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, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/zfs_context.h> 30 #include <sys/spa.h> 31 #include <sys/dmu.h> 32 #include <sys/space_map.h> 33 34 /* 35 * Space map routines. 36 * NOTE: caller is responsible for all locking. 37 */ 38 static int 39 space_map_seg_compare(const void *x1, const void *x2) 40 { 41 const space_seg_t *s1 = x1; 42 const space_seg_t *s2 = x2; 43 44 if (s1->ss_start < s2->ss_start) { 45 if (s1->ss_end > s2->ss_start) 46 return (0); 47 return (-1); 48 } 49 if (s1->ss_start > s2->ss_start) { 50 if (s1->ss_start < s2->ss_end) 51 return (0); 52 return (1); 53 } 54 return (0); 55 } 56 57 void 58 space_map_create(space_map_t *sm, uint64_t start, uint64_t size, uint64_t shift, 59 kmutex_t *lp) 60 { 61 avl_create(&sm->sm_root, space_map_seg_compare, 62 sizeof (space_seg_t), offsetof(struct space_seg, ss_node)); 63 sm->sm_start = start; 64 sm->sm_end = start + size; 65 sm->sm_size = size; 66 sm->sm_shift = shift; 67 sm->sm_space = 0; 68 sm->sm_lock = lp; 69 } 70 71 void 72 space_map_destroy(space_map_t *sm) 73 { 74 VERIFY3U(sm->sm_space, ==, 0); 75 avl_destroy(&sm->sm_root); 76 } 77 78 void 79 space_map_add(space_map_t *sm, uint64_t start, uint64_t size) 80 { 81 avl_index_t where; 82 space_seg_t ssearch, *ss_before, *ss_after, *ss; 83 uint64_t end = start + size; 84 int merge_before, merge_after; 85 86 ASSERT(MUTEX_HELD(sm->sm_lock)); 87 VERIFY(size != 0); 88 VERIFY3U(start, >=, sm->sm_start); 89 VERIFY3U(end, <=, sm->sm_end); 90 VERIFY(sm->sm_space + size <= sm->sm_size); 91 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0); 92 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0); 93 94 ssearch.ss_start = start; 95 ssearch.ss_end = end; 96 ss = avl_find(&sm->sm_root, &ssearch, &where); 97 98 /* Make sure we don't overlap with either of our neighbors */ 99 VERIFY(ss == NULL); 100 101 ss_before = avl_nearest(&sm->sm_root, where, AVL_BEFORE); 102 ss_after = avl_nearest(&sm->sm_root, where, AVL_AFTER); 103 104 merge_before = (ss_before != NULL && ss_before->ss_end == start); 105 merge_after = (ss_after != NULL && ss_after->ss_start == end); 106 107 if (merge_before && merge_after) { 108 avl_remove(&sm->sm_root, ss_before); 109 ss_after->ss_start = ss_before->ss_start; 110 kmem_free(ss_before, sizeof (*ss_before)); 111 } else if (merge_before) { 112 ss_before->ss_end = end; 113 } else if (merge_after) { 114 ss_after->ss_start = start; 115 } else { 116 ss = kmem_alloc(sizeof (*ss), KM_SLEEP); 117 ss->ss_start = start; 118 ss->ss_end = end; 119 avl_insert(&sm->sm_root, ss, where); 120 } 121 122 sm->sm_space += size; 123 } 124 125 void 126 space_map_remove(space_map_t *sm, uint64_t start, uint64_t size) 127 { 128 avl_index_t where; 129 space_seg_t ssearch, *ss, *newseg; 130 uint64_t end = start + size; 131 int left_over, right_over; 132 133 ASSERT(MUTEX_HELD(sm->sm_lock)); 134 VERIFY(size != 0); 135 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0); 136 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0); 137 138 ssearch.ss_start = start; 139 ssearch.ss_end = end; 140 ss = avl_find(&sm->sm_root, &ssearch, &where); 141 142 /* Make sure we completely overlap with someone */ 143 VERIFY(ss != NULL); 144 VERIFY3U(ss->ss_start, <=, start); 145 VERIFY3U(ss->ss_end, >=, end); 146 VERIFY(sm->sm_space - size <= sm->sm_size); 147 148 left_over = (ss->ss_start != start); 149 right_over = (ss->ss_end != end); 150 151 if (left_over && right_over) { 152 newseg = kmem_alloc(sizeof (*newseg), KM_SLEEP); 153 newseg->ss_start = end; 154 newseg->ss_end = ss->ss_end; 155 ss->ss_end = start; 156 avl_insert_here(&sm->sm_root, newseg, ss, AVL_AFTER); 157 } else if (left_over) { 158 ss->ss_end = start; 159 } else if (right_over) { 160 ss->ss_start = end; 161 } else { 162 avl_remove(&sm->sm_root, ss); 163 kmem_free(ss, sizeof (*ss)); 164 } 165 166 sm->sm_space -= size; 167 } 168 169 int 170 space_map_contains(space_map_t *sm, uint64_t start, uint64_t size) 171 { 172 avl_index_t where; 173 space_seg_t ssearch, *ss; 174 uint64_t end = start + size; 175 176 ASSERT(MUTEX_HELD(sm->sm_lock)); 177 VERIFY(size != 0); 178 VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0); 179 VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0); 180 181 ssearch.ss_start = start; 182 ssearch.ss_end = end; 183 ss = avl_find(&sm->sm_root, &ssearch, &where); 184 185 return (ss != NULL && ss->ss_start <= start && ss->ss_end >= end); 186 } 187 188 void 189 space_map_vacate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest) 190 { 191 space_seg_t *ss; 192 void *cookie = NULL; 193 194 ASSERT(MUTEX_HELD(sm->sm_lock)); 195 196 while ((ss = avl_destroy_nodes(&sm->sm_root, &cookie)) != NULL) { 197 if (func != NULL) 198 func(mdest, ss->ss_start, ss->ss_end - ss->ss_start); 199 kmem_free(ss, sizeof (*ss)); 200 } 201 sm->sm_space = 0; 202 } 203 204 void 205 space_map_iterate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest) 206 { 207 space_seg_t *ss; 208 209 for (ss = avl_first(&sm->sm_root); ss; ss = AVL_NEXT(&sm->sm_root, ss)) 210 func(mdest, ss->ss_start, ss->ss_end - ss->ss_start); 211 } 212 213 void 214 space_map_merge(space_map_t *src, space_map_t *dest) 215 { 216 space_map_vacate(src, space_map_add, dest); 217 } 218 219 void 220 space_map_excise(space_map_t *sm, uint64_t start, uint64_t size) 221 { 222 avl_tree_t *t = &sm->sm_root; 223 avl_index_t where; 224 space_seg_t *ss, search; 225 uint64_t end = start + size; 226 uint64_t rm_start, rm_end; 227 228 ASSERT(MUTEX_HELD(sm->sm_lock)); 229 230 search.ss_start = start; 231 search.ss_end = start; 232 233 for (;;) { 234 ss = avl_find(t, &search, &where); 235 236 if (ss == NULL) 237 ss = avl_nearest(t, where, AVL_AFTER); 238 239 if (ss == NULL || ss->ss_start >= end) 240 break; 241 242 rm_start = MAX(ss->ss_start, start); 243 rm_end = MIN(ss->ss_end, end); 244 245 space_map_remove(sm, rm_start, rm_end - rm_start); 246 } 247 } 248 249 /* 250 * Replace smd with the union of smd and sms. 251 */ 252 void 253 space_map_union(space_map_t *smd, space_map_t *sms) 254 { 255 avl_tree_t *t = &sms->sm_root; 256 space_seg_t *ss; 257 258 ASSERT(MUTEX_HELD(smd->sm_lock)); 259 260 /* 261 * For each source segment, remove any intersections with the 262 * destination, then add the source segment to the destination. 263 */ 264 for (ss = avl_first(t); ss != NULL; ss = AVL_NEXT(t, ss)) { 265 space_map_excise(smd, ss->ss_start, ss->ss_end - ss->ss_start); 266 space_map_add(smd, ss->ss_start, ss->ss_end - ss->ss_start); 267 } 268 } 269 270 int 271 space_map_load(space_map_t *sm, space_map_obj_t *smo, uint8_t maptype, 272 objset_t *os, uint64_t end, uint64_t space) 273 { 274 uint64_t *entry, *entry_map, *entry_map_end; 275 uint64_t bufsize, size, offset; 276 uint64_t mapstart = sm->sm_start; 277 278 ASSERT(MUTEX_HELD(sm->sm_lock)); 279 VERIFY3U(sm->sm_space, ==, 0); 280 281 bufsize = MIN(end, SPACE_MAP_CHUNKSIZE); 282 entry_map = kmem_alloc(bufsize, KM_SLEEP); 283 284 if (maptype == SM_FREE) { 285 space_map_add(sm, sm->sm_start, sm->sm_size); 286 space = sm->sm_size - space; 287 } 288 289 for (offset = 0; offset < end; offset += bufsize) { 290 size = MIN(end - offset, bufsize); 291 VERIFY(P2PHASE(size, sizeof (uint64_t)) == 0); 292 VERIFY(size != 0); 293 294 dprintf("object=%llu offset=%llx size=%llx\n", 295 smo->smo_object, offset, size); 296 dmu_read(os, smo->smo_object, offset, size, 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 } 317 318 void 319 space_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 } 390 391 void 392 space_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 dmu_free_range(os, smo->smo_object, 0, smo->smo_objsize, tx); 398 399 smo->smo_objsize = 0; 400 401 VERIFY3U(sm->sm_space, ==, smo->smo_alloc); 402 space_map_sync(sm, NULL, smo, SM_ALLOC, os, tx); 403 404 dprintf("write sm object %llu from %llu to %llu bytes in txg %llu\n", 405 smo->smo_object, oldsize, smo->smo_objsize, dmu_tx_get_txg(tx)); 406 } 407