1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2022 The FreeBSD Foundation 5 * 6 * This software was developed by Mark Johnston under sponsorship from 7 * the FreeBSD Foundation. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions are 11 * met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 #include <assert.h> 32 #include <string.h> 33 34 #include <util.h> 35 36 #include "zfs.h" 37 38 #define DNODES_PER_CHUNK (MAXBLOCKSIZE / sizeof(dnode_phys_t)) 39 40 struct objset_dnode_chunk { 41 dnode_phys_t buf[DNODES_PER_CHUNK]; 42 unsigned int nextfree; 43 STAILQ_ENTRY(objset_dnode_chunk) next; 44 }; 45 46 typedef struct zfs_objset { 47 /* Physical object set. */ 48 objset_phys_t *phys; 49 off_t osloc; 50 off_t osblksz; 51 blkptr_t osbp; /* set in objset_write() */ 52 53 /* Accounting. */ 54 off_t space; /* bytes allocated to this objset */ 55 56 /* dnode allocator. */ 57 uint64_t dnodecount; 58 STAILQ_HEAD(, objset_dnode_chunk) dnodechunks; 59 } zfs_objset_t; 60 61 static void 62 dnode_init(dnode_phys_t *dnode, uint8_t type, uint8_t bonustype, 63 uint16_t bonuslen) 64 { 65 dnode->dn_indblkshift = MAXBLOCKSHIFT; 66 dnode->dn_type = type; 67 dnode->dn_bonustype = bonustype; 68 dnode->dn_bonuslen = bonuslen; 69 dnode->dn_checksum = ZIO_CHECKSUM_FLETCHER_4; 70 dnode->dn_nlevels = 1; 71 dnode->dn_nblkptr = 1; 72 dnode->dn_flags = DNODE_FLAG_USED_BYTES; 73 } 74 75 zfs_objset_t * 76 objset_alloc(zfs_opt_t *zfs, uint64_t type) 77 { 78 struct objset_dnode_chunk *chunk; 79 zfs_objset_t *os; 80 81 os = ecalloc(1, sizeof(*os)); 82 os->osblksz = sizeof(objset_phys_t); 83 os->osloc = objset_space_alloc(zfs, os, &os->osblksz); 84 85 /* 86 * Object ID zero is always reserved for the meta dnode, which is 87 * embedded in the objset itself. 88 */ 89 STAILQ_INIT(&os->dnodechunks); 90 chunk = ecalloc(1, sizeof(*chunk)); 91 chunk->nextfree = 1; 92 STAILQ_INSERT_HEAD(&os->dnodechunks, chunk, next); 93 os->dnodecount = 1; 94 95 os->phys = ecalloc(1, os->osblksz); 96 os->phys->os_type = type; 97 98 dnode_init(&os->phys->os_meta_dnode, DMU_OT_DNODE, DMU_OT_NONE, 0); 99 os->phys->os_meta_dnode.dn_datablkszsec = 100 DNODE_BLOCK_SIZE >> MINBLOCKSHIFT; 101 102 return (os); 103 } 104 105 /* 106 * Write the dnode array and physical object set to disk. 107 */ 108 static void 109 _objset_write(zfs_opt_t *zfs, zfs_objset_t *os, struct dnode_cursor *c, 110 off_t loc) 111 { 112 struct objset_dnode_chunk *chunk, *tmp; 113 unsigned int total; 114 115 /* 116 * Write out the dnode array, i.e., the meta-dnode. For some reason its 117 * data blocks must be 16KB in size no matter how large the array is. 118 */ 119 total = 0; 120 STAILQ_FOREACH_SAFE(chunk, &os->dnodechunks, next, tmp) { 121 unsigned int i; 122 123 assert(chunk->nextfree <= os->dnodecount); 124 assert(chunk->nextfree <= DNODES_PER_CHUNK); 125 126 for (i = 0; i < chunk->nextfree; i += DNODES_PER_BLOCK) { 127 blkptr_t *bp; 128 uint64_t fill; 129 130 if (chunk->nextfree - i < DNODES_PER_BLOCK) 131 fill = DNODES_PER_BLOCK - (chunk->nextfree - i); 132 else 133 fill = 0; 134 bp = dnode_cursor_next(zfs, c, 135 (total + i) * sizeof(dnode_phys_t)); 136 vdev_pwrite_dnode_indir(zfs, &os->phys->os_meta_dnode, 137 0, fill, chunk->buf + i, DNODE_BLOCK_SIZE, loc, bp); 138 loc += DNODE_BLOCK_SIZE; 139 } 140 total += i; 141 142 free(chunk); 143 } 144 dnode_cursor_finish(zfs, c); 145 STAILQ_INIT(&os->dnodechunks); 146 147 /* 148 * Write the object set itself. The saved block pointer will be copied 149 * into the referencing DSL dataset or the uberblocks. 150 */ 151 vdev_pwrite_data(zfs, DMU_OT_OBJSET, ZIO_CHECKSUM_FLETCHER_4, 0, 1, 152 os->phys, os->osblksz, os->osloc, &os->osbp); 153 } 154 155 void 156 objset_write(zfs_opt_t *zfs, zfs_objset_t *os) 157 { 158 struct dnode_cursor *c; 159 off_t dnodeloc, dnodesz; 160 uint64_t dnodecount; 161 162 /* 163 * There is a chicken-and-egg problem here when writing the MOS: we 164 * cannot write space maps before we're finished allocating space from 165 * the vdev, and we can't write the MOS without having allocated space 166 * for indirect dnode blocks. Thus, rather than lazily allocating 167 * indirect blocks for the meta-dnode (which would be simpler), they are 168 * allocated up-front and before writing space maps. 169 */ 170 dnodecount = os->dnodecount; 171 if (os == zfs->mos) 172 dnodecount += zfs->mscount; 173 dnodesz = dnodecount * sizeof(dnode_phys_t); 174 c = dnode_cursor_init(zfs, os, &os->phys->os_meta_dnode, dnodesz, 175 DNODE_BLOCK_SIZE); 176 dnodesz = roundup2(dnodesz, DNODE_BLOCK_SIZE); 177 dnodeloc = objset_space_alloc(zfs, os, &dnodesz); 178 179 if (os == zfs->mos) { 180 vdev_spacemap_write(zfs); 181 182 /* 183 * We've finished allocating space, account for it in $MOS. 184 */ 185 dsl_dir_size_set(zfs->mosdsldir, os->space); 186 } 187 _objset_write(zfs, os, c, dnodeloc); 188 } 189 190 dnode_phys_t * 191 objset_dnode_bonus_alloc(zfs_objset_t *os, uint8_t type, uint8_t bonustype, 192 uint16_t bonuslen, uint64_t *idp) 193 { 194 struct objset_dnode_chunk *chunk; 195 dnode_phys_t *dnode; 196 197 assert(bonuslen <= DN_OLD_MAX_BONUSLEN); 198 assert(!STAILQ_EMPTY(&os->dnodechunks)); 199 200 chunk = STAILQ_LAST(&os->dnodechunks, objset_dnode_chunk, next); 201 if (chunk->nextfree == DNODES_PER_CHUNK) { 202 chunk = ecalloc(1, sizeof(*chunk)); 203 STAILQ_INSERT_TAIL(&os->dnodechunks, chunk, next); 204 } 205 *idp = os->dnodecount++; 206 dnode = &chunk->buf[chunk->nextfree++]; 207 dnode_init(dnode, type, bonustype, bonuslen); 208 dnode->dn_datablkszsec = os->osblksz >> MINBLOCKSHIFT; 209 return (dnode); 210 } 211 212 dnode_phys_t * 213 objset_dnode_alloc(zfs_objset_t *os, uint8_t type, uint64_t *idp) 214 { 215 return (objset_dnode_bonus_alloc(os, type, DMU_OT_NONE, 0, idp)); 216 } 217 218 /* 219 * Look up a physical dnode by ID. This is not used often so a linear search is 220 * fine. 221 */ 222 dnode_phys_t * 223 objset_dnode_lookup(zfs_objset_t *os, uint64_t id) 224 { 225 struct objset_dnode_chunk *chunk; 226 227 assert(id > 0); 228 assert(id < os->dnodecount); 229 230 STAILQ_FOREACH(chunk, &os->dnodechunks, next) { 231 if (id < DNODES_PER_CHUNK) 232 return (&chunk->buf[id]); 233 id -= DNODES_PER_CHUNK; 234 } 235 assert(0); 236 return (NULL); 237 } 238 239 off_t 240 objset_space_alloc(zfs_opt_t *zfs, zfs_objset_t *os, off_t *lenp) 241 { 242 off_t loc; 243 244 loc = vdev_space_alloc(zfs, lenp); 245 os->space += *lenp; 246 return (loc); 247 } 248 249 uint64_t 250 objset_space(const zfs_objset_t *os) 251 { 252 return (os->space); 253 } 254 255 void 256 objset_root_blkptr_copy(const zfs_objset_t *os, blkptr_t *bp) 257 { 258 memcpy(bp, &os->osbp, sizeof(blkptr_t)); 259 } 260