1 789 ahrens /* 2 789 ahrens * CDDL HEADER START 3 789 ahrens * 4 789 ahrens * The contents of this file are subject to the terms of the 5 1491 ahrens * Common Development and Distribution License (the "License"). 6 1491 ahrens * You may not use this file except in compliance with the License. 7 789 ahrens * 8 789 ahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 789 ahrens * or http://www.opensolaris.org/os/licensing. 10 789 ahrens * See the License for the specific language governing permissions 11 789 ahrens * and limitations under the License. 12 789 ahrens * 13 789 ahrens * When distributing Covered Code, include this CDDL HEADER in each 14 789 ahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 789 ahrens * If applicable, add the following below this CDDL HEADER, with the 16 789 ahrens * fields enclosed by brackets "[]" replaced with your own identifying 17 789 ahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18 789 ahrens * 19 789 ahrens * CDDL HEADER END 20 789 ahrens */ 21 789 ahrens /* 22 8582 Brendan * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 789 ahrens * Use is subject to license terms. 24 789 ahrens */ 25 789 ahrens 26 789 ahrens #include <sys/zfs_context.h> 27 789 ahrens #include <sys/dbuf.h> 28 789 ahrens #include <sys/dnode.h> 29 789 ahrens #include <sys/dmu.h> 30 789 ahrens #include <sys/dmu_impl.h> 31 789 ahrens #include <sys/dmu_tx.h> 32 789 ahrens #include <sys/dmu_objset.h> 33 789 ahrens #include <sys/dsl_dir.h> 34 789 ahrens #include <sys/dsl_dataset.h> 35 789 ahrens #include <sys/spa.h> 36 789 ahrens #include <sys/zio.h> 37 789 ahrens #include <sys/dmu_zfetch.h> 38 789 ahrens 39 789 ahrens static int free_range_compar(const void *node1, const void *node2); 40 789 ahrens 41 789 ahrens static kmem_cache_t *dnode_cache; 42 789 ahrens 43 789 ahrens static dnode_phys_t dnode_phys_zero; 44 789 ahrens 45 789 ahrens int zfs_default_bs = SPA_MINBLOCKSHIFT; 46 789 ahrens int zfs_default_ibs = DN_MAX_INDBLKSHIFT; 47 789 ahrens 48 789 ahrens /* ARGSUSED */ 49 789 ahrens static int 50 789 ahrens dnode_cons(void *arg, void *unused, int kmflag) 51 789 ahrens { 52 789 ahrens int i; 53 789 ahrens dnode_t *dn = arg; 54 789 ahrens bzero(dn, sizeof (dnode_t)); 55 789 ahrens 56 789 ahrens rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); 57 789 ahrens mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); 58 789 ahrens mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); 59 8214 Ricardo cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); 60 8214 Ricardo 61 789 ahrens refcount_create(&dn->dn_holds); 62 789 ahrens refcount_create(&dn->dn_tx_holds); 63 789 ahrens 64 789 ahrens for (i = 0; i < TXG_SIZE; i++) { 65 789 ahrens avl_create(&dn->dn_ranges[i], free_range_compar, 66 789 ahrens sizeof (free_range_t), 67 789 ahrens offsetof(struct free_range, fr_node)); 68 3547 maybee list_create(&dn->dn_dirty_records[i], 69 3547 maybee sizeof (dbuf_dirty_record_t), 70 3547 maybee offsetof(dbuf_dirty_record_t, dr_dirty_node)); 71 789 ahrens } 72 789 ahrens 73 789 ahrens list_create(&dn->dn_dbufs, sizeof (dmu_buf_impl_t), 74 789 ahrens offsetof(dmu_buf_impl_t, db_link)); 75 789 ahrens 76 789 ahrens return (0); 77 789 ahrens } 78 789 ahrens 79 789 ahrens /* ARGSUSED */ 80 789 ahrens static void 81 789 ahrens dnode_dest(void *arg, void *unused) 82 789 ahrens { 83 789 ahrens int i; 84 789 ahrens dnode_t *dn = arg; 85 789 ahrens 86 789 ahrens rw_destroy(&dn->dn_struct_rwlock); 87 789 ahrens mutex_destroy(&dn->dn_mtx); 88 789 ahrens mutex_destroy(&dn->dn_dbufs_mtx); 89 8214 Ricardo cv_destroy(&dn->dn_notxholds); 90 789 ahrens refcount_destroy(&dn->dn_holds); 91 789 ahrens refcount_destroy(&dn->dn_tx_holds); 92 789 ahrens 93 789 ahrens for (i = 0; i < TXG_SIZE; i++) { 94 789 ahrens avl_destroy(&dn->dn_ranges[i]); 95 3547 maybee list_destroy(&dn->dn_dirty_records[i]); 96 789 ahrens } 97 789 ahrens 98 789 ahrens list_destroy(&dn->dn_dbufs); 99 789 ahrens } 100 789 ahrens 101 789 ahrens void 102 789 ahrens dnode_init(void) 103 789 ahrens { 104 789 ahrens dnode_cache = kmem_cache_create("dnode_t", 105 789 ahrens sizeof (dnode_t), 106 789 ahrens 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); 107 789 ahrens } 108 789 ahrens 109 789 ahrens void 110 789 ahrens dnode_fini(void) 111 789 ahrens { 112 789 ahrens kmem_cache_destroy(dnode_cache); 113 789 ahrens } 114 789 ahrens 115 789 ahrens 116 873 ek110237 #ifdef ZFS_DEBUG 117 789 ahrens void 118 789 ahrens dnode_verify(dnode_t *dn) 119 789 ahrens { 120 789 ahrens int drop_struct_lock = FALSE; 121 789 ahrens 122 789 ahrens ASSERT(dn->dn_phys); 123 789 ahrens ASSERT(dn->dn_objset); 124 789 ahrens 125 789 ahrens ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 126 789 ahrens 127 789 ahrens if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) 128 789 ahrens return; 129 789 ahrens 130 789 ahrens if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 131 789 ahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 132 789 ahrens drop_struct_lock = TRUE; 133 789 ahrens } 134 789 ahrens if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { 135 789 ahrens int i; 136 789 ahrens ASSERT3U(dn->dn_indblkshift, >=, 0); 137 789 ahrens ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); 138 789 ahrens if (dn->dn_datablkshift) { 139 789 ahrens ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); 140 789 ahrens ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); 141 789 ahrens ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); 142 789 ahrens } 143 789 ahrens ASSERT3U(dn->dn_nlevels, <=, 30); 144 789 ahrens ASSERT3U(dn->dn_type, <=, DMU_OT_NUMTYPES); 145 789 ahrens ASSERT3U(dn->dn_nblkptr, >=, 1); 146 789 ahrens ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 147 789 ahrens ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); 148 789 ahrens ASSERT3U(dn->dn_datablksz, ==, 149 789 ahrens dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 150 789 ahrens ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); 151 789 ahrens ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + 152 789 ahrens dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); 153 789 ahrens for (i = 0; i < TXG_SIZE; i++) { 154 789 ahrens ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); 155 789 ahrens } 156 789 ahrens } 157 789 ahrens if (dn->dn_phys->dn_type != DMU_OT_NONE) 158 789 ahrens ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); 159 9396 Matthew ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL); 160 789 ahrens if (dn->dn_dbuf != NULL) { 161 789 ahrens ASSERT3P(dn->dn_phys, ==, 162 789 ahrens (dnode_phys_t *)dn->dn_dbuf->db.db_data + 163 789 ahrens (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); 164 789 ahrens } 165 789 ahrens if (drop_struct_lock) 166 789 ahrens rw_exit(&dn->dn_struct_rwlock); 167 873 ek110237 } 168 789 ahrens #endif 169 789 ahrens 170 789 ahrens void 171 789 ahrens dnode_byteswap(dnode_phys_t *dnp) 172 789 ahrens { 173 789 ahrens uint64_t *buf64 = (void*)&dnp->dn_blkptr; 174 789 ahrens int i; 175 789 ahrens 176 789 ahrens if (dnp->dn_type == DMU_OT_NONE) { 177 789 ahrens bzero(dnp, sizeof (dnode_phys_t)); 178 789 ahrens return; 179 789 ahrens } 180 789 ahrens 181 789 ahrens dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); 182 789 ahrens dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); 183 789 ahrens dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); 184 2082 eschrock dnp->dn_used = BSWAP_64(dnp->dn_used); 185 789 ahrens 186 789 ahrens /* 187 789 ahrens * dn_nblkptr is only one byte, so it's OK to read it in either 188 789 ahrens * byte order. We can't read dn_bouslen. 189 789 ahrens */ 190 789 ahrens ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); 191 789 ahrens ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); 192 789 ahrens for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) 193 789 ahrens buf64[i] = BSWAP_64(buf64[i]); 194 789 ahrens 195 789 ahrens /* 196 789 ahrens * OK to check dn_bonuslen for zero, because it won't matter if 197 789 ahrens * we have the wrong byte order. This is necessary because the 198 789 ahrens * dnode dnode is smaller than a regular dnode. 199 789 ahrens */ 200 789 ahrens if (dnp->dn_bonuslen != 0) { 201 789 ahrens /* 202 789 ahrens * Note that the bonus length calculated here may be 203 789 ahrens * longer than the actual bonus buffer. This is because 204 789 ahrens * we always put the bonus buffer after the last block 205 789 ahrens * pointer (instead of packing it against the end of the 206 789 ahrens * dnode buffer). 207 789 ahrens */ 208 789 ahrens int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); 209 789 ahrens size_t len = DN_MAX_BONUSLEN - off; 210 3882 ahrens ASSERT3U(dnp->dn_bonustype, <, DMU_OT_NUMTYPES); 211 789 ahrens dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); 212 789 ahrens } 213 789 ahrens } 214 789 ahrens 215 789 ahrens void 216 789 ahrens dnode_buf_byteswap(void *vbuf, size_t size) 217 789 ahrens { 218 789 ahrens dnode_phys_t *buf = vbuf; 219 789 ahrens int i; 220 789 ahrens 221 789 ahrens ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); 222 789 ahrens ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); 223 789 ahrens 224 789 ahrens size >>= DNODE_SHIFT; 225 789 ahrens for (i = 0; i < size; i++) { 226 789 ahrens dnode_byteswap(buf); 227 789 ahrens buf++; 228 789 ahrens } 229 789 ahrens } 230 789 ahrens 231 789 ahrens static int 232 789 ahrens free_range_compar(const void *node1, const void *node2) 233 789 ahrens { 234 789 ahrens const free_range_t *rp1 = node1; 235 789 ahrens const free_range_t *rp2 = node2; 236 789 ahrens 237 789 ahrens if (rp1->fr_blkid < rp2->fr_blkid) 238 789 ahrens return (-1); 239 789 ahrens else if (rp1->fr_blkid > rp2->fr_blkid) 240 789 ahrens return (1); 241 789 ahrens else return (0); 242 789 ahrens } 243 789 ahrens 244 4944 maybee void 245 4944 maybee dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) 246 4944 maybee { 247 4944 maybee ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); 248 4944 maybee 249 4944 maybee dnode_setdirty(dn, tx); 250 4944 maybee rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 251 4944 maybee ASSERT3U(newsize, <=, DN_MAX_BONUSLEN - 252 4944 maybee (dn->dn_nblkptr-1) * sizeof (blkptr_t)); 253 4944 maybee dn->dn_bonuslen = newsize; 254 4944 maybee if (newsize == 0) 255 4944 maybee dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; 256 4944 maybee else 257 4944 maybee dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; 258 4944 maybee rw_exit(&dn->dn_struct_rwlock); 259 4944 maybee } 260 4944 maybee 261 789 ahrens static void 262 789 ahrens dnode_setdblksz(dnode_t *dn, int size) 263 789 ahrens { 264 789 ahrens ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0); 265 789 ahrens ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 266 789 ahrens ASSERT3U(size, >=, SPA_MINBLOCKSIZE); 267 789 ahrens ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 268 789 ahrens 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); 269 789 ahrens dn->dn_datablksz = size; 270 789 ahrens dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; 271 789 ahrens dn->dn_datablkshift = ISP2(size) ? highbit(size - 1) : 0; 272 789 ahrens } 273 789 ahrens 274 789 ahrens static dnode_t * 275 10298 Matthew dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, 276 789 ahrens uint64_t object) 277 789 ahrens { 278 789 ahrens dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); 279 789 ahrens (void) dnode_cons(dn, NULL, 0); /* XXX */ 280 789 ahrens 281 789 ahrens dn->dn_objset = os; 282 789 ahrens dn->dn_object = object; 283 789 ahrens dn->dn_dbuf = db; 284 789 ahrens dn->dn_phys = dnp; 285 789 ahrens 286 789 ahrens if (dnp->dn_datablkszsec) 287 789 ahrens dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 288 789 ahrens dn->dn_indblkshift = dnp->dn_indblkshift; 289 789 ahrens dn->dn_nlevels = dnp->dn_nlevels; 290 789 ahrens dn->dn_type = dnp->dn_type; 291 789 ahrens dn->dn_nblkptr = dnp->dn_nblkptr; 292 789 ahrens dn->dn_checksum = dnp->dn_checksum; 293 789 ahrens dn->dn_compress = dnp->dn_compress; 294 789 ahrens dn->dn_bonustype = dnp->dn_bonustype; 295 789 ahrens dn->dn_bonuslen = dnp->dn_bonuslen; 296 789 ahrens dn->dn_maxblkid = dnp->dn_maxblkid; 297 789 ahrens 298 789 ahrens dmu_zfetch_init(&dn->dn_zfetch, dn); 299 789 ahrens 300 789 ahrens ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 301 789 ahrens mutex_enter(&os->os_lock); 302 789 ahrens list_insert_head(&os->os_dnodes, dn); 303 789 ahrens mutex_exit(&os->os_lock); 304 789 ahrens 305 8582 Brendan arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER); 306 789 ahrens return (dn); 307 789 ahrens } 308 789 ahrens 309 789 ahrens static void 310 789 ahrens dnode_destroy(dnode_t *dn) 311 789 ahrens { 312 10298 Matthew objset_t *os = dn->dn_objset; 313 2885 ahrens 314 2885 ahrens #ifdef ZFS_DEBUG 315 2885 ahrens int i; 316 2885 ahrens 317 2885 ahrens for (i = 0; i < TXG_SIZE; i++) { 318 2885 ahrens ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 319 3547 maybee ASSERT(NULL == list_head(&dn->dn_dirty_records[i])); 320 2885 ahrens ASSERT(0 == avl_numnodes(&dn->dn_ranges[i])); 321 2885 ahrens } 322 2885 ahrens ASSERT(NULL == list_head(&dn->dn_dbufs)); 323 2885 ahrens #endif 324 9396 Matthew ASSERT(dn->dn_oldphys == NULL); 325 789 ahrens 326 789 ahrens mutex_enter(&os->os_lock); 327 789 ahrens list_remove(&os->os_dnodes, dn); 328 789 ahrens mutex_exit(&os->os_lock); 329 789 ahrens 330 789 ahrens if (dn->dn_dirtyctx_firstset) { 331 789 ahrens kmem_free(dn->dn_dirtyctx_firstset, 1); 332 789 ahrens dn->dn_dirtyctx_firstset = NULL; 333 789 ahrens } 334 789 ahrens dmu_zfetch_rele(&dn->dn_zfetch); 335 1544 eschrock if (dn->dn_bonus) { 336 1544 eschrock mutex_enter(&dn->dn_bonus->db_mtx); 337 1544 eschrock dbuf_evict(dn->dn_bonus); 338 1544 eschrock dn->dn_bonus = NULL; 339 1544 eschrock } 340 789 ahrens kmem_cache_free(dnode_cache, dn); 341 8582 Brendan arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER); 342 789 ahrens } 343 789 ahrens 344 789 ahrens void 345 789 ahrens dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 346 1599 ahrens dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 347 789 ahrens { 348 789 ahrens int i; 349 789 ahrens 350 789 ahrens if (blocksize == 0) 351 789 ahrens blocksize = 1 << zfs_default_bs; 352 1402 ahrens else if (blocksize > SPA_MAXBLOCKSIZE) 353 1402 ahrens blocksize = SPA_MAXBLOCKSIZE; 354 1402 ahrens else 355 1402 ahrens blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); 356 789 ahrens 357 789 ahrens if (ibs == 0) 358 789 ahrens ibs = zfs_default_ibs; 359 789 ahrens 360 789 ahrens ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); 361 789 ahrens 362 789 ahrens dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, 363 789 ahrens dn->dn_object, tx->tx_txg, blocksize, ibs); 364 789 ahrens 365 789 ahrens ASSERT(dn->dn_type == DMU_OT_NONE); 366 789 ahrens ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); 367 789 ahrens ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); 368 789 ahrens ASSERT(ot != DMU_OT_NONE); 369 789 ahrens ASSERT3U(ot, <, DMU_OT_NUMTYPES); 370 789 ahrens ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 371 789 ahrens (bonustype != DMU_OT_NONE && bonuslen != 0)); 372 789 ahrens ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 373 789 ahrens ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 374 789 ahrens ASSERT(dn->dn_type == DMU_OT_NONE); 375 789 ahrens ASSERT3U(dn->dn_maxblkid, ==, 0); 376 789 ahrens ASSERT3U(dn->dn_allocated_txg, ==, 0); 377 789 ahrens ASSERT3U(dn->dn_assigned_txg, ==, 0); 378 789 ahrens ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 379 789 ahrens ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); 380 789 ahrens ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 381 789 ahrens 382 789 ahrens for (i = 0; i < TXG_SIZE; i++) { 383 789 ahrens ASSERT3U(dn->dn_next_nlevels[i], ==, 0); 384 789 ahrens ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); 385 4944 maybee ASSERT3U(dn->dn_next_bonuslen[i], ==, 0); 386 1596 ahrens ASSERT3U(dn->dn_next_blksz[i], ==, 0); 387 1596 ahrens ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 388 3547 maybee ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); 389 789 ahrens ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0); 390 789 ahrens } 391 789 ahrens 392 789 ahrens dn->dn_type = ot; 393 789 ahrens dnode_setdblksz(dn, blocksize); 394 789 ahrens dn->dn_indblkshift = ibs; 395 789 ahrens dn->dn_nlevels = 1; 396 789 ahrens dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 397 789 ahrens dn->dn_bonustype = bonustype; 398 789 ahrens dn->dn_bonuslen = bonuslen; 399 789 ahrens dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 400 789 ahrens dn->dn_compress = ZIO_COMPRESS_INHERIT; 401 789 ahrens dn->dn_dirtyctx = 0; 402 789 ahrens 403 789 ahrens dn->dn_free_txg = 0; 404 789 ahrens if (dn->dn_dirtyctx_firstset) { 405 789 ahrens kmem_free(dn->dn_dirtyctx_firstset, 1); 406 789 ahrens dn->dn_dirtyctx_firstset = NULL; 407 789 ahrens } 408 789 ahrens 409 789 ahrens dn->dn_allocated_txg = tx->tx_txg; 410 1599 ahrens 411 789 ahrens dnode_setdirty(dn, tx); 412 1599 ahrens dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; 413 4944 maybee dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; 414 1599 ahrens dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; 415 789 ahrens } 416 789 ahrens 417 789 ahrens void 418 789 ahrens dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 419 789 ahrens dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 420 789 ahrens { 421 8986 Mark int nblkptr; 422 1596 ahrens 423 789 ahrens ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); 424 789 ahrens ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); 425 789 ahrens ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0); 426 1544 eschrock ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 427 789 ahrens ASSERT(tx->tx_txg != 0); 428 789 ahrens ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 429 789 ahrens (bonustype != DMU_OT_NONE && bonuslen != 0)); 430 789 ahrens ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 431 789 ahrens ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 432 1596 ahrens 433 1544 eschrock /* clean up any unreferenced dbufs */ 434 4944 maybee dnode_evict_dbufs(dn); 435 789 ahrens 436 8986 Mark rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 437 8986 Mark dnode_setdirty(dn, tx); 438 8986 Mark if (dn->dn_datablksz != blocksize) { 439 8986 Mark /* change blocksize */ 440 8986 Mark ASSERT(dn->dn_maxblkid == 0 && 441 8986 Mark (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || 442 8986 Mark dnode_block_freed(dn, 0))); 443 8986 Mark dnode_setdblksz(dn, blocksize); 444 8986 Mark dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; 445 789 ahrens } 446 8986 Mark if (dn->dn_bonuslen != bonuslen) 447 8986 Mark dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; 448 8644 Mark nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 449 8644 Mark if (dn->dn_nblkptr != nblkptr) 450 8644 Mark dn->dn_next_nblkptr[tx->tx_txg&TXG_MASK] = nblkptr; 451 789 ahrens rw_exit(&dn->dn_struct_rwlock); 452 789 ahrens 453 789 ahrens /* change type */ 454 789 ahrens dn->dn_type = ot; 455 789 ahrens 456 789 ahrens /* change bonus size and type */ 457 789 ahrens mutex_enter(&dn->dn_mtx); 458 789 ahrens dn->dn_bonustype = bonustype; 459 789 ahrens dn->dn_bonuslen = bonuslen; 460 8644 Mark dn->dn_nblkptr = nblkptr; 461 789 ahrens dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 462 789 ahrens dn->dn_compress = ZIO_COMPRESS_INHERIT; 463 789 ahrens ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 464 3087 ahrens 465 8644 Mark /* fix up the bonus db_size */ 466 8644 Mark if (dn->dn_bonus) { 467 4944 maybee dn->dn_bonus->db.db_size = 468 4944 maybee DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); 469 4944 maybee ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); 470 4944 maybee } 471 789 ahrens 472 789 ahrens dn->dn_allocated_txg = tx->tx_txg; 473 789 ahrens mutex_exit(&dn->dn_mtx); 474 789 ahrens } 475 789 ahrens 476 789 ahrens void 477 789 ahrens dnode_special_close(dnode_t *dn) 478 789 ahrens { 479 1544 eschrock /* 480 1544 eschrock * Wait for final references to the dnode to clear. This can 481 1544 eschrock * only happen if the arc is asyncronously evicting state that 482 1544 eschrock * has a hold on this dnode while we are trying to evict this 483 1544 eschrock * dnode. 484 1544 eschrock */ 485 1544 eschrock while (refcount_count(&dn->dn_holds) > 0) 486 1544 eschrock delay(1); 487 789 ahrens dnode_destroy(dn); 488 789 ahrens } 489 789 ahrens 490 789 ahrens dnode_t * 491 10298 Matthew dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object) 492 789 ahrens { 493 789 ahrens dnode_t *dn = dnode_create(os, dnp, NULL, object); 494 873 ek110237 DNODE_VERIFY(dn); 495 789 ahrens return (dn); 496 789 ahrens } 497 789 ahrens 498 789 ahrens static void 499 789 ahrens dnode_buf_pageout(dmu_buf_t *db, void *arg) 500 789 ahrens { 501 789 ahrens dnode_t **children_dnodes = arg; 502 789 ahrens int i; 503 789 ahrens int epb = db->db_size >> DNODE_SHIFT; 504 789 ahrens 505 789 ahrens for (i = 0; i < epb; i++) { 506 789 ahrens dnode_t *dn = children_dnodes[i]; 507 789 ahrens int n; 508 789 ahrens 509 789 ahrens if (dn == NULL) 510 789 ahrens continue; 511 789 ahrens #ifdef ZFS_DEBUG 512 789 ahrens /* 513 789 ahrens * If there are holds on this dnode, then there should 514 789 ahrens * be holds on the dnode's containing dbuf as well; thus 515 789 ahrens * it wouldn't be eligable for eviction and this function 516 789 ahrens * would not have been called. 517 789 ahrens */ 518 789 ahrens ASSERT(refcount_is_zero(&dn->dn_holds)); 519 789 ahrens ASSERT(list_head(&dn->dn_dbufs) == NULL); 520 789 ahrens ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 521 789 ahrens 522 789 ahrens for (n = 0; n < TXG_SIZE; n++) 523 1596 ahrens ASSERT(!list_link_active(&dn->dn_dirty_link[n])); 524 789 ahrens #endif 525 789 ahrens children_dnodes[i] = NULL; 526 789 ahrens dnode_destroy(dn); 527 789 ahrens } 528 789 ahrens kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 529 789 ahrens } 530 789 ahrens 531 789 ahrens /* 532 1544 eschrock * errors: 533 1544 eschrock * EINVAL - invalid object number. 534 1544 eschrock * EIO - i/o error. 535 1544 eschrock * succeeds even for free dnodes. 536 789 ahrens */ 537 1544 eschrock int 538 10298 Matthew dnode_hold_impl(objset_t *os, uint64_t object, int flag, 539 1544 eschrock void *tag, dnode_t **dnp) 540 789 ahrens { 541 1544 eschrock int epb, idx, err; 542 789 ahrens int drop_struct_lock = FALSE; 543 1544 eschrock int type; 544 789 ahrens uint64_t blk; 545 789 ahrens dnode_t *mdn, *dn; 546 789 ahrens dmu_buf_impl_t *db; 547 789 ahrens dnode_t **children_dnodes; 548 7754 Jeff 549 7754 Jeff /* 550 7754 Jeff * If you are holding the spa config lock as writer, you shouldn't 551 7754 Jeff * be asking the DMU to do *anything*. 552 7754 Jeff */ 553 7754 Jeff ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0); 554 789 ahrens 555 9396 Matthew if (object == DMU_USERUSED_OBJECT || object == DMU_GROUPUSED_OBJECT) { 556 9396 Matthew dn = (object == DMU_USERUSED_OBJECT) ? 557 9396 Matthew os->os_userused_dnode : os->os_groupused_dnode; 558 9396 Matthew if (dn == NULL) 559 9396 Matthew return (ENOENT); 560 9396 Matthew type = dn->dn_type; 561 9396 Matthew if ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) 562 9396 Matthew return (ENOENT); 563 9396 Matthew if ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE) 564 9396 Matthew return (EEXIST); 565 9396 Matthew DNODE_VERIFY(dn); 566 9396 Matthew (void) refcount_add(&dn->dn_holds, tag); 567 9396 Matthew *dnp = dn; 568 9396 Matthew return (0); 569 9396 Matthew } 570 9396 Matthew 571 789 ahrens if (object == 0 || object >= DN_MAX_OBJECT) 572 1544 eschrock return (EINVAL); 573 789 ahrens 574 789 ahrens mdn = os->os_meta_dnode; 575 789 ahrens 576 873 ek110237 DNODE_VERIFY(mdn); 577 789 ahrens 578 789 ahrens if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { 579 789 ahrens rw_enter(&mdn->dn_struct_rwlock, RW_READER); 580 789 ahrens drop_struct_lock = TRUE; 581 789 ahrens } 582 789 ahrens 583 789 ahrens blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); 584 789 ahrens 585 1544 eschrock db = dbuf_hold(mdn, blk, FTAG); 586 789 ahrens if (drop_struct_lock) 587 789 ahrens rw_exit(&mdn->dn_struct_rwlock); 588 1544 eschrock if (db == NULL) 589 1544 eschrock return (EIO); 590 1544 eschrock err = dbuf_read(db, NULL, DB_RF_CANFAIL); 591 1544 eschrock if (err) { 592 1544 eschrock dbuf_rele(db, FTAG); 593 1544 eschrock return (err); 594 1544 eschrock } 595 789 ahrens 596 789 ahrens ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); 597 789 ahrens epb = db->db.db_size >> DNODE_SHIFT; 598 789 ahrens 599 789 ahrens idx = object & (epb-1); 600 789 ahrens 601 789 ahrens children_dnodes = dmu_buf_get_user(&db->db); 602 789 ahrens if (children_dnodes == NULL) { 603 789 ahrens dnode_t **winner; 604 789 ahrens children_dnodes = kmem_zalloc(epb * sizeof (dnode_t *), 605 789 ahrens KM_SLEEP); 606 789 ahrens if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, 607 789 ahrens dnode_buf_pageout)) { 608 789 ahrens kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 609 789 ahrens children_dnodes = winner; 610 789 ahrens } 611 789 ahrens } 612 789 ahrens 613 789 ahrens if ((dn = children_dnodes[idx]) == NULL) { 614 4309 maybee dnode_phys_t *dnp = (dnode_phys_t *)db->db.db_data+idx; 615 789 ahrens dnode_t *winner; 616 4309 maybee 617 4309 maybee dn = dnode_create(os, dnp, db, object); 618 789 ahrens winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn); 619 789 ahrens if (winner != NULL) { 620 789 ahrens dnode_destroy(dn); 621 789 ahrens dn = winner; 622 789 ahrens } 623 789 ahrens } 624 789 ahrens 625 789 ahrens mutex_enter(&dn->dn_mtx); 626 1544 eschrock type = dn->dn_type; 627 789 ahrens if (dn->dn_free_txg || 628 1544 eschrock ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || 629 9396 Matthew ((flag & DNODE_MUST_BE_FREE) && 630 9396 Matthew (type != DMU_OT_NONE || dn->dn_oldphys))) { 631 789 ahrens mutex_exit(&dn->dn_mtx); 632 1544 eschrock dbuf_rele(db, FTAG); 633 1544 eschrock return (type == DMU_OT_NONE ? ENOENT : EEXIST); 634 789 ahrens } 635 789 ahrens mutex_exit(&dn->dn_mtx); 636 789 ahrens 637 1544 eschrock if (refcount_add(&dn->dn_holds, tag) == 1) 638 789 ahrens dbuf_add_ref(db, dn); 639 789 ahrens 640 873 ek110237 DNODE_VERIFY(dn); 641 789 ahrens ASSERT3P(dn->dn_dbuf, ==, db); 642 789 ahrens ASSERT3U(dn->dn_object, ==, object); 643 1544 eschrock dbuf_rele(db, FTAG); 644 789 ahrens 645 1544 eschrock *dnp = dn; 646 1544 eschrock return (0); 647 789 ahrens } 648 789 ahrens 649 789 ahrens /* 650 789 ahrens * Return held dnode if the object is allocated, NULL if not. 651 789 ahrens */ 652 1544 eschrock int 653 10298 Matthew dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp) 654 789 ahrens { 655 1544 eschrock return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); 656 789 ahrens } 657 789 ahrens 658 4944 maybee /* 659 4944 maybee * Can only add a reference if there is already at least one 660 4944 maybee * reference on the dnode. Returns FALSE if unable to add a 661 4944 maybee * new reference. 662 4944 maybee */ 663 4944 maybee boolean_t 664 1544 eschrock dnode_add_ref(dnode_t *dn, void *tag) 665 789 ahrens { 666 4944 maybee mutex_enter(&dn->dn_mtx); 667 4944 maybee if (refcount_is_zero(&dn->dn_holds)) { 668 4944 maybee mutex_exit(&dn->dn_mtx); 669 4944 maybee return (FALSE); 670 4944 maybee } 671 4944 maybee VERIFY(1 < refcount_add(&dn->dn_holds, tag)); 672 4944 maybee mutex_exit(&dn->dn_mtx); 673 4944 maybee return (TRUE); 674 789 ahrens } 675 789 ahrens 676 789 ahrens void 677 1544 eschrock dnode_rele(dnode_t *dn, void *tag) 678 789 ahrens { 679 789 ahrens uint64_t refs; 680 789 ahrens 681 4944 maybee mutex_enter(&dn->dn_mtx); 682 1544 eschrock refs = refcount_remove(&dn->dn_holds, tag); 683 4944 maybee mutex_exit(&dn->dn_mtx); 684 789 ahrens /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ 685 789 ahrens if (refs == 0 && dn->dn_dbuf) 686 1544 eschrock dbuf_rele(dn->dn_dbuf, dn); 687 789 ahrens } 688 789 ahrens 689 789 ahrens void 690 789 ahrens dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) 691 789 ahrens { 692 10298 Matthew objset_t *os = dn->dn_objset; 693 789 ahrens uint64_t txg = tx->tx_txg; 694 789 ahrens 695 9396 Matthew if (DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 696 9396 Matthew dsl_dataset_dirty(os->os_dsl_dataset, tx); 697 789 ahrens return; 698 9396 Matthew } 699 789 ahrens 700 873 ek110237 DNODE_VERIFY(dn); 701 789 ahrens 702 789 ahrens #ifdef ZFS_DEBUG 703 789 ahrens mutex_enter(&dn->dn_mtx); 704 789 ahrens ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); 705 789 ahrens /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */ 706 789 ahrens mutex_exit(&dn->dn_mtx); 707 789 ahrens #endif 708 789 ahrens 709 789 ahrens mutex_enter(&os->os_lock); 710 789 ahrens 711 789 ahrens /* 712 789 ahrens * If we are already marked dirty, we're done. 713 789 ahrens */ 714 1596 ahrens if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { 715 789 ahrens mutex_exit(&os->os_lock); 716 789 ahrens return; 717 789 ahrens } 718 789 ahrens 719 789 ahrens ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); 720 789 ahrens ASSERT(dn->dn_datablksz != 0); 721 4944 maybee ASSERT3U(dn->dn_next_bonuslen[txg&TXG_MASK], ==, 0); 722 1599 ahrens ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0); 723 789 ahrens 724 789 ahrens dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", 725 789 ahrens dn->dn_object, txg); 726 789 ahrens 727 789 ahrens if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { 728 789 ahrens list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); 729 789 ahrens } else { 730 789 ahrens list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); 731 789 ahrens } 732 789 ahrens 733 789 ahrens mutex_exit(&os->os_lock); 734 789 ahrens 735 789 ahrens /* 736 789 ahrens * The dnode maintains a hold on its containing dbuf as 737 789 ahrens * long as there are holds on it. Each instantiated child 738 789 ahrens * dbuf maintaines a hold on the dnode. When the last child 739 789 ahrens * drops its hold, the dnode will drop its hold on the 740 789 ahrens * containing dbuf. We add a "dirty hold" here so that the 741 789 ahrens * dnode will hang around after we finish processing its 742 789 ahrens * children. 743 789 ahrens */ 744 4944 maybee VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); 745 789 ahrens 746 3547 maybee (void) dbuf_dirty(dn->dn_dbuf, tx); 747 789 ahrens 748 789 ahrens dsl_dataset_dirty(os->os_dsl_dataset, tx); 749 789 ahrens } 750 789 ahrens 751 789 ahrens void 752 789 ahrens dnode_free(dnode_t *dn, dmu_tx_t *tx) 753 789 ahrens { 754 1596 ahrens int txgoff = tx->tx_txg & TXG_MASK; 755 1596 ahrens 756 789 ahrens dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); 757 789 ahrens 758 789 ahrens /* we should be the only holder... hopefully */ 759 789 ahrens /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ 760 789 ahrens 761 789 ahrens mutex_enter(&dn->dn_mtx); 762 789 ahrens if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { 763 789 ahrens mutex_exit(&dn->dn_mtx); 764 789 ahrens return; 765 789 ahrens } 766 789 ahrens dn->dn_free_txg = tx->tx_txg; 767 789 ahrens mutex_exit(&dn->dn_mtx); 768 789 ahrens 769 789 ahrens /* 770 789 ahrens * If the dnode is already dirty, it needs to be moved from 771 789 ahrens * the dirty list to the free list. 772 789 ahrens */ 773 789 ahrens mutex_enter(&dn->dn_objset->os_lock); 774 1596 ahrens if (list_link_active(&dn->dn_dirty_link[txgoff])) { 775 1596 ahrens list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); 776 1596 ahrens list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); 777 789 ahrens mutex_exit(&dn->dn_objset->os_lock); 778 789 ahrens } else { 779 789 ahrens mutex_exit(&dn->dn_objset->os_lock); 780 789 ahrens dnode_setdirty(dn, tx); 781 789 ahrens } 782 789 ahrens } 783 789 ahrens 784 789 ahrens /* 785 789 ahrens * Try to change the block size for the indicated dnode. This can only 786 789 ahrens * succeed if there are no blocks allocated or dirty beyond first block 787 789 ahrens */ 788 789 ahrens int 789 789 ahrens dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) 790 789 ahrens { 791 789 ahrens dmu_buf_impl_t *db, *db_next; 792 6992 maybee int err; 793 789 ahrens 794 789 ahrens if (size == 0) 795 789 ahrens size = SPA_MINBLOCKSIZE; 796 789 ahrens if (size > SPA_MAXBLOCKSIZE) 797 789 ahrens size = SPA_MAXBLOCKSIZE; 798 789 ahrens else 799 789 ahrens size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); 800 789 ahrens 801 2445 ahrens if (ibs == dn->dn_indblkshift) 802 2445 ahrens ibs = 0; 803 789 ahrens 804 2445 ahrens if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) 805 789 ahrens return (0); 806 789 ahrens 807 789 ahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 808 789 ahrens 809 789 ahrens /* Check for any allocated blocks beyond the first */ 810 789 ahrens if (dn->dn_phys->dn_maxblkid != 0) 811 2445 ahrens goto fail; 812 789 ahrens 813 789 ahrens mutex_enter(&dn->dn_dbufs_mtx); 814 789 ahrens for (db = list_head(&dn->dn_dbufs); db; db = db_next) { 815 789 ahrens db_next = list_next(&dn->dn_dbufs, db); 816 789 ahrens 817 6992 maybee if (db->db_blkid != 0 && db->db_blkid != DB_BONUS_BLKID) { 818 789 ahrens mutex_exit(&dn->dn_dbufs_mtx); 819 2445 ahrens goto fail; 820 789 ahrens } 821 789 ahrens } 822 789 ahrens mutex_exit(&dn->dn_dbufs_mtx); 823 2445 ahrens 824 2445 ahrens if (ibs && dn->dn_nlevels != 1) 825 2445 ahrens goto fail; 826 789 ahrens 827 6992 maybee /* resize the old block */ 828 6992 maybee err = dbuf_hold_impl(dn, 0, 0, TRUE, FTAG, &db); 829 6992 maybee if (err == 0) 830 1596 ahrens dbuf_new_size(db, size, tx); 831 6992 maybee else if (err != ENOENT) 832 6992 maybee goto fail; 833 789 ahrens 834 789 ahrens dnode_setdblksz(dn, size); 835 1596 ahrens dnode_setdirty(dn, tx); 836 1596 ahrens dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; 837 2445 ahrens if (ibs) { 838 2445 ahrens dn->dn_indblkshift = ibs; 839 2445 ahrens dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 840 2445 ahrens } 841 6992 maybee /* rele after we have fixed the blocksize in the dnode */ 842 1596 ahrens if (db) 843 1596 ahrens dbuf_rele(db, FTAG); 844 789 ahrens 845 789 ahrens rw_exit(&dn->dn_struct_rwlock); 846 2445 ahrens return (0); 847 2445 ahrens 848 2445 ahrens fail: 849 2445 ahrens rw_exit(&dn->dn_struct_rwlock); 850 2445 ahrens return (ENOTSUP); 851 789 ahrens } 852 789 ahrens 853 7332 Jonathan /* read-holding callers must not rely on the lock being continuously held */ 854 789 ahrens void 855 7332 Jonathan dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read) 856 789 ahrens { 857 789 ahrens uint64_t txgoff = tx->tx_txg & TXG_MASK; 858 1596 ahrens int epbs, new_nlevels; 859 789 ahrens uint64_t sz; 860 789 ahrens 861 1596 ahrens ASSERT(blkid != DB_BONUS_BLKID); 862 789 ahrens 863 7332 Jonathan ASSERT(have_read ? 864 7332 Jonathan RW_READ_HELD(&dn->dn_struct_rwlock) : 865 7332 Jonathan RW_WRITE_HELD(&dn->dn_struct_rwlock)); 866 7332 Jonathan 867 7332 Jonathan /* 868 7332 Jonathan * if we have a read-lock, check to see if we need to do any work 869 7332 Jonathan * before upgrading to a write-lock. 870 7332 Jonathan */ 871 7332 Jonathan if (have_read) { 872 7332 Jonathan if (blkid <= dn->dn_maxblkid) 873 7332 Jonathan return; 874 7332 Jonathan 875 7332 Jonathan if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { 876 7332 Jonathan rw_exit(&dn->dn_struct_rwlock); 877 7332 Jonathan rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 878 7332 Jonathan } 879 789 ahrens } 880 789 ahrens 881 1596 ahrens if (blkid <= dn->dn_maxblkid) 882 1596 ahrens goto out; 883 1596 ahrens 884 1596 ahrens dn->dn_maxblkid = blkid; 885 789 ahrens 886 789 ahrens /* 887 1596 ahrens * Compute the number of levels necessary to support the new maxblkid. 888 789 ahrens */ 889 789 ahrens new_nlevels = 1; 890 789 ahrens epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 891 1596 ahrens for (sz = dn->dn_nblkptr; 892 1596 ahrens sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) 893 1596 ahrens new_nlevels++; 894 789 ahrens 895 1596 ahrens if (new_nlevels > dn->dn_nlevels) { 896 1596 ahrens int old_nlevels = dn->dn_nlevels; 897 1596 ahrens dmu_buf_impl_t *db; 898 3547 maybee list_t *list; 899 3547 maybee dbuf_dirty_record_t *new, *dr, *dr_next; 900 789 ahrens 901 1596 ahrens dn->dn_nlevels = new_nlevels; 902 1596 ahrens 903 1596 ahrens ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); 904 789 ahrens dn->dn_next_nlevels[txgoff] = new_nlevels; 905 789 ahrens 906 3547 maybee /* dirty the left indirects */ 907 1596 ahrens db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); 908 3547 maybee new = dbuf_dirty(db, tx); 909 1544 eschrock dbuf_rele(db, FTAG); 910 1596 ahrens 911 3547 maybee /* transfer the dirty records to the new indirect */ 912 3547 maybee mutex_enter(&dn->dn_mtx); 913 3547 maybee mutex_enter(&new->dt.di.dr_mtx); 914 3547 maybee list = &dn->dn_dirty_records[txgoff]; 915 3547 maybee for (dr = list_head(list); dr; dr = dr_next) { 916 3547 maybee dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); 917 3547 maybee if (dr->dr_dbuf->db_level != new_nlevels-1 && 918 3547 maybee dr->dr_dbuf->db_blkid != DB_BONUS_BLKID) { 919 3547 maybee ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); 920 3547 maybee list_remove(&dn->dn_dirty_records[txgoff], dr); 921 3547 maybee list_insert_tail(&new->dt.di.dr_children, dr); 922 3547 maybee dr->dr_parent = new; 923 3547 maybee } 924 3547 maybee } 925 3547 maybee mutex_exit(&new->dt.di.dr_mtx); 926 3547 maybee mutex_exit(&dn->dn_mtx); 927 789 ahrens } 928 789 ahrens 929 789 ahrens out: 930 7332 Jonathan if (have_read) 931 7332 Jonathan rw_downgrade(&dn->dn_struct_rwlock); 932 789 ahrens } 933 789 ahrens 934 789 ahrens void 935 789 ahrens dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) 936 789 ahrens { 937 789 ahrens avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 938 789 ahrens avl_index_t where; 939 789 ahrens free_range_t *rp; 940 789 ahrens free_range_t rp_tofind; 941 789 ahrens uint64_t endblk = blkid + nblks; 942 789 ahrens 943 789 ahrens ASSERT(MUTEX_HELD(&dn->dn_mtx)); 944 789 ahrens ASSERT(nblks <= UINT64_MAX - blkid); /* no overflow */ 945 789 ahrens 946 789 ahrens dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 947 789 ahrens blkid, nblks, tx->tx_txg); 948 789 ahrens rp_tofind.fr_blkid = blkid; 949 789 ahrens rp = avl_find(tree, &rp_tofind, &where); 950 789 ahrens if (rp == NULL) 951 789 ahrens rp = avl_nearest(tree, where, AVL_BEFORE); 952 789 ahrens if (rp == NULL) 953 789 ahrens rp = avl_nearest(tree, where, AVL_AFTER); 954 789 ahrens 955 789 ahrens while (rp && (rp->fr_blkid <= blkid + nblks)) { 956 789 ahrens uint64_t fr_endblk = rp->fr_blkid + rp->fr_nblks; 957 789 ahrens free_range_t *nrp = AVL_NEXT(tree, rp); 958 789 ahrens 959 789 ahrens if (blkid <= rp->fr_blkid && endblk >= fr_endblk) { 960 789 ahrens /* clear this entire range */ 961 789 ahrens avl_remove(tree, rp); 962 789 ahrens kmem_free(rp, sizeof (free_range_t)); 963 789 ahrens } else if (blkid <= rp->fr_blkid && 964 789 ahrens endblk > rp->fr_blkid && endblk < fr_endblk) { 965 789 ahrens /* clear the beginning of this range */ 966 789 ahrens rp->fr_blkid = endblk; 967 789 ahrens rp->fr_nblks = fr_endblk - endblk; 968 789 ahrens } else if (blkid > rp->fr_blkid && blkid < fr_endblk && 969 789 ahrens endblk >= fr_endblk) { 970 789 ahrens /* clear the end of this range */ 971 789 ahrens rp->fr_nblks = blkid - rp->fr_blkid; 972 789 ahrens } else if (blkid > rp->fr_blkid && endblk < fr_endblk) { 973 789 ahrens /* clear a chunk out of this range */ 974 789 ahrens free_range_t *new_rp = 975 789 ahrens kmem_alloc(sizeof (free_range_t), KM_SLEEP); 976 789 ahrens 977 789 ahrens new_rp->fr_blkid = endblk; 978 789 ahrens new_rp->fr_nblks = fr_endblk - endblk; 979 789 ahrens avl_insert_here(tree, new_rp, rp, AVL_AFTER); 980 789 ahrens rp->fr_nblks = blkid - rp->fr_blkid; 981 789 ahrens } 982 789 ahrens /* there may be no overlap */ 983 789 ahrens rp = nrp; 984 789 ahrens } 985 789 ahrens } 986 789 ahrens 987 789 ahrens void 988 789 ahrens dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) 989 789 ahrens { 990 789 ahrens dmu_buf_impl_t *db; 991 2445 ahrens uint64_t blkoff, blkid, nblks; 992 6992 maybee int blksz, blkshift, head, tail; 993 789 ahrens int trunc = FALSE; 994 6992 maybee int epbs; 995 789 ahrens 996 789 ahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 997 789 ahrens blksz = dn->dn_datablksz; 998 6992 maybee blkshift = dn->dn_datablkshift; 999 6992 maybee epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 1000 789 ahrens 1001 789 ahrens if (len == -1ULL) { 1002 789 ahrens len = UINT64_MAX - off; 1003 789 ahrens trunc = TRUE; 1004 789 ahrens } 1005 789 ahrens 1006 789 ahrens /* 1007 789 ahrens * First, block align the region to free: 1008 789 ahrens */ 1009 2445 ahrens if (ISP2(blksz)) { 1010 2445 ahrens head = P2NPHASE(off, blksz); 1011 2445 ahrens blkoff = P2PHASE(off, blksz); 1012 6992 maybee if ((off >> blkshift) > dn->dn_maxblkid) 1013 6992 maybee goto out; 1014 2445 ahrens } else { 1015 2445 ahrens ASSERT(dn->dn_maxblkid == 0); 1016 2445 ahrens if (off == 0 && len >= blksz) { 1017 6992 maybee /* Freeing the whole block; fast-track this request */ 1018 6992 maybee blkid = 0; 1019 6992 maybee nblks = 1; 1020 6992 maybee goto done; 1021 7385 Mark } else if (off >= blksz) { 1022 6992 maybee /* Freeing past end-of-data */ 1023 6992 maybee goto out; 1024 789 ahrens } else { 1025 2445 ahrens /* Freeing part of the block. */ 1026 789 ahrens head = blksz - off; 1027 789 ahrens ASSERT3U(head, >, 0); 1028 789 ahrens } 1029 2445 ahrens blkoff = off; 1030 789 ahrens } 1031 789 ahrens /* zero out any partial block data at the start of the range */ 1032 789 ahrens if (head) { 1033 2445 ahrens ASSERT3U(blkoff + head, ==, blksz); 1034 789 ahrens if (len < head) 1035 789 ahrens head = len; 1036 789 ahrens if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, 1037 789 ahrens FTAG, &db) == 0) { 1038 789 ahrens caddr_t data; 1039 789 ahrens 1040 789 ahrens /* don't dirty if it isn't on disk and isn't dirty */ 1041 3547 maybee if (db->db_last_dirty || 1042 789 ahrens (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 1043 789 ahrens rw_exit(&dn->dn_struct_rwlock); 1044 789 ahrens dbuf_will_dirty(db, tx); 1045 789 ahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1046 789 ahrens data = db->db.db_data; 1047 2445 ahrens bzero(data + blkoff, head); 1048 789 ahrens } 1049 1544 eschrock dbuf_rele(db, FTAG); 1050 789 ahrens } 1051 789 ahrens off += head; 1052 789 ahrens len -= head; 1053 789 ahrens } 1054 2445 ahrens 1055 2445 ahrens /* If the range was less than one block, we're done */ 1056 6992 maybee if (len == 0) 1057 789 ahrens goto out; 1058 789 ahrens 1059 6992 maybee /* If the remaining range is past end of file, we're done */ 1060 6992 maybee if ((off >> blkshift) > dn->dn_maxblkid) 1061 6992 maybee goto out; 1062 789 ahrens 1063 7385 Mark ASSERT(ISP2(blksz)); 1064 6992 maybee if (trunc) 1065 6992 maybee tail = 0; 1066 6992 maybee else 1067 6992 maybee tail = P2PHASE(len, blksz); 1068 789 ahrens 1069 6992 maybee ASSERT3U(P2PHASE(off, blksz), ==, 0); 1070 6992 maybee /* zero out any partial block data at the end of the range */ 1071 6992 maybee if (tail) { 1072 6992 maybee if (len < tail) 1073 6992 maybee tail = len; 1074 6992 maybee if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), 1075 6992 maybee TRUE, FTAG, &db) == 0) { 1076 6992 maybee /* don't dirty if not on disk and not dirty */ 1077 6992 maybee if (db->db_last_dirty || 1078 6992 maybee (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 1079 6992 maybee rw_exit(&dn->dn_struct_rwlock); 1080 6992 maybee dbuf_will_dirty(db, tx); 1081 6992 maybee rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1082 6992 maybee bzero(db->db.db_data, tail); 1083 6992 maybee } 1084 6992 maybee dbuf_rele(db, FTAG); 1085 6992 maybee } 1086 6992 maybee len -= tail; 1087 6992 maybee } 1088 6992 maybee 1089 6992 maybee /* If the range did not include a full block, we are done */ 1090 6992 maybee if (len == 0) 1091 6992 maybee goto out; 1092 6992 maybee 1093 6992 maybee ASSERT(IS_P2ALIGNED(off, blksz)); 1094 6992 maybee ASSERT(trunc || IS_P2ALIGNED(len, blksz)); 1095 6992 maybee blkid = off >> blkshift; 1096 6992 maybee nblks = len >> blkshift; 1097 6992 maybee if (trunc) 1098 6992 maybee nblks += 1; 1099 6992 maybee 1100 6992 maybee /* 1101 6992 maybee * Read in and mark all the level-1 indirects dirty, 1102 6992 maybee * so that they will stay in memory until syncing phase. 1103 7049 maybee * Always dirty the first and last indirect to make sure 1104 7049 maybee * we dirty all the partial indirects. 1105 6992 maybee */ 1106 6992 maybee if (dn->dn_nlevels > 1) { 1107 6992 maybee uint64_t i, first, last; 1108 6992 maybee int shift = epbs + dn->dn_datablkshift; 1109 6992 maybee 1110 6992 maybee first = blkid >> epbs; 1111 7049 maybee if (db = dbuf_hold_level(dn, 1, first, FTAG)) { 1112 7049 maybee dbuf_will_dirty(db, tx); 1113 7049 maybee dbuf_rele(db, FTAG); 1114 7049 maybee } 1115 6992 maybee if (trunc) 1116 6992 maybee last = dn->dn_maxblkid >> epbs; 1117 2445 ahrens else 1118 6992 maybee last = (blkid + nblks - 1) >> epbs; 1119 7049 maybee if (last > first && (db = dbuf_hold_level(dn, 1, last, FTAG))) { 1120 7049 maybee dbuf_will_dirty(db, tx); 1121 7049 maybee dbuf_rele(db, FTAG); 1122 7049 maybee } 1123 7049 maybee for (i = first + 1; i < last; i++) { 1124 6992 maybee uint64_t ibyte = i << shift; 1125 6992 maybee int err; 1126 2445 ahrens 1127 6992 maybee err = dnode_next_offset(dn, 1128 6992 maybee DNODE_FIND_HAVELOCK, &ibyte, 1, 1, 0); 1129 6992 maybee i = ibyte >> shift; 1130 7049 maybee if (err == ESRCH || i >= last) 1131 6992 maybee break; 1132 6992 maybee ASSERT(err == 0); 1133 6992 maybee db = dbuf_hold_level(dn, 1, i, FTAG); 1134 6992 maybee if (db) { 1135 6992 maybee dbuf_will_dirty(db, tx); 1136 2445 ahrens dbuf_rele(db, FTAG); 1137 789 ahrens } 1138 2445 ahrens } 1139 789 ahrens } 1140 6992 maybee done: 1141 6992 maybee /* 1142 6992 maybee * Add this range to the dnode range list. 1143 6992 maybee * We will finish up this free operation in the syncing phase. 1144 6992 maybee */ 1145 789 ahrens mutex_enter(&dn->dn_mtx); 1146 789 ahrens dnode_clear_range(dn, blkid, nblks, tx); 1147 789 ahrens { 1148 789 ahrens free_range_t *rp, *found; 1149 789 ahrens avl_index_t where; 1150 789 ahrens avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 1151 789 ahrens 1152 789 ahrens /* Add new range to dn_ranges */ 1153 789 ahrens rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP); 1154 789 ahrens rp->fr_blkid = blkid; 1155 789 ahrens rp->fr_nblks = nblks; 1156 789 ahrens found = avl_find(tree, rp, &where); 1157 789 ahrens ASSERT(found == NULL); 1158 789 ahrens avl_insert(tree, rp, where); 1159 789 ahrens dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 1160 789 ahrens blkid, nblks, tx->tx_txg); 1161 789 ahrens } 1162 789 ahrens mutex_exit(&dn->dn_mtx); 1163 789 ahrens 1164 6992 maybee dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); 1165 789 ahrens dnode_setdirty(dn, tx); 1166 789 ahrens out: 1167 6992 maybee if (trunc && dn->dn_maxblkid >= (off >> blkshift)) 1168 6992 maybee dn->dn_maxblkid = (off >> blkshift ? (off >> blkshift) - 1 : 0); 1169 6992 maybee 1170 789 ahrens rw_exit(&dn->dn_struct_rwlock); 1171 789 ahrens } 1172 789 ahrens 1173 789 ahrens /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ 1174 789 ahrens uint64_t 1175 789 ahrens dnode_block_freed(dnode_t *dn, uint64_t blkid) 1176 789 ahrens { 1177 789 ahrens free_range_t range_tofind; 1178 789 ahrens void *dp = spa_get_dsl(dn->dn_objset->os_spa); 1179 789 ahrens int i; 1180 789 ahrens 1181 789 ahrens if (blkid == DB_BONUS_BLKID) 1182 789 ahrens return (FALSE); 1183 789 ahrens 1184 789 ahrens /* 1185 789 ahrens * If we're in the process of opening the pool, dp will not be 1186 789 ahrens * set yet, but there shouldn't be anything dirty. 1187 789 ahrens */ 1188 789 ahrens if (dp == NULL) 1189 789 ahrens return (FALSE); 1190 789 ahrens 1191 789 ahrens if (dn->dn_free_txg) 1192 789 ahrens return (TRUE); 1193 789 ahrens 1194 789 ahrens range_tofind.fr_blkid = blkid; 1195 789 ahrens mutex_enter(&dn->dn_mtx); 1196 789 ahrens for (i = 0; i < TXG_SIZE; i++) { 1197 789 ahrens free_range_t *range_found; 1198 789 ahrens avl_index_t idx; 1199 789 ahrens 1200 789 ahrens range_found = avl_find(&dn->dn_ranges[i], &range_tofind, &idx); 1201 789 ahrens if (range_found) { 1202 789 ahrens ASSERT(range_found->fr_nblks > 0); 1203 789 ahrens break; 1204 789 ahrens } 1205 789 ahrens range_found = avl_nearest(&dn->dn_ranges[i], idx, AVL_BEFORE); 1206 789 ahrens if (range_found && 1207 789 ahrens range_found->fr_blkid + range_found->fr_nblks > blkid) 1208 789 ahrens break; 1209 789 ahrens } 1210 789 ahrens mutex_exit(&dn->dn_mtx); 1211 789 ahrens return (i < TXG_SIZE); 1212 789 ahrens } 1213 789 ahrens 1214 789 ahrens /* call from syncing context when we actually write/free space for this dnode */ 1215 789 ahrens void 1216 2082 eschrock dnode_diduse_space(dnode_t *dn, int64_t delta) 1217 789 ahrens { 1218 2082 eschrock uint64_t space; 1219 2082 eschrock dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", 1220 789 ahrens dn, dn->dn_phys, 1221 2082 eschrock (u_longlong_t)dn->dn_phys->dn_used, 1222 2082 eschrock (longlong_t)delta); 1223 789 ahrens 1224 789 ahrens mutex_enter(&dn->dn_mtx); 1225 2082 eschrock space = DN_USED_BYTES(dn->dn_phys); 1226 2082 eschrock if (delta > 0) { 1227 2082 eschrock ASSERT3U(space + delta, >=, space); /* no overflow */ 1228 789 ahrens } else { 1229 2082 eschrock ASSERT3U(space, >=, -delta); /* no underflow */ 1230 2082 eschrock } 1231 2082 eschrock space += delta; 1232 4577 ahrens if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { 1233 2082 eschrock ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); 1234 2082 eschrock ASSERT3U(P2PHASE(space, 1<<DEV_BSHIFT), ==, 0); 1235 2082 eschrock dn->dn_phys->dn_used = space >> DEV_BSHIFT; 1236 2082 eschrock } else { 1237 2082 eschrock dn->dn_phys->dn_used = space; 1238 2082 eschrock dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; 1239 789 ahrens } 1240 789 ahrens mutex_exit(&dn->dn_mtx); 1241 789 ahrens } 1242 789 ahrens 1243 789 ahrens /* 1244 789 ahrens * Call when we think we're going to write/free space in open context. 1245 789 ahrens * Be conservative (ie. OK to write less than this or free more than 1246 789 ahrens * this, but don't write more or free less). 1247 789 ahrens */ 1248 789 ahrens void 1249 789 ahrens dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) 1250 789 ahrens { 1251 10298 Matthew objset_t *os = dn->dn_objset; 1252 789 ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 1253 789 ahrens 1254 789 ahrens if (space > 0) 1255 789 ahrens space = spa_get_asize(os->os_spa, space); 1256 789 ahrens 1257 789 ahrens if (ds) 1258 789 ahrens dsl_dir_willuse_space(ds->ds_dir, space, tx); 1259 789 ahrens 1260 789 ahrens dmu_tx_willuse_space(tx, space); 1261 789 ahrens } 1262 789 ahrens 1263 9950 Mark /* 1264 9950 Mark * This function scans a block at the indicated "level" looking for 1265 9950 Mark * a hole or data (depending on 'flags'). If level > 0, then we are 1266 9950 Mark * scanning an indirect block looking at its pointers. If level == 0, 1267 9950 Mark * then we are looking at a block of dnodes. If we don't find what we 1268 9950 Mark * are looking for in the block, we return ESRCH. Otherwise, return 1269 9950 Mark * with *offset pointing to the beginning (if searching forwards) or 1270 9950 Mark * end (if searching backwards) of the range covered by the block 1271 9950 Mark * pointer we matched on (or dnode). 1272 9950 Mark * 1273 9950 Mark * The basic search algorithm used below by dnode_next_offset() is to 1274 9950 Mark * use this function to search up the block tree (widen the search) until 1275 9950 Mark * we find something (i.e., we don't return ESRCH) and then search back 1276 9950 Mark * down the tree (narrow the search) until we reach our original search 1277 9950 Mark * level. 1278 9950 Mark */ 1279 789 ahrens static int 1280 6992 maybee dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, 1281 3025 ahrens int lvl, uint64_t blkfill, uint64_t txg) 1282 789 ahrens { 1283 789 ahrens dmu_buf_impl_t *db = NULL; 1284 789 ahrens void *data = NULL; 1285 789 ahrens uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 1286 789 ahrens uint64_t epb = 1ULL << epbs; 1287 789 ahrens uint64_t minfill, maxfill; 1288 6992 maybee boolean_t hole; 1289 6992 maybee int i, inc, error, span; 1290 789 ahrens 1291 789 ahrens dprintf("probing object %llu offset %llx level %d of %u\n", 1292 789 ahrens dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); 1293 6992 maybee 1294 9396 Matthew hole = ((flags & DNODE_FIND_HOLE) != 0); 1295 6992 maybee inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; 1296 7385 Mark ASSERT(txg == 0 || !hole); 1297 789 ahrens 1298 789 ahrens if (lvl == dn->dn_phys->dn_nlevels) { 1299 789 ahrens error = 0; 1300 789 ahrens epb = dn->dn_phys->dn_nblkptr; 1301 789 ahrens data = dn->dn_phys->dn_blkptr; 1302 789 ahrens } else { 1303 789 ahrens uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); 1304 789 ahrens error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); 1305 789 ahrens if (error) { 1306 7385 Mark if (error != ENOENT) 1307 7385 Mark return (error); 1308 7385 Mark if (hole) 1309 7385 Mark return (0); 1310 7385 Mark /* 1311 7385 Mark * This can only happen when we are searching up 1312 7385 Mark * the block tree for data. We don't really need to 1313 7385 Mark * adjust the offset, as we will just end up looking 1314 7385 Mark * at the pointer to this block in its parent, and its 1315 7385 Mark * going to be unallocated, so we will skip over it. 1316 7385 Mark */ 1317 7385 Mark return (ESRCH); 1318 789 ahrens } 1319 1793 ahrens error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); 1320 1793 ahrens if (error) { 1321 1793 ahrens dbuf_rele(db, FTAG); 1322 1793 ahrens return (error); 1323 1793 ahrens } 1324 789 ahrens data = db->db.db_data; 1325 789 ahrens } 1326 789 ahrens 1327 3025 ahrens if (db && txg && 1328 3025 ahrens (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg)) { 1329 7385 Mark /* 1330 7385 Mark * This can only happen when we are searching up the tree 1331 7385 Mark * and these conditions mean that we need to keep climbing. 1332 7385 Mark */ 1333 3025 ahrens error = ESRCH; 1334 3025 ahrens } else if (lvl == 0) { 1335 789 ahrens dnode_phys_t *dnp = data; 1336 789 ahrens span = DNODE_SHIFT; 1337 789 ahrens ASSERT(dn->dn_type == DMU_OT_DNODE); 1338 789 ahrens 1339 6992 maybee for (i = (*offset >> span) & (blkfill - 1); 1340 6992 maybee i >= 0 && i < blkfill; i += inc) { 1341 9409 Jonathan if ((dnp[i].dn_type == DMU_OT_NONE) == hole) 1342 789 ahrens break; 1343 6992 maybee *offset += (1ULL << span) * inc; 1344 789 ahrens } 1345 6992 maybee if (i < 0 || i == blkfill) 1346 789 ahrens error = ESRCH; 1347 789 ahrens } else { 1348 789 ahrens blkptr_t *bp = data; 1349 9950 Mark uint64_t start = *offset; 1350 789 ahrens span = (lvl - 1) * epbs + dn->dn_datablkshift; 1351 789 ahrens minfill = 0; 1352 789 ahrens maxfill = blkfill << ((lvl - 1) * epbs); 1353 789 ahrens 1354 789 ahrens if (hole) 1355 789 ahrens maxfill--; 1356 789 ahrens else 1357 789 ahrens minfill++; 1358 789 ahrens 1359 9950 Mark *offset = *offset >> span; 1360 9950 Mark for (i = BF64_GET(*offset, 0, epbs); 1361 6992 maybee i >= 0 && i < epb; i += inc) { 1362 789 ahrens if (bp[i].blk_fill >= minfill && 1363 3025 ahrens bp[i].blk_fill <= maxfill && 1364 7385 Mark (hole || bp[i].blk_birth > txg)) 1365 789 ahrens break; 1366 9950 Mark if (inc > 0 || *offset > 0) 1367 9950 Mark *offset += inc; 1368 789 ahrens } 1369 9950 Mark *offset = *offset << span; 1370 9950 Mark if (inc < 0) { 1371 9950 Mark /* traversing backwards; position offset at the end */ 1372 9950 Mark ASSERT3U(*offset, <=, start); 1373 9950 Mark *offset = MIN(*offset + (1ULL << span) - 1, start); 1374 9950 Mark } else if (*offset < start) { 1375 9950 Mark *offset = start; 1376 9950 Mark } 1377 9950 Mark if (i < 0 || i >= epb) 1378 789 ahrens error = ESRCH; 1379 789 ahrens } 1380 789 ahrens 1381 789 ahrens if (db) 1382 1544 eschrock dbuf_rele(db, FTAG); 1383 789 ahrens 1384 789 ahrens return (error); 1385 789 ahrens } 1386 789 ahrens 1387 789 ahrens /* 1388 789 ahrens * Find the next hole, data, or sparse region at or after *offset. 1389 789 ahrens * The value 'blkfill' tells us how many items we expect to find 1390 789 ahrens * in an L0 data block; this value is 1 for normal objects, 1391 789 ahrens * DNODES_PER_BLOCK for the meta dnode, and some fraction of 1392 789 ahrens * DNODES_PER_BLOCK when searching for sparse regions thereof. 1393 3025 ahrens * 1394 789 ahrens * Examples: 1395 789 ahrens * 1396 6992 maybee * dnode_next_offset(dn, flags, offset, 1, 1, 0); 1397 6992 maybee * Finds the next/previous hole/data in a file. 1398 789 ahrens * Used in dmu_offset_next(). 1399 789 ahrens * 1400 6992 maybee * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); 1401 789 ahrens * Finds the next free/allocated dnode an objset's meta-dnode. 1402 3025 ahrens * Only finds objects that have new contents since txg (ie. 1403 3025 ahrens * bonus buffer changes and content removal are ignored). 1404 789 ahrens * Used in dmu_object_next(). 1405 789 ahrens * 1406 6992 maybee * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); 1407 789 ahrens * Finds the next L2 meta-dnode bp that's at most 1/4 full. 1408 789 ahrens * Used in dmu_object_alloc(). 1409 789 ahrens */ 1410 789 ahrens int 1411 6992 maybee dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, 1412 3025 ahrens int minlvl, uint64_t blkfill, uint64_t txg) 1413 789 ahrens { 1414 6992 maybee uint64_t initial_offset = *offset; 1415 789 ahrens int lvl, maxlvl; 1416 789 ahrens int error = 0; 1417 789 ahrens 1418 6992 maybee if (!(flags & DNODE_FIND_HAVELOCK)) 1419 6992 maybee rw_enter(&dn->dn_struct_rwlock, RW_READER); 1420 789 ahrens 1421 789 ahrens if (dn->dn_phys->dn_nlevels == 0) { 1422 6992 maybee error = ESRCH; 1423 6992 maybee goto out; 1424 789 ahrens } 1425 789 ahrens 1426 789 ahrens if (dn->dn_datablkshift == 0) { 1427 789 ahrens if (*offset < dn->dn_datablksz) { 1428 6992 maybee if (flags & DNODE_FIND_HOLE) 1429 789 ahrens *offset = dn->dn_datablksz; 1430 789 ahrens } else { 1431 789 ahrens error = ESRCH; 1432 789 ahrens } 1433 6992 maybee goto out; 1434 789 ahrens } 1435 789 ahrens 1436 789 ahrens maxlvl = dn->dn_phys->dn_nlevels; 1437 789 ahrens 1438 789 ahrens for (lvl = minlvl; lvl <= maxlvl; lvl++) { 1439 3025 ahrens error = dnode_next_offset_level(dn, 1440 6992 maybee flags, offset, lvl, blkfill, txg); 1441 1793 ahrens if (error != ESRCH) 1442 789 ahrens break; 1443 789 ahrens } 1444 789 ahrens 1445 6992 maybee while (error == 0 && --lvl >= minlvl) { 1446 3025 ahrens error = dnode_next_offset_level(dn, 1447 6992 maybee flags, offset, lvl, blkfill, txg); 1448 3025 ahrens } 1449 789 ahrens 1450 6992 maybee if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? 1451 6992 maybee initial_offset < *offset : initial_offset > *offset)) 1452 1793 ahrens error = ESRCH; 1453 6992 maybee out: 1454 6992 maybee if (!(flags & DNODE_FIND_HAVELOCK)) 1455 6992 maybee rw_exit(&dn->dn_struct_rwlock); 1456 789 ahrens 1457 789 ahrens return (error); 1458 789 ahrens } 1459
Indexes created Sat Nov 28 14:09:03 UTC 2009
Terms of Use |
Privacy |
Trademarks |
Copyright Policy |
Site Guidelines |
Help
Your use of this web site or any of its content or software indicates your agreement to be bound by these Terms of Use.
Copyright © 1995-2008 Sun Microsystems, Inc.