1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 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 <sys/param.h> 32 #include <assert.h> 33 #include <stdlib.h> 34 #include <string.h> 35 36 #include <util.h> 37 38 #include "zfs.h" 39 40 #define DNODES_PER_CHUNK (MAXBLOCKSIZE / sizeof(dnode_phys_t)) 41 42 struct objset_dnode_chunk { 43 dnode_phys_t buf[DNODES_PER_CHUNK]; 44 unsigned int nextfree; 45 STAILQ_ENTRY(objset_dnode_chunk) next; 46 }; 47 48 typedef struct zfs_objset { 49 /* Physical object set. */ 50 objset_phys_t *phys; 51 off_t osloc; 52 off_t osblksz; 53 blkptr_t osbp; /* set in objset_write() */ 54 55 /* Accounting. */ 56 off_t space; /* bytes allocated to this objset */ 57 58 /* dnode allocator. */ 59 uint64_t dnodecount; 60 STAILQ_HEAD(, objset_dnode_chunk) dnodechunks; 61 } zfs_objset_t; 62 63 static void 64 dnode_init(dnode_phys_t *dnode, uint8_t type, uint8_t bonustype, 65 uint16_t bonuslen) 66 { 67 dnode->dn_indblkshift = MAXBLOCKSHIFT; 68 dnode->dn_type = type; 69 dnode->dn_bonustype = bonustype; 70 dnode->dn_bonuslen = bonuslen; 71 dnode->dn_checksum = ZIO_CHECKSUM_FLETCHER_4; 72 dnode->dn_nlevels = 1; 73 dnode->dn_nblkptr = 1; 74 dnode->dn_flags = DNODE_FLAG_USED_BYTES; 75 } 76 77 zfs_objset_t * 78 objset_alloc(zfs_opt_t *zfs, uint64_t type) 79 { 80 struct objset_dnode_chunk *chunk; 81 zfs_objset_t *os; 82 83 os = ecalloc(1, sizeof(*os)); 84 os->osblksz = sizeof(objset_phys_t); 85 os->osloc = objset_space_alloc(zfs, os, &os->osblksz); 86 87 /* 88 * Object ID zero is always reserved for the meta dnode, which is 89 * embedded in the objset itself. 90 */ 91 STAILQ_INIT(&os->dnodechunks); 92 chunk = ecalloc(1, sizeof(*chunk)); 93 chunk->nextfree = 1; 94 STAILQ_INSERT_HEAD(&os->dnodechunks, chunk, next); 95 os->dnodecount = 1; 96 97 os->phys = ecalloc(1, os->osblksz); 98 os->phys->os_type = type; 99 100 dnode_init(&os->phys->os_meta_dnode, DMU_OT_DNODE, DMU_OT_NONE, 0); 101 os->phys->os_meta_dnode.dn_datablkszsec = 102 DNODE_BLOCK_SIZE >> MINBLOCKSHIFT; 103 104 return (os); 105 } 106 107 /* 108 * Write the dnode array and physical object set to disk. 109 */ 110 static void 111 _objset_write(zfs_opt_t *zfs, zfs_objset_t *os, struct dnode_cursor *c, 112 off_t loc) 113 { 114 struct objset_dnode_chunk *chunk, *tmp; 115 unsigned int total; 116 117 /* 118 * Write out the dnode array, i.e., the meta-dnode. For some reason its 119 * data blocks must be 16KB in size no matter how large the array is. 120 */ 121 total = 0; 122 STAILQ_FOREACH_SAFE(chunk, &os->dnodechunks, next, tmp) { 123 unsigned int i; 124 125 assert(chunk->nextfree > 0); 126 assert(chunk->nextfree <= os->dnodecount); 127 assert(chunk->nextfree <= DNODES_PER_CHUNK); 128 129 for (i = 0; i < chunk->nextfree; i += DNODES_PER_BLOCK) { 130 blkptr_t *bp; 131 uint64_t fill; 132 133 if (chunk->nextfree - i < DNODES_PER_BLOCK) 134 fill = DNODES_PER_BLOCK - (chunk->nextfree - i); 135 else 136 fill = 0; 137 bp = dnode_cursor_next(zfs, c, 138 (total + i) * sizeof(dnode_phys_t)); 139 vdev_pwrite_dnode_indir(zfs, &os->phys->os_meta_dnode, 140 0, fill, chunk->buf + i, DNODE_BLOCK_SIZE, loc, bp); 141 loc += DNODE_BLOCK_SIZE; 142 } 143 total += i; 144 145 free(chunk); 146 } 147 dnode_cursor_finish(zfs, c); 148 STAILQ_INIT(&os->dnodechunks); 149 150 /* 151 * Write the object set itself. The saved block pointer will be copied 152 * into the referencing DSL dataset or the uberblocks. 153 */ 154 vdev_pwrite_data(zfs, DMU_OT_OBJSET, ZIO_CHECKSUM_FLETCHER_4, 0, 155 os->dnodecount - 1, os->phys, os->osblksz, os->osloc, &os->osbp); 156 } 157 158 void 159 objset_write(zfs_opt_t *zfs, zfs_objset_t *os) 160 { 161 struct dnode_cursor *c; 162 off_t dnodeloc, dnodesz; 163 uint64_t dnodecount; 164 165 /* 166 * There is a chicken-and-egg problem here when writing the MOS: we 167 * cannot write space maps before we're finished allocating space from 168 * the vdev, and we can't write the MOS without having allocated space 169 * for indirect dnode blocks. Thus, rather than lazily allocating 170 * indirect blocks for the meta-dnode (which would be simpler), they are 171 * allocated up-front and before writing space maps. 172 */ 173 dnodecount = os->dnodecount; 174 if (os == zfs->mos) 175 dnodecount += zfs->mscount; 176 dnodesz = dnodecount * sizeof(dnode_phys_t); 177 c = dnode_cursor_init(zfs, os, &os->phys->os_meta_dnode, dnodesz, 178 DNODE_BLOCK_SIZE); 179 dnodesz = roundup2(dnodesz, DNODE_BLOCK_SIZE); 180 dnodeloc = objset_space_alloc(zfs, os, &dnodesz); 181 182 if (os == zfs->mos) { 183 vdev_spacemap_write(zfs); 184 185 /* 186 * We've finished allocating space, account for it in $MOS and 187 * in the parent directory. 188 */ 189 dsl_dir_root_finalize(zfs, os->space); 190 } 191 _objset_write(zfs, os, c, dnodeloc); 192 } 193 194 dnode_phys_t * 195 objset_dnode_bonus_alloc(zfs_objset_t *os, uint8_t type, uint8_t bonustype, 196 uint16_t bonuslen, uint64_t *idp) 197 { 198 struct objset_dnode_chunk *chunk; 199 dnode_phys_t *dnode; 200 201 assert(bonuslen <= DN_OLD_MAX_BONUSLEN); 202 assert(!STAILQ_EMPTY(&os->dnodechunks)); 203 204 chunk = STAILQ_LAST(&os->dnodechunks, objset_dnode_chunk, next); 205 if (chunk->nextfree == DNODES_PER_CHUNK) { 206 chunk = ecalloc(1, sizeof(*chunk)); 207 STAILQ_INSERT_TAIL(&os->dnodechunks, chunk, next); 208 } 209 *idp = os->dnodecount++; 210 dnode = &chunk->buf[chunk->nextfree++]; 211 dnode_init(dnode, type, bonustype, bonuslen); 212 dnode->dn_datablkszsec = os->osblksz >> MINBLOCKSHIFT; 213 return (dnode); 214 } 215 216 dnode_phys_t * 217 objset_dnode_alloc(zfs_objset_t *os, uint8_t type, uint64_t *idp) 218 { 219 return (objset_dnode_bonus_alloc(os, type, DMU_OT_NONE, 0, idp)); 220 } 221 222 /* 223 * Look up a physical dnode by ID. This is not used often so a linear search is 224 * fine. 225 */ 226 dnode_phys_t * 227 objset_dnode_lookup(zfs_objset_t *os, uint64_t id) 228 { 229 struct objset_dnode_chunk *chunk; 230 231 assert(id > 0); 232 assert(id < os->dnodecount); 233 234 STAILQ_FOREACH(chunk, &os->dnodechunks, next) { 235 if (id < DNODES_PER_CHUNK) 236 return (&chunk->buf[id]); 237 id -= DNODES_PER_CHUNK; 238 } 239 assert(0); 240 return (NULL); 241 } 242 243 off_t 244 objset_space_alloc(zfs_opt_t *zfs, zfs_objset_t *os, off_t *lenp) 245 { 246 off_t loc; 247 248 loc = vdev_space_alloc(zfs, lenp); 249 os->space += *lenp; 250 return (loc); 251 } 252 253 uint64_t 254 objset_space(const zfs_objset_t *os) 255 { 256 return (os->space); 257 } 258 259 void 260 objset_root_blkptr_copy(const zfs_objset_t *os, blkptr_t *bp) 261 { 262 memcpy(bp, &os->osbp, sizeof(blkptr_t)); 263 } 264