (s1->ss_end > s2->ss_start)
47 if (s1->ss_start > s2->ss_start) {
48 if (s1->ss_start < s2->ss_end)
56 space_map_create(space_map_t *sm, uint64_t start, uint64_t size, uint8_t shift,
59 bzero(sm, sizeof (*sm));
61 cv_init(&sm->sm_load_cv, NULL, CV_DEFAULT, NULL);
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);
78 cv_destroy(&sm->sm_load_cv);
82 space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
85 space_seg_t ssearch, *ss_before, *ss_after, *ss;
86 uint64_t end = start + size;
87 int merge_before, merge_after;
89 ASSERT(MUTEX_HELD(sm->sm_lock));
91 VERIFY3U(start, >=, sm->sm_start);
92 VERIFY3U(end, <=, sm->sm_start + sm->sm_size);
93 VERIFY(sm->sm_space + size <= sm->sm_size);
94 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
95 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
97 ssearch.ss_start = start;
99 ss = avl_find(&sm->sm_root, &ssearch, &where);
101 if (ss != NULL && ss->ss_start <= start && ss->ss_end >= end) {
102 zfs_panic_recover("zfs: allocating allocated segment"
103 "(offset=%llu size=%llu)\n",
104 (longlong_t)start, (longlong_t)size);
108 /* Make sure we don't overlap with either of our neighbors */
111 ss_before = avl_nearest(&sm->sm_root, where, AVL_BEFORE);
112 ss_after = avl_nearest(&sm->sm_root, where, AVL_AFTER);
114 merge_before = (ss_before != NULL && ss_before->ss_end == start);
115 merge_after = (ss_after != NULL && ss_after->ss_start == end);
117 if (merge_before && merge_after) {
118 avl_remove(&sm->sm_root, ss_before);
119 ss_after->ss_start = ss_before->ss_start;
120 kmem_free(ss_before, sizeof (*ss_before));
121 } else if (merge_before) {
122 ss_before->ss_end = end;
123 } else if (merge_after) {
124 ss_after->ss_start = start;
126 ss = kmem_alloc(sizeof (*ss), KM_SLEEP);
127 ss->ss_start = start;
129 avl_insert(&sm->sm_root, ss, where);
132 sm->sm_space += size;
136 space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
139 space_seg_t ssearch, *ss, *newseg;
140 uint64_t end = start + size;
141 int left_over, right_over;
143 ASSERT(MUTEX_HELD(sm->sm_lock));
145 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
146 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
148 ssearch.ss_start = start;
149 ssearch.ss_end = end;
150 ss = avl_find(&sm->sm_root, &ssearch, &where);
152 /* Make sure we completely overlap with someone */
154 zfs_panic_recover("zfs: freeing free segment "
155 "(offset=%llu size=%llu)",
156 (longlong_t)start, (longlong_t)size);
159 VERIFY3U(ss->ss_start, <=, start);
160 VERIFY3U(ss->ss_end, >=, end);
161 VERIFY(sm->sm_space - size <= sm->sm_size);
163 left_over = (ss->ss_start != start);
164 right_over = (ss->ss_end != end);
166 if (left_over && right_over) {
167 newseg = kmem_alloc(sizeof (*newseg), KM_SLEEP);
168 newseg->ss_start = end;
169 newseg->ss_end = ss->ss_end;
171 avl_insert_here(&sm->sm_root, newseg, ss, AVL_AFTER);
172 } else if (left_over) {
174 } else if (right_over) {
177 avl_remove(&sm->sm_root, ss);
178 kmem_free(ss, sizeof (*ss));
181 sm->sm_space -= size;
185 space_map_contains(space_map_t *sm, uint64_t start, uint64_t size)
188 space_seg_t ssearch, *ss;
189 uint64_t end = start + size;
191 ASSERT(MUTEX_HELD(sm->sm_lock));
193 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
194 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
196 ssearch.ss_start = start;
197 ssearch.ss_end = end;
198 ss = avl_find(&sm->sm_root, &ssearch, &where);
200 return (ss != NULL && ss->ss_start <= start && ss->ss_end >= end);
204 space_map_vacate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
209 ASSERT(MUTEX_HELD(sm->sm_lock));
211 while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
213 func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
214 kmem_free(ss, sizeof (*ss));
220 space_map_walk(space_map_t *sm, space_map_func_t *func, space_map_t *mdest)
224 for (ss = avl_first(&sm->sm_root); ss; ss = AVL_NEXT(&sm->sm_root, ss))
225 func(mdest, ss->ss_start, ss->ss_end - ss->ss_start);
229 space_map_excise(space_map_t *sm, uint64_t start, uint64_t size)
231 avl_tree_t *t = &sm->sm_root;
233 space_seg_t *ss, search;
234 uint64_t end = start + size;
235 uint64_t rm_start, rm_end;
237 ASSERT(MUTEX_HELD(sm->sm_lock));
239 search.ss_start = start;
240 search.ss_end = start;
243 ss = avl_find(t, &search, &where);
246 ss = avl_nearest(t, where, AVL_AFTER);
248 if (ss == NULL || ss->ss_start >= end)
251 rm_start = MAX(ss->ss_start, start);
252 rm_end = MIN(ss->ss_end, end);
254 space_map_remove(sm, rm_start, rm_end - rm_start);
259 * Replace smd with the union of smd and sms.
262 space_map_union(space_map_t *smd, space_map_t *sms)
264 avl_tree_t *t = &sms->sm_root;
267 ASSERT(MUTEX_HELD(smd->sm_lock));
270 * For each source segment, remove any intersections with the
271 * destination, then add the source segment to the destination.
273 for (ss = avl_first(t); ss != NULL; ss = AVL_NEXT(t, ss)) {
274 space_map_excise(smd, ss->ss_start, ss->ss_end - ss->ss_start);
275 space_map_add(smd, ss->ss_start, ss->ss_end - ss->ss_start);
280 * Wait for any in-progress space_map_load() to complete.
283 space_map_load_wait(space_map_t *sm)
285 ASSERT(MUTEX_HELD(sm->sm_lock));
287 while (sm->sm_loading)
288 cv_wait(&sm->sm_load_cv, sm->sm_lock);
292 * Note: space_map_load() will drop sm_lock across dmu_read() calls.
293 * The caller must be OK with this.
296 space_map_load(space_map_t *sm, space_map_ops_t *ops, uint8_t maptype,
297 space_map_obj_t *smo, objset_t *os)
299 uint64_t *entry, *entry_map, *entry_map_end;
300 uint64_t bufsize, size, offset, end, space;
301 uint64_t mapstart = sm->sm_start;
304 ASSERT(MUTEX_HELD(sm->sm_lock));
306 space_map_load_wait(sm);
311 sm->sm_loading = B_TRUE;
312 end = smo->smo_objsize;
313 space = smo->smo_alloc;
315 ASSERT(sm->sm_ops == NULL);
316 VERIFY3U(sm->sm_space, ==, 0);
318 if (maptype == SM_FREE) {
319 space_map_add(sm, sm->sm_start, sm->sm_size);
320 space = sm->sm_size - space;
323 bufsize = 1ULL << SPACE_MAP_BLOCKSHIFT;
324 entry_map = zio_buf_alloc(bufsize);
326 mutex_exit(sm->sm_lock);
328 dmu_prefetch(os, smo->smo_object, bufsize, end - bufsize);
329 mutex_enter(sm->sm_lock);
331 for (offset = 0; offset < end; offset += bufsize) {
332 size = MIN(end - offset, bufsize);
333 VERIFY(P2PHASE(size, sizeof (uint64_t)) == 0);
336 dprintf("object=%llu offset=%llx size=%llx\n",
337 smo->smo_object, offset, size);
339 mutex_exit(sm->sm_lock);
340 error = dmu_read(os, smo->smo_object, offset, size, entry_map);
341 mutex_enter(sm->sm_lock);
345 entry_map_end = entry_map + (size / sizeof (uint64_t));
346 for (entry = entry_map; entry < entry_map_end; entry++) {
349 if (SM_DEBUG_DECODE(e)) /* Skip debug entries */
352 (SM_TYPE_DECODE(e) == maptype ?
353 space_map_add : space_map_remove)(sm,
354 (SM_OFFSET_DECODE(e) << sm->sm_shift) + mapstart,
355 SM_RUN_DECODE(e) << sm->sm_shift);
360 VERIFY3U(sm->sm_space, ==, space);
362 sm->sm_loaded = B_TRUE;
367 space_map_vacate(sm, NULL, NULL);
370 zio_buf_free(entry_map, bufsize);
372 sm->sm_loading = B_FALSE;
374 cv_broadcast(&sm->sm_load_cv);
380 space_map_unload(space_map_t *sm)
382 ASSERT(MUTEX_HELD(sm->sm_lock));
384 if (sm->sm_loaded && sm->sm_ops != NULL)
385 sm->sm_ops->smop_unload(sm);
387 sm->sm_loaded = B_FALSE;
390 space_map_vacate(sm, NULL, NULL);
394 space_map_alloc(space_map_t *sm, uint64_t size)
398 start = sm->sm_ops->smop_alloc(sm, size);
400 space_map_remove(sm, start, size);
405 space_map_claim(space_map_t *sm, uint64_t start, uint64_t size)
407 sm->sm_ops->smop_claim(sm, start, size);
408 space_map_remove(sm, start, size);
412 space_map_free(space_map_t *sm, uint64_t start, uint64_t size)
414 space_map_add(sm, start, size);
415 sm->sm_ops->smop_free(sm, start, size);
419 * Note: space_map_sync() will drop sm_lock across dmu_write() calls.
422 space_map_sync(space_map_t *sm, uint8_t maptype,
423 space_map_obj_t *smo, objset_t *os, dmu_tx_t *tx)
425 spa_t *spa = dmu_objset_spa(os);
428 uint64_t bufsize, start, size, run_len;
429 uint64_t *entry, *entry_map, *entry_map_end;
431 ASSERT(MUTEX_HELD(sm->sm_lock));
433 if (sm->sm_space == 0)
436 dprintf("object %4llu, txg %llu, pass %d, %c, count %lu, space %llx\n",
437 smo->smo_object, dmu_tx_get_txg(tx), spa_sync_pass(spa),
438 maptype == SM_ALLOC ? 'A' : 'F', avl_numnodes(&sm->sm_root),
441 if (maptype == SM_ALLOC)
442 smo->smo_alloc += sm->sm_space;
444 smo->smo_alloc -= sm->sm_space;
446 bufsize = (8 + avl_numnodes(&sm->sm_root)) * sizeof (uint64_t);
447 bufsize = MIN(bufsize, 1ULL << SPACE_MAP_BLOCKSHIFT);
448 entry_map = zio_buf_alloc(bufsize);
449 entry_map_end = entry_map + (bufsize / sizeof (uint64_t));
452 *entry++ = SM_DEBUG_ENCODE(1) |
453 SM_DEBUG_ACTION_ENCODE(maptype) |
454 SM_DEBUG_SYNCPASS_ENCODE(spa_sync_pass(spa)) |
455 SM_DEBUG_TXG_ENCODE(dmu_tx_get_txg(tx));
457 while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) {
458 size = ss->ss_end - ss->ss_start;
459 start = (ss->ss_start - sm->sm_start) >> sm->sm_shift;
461 sm->sm_space -= size;
462 size >>= sm->sm_shift;
465 run_len = MIN(size, SM_RUN_MAX);
467 if (entry == entry_map_end) {
468 mutex_exit(sm->sm_lock);
469 dmu_write(os, smo->smo_object, smo->smo_objsize,
470 bufsize, entry_map, tx);
471 mutex_enter(sm->sm_lock);
472 smo->smo_objsize += bufsize;
476 *entry++ = SM_OFFSET_ENCODE(start) |
477 SM_TYPE_ENCODE(maptype) |
478 SM_RUN_ENCODE(run_len);
483 kmem_free(ss, sizeof (*ss));
486 if (entry != entry_map) {
487 size = (entry - entry_map) * sizeof (uint64_t);
488 mutex_exit(sm->sm_lock);
489 dmu_write(os, smo->smo_object, smo->smo_objsize,
490 size, entry_map, tx);
491 mutex_enter(sm->sm_lock);
492 smo->smo_objsize += size;
495 zio_buf_free(entry_map, bufsize);
497 VERIFY3U(sm->sm_space, ==, 0);
501 space_map_truncate(space_map_obj_t *smo, objset_t *os, dmu_tx_t *tx)
503 VERIFY(dmu_free_range(os, smo->smo_object, 0, -1ULL, tx) == 0);
505 smo->smo_objsize = 0;