1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5*f65e61c0Sahrens * Common Development and Distribution License (the "License"). 6*f65e61c0Sahrens * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 223b83abddSahrens * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23fa9e4066Sahrens * Use is subject to license terms. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #pragma ident "%Z%%M% %I% %E% SMI" 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/zfs_context.h> 29fa9e4066Sahrens #include <sys/dbuf.h> 30fa9e4066Sahrens #include <sys/dnode.h> 31fa9e4066Sahrens #include <sys/dmu.h> 32fa9e4066Sahrens #include <sys/dmu_impl.h> 33fa9e4066Sahrens #include <sys/dmu_tx.h> 34fa9e4066Sahrens #include <sys/dmu_objset.h> 35fa9e4066Sahrens #include <sys/dsl_dir.h> 36fa9e4066Sahrens #include <sys/dsl_dataset.h> 37fa9e4066Sahrens #include <sys/spa.h> 38fa9e4066Sahrens #include <sys/zio.h> 39fa9e4066Sahrens #include <sys/dmu_zfetch.h> 40fa9e4066Sahrens 41fa9e4066Sahrens static int free_range_compar(const void *node1, const void *node2); 42fa9e4066Sahrens 43fa9e4066Sahrens static kmem_cache_t *dnode_cache; 44fa9e4066Sahrens 45fa9e4066Sahrens static dnode_phys_t dnode_phys_zero; 46fa9e4066Sahrens 47fa9e4066Sahrens int zfs_default_bs = SPA_MINBLOCKSHIFT; 48fa9e4066Sahrens int zfs_default_ibs = DN_MAX_INDBLKSHIFT; 49fa9e4066Sahrens 50fa9e4066Sahrens /* ARGSUSED */ 51fa9e4066Sahrens static int 52fa9e4066Sahrens dnode_cons(void *arg, void *unused, int kmflag) 53fa9e4066Sahrens { 54fa9e4066Sahrens int i; 55fa9e4066Sahrens dnode_t *dn = arg; 56fa9e4066Sahrens bzero(dn, sizeof (dnode_t)); 57fa9e4066Sahrens 58fa9e4066Sahrens rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); 59fa9e4066Sahrens mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); 60fa9e4066Sahrens mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); 61fa9e4066Sahrens refcount_create(&dn->dn_holds); 62fa9e4066Sahrens refcount_create(&dn->dn_tx_holds); 63fa9e4066Sahrens 64fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 65fa9e4066Sahrens avl_create(&dn->dn_ranges[i], free_range_compar, 66fa9e4066Sahrens sizeof (free_range_t), 67fa9e4066Sahrens offsetof(struct free_range, fr_node)); 68fa9e4066Sahrens list_create(&dn->dn_dirty_dbufs[i], 69fa9e4066Sahrens sizeof (dmu_buf_impl_t), 70fa9e4066Sahrens offsetof(dmu_buf_impl_t, db_dirty_node[i])); 71fa9e4066Sahrens } 72fa9e4066Sahrens 73fa9e4066Sahrens list_create(&dn->dn_dbufs, sizeof (dmu_buf_impl_t), 74fa9e4066Sahrens offsetof(dmu_buf_impl_t, db_link)); 75fa9e4066Sahrens 76fa9e4066Sahrens return (0); 77fa9e4066Sahrens } 78fa9e4066Sahrens 79fa9e4066Sahrens /* ARGSUSED */ 80fa9e4066Sahrens static void 81fa9e4066Sahrens dnode_dest(void *arg, void *unused) 82fa9e4066Sahrens { 83fa9e4066Sahrens int i; 84fa9e4066Sahrens dnode_t *dn = arg; 85fa9e4066Sahrens 86fa9e4066Sahrens rw_destroy(&dn->dn_struct_rwlock); 87fa9e4066Sahrens mutex_destroy(&dn->dn_mtx); 88fa9e4066Sahrens mutex_destroy(&dn->dn_dbufs_mtx); 89fa9e4066Sahrens refcount_destroy(&dn->dn_holds); 90fa9e4066Sahrens refcount_destroy(&dn->dn_tx_holds); 91fa9e4066Sahrens 92fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 93fa9e4066Sahrens avl_destroy(&dn->dn_ranges[i]); 94fa9e4066Sahrens list_destroy(&dn->dn_dirty_dbufs[i]); 95fa9e4066Sahrens } 96fa9e4066Sahrens 97fa9e4066Sahrens list_destroy(&dn->dn_dbufs); 98fa9e4066Sahrens } 99fa9e4066Sahrens 100fa9e4066Sahrens void 101fa9e4066Sahrens dnode_init(void) 102fa9e4066Sahrens { 103fa9e4066Sahrens dnode_cache = kmem_cache_create("dnode_t", 104fa9e4066Sahrens sizeof (dnode_t), 105fa9e4066Sahrens 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); 106fa9e4066Sahrens } 107fa9e4066Sahrens 108fa9e4066Sahrens void 109fa9e4066Sahrens dnode_fini(void) 110fa9e4066Sahrens { 111fa9e4066Sahrens kmem_cache_destroy(dnode_cache); 112fa9e4066Sahrens } 113fa9e4066Sahrens 114fa9e4066Sahrens 1159c9dc39aSek #ifdef ZFS_DEBUG 116fa9e4066Sahrens void 117fa9e4066Sahrens dnode_verify(dnode_t *dn) 118fa9e4066Sahrens { 119fa9e4066Sahrens int drop_struct_lock = FALSE; 120fa9e4066Sahrens 121fa9e4066Sahrens ASSERT(dn->dn_phys); 122fa9e4066Sahrens ASSERT(dn->dn_objset); 123fa9e4066Sahrens 124fa9e4066Sahrens ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 125fa9e4066Sahrens 126fa9e4066Sahrens if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) 127fa9e4066Sahrens return; 128fa9e4066Sahrens 129fa9e4066Sahrens if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 130fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 131fa9e4066Sahrens drop_struct_lock = TRUE; 132fa9e4066Sahrens } 133fa9e4066Sahrens if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { 134fa9e4066Sahrens int i; 135fa9e4066Sahrens ASSERT3U(dn->dn_indblkshift, >=, 0); 136fa9e4066Sahrens ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); 137fa9e4066Sahrens if (dn->dn_datablkshift) { 138fa9e4066Sahrens ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); 139fa9e4066Sahrens ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); 140fa9e4066Sahrens ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); 141fa9e4066Sahrens } 142fa9e4066Sahrens ASSERT3U(dn->dn_nlevels, <=, 30); 143fa9e4066Sahrens ASSERT3U(dn->dn_type, <=, DMU_OT_NUMTYPES); 144fa9e4066Sahrens ASSERT3U(dn->dn_nblkptr, >=, 1); 145fa9e4066Sahrens ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 146fa9e4066Sahrens ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); 147fa9e4066Sahrens ASSERT3U(dn->dn_datablksz, ==, 148fa9e4066Sahrens dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 149fa9e4066Sahrens ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); 150fa9e4066Sahrens ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + 151fa9e4066Sahrens dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); 152fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 153fa9e4066Sahrens ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); 154fa9e4066Sahrens } 155fa9e4066Sahrens } 156fa9e4066Sahrens if (dn->dn_phys->dn_type != DMU_OT_NONE) 157fa9e4066Sahrens ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); 158fa9e4066Sahrens ASSERT(IS_DNODE_DNODE(dn->dn_object) || dn->dn_dbuf); 159fa9e4066Sahrens if (dn->dn_dbuf != NULL) { 160fa9e4066Sahrens ASSERT3P(dn->dn_phys, ==, 161fa9e4066Sahrens (dnode_phys_t *)dn->dn_dbuf->db.db_data + 162fa9e4066Sahrens (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); 163fa9e4066Sahrens } 164fa9e4066Sahrens if (drop_struct_lock) 165fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 166fa9e4066Sahrens } 1679c9dc39aSek #endif 168fa9e4066Sahrens 169fa9e4066Sahrens void 170fa9e4066Sahrens dnode_byteswap(dnode_phys_t *dnp) 171fa9e4066Sahrens { 172fa9e4066Sahrens uint64_t *buf64 = (void*)&dnp->dn_blkptr; 173fa9e4066Sahrens int i; 174fa9e4066Sahrens 175fa9e4066Sahrens if (dnp->dn_type == DMU_OT_NONE) { 176fa9e4066Sahrens bzero(dnp, sizeof (dnode_phys_t)); 177fa9e4066Sahrens return; 178fa9e4066Sahrens } 179fa9e4066Sahrens 180fa9e4066Sahrens dnp->dn_type = BSWAP_8(dnp->dn_type); 181fa9e4066Sahrens dnp->dn_indblkshift = BSWAP_8(dnp->dn_indblkshift); 182fa9e4066Sahrens dnp->dn_nlevels = BSWAP_8(dnp->dn_nlevels); 183fa9e4066Sahrens dnp->dn_nblkptr = BSWAP_8(dnp->dn_nblkptr); 184fa9e4066Sahrens dnp->dn_bonustype = BSWAP_8(dnp->dn_bonustype); 185fa9e4066Sahrens dnp->dn_checksum = BSWAP_8(dnp->dn_checksum); 186fa9e4066Sahrens dnp->dn_compress = BSWAP_8(dnp->dn_compress); 187fa9e4066Sahrens dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); 188fa9e4066Sahrens dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); 189fa9e4066Sahrens dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); 190fa9e4066Sahrens dnp->dn_secphys = BSWAP_64(dnp->dn_secphys); 191fa9e4066Sahrens 192fa9e4066Sahrens /* 193fa9e4066Sahrens * dn_nblkptr is only one byte, so it's OK to read it in either 194fa9e4066Sahrens * byte order. We can't read dn_bouslen. 195fa9e4066Sahrens */ 196fa9e4066Sahrens ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); 197fa9e4066Sahrens ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); 198fa9e4066Sahrens for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) 199fa9e4066Sahrens buf64[i] = BSWAP_64(buf64[i]); 200fa9e4066Sahrens 201fa9e4066Sahrens /* 202fa9e4066Sahrens * OK to check dn_bonuslen for zero, because it won't matter if 203fa9e4066Sahrens * we have the wrong byte order. This is necessary because the 204fa9e4066Sahrens * dnode dnode is smaller than a regular dnode. 205fa9e4066Sahrens */ 206fa9e4066Sahrens if (dnp->dn_bonuslen != 0) { 207fa9e4066Sahrens /* 208fa9e4066Sahrens * Note that the bonus length calculated here may be 209fa9e4066Sahrens * longer than the actual bonus buffer. This is because 210fa9e4066Sahrens * we always put the bonus buffer after the last block 211fa9e4066Sahrens * pointer (instead of packing it against the end of the 212fa9e4066Sahrens * dnode buffer). 213fa9e4066Sahrens */ 214fa9e4066Sahrens int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); 215fa9e4066Sahrens size_t len = DN_MAX_BONUSLEN - off; 216fa9e4066Sahrens dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); 217fa9e4066Sahrens } 218fa9e4066Sahrens } 219fa9e4066Sahrens 220fa9e4066Sahrens void 221fa9e4066Sahrens dnode_buf_byteswap(void *vbuf, size_t size) 222fa9e4066Sahrens { 223fa9e4066Sahrens dnode_phys_t *buf = vbuf; 224fa9e4066Sahrens int i; 225fa9e4066Sahrens 226fa9e4066Sahrens ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); 227fa9e4066Sahrens ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); 228fa9e4066Sahrens 229fa9e4066Sahrens size >>= DNODE_SHIFT; 230fa9e4066Sahrens for (i = 0; i < size; i++) { 231fa9e4066Sahrens dnode_byteswap(buf); 232fa9e4066Sahrens buf++; 233fa9e4066Sahrens } 234fa9e4066Sahrens } 235fa9e4066Sahrens 236fa9e4066Sahrens static int 237fa9e4066Sahrens free_range_compar(const void *node1, const void *node2) 238fa9e4066Sahrens { 239fa9e4066Sahrens const free_range_t *rp1 = node1; 240fa9e4066Sahrens const free_range_t *rp2 = node2; 241fa9e4066Sahrens 242fa9e4066Sahrens if (rp1->fr_blkid < rp2->fr_blkid) 243fa9e4066Sahrens return (-1); 244fa9e4066Sahrens else if (rp1->fr_blkid > rp2->fr_blkid) 245fa9e4066Sahrens return (1); 246fa9e4066Sahrens else return (0); 247fa9e4066Sahrens } 248fa9e4066Sahrens 249fa9e4066Sahrens static void 250fa9e4066Sahrens dnode_setdblksz(dnode_t *dn, int size) 251fa9e4066Sahrens { 252fa9e4066Sahrens ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0); 253fa9e4066Sahrens ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 254fa9e4066Sahrens ASSERT3U(size, >=, SPA_MINBLOCKSIZE); 255fa9e4066Sahrens ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 256fa9e4066Sahrens 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); 257fa9e4066Sahrens dn->dn_datablksz = size; 258fa9e4066Sahrens dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; 259fa9e4066Sahrens dn->dn_datablkshift = ISP2(size) ? highbit(size - 1) : 0; 260fa9e4066Sahrens } 261fa9e4066Sahrens 262fa9e4066Sahrens static dnode_t * 263fa9e4066Sahrens dnode_create(objset_impl_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, 264fa9e4066Sahrens uint64_t object) 265fa9e4066Sahrens { 266fa9e4066Sahrens dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); 267fa9e4066Sahrens (void) dnode_cons(dn, NULL, 0); /* XXX */ 268fa9e4066Sahrens 269fa9e4066Sahrens dn->dn_objset = os; 270fa9e4066Sahrens dn->dn_object = object; 271fa9e4066Sahrens dn->dn_dbuf = db; 272fa9e4066Sahrens dn->dn_phys = dnp; 273fa9e4066Sahrens 274fa9e4066Sahrens if (dnp->dn_datablkszsec) 275fa9e4066Sahrens dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 276fa9e4066Sahrens dn->dn_indblkshift = dnp->dn_indblkshift; 277fa9e4066Sahrens dn->dn_nlevels = dnp->dn_nlevels; 278fa9e4066Sahrens dn->dn_type = dnp->dn_type; 279fa9e4066Sahrens dn->dn_nblkptr = dnp->dn_nblkptr; 280fa9e4066Sahrens dn->dn_checksum = dnp->dn_checksum; 281fa9e4066Sahrens dn->dn_compress = dnp->dn_compress; 282fa9e4066Sahrens dn->dn_bonustype = dnp->dn_bonustype; 283fa9e4066Sahrens dn->dn_bonuslen = dnp->dn_bonuslen; 284fa9e4066Sahrens dn->dn_maxblkid = dnp->dn_maxblkid; 285fa9e4066Sahrens 286fa9e4066Sahrens dmu_zfetch_init(&dn->dn_zfetch, dn); 287fa9e4066Sahrens 288fa9e4066Sahrens ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 289fa9e4066Sahrens mutex_enter(&os->os_lock); 290fa9e4066Sahrens list_insert_head(&os->os_dnodes, dn); 291fa9e4066Sahrens mutex_exit(&os->os_lock); 292fa9e4066Sahrens 293fa9e4066Sahrens return (dn); 294fa9e4066Sahrens } 295fa9e4066Sahrens 296fa9e4066Sahrens static void 297fa9e4066Sahrens dnode_destroy(dnode_t *dn) 298fa9e4066Sahrens { 299fa9e4066Sahrens objset_impl_t *os = dn->dn_objset; 300fa9e4066Sahrens 301fa9e4066Sahrens mutex_enter(&os->os_lock); 302fa9e4066Sahrens list_remove(&os->os_dnodes, dn); 303fa9e4066Sahrens mutex_exit(&os->os_lock); 304fa9e4066Sahrens 305fa9e4066Sahrens if (dn->dn_dirtyctx_firstset) { 306fa9e4066Sahrens kmem_free(dn->dn_dirtyctx_firstset, 1); 307fa9e4066Sahrens dn->dn_dirtyctx_firstset = NULL; 308fa9e4066Sahrens } 309fa9e4066Sahrens dmu_zfetch_rele(&dn->dn_zfetch); 310fa9e4066Sahrens kmem_cache_free(dnode_cache, dn); 311fa9e4066Sahrens } 312fa9e4066Sahrens 313fa9e4066Sahrens void 314fa9e4066Sahrens dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 315fa9e4066Sahrens dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 316fa9e4066Sahrens { 317fa9e4066Sahrens int i; 318fa9e4066Sahrens 319fa9e4066Sahrens if (blocksize == 0) 320fa9e4066Sahrens blocksize = 1 << zfs_default_bs; 3213b83abddSahrens else if (blocksize > SPA_MAXBLOCKSIZE) 3223b83abddSahrens blocksize = SPA_MAXBLOCKSIZE; 3233b83abddSahrens else 3243b83abddSahrens blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); 325fa9e4066Sahrens 326fa9e4066Sahrens if (ibs == 0) 327fa9e4066Sahrens ibs = zfs_default_ibs; 328fa9e4066Sahrens 329fa9e4066Sahrens ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); 330fa9e4066Sahrens 331fa9e4066Sahrens dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, 332fa9e4066Sahrens dn->dn_object, tx->tx_txg, blocksize, ibs); 333fa9e4066Sahrens 334fa9e4066Sahrens ASSERT(dn->dn_type == DMU_OT_NONE); 335fa9e4066Sahrens ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); 336fa9e4066Sahrens ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); 337fa9e4066Sahrens ASSERT(ot != DMU_OT_NONE); 338fa9e4066Sahrens ASSERT3U(ot, <, DMU_OT_NUMTYPES); 339fa9e4066Sahrens ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 340fa9e4066Sahrens (bonustype != DMU_OT_NONE && bonuslen != 0)); 341fa9e4066Sahrens ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 342fa9e4066Sahrens ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 343fa9e4066Sahrens ASSERT(dn->dn_type == DMU_OT_NONE); 344fa9e4066Sahrens ASSERT3U(dn->dn_maxblkid, ==, 0); 345fa9e4066Sahrens ASSERT3U(dn->dn_allocated_txg, ==, 0); 346fa9e4066Sahrens ASSERT3U(dn->dn_assigned_txg, ==, 0); 347fa9e4066Sahrens ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 348fa9e4066Sahrens ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); 349fa9e4066Sahrens ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 350fa9e4066Sahrens 351fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 352fa9e4066Sahrens ASSERT3U(dn->dn_next_nlevels[i], ==, 0); 353fa9e4066Sahrens ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); 354fa9e4066Sahrens ASSERT3U(dn->dn_dirtyblksz[i], ==, 0); 355fa9e4066Sahrens ASSERT3P(list_head(&dn->dn_dirty_dbufs[i]), ==, NULL); 356fa9e4066Sahrens ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0); 357fa9e4066Sahrens } 358fa9e4066Sahrens 359fa9e4066Sahrens dn->dn_type = ot; 360fa9e4066Sahrens dnode_setdblksz(dn, blocksize); 361fa9e4066Sahrens dn->dn_indblkshift = ibs; 362fa9e4066Sahrens dn->dn_nlevels = 1; 363fa9e4066Sahrens dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 364fa9e4066Sahrens dn->dn_bonustype = bonustype; 365fa9e4066Sahrens dn->dn_bonuslen = bonuslen; 366fa9e4066Sahrens dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 367fa9e4066Sahrens dn->dn_compress = ZIO_COMPRESS_INHERIT; 368fa9e4066Sahrens dn->dn_dirtyctx = 0; 369fa9e4066Sahrens 370fa9e4066Sahrens dn->dn_free_txg = 0; 371fa9e4066Sahrens if (dn->dn_dirtyctx_firstset) { 372fa9e4066Sahrens kmem_free(dn->dn_dirtyctx_firstset, 1); 373fa9e4066Sahrens dn->dn_dirtyctx_firstset = NULL; 374fa9e4066Sahrens } 375fa9e4066Sahrens 376fa9e4066Sahrens dn->dn_allocated_txg = tx->tx_txg; 377fa9e4066Sahrens dnode_setdirty(dn, tx); 378fa9e4066Sahrens } 379fa9e4066Sahrens 380fa9e4066Sahrens void 381fa9e4066Sahrens dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 382fa9e4066Sahrens dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 383fa9e4066Sahrens { 384fa9e4066Sahrens dmu_buf_impl_t *db = NULL; 385fa9e4066Sahrens 386fa9e4066Sahrens ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); 387fa9e4066Sahrens ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); 388fa9e4066Sahrens ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0); 389fa9e4066Sahrens ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 390fa9e4066Sahrens ASSERT(!(dn->dn_object & DMU_PRIVATE_OBJECT) || dmu_tx_private_ok(tx)); 391fa9e4066Sahrens ASSERT(tx->tx_txg != 0); 392fa9e4066Sahrens ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 393fa9e4066Sahrens (bonustype != DMU_OT_NONE && bonuslen != 0)); 394fa9e4066Sahrens ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 395fa9e4066Sahrens ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 396fa9e4066Sahrens ASSERT(dn->dn_dirtyblksz[0] == 0); 397fa9e4066Sahrens ASSERT(dn->dn_dirtyblksz[1] == 0); 398fa9e4066Sahrens ASSERT(dn->dn_dirtyblksz[2] == 0); 399fa9e4066Sahrens ASSERT(dn->dn_dirtyblksz[3] == 0); 400fa9e4066Sahrens 401fa9e4066Sahrens /* 402fa9e4066Sahrens * XXX I should really have a generation number to tell if we 403fa9e4066Sahrens * need to do this... 404fa9e4066Sahrens */ 405fa9e4066Sahrens if (blocksize != dn->dn_datablksz || 406fa9e4066Sahrens dn->dn_bonustype != bonustype || dn->dn_bonuslen != bonuslen) { 407fa9e4066Sahrens /* free all old data */ 408fa9e4066Sahrens dnode_free_range(dn, 0, -1ULL, tx); 409fa9e4066Sahrens } 410fa9e4066Sahrens 411fa9e4066Sahrens /* change blocksize */ 412fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 413fa9e4066Sahrens dnode_setdblksz(dn, blocksize); 414fa9e4066Sahrens dnode_setdirty(dn, tx); 415fa9e4066Sahrens /* don't need dd_dirty_mtx, dnode is already dirty */ 416fa9e4066Sahrens ASSERT(dn->dn_dirtyblksz[tx->tx_txg&TXG_MASK] != 0); 417fa9e4066Sahrens dn->dn_dirtyblksz[tx->tx_txg&TXG_MASK] = blocksize; 418fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 419fa9e4066Sahrens 420fa9e4066Sahrens /* change type */ 421fa9e4066Sahrens dn->dn_type = ot; 422fa9e4066Sahrens 423fa9e4066Sahrens if (dn->dn_bonuslen != bonuslen) { 424fa9e4066Sahrens /* change bonus size */ 425fa9e4066Sahrens if (bonuslen == 0) 426fa9e4066Sahrens bonuslen = 1; /* XXX */ 427fa9e4066Sahrens db = dbuf_hold_bonus(dn, FTAG); 428fa9e4066Sahrens dbuf_read(db); 429fa9e4066Sahrens mutex_enter(&db->db_mtx); 430fa9e4066Sahrens ASSERT3U(db->db.db_size, ==, dn->dn_bonuslen); 431fa9e4066Sahrens ASSERT(db->db.db_data != NULL); 432fa9e4066Sahrens db->db.db_size = bonuslen; 433fa9e4066Sahrens mutex_exit(&db->db_mtx); 434fa9e4066Sahrens dbuf_dirty(db, tx); 435fa9e4066Sahrens } 436fa9e4066Sahrens 437fa9e4066Sahrens /* change bonus size and type */ 438fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 439fa9e4066Sahrens dn->dn_bonustype = bonustype; 440fa9e4066Sahrens dn->dn_bonuslen = bonuslen; 441fa9e4066Sahrens dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 442fa9e4066Sahrens dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 443fa9e4066Sahrens dn->dn_compress = ZIO_COMPRESS_INHERIT; 444fa9e4066Sahrens ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 445fa9e4066Sahrens 446fa9e4066Sahrens dn->dn_allocated_txg = tx->tx_txg; 447fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 448fa9e4066Sahrens 449fa9e4066Sahrens if (db) 450fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 451fa9e4066Sahrens } 452fa9e4066Sahrens 453fa9e4066Sahrens void 454fa9e4066Sahrens dnode_special_close(dnode_t *dn) 455fa9e4066Sahrens { 456fa9e4066Sahrens dnode_destroy(dn); 457fa9e4066Sahrens } 458fa9e4066Sahrens 459fa9e4066Sahrens dnode_t * 460fa9e4066Sahrens dnode_special_open(objset_impl_t *os, dnode_phys_t *dnp, uint64_t object) 461fa9e4066Sahrens { 462fa9e4066Sahrens dnode_t *dn = dnode_create(os, dnp, NULL, object); 4639c9dc39aSek DNODE_VERIFY(dn); 464fa9e4066Sahrens return (dn); 465fa9e4066Sahrens } 466fa9e4066Sahrens 467fa9e4066Sahrens static void 468fa9e4066Sahrens dnode_buf_pageout(dmu_buf_t *db, void *arg) 469fa9e4066Sahrens { 470fa9e4066Sahrens dnode_t **children_dnodes = arg; 471fa9e4066Sahrens int i; 472fa9e4066Sahrens int epb = db->db_size >> DNODE_SHIFT; 473fa9e4066Sahrens 474fa9e4066Sahrens for (i = 0; i < epb; i++) { 475fa9e4066Sahrens dnode_t *dn = children_dnodes[i]; 476fa9e4066Sahrens int n; 477fa9e4066Sahrens 478fa9e4066Sahrens if (dn == NULL) 479fa9e4066Sahrens continue; 480fa9e4066Sahrens #ifdef ZFS_DEBUG 481fa9e4066Sahrens /* 482fa9e4066Sahrens * If there are holds on this dnode, then there should 483fa9e4066Sahrens * be holds on the dnode's containing dbuf as well; thus 484fa9e4066Sahrens * it wouldn't be eligable for eviction and this function 485fa9e4066Sahrens * would not have been called. 486fa9e4066Sahrens */ 487fa9e4066Sahrens ASSERT(refcount_is_zero(&dn->dn_holds)); 488fa9e4066Sahrens ASSERT(list_head(&dn->dn_dbufs) == NULL); 489fa9e4066Sahrens ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 490fa9e4066Sahrens 491fa9e4066Sahrens for (n = 0; n < TXG_SIZE; n++) 492fa9e4066Sahrens ASSERT(dn->dn_dirtyblksz[n] == 0); 493fa9e4066Sahrens #endif 494fa9e4066Sahrens children_dnodes[i] = NULL; 495fa9e4066Sahrens dnode_destroy(dn); 496fa9e4066Sahrens } 497fa9e4066Sahrens kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 498fa9e4066Sahrens } 499fa9e4066Sahrens 500fa9e4066Sahrens /* 501fa9e4066Sahrens * Returns held dnode if the object number is valid, NULL if not. 502fa9e4066Sahrens * Note that this will succeed even for free dnodes. 503fa9e4066Sahrens */ 504fa9e4066Sahrens dnode_t * 505fa9e4066Sahrens dnode_hold_impl(objset_impl_t *os, uint64_t object, int flag, void *ref) 506fa9e4066Sahrens { 507fa9e4066Sahrens int epb, idx; 508fa9e4066Sahrens int drop_struct_lock = FALSE; 509fa9e4066Sahrens uint64_t blk; 510fa9e4066Sahrens dnode_t *mdn, *dn; 511fa9e4066Sahrens dmu_buf_impl_t *db; 512fa9e4066Sahrens dnode_t **children_dnodes; 513fa9e4066Sahrens 514fa9e4066Sahrens if (object == 0 || object >= DN_MAX_OBJECT) 515fa9e4066Sahrens return (NULL); 516fa9e4066Sahrens 517fa9e4066Sahrens mdn = os->os_meta_dnode; 518fa9e4066Sahrens 5199c9dc39aSek DNODE_VERIFY(mdn); 520fa9e4066Sahrens 521fa9e4066Sahrens if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { 522fa9e4066Sahrens rw_enter(&mdn->dn_struct_rwlock, RW_READER); 523fa9e4066Sahrens drop_struct_lock = TRUE; 524fa9e4066Sahrens } 525fa9e4066Sahrens 526fa9e4066Sahrens blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); 527fa9e4066Sahrens 528fa9e4066Sahrens db = dbuf_hold(mdn, blk); 529fa9e4066Sahrens if (drop_struct_lock) 530fa9e4066Sahrens rw_exit(&mdn->dn_struct_rwlock); 531fa9e4066Sahrens dbuf_read(db); 532fa9e4066Sahrens 533fa9e4066Sahrens ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); 534fa9e4066Sahrens epb = db->db.db_size >> DNODE_SHIFT; 535fa9e4066Sahrens 536fa9e4066Sahrens idx = object & (epb-1); 537fa9e4066Sahrens 538fa9e4066Sahrens children_dnodes = dmu_buf_get_user(&db->db); 539fa9e4066Sahrens if (children_dnodes == NULL) { 540fa9e4066Sahrens dnode_t **winner; 541fa9e4066Sahrens children_dnodes = kmem_zalloc(epb * sizeof (dnode_t *), 542fa9e4066Sahrens KM_SLEEP); 543fa9e4066Sahrens if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, 544fa9e4066Sahrens dnode_buf_pageout)) { 545fa9e4066Sahrens kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 546fa9e4066Sahrens children_dnodes = winner; 547fa9e4066Sahrens } 548fa9e4066Sahrens } 549fa9e4066Sahrens 550fa9e4066Sahrens if ((dn = children_dnodes[idx]) == NULL) { 551fa9e4066Sahrens dnode_t *winner; 552fa9e4066Sahrens dn = dnode_create(os, (dnode_phys_t *)db->db.db_data+idx, 553fa9e4066Sahrens db, object); 554fa9e4066Sahrens winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn); 555fa9e4066Sahrens if (winner != NULL) { 556fa9e4066Sahrens dnode_destroy(dn); 557fa9e4066Sahrens dn = winner; 558fa9e4066Sahrens } 559fa9e4066Sahrens } 560fa9e4066Sahrens 561fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 562fa9e4066Sahrens if (dn->dn_free_txg || 563fa9e4066Sahrens ((flag & DNODE_MUST_BE_ALLOCATED) && dn->dn_type == DMU_OT_NONE) || 564fa9e4066Sahrens ((flag & DNODE_MUST_BE_FREE) && dn->dn_type != DMU_OT_NONE)) { 565fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 566fa9e4066Sahrens dbuf_rele(db); 567fa9e4066Sahrens return (NULL); 568fa9e4066Sahrens } 569fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 570fa9e4066Sahrens 571fa9e4066Sahrens if (refcount_add(&dn->dn_holds, ref) == 1) 572fa9e4066Sahrens dbuf_add_ref(db, dn); 573fa9e4066Sahrens 5749c9dc39aSek DNODE_VERIFY(dn); 575fa9e4066Sahrens ASSERT3P(dn->dn_dbuf, ==, db); 576fa9e4066Sahrens ASSERT3U(dn->dn_object, ==, object); 577fa9e4066Sahrens dbuf_rele(db); 578fa9e4066Sahrens 579fa9e4066Sahrens return (dn); 580fa9e4066Sahrens } 581fa9e4066Sahrens 582fa9e4066Sahrens /* 583fa9e4066Sahrens * Return held dnode if the object is allocated, NULL if not. 584fa9e4066Sahrens */ 585fa9e4066Sahrens dnode_t * 586fa9e4066Sahrens dnode_hold(objset_impl_t *os, uint64_t object, void *ref) 587fa9e4066Sahrens { 588fa9e4066Sahrens return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, ref)); 589fa9e4066Sahrens } 590fa9e4066Sahrens 591fa9e4066Sahrens void 592fa9e4066Sahrens dnode_add_ref(dnode_t *dn, void *ref) 593fa9e4066Sahrens { 594fa9e4066Sahrens ASSERT(refcount_count(&dn->dn_holds) > 0); 595fa9e4066Sahrens (void) refcount_add(&dn->dn_holds, ref); 596fa9e4066Sahrens } 597fa9e4066Sahrens 598fa9e4066Sahrens void 599fa9e4066Sahrens dnode_rele(dnode_t *dn, void *ref) 600fa9e4066Sahrens { 601fa9e4066Sahrens uint64_t refs; 602fa9e4066Sahrens 603fa9e4066Sahrens refs = refcount_remove(&dn->dn_holds, ref); 604fa9e4066Sahrens /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ 605fa9e4066Sahrens if (refs == 0 && dn->dn_dbuf) 606fa9e4066Sahrens dbuf_remove_ref(dn->dn_dbuf, dn); 607fa9e4066Sahrens } 608fa9e4066Sahrens 609fa9e4066Sahrens void 610fa9e4066Sahrens dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) 611fa9e4066Sahrens { 612fa9e4066Sahrens objset_impl_t *os = dn->dn_objset; 613fa9e4066Sahrens uint64_t txg = tx->tx_txg; 614fa9e4066Sahrens 615fa9e4066Sahrens if (IS_DNODE_DNODE(dn->dn_object)) 616fa9e4066Sahrens return; 617fa9e4066Sahrens 6189c9dc39aSek DNODE_VERIFY(dn); 619fa9e4066Sahrens 620fa9e4066Sahrens #ifdef ZFS_DEBUG 621fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 622fa9e4066Sahrens ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); 623fa9e4066Sahrens /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */ 624fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 625fa9e4066Sahrens #endif 626fa9e4066Sahrens 627fa9e4066Sahrens mutex_enter(&os->os_lock); 628fa9e4066Sahrens 629fa9e4066Sahrens /* 630fa9e4066Sahrens * If we are already marked dirty, we're done. 631fa9e4066Sahrens */ 632fa9e4066Sahrens if (dn->dn_dirtyblksz[txg&TXG_MASK] > 0) { 633fa9e4066Sahrens mutex_exit(&os->os_lock); 634fa9e4066Sahrens return; 635fa9e4066Sahrens } 636fa9e4066Sahrens 637fa9e4066Sahrens ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); 638fa9e4066Sahrens ASSERT(dn->dn_datablksz != 0); 639fa9e4066Sahrens dn->dn_dirtyblksz[txg&TXG_MASK] = dn->dn_datablksz; 640fa9e4066Sahrens 641fa9e4066Sahrens dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", 642fa9e4066Sahrens dn->dn_object, txg); 643fa9e4066Sahrens 644fa9e4066Sahrens if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { 645fa9e4066Sahrens list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); 646fa9e4066Sahrens } else { 647fa9e4066Sahrens list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); 648fa9e4066Sahrens } 649fa9e4066Sahrens 650fa9e4066Sahrens mutex_exit(&os->os_lock); 651fa9e4066Sahrens 652fa9e4066Sahrens /* 653fa9e4066Sahrens * The dnode maintains a hold on its containing dbuf as 654fa9e4066Sahrens * long as there are holds on it. Each instantiated child 655fa9e4066Sahrens * dbuf maintaines a hold on the dnode. When the last child 656fa9e4066Sahrens * drops its hold, the dnode will drop its hold on the 657fa9e4066Sahrens * containing dbuf. We add a "dirty hold" here so that the 658fa9e4066Sahrens * dnode will hang around after we finish processing its 659fa9e4066Sahrens * children. 660fa9e4066Sahrens */ 661fa9e4066Sahrens (void) refcount_add(&dn->dn_holds, (void *)(uintptr_t)tx->tx_txg); 662fa9e4066Sahrens 663fa9e4066Sahrens dbuf_dirty(dn->dn_dbuf, tx); 664fa9e4066Sahrens 665fa9e4066Sahrens dsl_dataset_dirty(os->os_dsl_dataset, tx); 666fa9e4066Sahrens } 667fa9e4066Sahrens 668fa9e4066Sahrens void 669fa9e4066Sahrens dnode_free(dnode_t *dn, dmu_tx_t *tx) 670fa9e4066Sahrens { 671fa9e4066Sahrens dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); 672fa9e4066Sahrens 673fa9e4066Sahrens /* we should be the only holder... hopefully */ 674fa9e4066Sahrens /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ 675fa9e4066Sahrens 676fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 677fa9e4066Sahrens if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { 678fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 679fa9e4066Sahrens return; 680fa9e4066Sahrens } 681fa9e4066Sahrens dn->dn_free_txg = tx->tx_txg; 682fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 683fa9e4066Sahrens 684fa9e4066Sahrens /* 685fa9e4066Sahrens * If the dnode is already dirty, it needs to be moved from 686fa9e4066Sahrens * the dirty list to the free list. 687fa9e4066Sahrens */ 688fa9e4066Sahrens mutex_enter(&dn->dn_objset->os_lock); 689fa9e4066Sahrens if (dn->dn_dirtyblksz[tx->tx_txg&TXG_MASK] > 0) { 690fa9e4066Sahrens list_remove( 691fa9e4066Sahrens &dn->dn_objset->os_dirty_dnodes[tx->tx_txg&TXG_MASK], dn); 692fa9e4066Sahrens list_insert_tail( 693fa9e4066Sahrens &dn->dn_objset->os_free_dnodes[tx->tx_txg&TXG_MASK], dn); 694fa9e4066Sahrens mutex_exit(&dn->dn_objset->os_lock); 695fa9e4066Sahrens } else { 696fa9e4066Sahrens mutex_exit(&dn->dn_objset->os_lock); 697fa9e4066Sahrens dnode_setdirty(dn, tx); 698fa9e4066Sahrens } 699fa9e4066Sahrens } 700fa9e4066Sahrens 701fa9e4066Sahrens /* 702fa9e4066Sahrens * Try to change the block size for the indicated dnode. This can only 703fa9e4066Sahrens * succeed if there are no blocks allocated or dirty beyond first block 704fa9e4066Sahrens */ 705fa9e4066Sahrens int 706fa9e4066Sahrens dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) 707fa9e4066Sahrens { 708fa9e4066Sahrens dmu_buf_impl_t *db, *db_next; 709fa9e4066Sahrens int have_db0 = FALSE; 710fa9e4066Sahrens int err = ENOTSUP; 711fa9e4066Sahrens 712fa9e4066Sahrens if (size == 0) 713fa9e4066Sahrens size = SPA_MINBLOCKSIZE; 714fa9e4066Sahrens if (size > SPA_MAXBLOCKSIZE) 715fa9e4066Sahrens size = SPA_MAXBLOCKSIZE; 716fa9e4066Sahrens else 717fa9e4066Sahrens size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); 718fa9e4066Sahrens 719fa9e4066Sahrens if (ibs == 0) 720fa9e4066Sahrens ibs = dn->dn_indblkshift; 721fa9e4066Sahrens 722fa9e4066Sahrens if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && 723fa9e4066Sahrens ibs == dn->dn_indblkshift) 724fa9e4066Sahrens return (0); 725fa9e4066Sahrens 726fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 727fa9e4066Sahrens 728fa9e4066Sahrens /* Check for any allocated blocks beyond the first */ 729fa9e4066Sahrens if (dn->dn_phys->dn_maxblkid != 0) 730fa9e4066Sahrens goto end; 731fa9e4066Sahrens 732fa9e4066Sahrens /* 733fa9e4066Sahrens * Any buffers allocated for blocks beyond the first 734fa9e4066Sahrens * must be evictable/evicted, because they're the wrong size. 735fa9e4066Sahrens */ 736fa9e4066Sahrens mutex_enter(&dn->dn_dbufs_mtx); 737fa9e4066Sahrens /* 738fa9e4066Sahrens * Since we have the dn_dbufs_mtx, nothing can be 739fa9e4066Sahrens * removed from dn_dbufs. Since we have dn_struct_rwlock/w, 740fa9e4066Sahrens * nothing can be added to dn_dbufs. 741fa9e4066Sahrens */ 742fa9e4066Sahrens for (db = list_head(&dn->dn_dbufs); db; db = db_next) { 743fa9e4066Sahrens db_next = list_next(&dn->dn_dbufs, db); 744fa9e4066Sahrens 745fa9e4066Sahrens if (db->db_blkid == 0) { 746fa9e4066Sahrens have_db0 = TRUE; 747fa9e4066Sahrens } else if (db->db_blkid != DB_BONUS_BLKID) { 748fa9e4066Sahrens mutex_exit(&dn->dn_dbufs_mtx); 749fa9e4066Sahrens goto end; 750fa9e4066Sahrens } 751fa9e4066Sahrens } 752fa9e4066Sahrens mutex_exit(&dn->dn_dbufs_mtx); 753fa9e4066Sahrens 754fa9e4066Sahrens /* Fast-track if there is no data in the file */ 755fa9e4066Sahrens if (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) && !have_db0) { 756fa9e4066Sahrens dnode_setdblksz(dn, size); 757fa9e4066Sahrens dn->dn_indblkshift = ibs; 758fa9e4066Sahrens dnode_setdirty(dn, tx); 759fa9e4066Sahrens /* don't need dd_dirty_mtx, dnode is already dirty */ 760fa9e4066Sahrens dn->dn_dirtyblksz[tx->tx_txg&TXG_MASK] = size; 761fa9e4066Sahrens dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 762fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 763fa9e4066Sahrens return (0); 764fa9e4066Sahrens } 765fa9e4066Sahrens 766fa9e4066Sahrens /* obtain the old block */ 767fa9e4066Sahrens db = dbuf_hold(dn, 0); 768fa9e4066Sahrens 769fa9e4066Sahrens dbuf_new_size(db, size, tx); 770fa9e4066Sahrens 771fa9e4066Sahrens dnode_setdblksz(dn, size); 772fa9e4066Sahrens dn->dn_indblkshift = ibs; 773fa9e4066Sahrens /* don't need dd_dirty_mtx, dnode is already dirty */ 774fa9e4066Sahrens dn->dn_dirtyblksz[tx->tx_txg&TXG_MASK] = size; 775fa9e4066Sahrens dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 776fa9e4066Sahrens dbuf_rele(db); 777fa9e4066Sahrens 778fa9e4066Sahrens err = 0; 779fa9e4066Sahrens end: 780fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 781fa9e4066Sahrens return (err); 782fa9e4066Sahrens } 783fa9e4066Sahrens 784fa9e4066Sahrens uint64_t 785fa9e4066Sahrens dnode_max_nonzero_offset(dnode_t *dn) 786fa9e4066Sahrens { 787fa9e4066Sahrens if (dn->dn_phys->dn_maxblkid == 0 && 788fa9e4066Sahrens BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0])) 789fa9e4066Sahrens return (0); 790fa9e4066Sahrens else 791fa9e4066Sahrens return ((dn->dn_phys->dn_maxblkid+1) * dn->dn_datablksz); 792fa9e4066Sahrens } 793fa9e4066Sahrens 794fa9e4066Sahrens void 795fa9e4066Sahrens dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx) 796fa9e4066Sahrens { 797fa9e4066Sahrens uint64_t txgoff = tx->tx_txg & TXG_MASK; 798fa9e4066Sahrens int drop_struct_lock = FALSE; 799fa9e4066Sahrens int epbs, old_nlevels, new_nlevels; 800fa9e4066Sahrens uint64_t sz; 801fa9e4066Sahrens 802fa9e4066Sahrens if (blkid == DB_BONUS_BLKID) 803fa9e4066Sahrens return; 804fa9e4066Sahrens 805fa9e4066Sahrens if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 806fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 807fa9e4066Sahrens drop_struct_lock = TRUE; 808fa9e4066Sahrens } 809fa9e4066Sahrens 810fa9e4066Sahrens if (blkid > dn->dn_maxblkid) 811fa9e4066Sahrens dn->dn_maxblkid = blkid; 812fa9e4066Sahrens 813fa9e4066Sahrens /* 814fa9e4066Sahrens * Compute the number of levels necessary to support the 815fa9e4066Sahrens * new blkid. 816fa9e4066Sahrens */ 817fa9e4066Sahrens new_nlevels = 1; 818fa9e4066Sahrens epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 819fa9e4066Sahrens 820fa9e4066Sahrens for (sz = dn->dn_nblkptr; sz <= blkid && sz >= dn->dn_nblkptr; 821fa9e4066Sahrens sz <<= epbs) 822fa9e4066Sahrens new_nlevels++; 823fa9e4066Sahrens old_nlevels = dn->dn_nlevels; 824fa9e4066Sahrens 825fa9e4066Sahrens if (new_nlevels > dn->dn_next_nlevels[txgoff]) 826fa9e4066Sahrens dn->dn_next_nlevels[txgoff] = new_nlevels; 827fa9e4066Sahrens 828fa9e4066Sahrens if (new_nlevels > old_nlevels) { 829fa9e4066Sahrens dprintf("dn %p increasing nlevels from %u to %u\n", 830fa9e4066Sahrens dn, dn->dn_nlevels, new_nlevels); 831fa9e4066Sahrens dn->dn_nlevels = new_nlevels; 832fa9e4066Sahrens } 833fa9e4066Sahrens 834fa9e4066Sahrens /* 835fa9e4066Sahrens * Dirty the left indirects. 836fa9e4066Sahrens * Note: the caller should have just dnode_use_space()'d one 837fa9e4066Sahrens * data block's worth, so we could subtract that out of 838fa9e4066Sahrens * dn_inflight_data to determine if there is any dirty data 839fa9e4066Sahrens * besides this block. 840fa9e4066Sahrens * We don't strictly need to dirty them unless there's 841fa9e4066Sahrens * *something* in the object (eg. on disk or dirty)... 842fa9e4066Sahrens */ 843fa9e4066Sahrens if (new_nlevels > old_nlevels) { 844fa9e4066Sahrens dmu_buf_impl_t *db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); 845fa9e4066Sahrens dprintf("dn %p dirtying left indirects\n", dn); 846fa9e4066Sahrens dbuf_dirty(db, tx); 847fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 848fa9e4066Sahrens } 849fa9e4066Sahrens #ifdef ZFS_DEBUG 850fa9e4066Sahrens else if (old_nlevels > 1 && new_nlevels > old_nlevels) { 851fa9e4066Sahrens dmu_buf_impl_t *db; 852fa9e4066Sahrens int i; 853fa9e4066Sahrens 854fa9e4066Sahrens for (i = 0; i < dn->dn_nblkptr; i++) { 855fa9e4066Sahrens db = dbuf_hold_level(dn, old_nlevels-1, i, FTAG); 856fa9e4066Sahrens ASSERT(! 857fa9e4066Sahrens list_link_active(&db->db_dirty_node[txgoff])); 858fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 859fa9e4066Sahrens } 860fa9e4066Sahrens } 861fa9e4066Sahrens #endif 862fa9e4066Sahrens 863fa9e4066Sahrens dprintf("dn %p done\n", dn); 864fa9e4066Sahrens 865fa9e4066Sahrens out: 866fa9e4066Sahrens if (drop_struct_lock) 867fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 868fa9e4066Sahrens } 869fa9e4066Sahrens 870fa9e4066Sahrens void 871fa9e4066Sahrens dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) 872fa9e4066Sahrens { 873fa9e4066Sahrens avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 874fa9e4066Sahrens avl_index_t where; 875fa9e4066Sahrens free_range_t *rp; 876fa9e4066Sahrens free_range_t rp_tofind; 877fa9e4066Sahrens uint64_t endblk = blkid + nblks; 878fa9e4066Sahrens 879fa9e4066Sahrens ASSERT(MUTEX_HELD(&dn->dn_mtx)); 880fa9e4066Sahrens ASSERT(nblks <= UINT64_MAX - blkid); /* no overflow */ 881fa9e4066Sahrens 882fa9e4066Sahrens dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 883fa9e4066Sahrens blkid, nblks, tx->tx_txg); 884fa9e4066Sahrens rp_tofind.fr_blkid = blkid; 885fa9e4066Sahrens rp = avl_find(tree, &rp_tofind, &where); 886fa9e4066Sahrens if (rp == NULL) 887fa9e4066Sahrens rp = avl_nearest(tree, where, AVL_BEFORE); 888fa9e4066Sahrens if (rp == NULL) 889fa9e4066Sahrens rp = avl_nearest(tree, where, AVL_AFTER); 890fa9e4066Sahrens 891fa9e4066Sahrens while (rp && (rp->fr_blkid <= blkid + nblks)) { 892fa9e4066Sahrens uint64_t fr_endblk = rp->fr_blkid + rp->fr_nblks; 893fa9e4066Sahrens free_range_t *nrp = AVL_NEXT(tree, rp); 894fa9e4066Sahrens 895fa9e4066Sahrens if (blkid <= rp->fr_blkid && endblk >= fr_endblk) { 896fa9e4066Sahrens /* clear this entire range */ 897fa9e4066Sahrens avl_remove(tree, rp); 898fa9e4066Sahrens kmem_free(rp, sizeof (free_range_t)); 899fa9e4066Sahrens } else if (blkid <= rp->fr_blkid && 900fa9e4066Sahrens endblk > rp->fr_blkid && endblk < fr_endblk) { 901fa9e4066Sahrens /* clear the beginning of this range */ 902fa9e4066Sahrens rp->fr_blkid = endblk; 903fa9e4066Sahrens rp->fr_nblks = fr_endblk - endblk; 904fa9e4066Sahrens } else if (blkid > rp->fr_blkid && blkid < fr_endblk && 905fa9e4066Sahrens endblk >= fr_endblk) { 906fa9e4066Sahrens /* clear the end of this range */ 907fa9e4066Sahrens rp->fr_nblks = blkid - rp->fr_blkid; 908fa9e4066Sahrens } else if (blkid > rp->fr_blkid && endblk < fr_endblk) { 909fa9e4066Sahrens /* clear a chunk out of this range */ 910fa9e4066Sahrens free_range_t *new_rp = 911fa9e4066Sahrens kmem_alloc(sizeof (free_range_t), KM_SLEEP); 912fa9e4066Sahrens 913fa9e4066Sahrens new_rp->fr_blkid = endblk; 914fa9e4066Sahrens new_rp->fr_nblks = fr_endblk - endblk; 915fa9e4066Sahrens avl_insert_here(tree, new_rp, rp, AVL_AFTER); 916fa9e4066Sahrens rp->fr_nblks = blkid - rp->fr_blkid; 917fa9e4066Sahrens } 918fa9e4066Sahrens /* there may be no overlap */ 919fa9e4066Sahrens rp = nrp; 920fa9e4066Sahrens } 921fa9e4066Sahrens } 922fa9e4066Sahrens 923fa9e4066Sahrens void 924fa9e4066Sahrens dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) 925fa9e4066Sahrens { 926fa9e4066Sahrens dmu_buf_impl_t *db; 927fa9e4066Sahrens uint64_t start, objsize, blkid, nblks; 928fa9e4066Sahrens int blkshift, blksz, tail, head, epbs; 929fa9e4066Sahrens int trunc = FALSE; 930fa9e4066Sahrens 931fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 932fa9e4066Sahrens blksz = dn->dn_datablksz; 933fa9e4066Sahrens blkshift = dn->dn_datablkshift; 934fa9e4066Sahrens epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 935fa9e4066Sahrens 936fa9e4066Sahrens /* If the range is past the end of the file, this is a no-op */ 937fa9e4066Sahrens objsize = blksz * (dn->dn_maxblkid+1); 938fa9e4066Sahrens if (off >= objsize) 939fa9e4066Sahrens goto out; 940fa9e4066Sahrens if (len == -1ULL) { 941fa9e4066Sahrens len = UINT64_MAX - off; 942fa9e4066Sahrens trunc = TRUE; 943fa9e4066Sahrens } 944fa9e4066Sahrens 945fa9e4066Sahrens /* 946fa9e4066Sahrens * First, block align the region to free: 947fa9e4066Sahrens */ 948fa9e4066Sahrens if (dn->dn_maxblkid == 0) { 949fa9e4066Sahrens if (off == 0) { 950fa9e4066Sahrens head = 0; 951fa9e4066Sahrens } else { 952fa9e4066Sahrens head = blksz - off; 953fa9e4066Sahrens ASSERT3U(head, >, 0); 954fa9e4066Sahrens } 955fa9e4066Sahrens start = off; 956fa9e4066Sahrens } else { 957fa9e4066Sahrens ASSERT(ISP2(blksz)); 958fa9e4066Sahrens head = P2NPHASE(off, blksz); 959fa9e4066Sahrens start = P2PHASE(off, blksz); 960fa9e4066Sahrens } 961fa9e4066Sahrens /* zero out any partial block data at the start of the range */ 962fa9e4066Sahrens if (head) { 963fa9e4066Sahrens ASSERT3U(start + head, ==, blksz); 964fa9e4066Sahrens if (len < head) 965fa9e4066Sahrens head = len; 966fa9e4066Sahrens if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, 967fa9e4066Sahrens FTAG, &db) == 0) { 968fa9e4066Sahrens caddr_t data; 969fa9e4066Sahrens 970fa9e4066Sahrens /* don't dirty if it isn't on disk and isn't dirty */ 971fa9e4066Sahrens if (db->db_dirtied || 972fa9e4066Sahrens (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 973fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 974fa9e4066Sahrens dbuf_will_dirty(db, tx); 975fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 976fa9e4066Sahrens data = db->db.db_data; 977fa9e4066Sahrens bzero(data + start, head); 978fa9e4066Sahrens } 979fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 980fa9e4066Sahrens } 981fa9e4066Sahrens off += head; 982fa9e4066Sahrens len -= head; 983fa9e4066Sahrens } 984fa9e4066Sahrens /* If the range was less than one block, we are done */ 985fa9e4066Sahrens if (len == 0) 986fa9e4066Sahrens goto out; 987fa9e4066Sahrens 988fa9e4066Sahrens /* If the remaining range is past the end of the file, we are done */ 989fa9e4066Sahrens if (off > dn->dn_maxblkid << blkshift) 990fa9e4066Sahrens goto out; 991fa9e4066Sahrens 992fa9e4066Sahrens if (off + len == UINT64_MAX) 993fa9e4066Sahrens tail = 0; 994fa9e4066Sahrens else 995fa9e4066Sahrens tail = P2PHASE(len, blksz); 996fa9e4066Sahrens 997fa9e4066Sahrens ASSERT3U(P2PHASE(off, blksz), ==, 0); 998fa9e4066Sahrens /* zero out any partial block data at the end of the range */ 999fa9e4066Sahrens if (tail) { 1000fa9e4066Sahrens if (len < tail) 1001fa9e4066Sahrens tail = len; 1002fa9e4066Sahrens if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), 1003fa9e4066Sahrens TRUE, FTAG, &db) == 0) { 1004fa9e4066Sahrens /* don't dirty if it isn't on disk and isn't dirty */ 1005fa9e4066Sahrens if (db->db_dirtied || 1006fa9e4066Sahrens (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 1007fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1008fa9e4066Sahrens dbuf_will_dirty(db, tx); 1009fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1010fa9e4066Sahrens bzero(db->db.db_data, tail); 1011fa9e4066Sahrens } 1012fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 1013fa9e4066Sahrens } 1014fa9e4066Sahrens len -= tail; 1015fa9e4066Sahrens } 1016fa9e4066Sahrens /* If the range did not include a full block, we are done */ 1017fa9e4066Sahrens if (len == 0) 1018fa9e4066Sahrens goto out; 1019fa9e4066Sahrens 1020fa9e4066Sahrens /* dirty the left indirects */ 1021fa9e4066Sahrens if (dn->dn_nlevels > 1 && off != 0) { 1022fa9e4066Sahrens db = dbuf_hold_level(dn, 1, 1023fa9e4066Sahrens (off - head) >> (blkshift + epbs), FTAG); 1024fa9e4066Sahrens dbuf_will_dirty(db, tx); 1025fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 1026fa9e4066Sahrens } 1027fa9e4066Sahrens 1028fa9e4066Sahrens /* dirty the right indirects */ 1029fa9e4066Sahrens if (dn->dn_nlevels > 1 && !trunc) { 1030fa9e4066Sahrens db = dbuf_hold_level(dn, 1, 1031fa9e4066Sahrens (off + len + tail - 1) >> (blkshift + epbs), FTAG); 1032fa9e4066Sahrens dbuf_will_dirty(db, tx); 1033fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 1034fa9e4066Sahrens } 1035fa9e4066Sahrens 1036fa9e4066Sahrens /* 1037fa9e4066Sahrens * Finally, add this range to the dnode range list, we 1038fa9e4066Sahrens * will finish up this free operation in the syncing phase. 1039fa9e4066Sahrens */ 1040fa9e4066Sahrens ASSERT(IS_P2ALIGNED(off, 1<<blkshift)); 1041fa9e4066Sahrens ASSERT(off + len == UINT64_MAX || IS_P2ALIGNED(len, 1<<blkshift)); 1042fa9e4066Sahrens blkid = off >> blkshift; 1043fa9e4066Sahrens nblks = len >> blkshift; 1044fa9e4066Sahrens 1045fa9e4066Sahrens if (trunc) 1046fa9e4066Sahrens dn->dn_maxblkid = (blkid ? blkid - 1 : 0); 1047fa9e4066Sahrens 1048fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 1049fa9e4066Sahrens dnode_clear_range(dn, blkid, nblks, tx); 1050fa9e4066Sahrens { 1051fa9e4066Sahrens free_range_t *rp, *found; 1052fa9e4066Sahrens avl_index_t where; 1053fa9e4066Sahrens avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 1054fa9e4066Sahrens 1055fa9e4066Sahrens /* Add new range to dn_ranges */ 1056fa9e4066Sahrens rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP); 1057fa9e4066Sahrens rp->fr_blkid = blkid; 1058fa9e4066Sahrens rp->fr_nblks = nblks; 1059fa9e4066Sahrens found = avl_find(tree, rp, &where); 1060fa9e4066Sahrens ASSERT(found == NULL); 1061fa9e4066Sahrens avl_insert(tree, rp, where); 1062fa9e4066Sahrens dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 1063fa9e4066Sahrens blkid, nblks, tx->tx_txg); 1064fa9e4066Sahrens } 1065fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 1066fa9e4066Sahrens 1067fa9e4066Sahrens dbuf_free_range(dn, blkid, nblks, tx); 1068fa9e4066Sahrens dnode_setdirty(dn, tx); 1069fa9e4066Sahrens out: 1070fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1071fa9e4066Sahrens } 1072fa9e4066Sahrens 1073fa9e4066Sahrens /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ 1074fa9e4066Sahrens uint64_t 1075fa9e4066Sahrens dnode_block_freed(dnode_t *dn, uint64_t blkid) 1076fa9e4066Sahrens { 1077fa9e4066Sahrens free_range_t range_tofind; 1078fa9e4066Sahrens void *dp = spa_get_dsl(dn->dn_objset->os_spa); 1079fa9e4066Sahrens int i; 1080fa9e4066Sahrens 1081fa9e4066Sahrens if (blkid == DB_BONUS_BLKID) 1082fa9e4066Sahrens return (FALSE); 1083fa9e4066Sahrens 1084fa9e4066Sahrens /* 1085fa9e4066Sahrens * If we're in the process of opening the pool, dp will not be 1086fa9e4066Sahrens * set yet, but there shouldn't be anything dirty. 1087fa9e4066Sahrens */ 1088fa9e4066Sahrens if (dp == NULL) 1089fa9e4066Sahrens return (FALSE); 1090fa9e4066Sahrens 1091fa9e4066Sahrens if (dn->dn_free_txg) 1092fa9e4066Sahrens return (TRUE); 1093fa9e4066Sahrens 1094fa9e4066Sahrens /* 1095fa9e4066Sahrens * If dn_datablkshift is not set, then there's only a single 1096fa9e4066Sahrens * block, in which case there will never be a free range so it 1097fa9e4066Sahrens * won't matter. 1098fa9e4066Sahrens */ 1099fa9e4066Sahrens range_tofind.fr_blkid = blkid; 1100fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 1101fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 1102fa9e4066Sahrens free_range_t *range_found; 1103fa9e4066Sahrens avl_index_t idx; 1104fa9e4066Sahrens 1105fa9e4066Sahrens range_found = avl_find(&dn->dn_ranges[i], &range_tofind, &idx); 1106fa9e4066Sahrens if (range_found) { 1107fa9e4066Sahrens ASSERT(range_found->fr_nblks > 0); 1108fa9e4066Sahrens break; 1109fa9e4066Sahrens } 1110fa9e4066Sahrens range_found = avl_nearest(&dn->dn_ranges[i], idx, AVL_BEFORE); 1111fa9e4066Sahrens if (range_found && 1112fa9e4066Sahrens range_found->fr_blkid + range_found->fr_nblks > blkid) 1113fa9e4066Sahrens break; 1114fa9e4066Sahrens } 1115fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 1116fa9e4066Sahrens return (i < TXG_SIZE); 1117fa9e4066Sahrens } 1118fa9e4066Sahrens 1119fa9e4066Sahrens /* call from syncing context when we actually write/free space for this dnode */ 1120fa9e4066Sahrens void 1121fa9e4066Sahrens dnode_diduse_space(dnode_t *dn, int64_t space) 1122fa9e4066Sahrens { 1123fa9e4066Sahrens uint64_t sectors; 1124fa9e4066Sahrens 1125fa9e4066Sahrens dprintf_dnode(dn, "dn=%p dnp=%p secphys=%llu space=%lld\n", 1126fa9e4066Sahrens dn, dn->dn_phys, 1127fa9e4066Sahrens (u_longlong_t)dn->dn_phys->dn_secphys, 1128fa9e4066Sahrens (longlong_t)space); 1129fa9e4066Sahrens 1130fa9e4066Sahrens ASSERT(P2PHASE(space, 1<<DEV_BSHIFT) == 0); 1131fa9e4066Sahrens 1132fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 1133fa9e4066Sahrens if (space > 0) { 1134fa9e4066Sahrens sectors = space >> DEV_BSHIFT; 1135fa9e4066Sahrens ASSERT3U(dn->dn_phys->dn_secphys + sectors, >=, 1136fa9e4066Sahrens dn->dn_phys->dn_secphys); 1137fa9e4066Sahrens dn->dn_phys->dn_secphys += sectors; 1138fa9e4066Sahrens } else { 1139fa9e4066Sahrens sectors = -space >> DEV_BSHIFT; 1140fa9e4066Sahrens ASSERT3U(dn->dn_phys->dn_secphys, >=, sectors); 1141fa9e4066Sahrens dn->dn_phys->dn_secphys -= sectors; 1142fa9e4066Sahrens } 1143fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 1144fa9e4066Sahrens } 1145fa9e4066Sahrens 1146fa9e4066Sahrens /* 1147fa9e4066Sahrens * Call when we think we're going to write/free space in open context. 1148fa9e4066Sahrens * Be conservative (ie. OK to write less than this or free more than 1149fa9e4066Sahrens * this, but don't write more or free less). 1150fa9e4066Sahrens */ 1151fa9e4066Sahrens void 1152fa9e4066Sahrens dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) 1153fa9e4066Sahrens { 1154fa9e4066Sahrens objset_impl_t *os = dn->dn_objset; 1155fa9e4066Sahrens dsl_dataset_t *ds = os->os_dsl_dataset; 1156fa9e4066Sahrens 1157fa9e4066Sahrens if (space > 0) 1158fa9e4066Sahrens space = spa_get_asize(os->os_spa, space); 1159fa9e4066Sahrens 1160fa9e4066Sahrens if (ds) 1161fa9e4066Sahrens dsl_dir_willuse_space(ds->ds_dir, space, tx); 1162fa9e4066Sahrens 1163fa9e4066Sahrens dmu_tx_willuse_space(tx, space); 1164fa9e4066Sahrens } 1165fa9e4066Sahrens 1166fa9e4066Sahrens static int 1167fa9e4066Sahrens dnode_next_offset_level(dnode_t *dn, boolean_t hole, uint64_t *offset, 1168fa9e4066Sahrens int lvl, uint64_t blkfill) 1169fa9e4066Sahrens { 1170fa9e4066Sahrens dmu_buf_impl_t *db = NULL; 1171fa9e4066Sahrens void *data = NULL; 1172fa9e4066Sahrens uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 1173fa9e4066Sahrens uint64_t epb = 1ULL << epbs; 1174fa9e4066Sahrens uint64_t minfill, maxfill; 1175fa9e4066Sahrens int i, error, span; 1176fa9e4066Sahrens 1177fa9e4066Sahrens dprintf("probing object %llu offset %llx level %d of %u\n", 1178fa9e4066Sahrens dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); 1179fa9e4066Sahrens 1180fa9e4066Sahrens if (lvl == dn->dn_phys->dn_nlevels) { 1181fa9e4066Sahrens error = 0; 1182fa9e4066Sahrens epb = dn->dn_phys->dn_nblkptr; 1183fa9e4066Sahrens data = dn->dn_phys->dn_blkptr; 1184fa9e4066Sahrens } else { 1185fa9e4066Sahrens uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); 1186fa9e4066Sahrens error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); 1187fa9e4066Sahrens if (error) { 1188fa9e4066Sahrens if (error == ENOENT) 1189fa9e4066Sahrens return (hole ? 0 : ESRCH); 1190fa9e4066Sahrens return (error); 1191fa9e4066Sahrens } 1192fa9e4066Sahrens dbuf_read_havestruct(db); 1193fa9e4066Sahrens data = db->db.db_data; 1194fa9e4066Sahrens } 1195fa9e4066Sahrens 1196fa9e4066Sahrens if (lvl == 0) { 1197fa9e4066Sahrens dnode_phys_t *dnp = data; 1198fa9e4066Sahrens span = DNODE_SHIFT; 1199fa9e4066Sahrens ASSERT(dn->dn_type == DMU_OT_DNODE); 1200fa9e4066Sahrens 1201fa9e4066Sahrens for (i = (*offset >> span) & (blkfill - 1); i < blkfill; i++) { 1202fa9e4066Sahrens if (!dnp[i].dn_type == hole) 1203fa9e4066Sahrens break; 1204fa9e4066Sahrens *offset += 1ULL << span; 1205fa9e4066Sahrens } 1206fa9e4066Sahrens if (i == blkfill) 1207fa9e4066Sahrens error = ESRCH; 1208fa9e4066Sahrens } else { 1209fa9e4066Sahrens blkptr_t *bp = data; 1210fa9e4066Sahrens span = (lvl - 1) * epbs + dn->dn_datablkshift; 1211fa9e4066Sahrens minfill = 0; 1212fa9e4066Sahrens maxfill = blkfill << ((lvl - 1) * epbs); 1213fa9e4066Sahrens 1214fa9e4066Sahrens if (hole) 1215fa9e4066Sahrens maxfill--; 1216fa9e4066Sahrens else 1217fa9e4066Sahrens minfill++; 1218fa9e4066Sahrens 1219fa9e4066Sahrens for (i = (*offset >> span) & ((1ULL << epbs) - 1); 1220fa9e4066Sahrens i < epb; i++) { 1221fa9e4066Sahrens if (bp[i].blk_fill >= minfill && 1222fa9e4066Sahrens bp[i].blk_fill <= maxfill) 1223fa9e4066Sahrens break; 1224fa9e4066Sahrens *offset += 1ULL << span; 1225fa9e4066Sahrens } 1226fa9e4066Sahrens if (i >= epb) 1227fa9e4066Sahrens error = ESRCH; 1228fa9e4066Sahrens } 1229fa9e4066Sahrens 1230fa9e4066Sahrens if (db) 1231fa9e4066Sahrens dbuf_remove_ref(db, FTAG); 1232fa9e4066Sahrens 1233fa9e4066Sahrens return (error); 1234fa9e4066Sahrens } 1235fa9e4066Sahrens 1236fa9e4066Sahrens /* 1237fa9e4066Sahrens * Find the next hole, data, or sparse region at or after *offset. 1238fa9e4066Sahrens * The value 'blkfill' tells us how many items we expect to find 1239fa9e4066Sahrens * in an L0 data block; this value is 1 for normal objects, 1240fa9e4066Sahrens * DNODES_PER_BLOCK for the meta dnode, and some fraction of 1241fa9e4066Sahrens * DNODES_PER_BLOCK when searching for sparse regions thereof. 1242fa9e4066Sahrens * Examples: 1243fa9e4066Sahrens * 1244fa9e4066Sahrens * dnode_next_offset(dn, hole, offset, 1, 1); 1245fa9e4066Sahrens * Finds the next hole/data in a file. 1246fa9e4066Sahrens * Used in dmu_offset_next(). 1247fa9e4066Sahrens * 1248fa9e4066Sahrens * dnode_next_offset(mdn, hole, offset, 0, DNODES_PER_BLOCK); 1249fa9e4066Sahrens * Finds the next free/allocated dnode an objset's meta-dnode. 1250fa9e4066Sahrens * Used in dmu_object_next(). 1251fa9e4066Sahrens * 1252fa9e4066Sahrens * dnode_next_offset(mdn, TRUE, offset, 2, DNODES_PER_BLOCK >> 2); 1253fa9e4066Sahrens * Finds the next L2 meta-dnode bp that's at most 1/4 full. 1254fa9e4066Sahrens * Used in dmu_object_alloc(). 1255fa9e4066Sahrens */ 1256fa9e4066Sahrens int 1257fa9e4066Sahrens dnode_next_offset(dnode_t *dn, boolean_t hole, uint64_t *offset, 1258fa9e4066Sahrens int minlvl, uint64_t blkfill) 1259fa9e4066Sahrens { 1260fa9e4066Sahrens int lvl, maxlvl; 1261fa9e4066Sahrens int error = 0; 1262fa9e4066Sahrens uint64_t initial_offset = *offset; 1263fa9e4066Sahrens 1264fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 1265fa9e4066Sahrens 1266fa9e4066Sahrens if (dn->dn_phys->dn_nlevels == 0) { 1267fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1268fa9e4066Sahrens return (ESRCH); 1269fa9e4066Sahrens } 1270fa9e4066Sahrens 1271fa9e4066Sahrens if (dn->dn_datablkshift == 0) { 1272fa9e4066Sahrens if (*offset < dn->dn_datablksz) { 1273fa9e4066Sahrens if (hole) 1274fa9e4066Sahrens *offset = dn->dn_datablksz; 1275fa9e4066Sahrens } else { 1276fa9e4066Sahrens error = ESRCH; 1277fa9e4066Sahrens } 1278fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1279fa9e4066Sahrens return (error); 1280fa9e4066Sahrens } 1281fa9e4066Sahrens 1282fa9e4066Sahrens maxlvl = dn->dn_phys->dn_nlevels; 1283fa9e4066Sahrens 1284fa9e4066Sahrens for (lvl = minlvl; lvl <= maxlvl; lvl++) { 1285fa9e4066Sahrens error = dnode_next_offset_level(dn, hole, offset, lvl, blkfill); 1286fa9e4066Sahrens if (error == 0) 1287fa9e4066Sahrens break; 1288fa9e4066Sahrens } 1289fa9e4066Sahrens 1290fa9e4066Sahrens while (--lvl >= minlvl && error == 0) 1291fa9e4066Sahrens error = dnode_next_offset_level(dn, hole, offset, lvl, blkfill); 1292fa9e4066Sahrens 1293fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1294fa9e4066Sahrens 1295fa9e4066Sahrens if (initial_offset > *offset) 1296fa9e4066Sahrens return (ESRCH); 1297fa9e4066Sahrens 1298fa9e4066Sahrens return (error); 1299fa9e4066Sahrens } 1300