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 1544 eschrock * Common Development and Distribution License (the "License"). 6 1544 eschrock * 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 8525 Eric * 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 #ifndef _SYS_SPA_H 27 789 ahrens #define _SYS_SPA_H 28 789 ahrens 29 789 ahrens #include <sys/avl.h> 30 789 ahrens #include <sys/zfs_context.h> 31 789 ahrens #include <sys/nvpair.h> 32 789 ahrens #include <sys/sysmacros.h> 33 789 ahrens #include <sys/types.h> 34 789 ahrens #include <sys/fs/zfs.h> 35 789 ahrens 36 789 ahrens #ifdef __cplusplus 37 789 ahrens extern "C" { 38 789 ahrens #endif 39 789 ahrens 40 789 ahrens /* 41 789 ahrens * Forward references that lots of things need. 42 789 ahrens */ 43 789 ahrens typedef struct spa spa_t; 44 789 ahrens typedef struct vdev vdev_t; 45 789 ahrens typedef struct metaslab metaslab_t; 46 10922 Jeff typedef struct metaslab_group metaslab_group_t; 47 10922 Jeff typedef struct metaslab_class metaslab_class_t; 48 10922 Jeff typedef struct zio zio_t; 49 789 ahrens typedef struct zilog zilog_t; 50 5450 brendan typedef struct spa_aux_vdev spa_aux_vdev_t; 51 10922 Jeff typedef struct ddt ddt_t; 52 10922 Jeff typedef struct ddt_entry ddt_entry_t; 53 789 ahrens struct dsl_pool; 54 789 ahrens 55 789 ahrens /* 56 789 ahrens * General-purpose 32-bit and 64-bit bitfield encodings. 57 789 ahrens */ 58 789 ahrens #define BF32_DECODE(x, low, len) P2PHASE((x) >> (low), 1U << (len)) 59 789 ahrens #define BF64_DECODE(x, low, len) P2PHASE((x) >> (low), 1ULL << (len)) 60 789 ahrens #define BF32_ENCODE(x, low, len) (P2PHASE((x), 1U << (len)) << (low)) 61 789 ahrens #define BF64_ENCODE(x, low, len) (P2PHASE((x), 1ULL << (len)) << (low)) 62 789 ahrens 63 789 ahrens #define BF32_GET(x, low, len) BF32_DECODE(x, low, len) 64 789 ahrens #define BF64_GET(x, low, len) BF64_DECODE(x, low, len) 65 789 ahrens 66 789 ahrens #define BF32_SET(x, low, len, val) \ 67 2856 nd150628 ((x) ^= BF32_ENCODE((x >> low) ^ (val), low, len)) 68 789 ahrens #define BF64_SET(x, low, len, val) \ 69 2856 nd150628 ((x) ^= BF64_ENCODE((x >> low) ^ (val), low, len)) 70 789 ahrens 71 789 ahrens #define BF32_GET_SB(x, low, len, shift, bias) \ 72 789 ahrens ((BF32_GET(x, low, len) + (bias)) << (shift)) 73 789 ahrens #define BF64_GET_SB(x, low, len, shift, bias) \ 74 789 ahrens ((BF64_GET(x, low, len) + (bias)) << (shift)) 75 789 ahrens 76 789 ahrens #define BF32_SET_SB(x, low, len, shift, bias, val) \ 77 789 ahrens BF32_SET(x, low, len, ((val) >> (shift)) - (bias)) 78 789 ahrens #define BF64_SET_SB(x, low, len, shift, bias, val) \ 79 789 ahrens BF64_SET(x, low, len, ((val) >> (shift)) - (bias)) 80 789 ahrens 81 789 ahrens /* 82 789 ahrens * We currently support nine block sizes, from 512 bytes to 128K. 83 789 ahrens * We could go higher, but the benefits are near-zero and the cost 84 789 ahrens * of COWing a giant block to modify one byte would become excessive. 85 789 ahrens */ 86 789 ahrens #define SPA_MINBLOCKSHIFT 9 87 789 ahrens #define SPA_MAXBLOCKSHIFT 17 88 789 ahrens #define SPA_MINBLOCKSIZE (1ULL << SPA_MINBLOCKSHIFT) 89 789 ahrens #define SPA_MAXBLOCKSIZE (1ULL << SPA_MAXBLOCKSHIFT) 90 789 ahrens 91 789 ahrens #define SPA_BLOCKSIZES (SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1) 92 7497 Tim 93 7497 Tim /* 94 7497 Tim * Size of block to hold the configuration data (a packed nvlist) 95 7497 Tim */ 96 7497 Tim #define SPA_CONFIG_BLOCKSIZE (1 << 14) 97 789 ahrens 98 789 ahrens /* 99 789 ahrens * The DVA size encodings for LSIZE and PSIZE support blocks up to 32MB. 100 789 ahrens * The ASIZE encoding should be at least 64 times larger (6 more bits) 101 789 ahrens * to support up to 4-way RAID-Z mirror mode with worst-case gang block 102 789 ahrens * overhead, three DVAs per bp, plus one more bit in case we do anything 103 789 ahrens * else that expands the ASIZE. 104 789 ahrens */ 105 789 ahrens #define SPA_LSIZEBITS 16 /* LSIZE up to 32M (2^16 * 512) */ 106 789 ahrens #define SPA_PSIZEBITS 16 /* PSIZE up to 32M (2^16 * 512) */ 107 789 ahrens #define SPA_ASIZEBITS 24 /* ASIZE up to 64 times larger */ 108 789 ahrens 109 789 ahrens /* 110 789 ahrens * All SPA data is represented by 128-bit data virtual addresses (DVAs). 111 789 ahrens * The members of the dva_t should be considered opaque outside the SPA. 112 789 ahrens */ 113 789 ahrens typedef struct dva { 114 789 ahrens uint64_t dva_word[2]; 115 789 ahrens } dva_t; 116 789 ahrens 117 789 ahrens /* 118 789 ahrens * Each block has a 256-bit checksum -- strong enough for cryptographic hashes. 119 789 ahrens */ 120 789 ahrens typedef struct zio_cksum { 121 789 ahrens uint64_t zc_word[4]; 122 789 ahrens } zio_cksum_t; 123 789 ahrens 124 789 ahrens /* 125 789 ahrens * Each block is described by its DVAs, time of birth, checksum, etc. 126 789 ahrens * The word-by-word, bit-by-bit layout of the blkptr is as follows: 127 789 ahrens * 128 789 ahrens * 64 56 48 40 32 24 16 8 0 129 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 130 789 ahrens * 0 | vdev1 | GRID | ASIZE | 131 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 132 789 ahrens * 1 |G| offset1 | 133 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 134 789 ahrens * 2 | vdev2 | GRID | ASIZE | 135 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 136 789 ahrens * 3 |G| offset2 | 137 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 138 789 ahrens * 4 | vdev3 | GRID | ASIZE | 139 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 140 789 ahrens * 5 |G| offset3 | 141 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 142 10922 Jeff * 6 |BDX|lvl| type | cksum | comp | PSIZE | LSIZE | 143 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 144 789 ahrens * 7 | padding | 145 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 146 789 ahrens * 8 | padding | 147 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 148 10922 Jeff * 9 | physical birth txg | 149 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 150 10922 Jeff * a | logical birth txg | 151 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 152 789 ahrens * b | fill count | 153 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 154 789 ahrens * c | checksum[0] | 155 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 156 789 ahrens * d | checksum[1] | 157 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 158 789 ahrens * e | checksum[2] | 159 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 160 789 ahrens * f | checksum[3] | 161 789 ahrens * +-------+-------+-------+-------+-------+-------+-------+-------+ 162 789 ahrens * 163 789 ahrens * Legend: 164 789 ahrens * 165 789 ahrens * vdev virtual device ID 166 789 ahrens * offset offset into virtual device 167 789 ahrens * LSIZE logical size 168 789 ahrens * PSIZE physical size (after compression) 169 789 ahrens * ASIZE allocated size (including RAID-Z parity and gang block headers) 170 789 ahrens * GRID RAID-Z layout information (reserved for future use) 171 789 ahrens * cksum checksum function 172 789 ahrens * comp compression function 173 789 ahrens * G gang block indicator 174 10922 Jeff * B byteorder (endianness) 175 10922 Jeff * D dedup 176 10922 Jeff * X unused 177 10922 Jeff * lvl level of indirection 178 789 ahrens * type DMU object type 179 10922 Jeff * phys birth txg of block allocation; zero if same as logical birth txg 180 10922 Jeff * log. birth transaction group in which the block was logically born 181 789 ahrens * fill count number of non-zero blocks under this bp 182 789 ahrens * checksum[4] 256-bit checksum of the data this bp describes 183 789 ahrens */ 184 789 ahrens #define SPA_BLKPTRSHIFT 7 /* blkptr_t is 128 bytes */ 185 789 ahrens #define SPA_DVAS_PER_BP 3 /* Number of DVAs in a bp */ 186 10922 Jeff 187 10922 Jeff typedef struct blkptr { 188 10922 Jeff dva_t blk_dva[SPA_DVAS_PER_BP]; /* Data Virtual Addresses */ 189 10922 Jeff uint64_t blk_prop; /* size, compression, type, etc */ 190 10922 Jeff uint64_t blk_pad[2]; /* Extra space for the future */ 191 10922 Jeff uint64_t blk_phys_birth; /* txg when block was allocated */ 192 10922 Jeff uint64_t blk_birth; /* transaction group at birth */ 193 10922 Jeff uint64_t blk_fill; /* fill count */ 194 10922 Jeff zio_cksum_t blk_cksum; /* 256-bit checksum */ 195 10922 Jeff } blkptr_t; 196 789 ahrens 197 789 ahrens /* 198 789 ahrens * Macros to get and set fields in a bp or DVA. 199 789 ahrens */ 200 789 ahrens #define DVA_GET_ASIZE(dva) \ 201 789 ahrens BF64_GET_SB((dva)->dva_word[0], 0, 24, SPA_MINBLOCKSHIFT, 0) 202 789 ahrens #define DVA_SET_ASIZE(dva, x) \ 203 789 ahrens BF64_SET_SB((dva)->dva_word[0], 0, 24, SPA_MINBLOCKSHIFT, 0, x) 204 789 ahrens 205 789 ahrens #define DVA_GET_GRID(dva) BF64_GET((dva)->dva_word[0], 24, 8) 206 789 ahrens #define DVA_SET_GRID(dva, x) BF64_SET((dva)->dva_word[0], 24, 8, x) 207 789 ahrens 208 789 ahrens #define DVA_GET_VDEV(dva) BF64_GET((dva)->dva_word[0], 32, 32) 209 789 ahrens #define DVA_SET_VDEV(dva, x) BF64_SET((dva)->dva_word[0], 32, 32, x) 210 789 ahrens 211 789 ahrens #define DVA_GET_OFFSET(dva) \ 212 789 ahrens BF64_GET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0) 213 789 ahrens #define DVA_SET_OFFSET(dva, x) \ 214 789 ahrens BF64_SET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0, x) 215 789 ahrens 216 789 ahrens #define DVA_GET_GANG(dva) BF64_GET((dva)->dva_word[1], 63, 1) 217 789 ahrens #define DVA_SET_GANG(dva, x) BF64_SET((dva)->dva_word[1], 63, 1, x) 218 789 ahrens 219 789 ahrens #define BP_GET_LSIZE(bp) \ 220 10800 Neil BF64_GET_SB((bp)->blk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1) 221 789 ahrens #define BP_SET_LSIZE(bp, x) \ 222 789 ahrens BF64_SET_SB((bp)->blk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1, x) 223 789 ahrens 224 789 ahrens #define BP_GET_PSIZE(bp) \ 225 789 ahrens BF64_GET_SB((bp)->blk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1) 226 789 ahrens #define BP_SET_PSIZE(bp, x) \ 227 789 ahrens BF64_SET_SB((bp)->blk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1, x) 228 789 ahrens 229 10922 Jeff #define BP_GET_COMPRESS(bp) BF64_GET((bp)->blk_prop, 32, 8) 230 10922 Jeff #define BP_SET_COMPRESS(bp, x) BF64_SET((bp)->blk_prop, 32, 8, x) 231 789 ahrens 232 10922 Jeff #define BP_GET_CHECKSUM(bp) BF64_GET((bp)->blk_prop, 40, 8) 233 10922 Jeff #define BP_SET_CHECKSUM(bp, x) BF64_SET((bp)->blk_prop, 40, 8, x) 234 789 ahrens 235 10922 Jeff #define BP_GET_TYPE(bp) BF64_GET((bp)->blk_prop, 48, 8) 236 10922 Jeff #define BP_SET_TYPE(bp, x) BF64_SET((bp)->blk_prop, 48, 8, x) 237 789 ahrens 238 10922 Jeff #define BP_GET_LEVEL(bp) BF64_GET((bp)->blk_prop, 56, 5) 239 10922 Jeff #define BP_SET_LEVEL(bp, x) BF64_SET((bp)->blk_prop, 56, 5, x) 240 789 ahrens 241 10922 Jeff #define BP_GET_PROP_BIT_61(bp) BF64_GET((bp)->blk_prop, 61, 1) 242 10922 Jeff #define BP_SET_PROP_BIT_61(bp, x) BF64_SET((bp)->blk_prop, 61, 1, x) 243 10922 Jeff 244 10922 Jeff #define BP_GET_DEDUP(bp) BF64_GET((bp)->blk_prop, 62, 1) 245 10922 Jeff #define BP_SET_DEDUP(bp, x) BF64_SET((bp)->blk_prop, 62, 1, x) 246 10922 Jeff 247 10922 Jeff #define BP_GET_BYTEORDER(bp) (0 - BF64_GET((bp)->blk_prop, 63, 1)) 248 10922 Jeff #define BP_SET_BYTEORDER(bp, x) BF64_SET((bp)->blk_prop, 63, 1, x) 249 10922 Jeff 250 10922 Jeff #define BP_PHYSICAL_BIRTH(bp) \ 251 10922 Jeff ((bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth) 252 10922 Jeff 253 10922 Jeff #define BP_SET_BIRTH(bp, logical, physical) \ 254 10922 Jeff { \ 255 10922 Jeff (bp)->blk_birth = (logical); \ 256 10922 Jeff (bp)->blk_phys_birth = ((logical) == (physical) ? 0 : (physical)); \ 257 10922 Jeff } 258 789 ahrens 259 789 ahrens #define BP_GET_ASIZE(bp) \ 260 789 ahrens (DVA_GET_ASIZE(&(bp)->blk_dva[0]) + DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \ 261 2082 eschrock DVA_GET_ASIZE(&(bp)->blk_dva[2])) 262 2082 eschrock 263 2082 eschrock #define BP_GET_UCSIZE(bp) \ 264 2082 eschrock ((BP_GET_LEVEL(bp) > 0 || dmu_ot[BP_GET_TYPE(bp)].ot_metadata) ? \ 265 2082 eschrock BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp)); 266 789 ahrens 267 1775 billm #define BP_GET_NDVAS(bp) \ 268 1775 billm (!!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \ 269 1775 billm !!DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \ 270 1775 billm !!DVA_GET_ASIZE(&(bp)->blk_dva[2])) 271 1775 billm 272 1775 billm #define BP_COUNT_GANG(bp) \ 273 1775 billm (DVA_GET_GANG(&(bp)->blk_dva[0]) + \ 274 1775 billm DVA_GET_GANG(&(bp)->blk_dva[1]) + \ 275 1775 billm DVA_GET_GANG(&(bp)->blk_dva[2])) 276 1775 billm 277 789 ahrens #define DVA_EQUAL(dva1, dva2) \ 278 789 ahrens ((dva1)->dva_word[1] == (dva2)->dva_word[1] && \ 279 789 ahrens (dva1)->dva_word[0] == (dva2)->dva_word[0]) 280 789 ahrens 281 10922 Jeff #define BP_EQUAL(bp1, bp2) \ 282 10922 Jeff (BP_PHYSICAL_BIRTH(bp1) == BP_PHYSICAL_BIRTH(bp2) && \ 283 10922 Jeff DVA_EQUAL(&(bp1)->blk_dva[0], &(bp2)->blk_dva[0]) && \ 284 10922 Jeff DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) && \ 285 10922 Jeff DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2])) 286 10922 Jeff 287 3093 ahrens #define ZIO_CHECKSUM_EQUAL(zc1, zc2) \ 288 3093 ahrens (0 == (((zc1).zc_word[0] - (zc2).zc_word[0]) | \ 289 3093 ahrens ((zc1).zc_word[1] - (zc2).zc_word[1]) | \ 290 3093 ahrens ((zc1).zc_word[2] - (zc2).zc_word[2]) | \ 291 3093 ahrens ((zc1).zc_word[3] - (zc2).zc_word[3]))) 292 3093 ahrens 293 789 ahrens #define DVA_IS_VALID(dva) (DVA_GET_ASIZE(dva) != 0) 294 789 ahrens 295 789 ahrens #define ZIO_SET_CHECKSUM(zcp, w0, w1, w2, w3) \ 296 789 ahrens { \ 297 789 ahrens (zcp)->zc_word[0] = w0; \ 298 789 ahrens (zcp)->zc_word[1] = w1; \ 299 789 ahrens (zcp)->zc_word[2] = w2; \ 300 789 ahrens (zcp)->zc_word[3] = w3; \ 301 789 ahrens } 302 789 ahrens 303 1775 billm #define BP_IDENTITY(bp) (&(bp)->blk_dva[0]) 304 1775 billm #define BP_IS_GANG(bp) DVA_GET_GANG(BP_IDENTITY(bp)) 305 789 ahrens #define BP_IS_HOLE(bp) ((bp)->blk_birth == 0) 306 789 ahrens 307 7754 Jeff #define BP_ZERO(bp) \ 308 789 ahrens { \ 309 789 ahrens (bp)->blk_dva[0].dva_word[0] = 0; \ 310 789 ahrens (bp)->blk_dva[0].dva_word[1] = 0; \ 311 789 ahrens (bp)->blk_dva[1].dva_word[0] = 0; \ 312 789 ahrens (bp)->blk_dva[1].dva_word[1] = 0; \ 313 789 ahrens (bp)->blk_dva[2].dva_word[0] = 0; \ 314 789 ahrens (bp)->blk_dva[2].dva_word[1] = 0; \ 315 789 ahrens (bp)->blk_prop = 0; \ 316 789 ahrens (bp)->blk_pad[0] = 0; \ 317 789 ahrens (bp)->blk_pad[1] = 0; \ 318 10922 Jeff (bp)->blk_phys_birth = 0; \ 319 7754 Jeff (bp)->blk_birth = 0; \ 320 789 ahrens (bp)->blk_fill = 0; \ 321 789 ahrens ZIO_SET_CHECKSUM(&(bp)->blk_cksum, 0, 0, 0, 0); \ 322 789 ahrens } 323 789 ahrens 324 789 ahrens /* 325 789 ahrens * Note: the byteorder is either 0 or -1, both of which are palindromes. 326 789 ahrens * This simplifies the endianness handling a bit. 327 789 ahrens */ 328 789 ahrens #ifdef _BIG_ENDIAN 329 789 ahrens #define ZFS_HOST_BYTEORDER (0ULL) 330 789 ahrens #else 331 789 ahrens #define ZFS_HOST_BYTEORDER (-1ULL) 332 789 ahrens #endif 333 789 ahrens 334 789 ahrens #define BP_SHOULD_BYTESWAP(bp) (BP_GET_BYTEORDER(bp) != ZFS_HOST_BYTEORDER) 335 789 ahrens 336 1775 billm #define BP_SPRINTF_LEN 320 337 896 maybee 338 10922 Jeff /* 339 10922 Jeff * This macro allows code sharing between zfs, libzpool, and mdb. 340 10922 Jeff * 'func' is either snprintf() or mdb_snprintf(). 341 10922 Jeff * 'ws' (whitespace) can be ' ' for single-line format, '\n' for multi-line. 342 10922 Jeff */ 343 10922 Jeff #define SPRINTF_BLKPTR(func, ws, buf, bp, type, checksum, compress) \ 344 10922 Jeff { \ 345 10922 Jeff static const char *copyname[] = \ 346 10922 Jeff { "zero", "single", "double", "triple" }; \ 347 10922 Jeff int size = BP_SPRINTF_LEN; \ 348 10922 Jeff int len = 0; \ 349 10922 Jeff int copies = 0; \ 350 10922 Jeff \ 351 10922 Jeff if (bp == NULL) { \ 352 10922 Jeff len = func(buf + len, size - len, "<NULL>"); \ 353 10922 Jeff } else if (BP_IS_HOLE(bp)) { \ 354 10922 Jeff len = func(buf + len, size - len, "<hole>"); \ 355 10922 Jeff } else { \ 356 10922 Jeff for (int d = 0; d < BP_GET_NDVAS(bp); d++) { \ 357 10922 Jeff const dva_t *dva = &bp->blk_dva[d]; \ 358 10922 Jeff if (DVA_IS_VALID(dva)) \ 359 10922 Jeff copies++; \ 360 10922 Jeff len += func(buf + len, size - len, \ 361 10922 Jeff "DVA[%d]=<%llu:%llx:%llx>%c", d, \ 362 10922 Jeff (u_longlong_t)DVA_GET_VDEV(dva), \ 363 10922 Jeff (u_longlong_t)DVA_GET_OFFSET(dva), \ 364 10922 Jeff (u_longlong_t)DVA_GET_ASIZE(dva), \ 365 10922 Jeff ws); \ 366 10922 Jeff } \ 367 10922 Jeff if (BP_IS_GANG(bp) && \ 368 10922 Jeff DVA_GET_ASIZE(&bp->blk_dva[2]) <= \ 369 10922 Jeff DVA_GET_ASIZE(&bp->blk_dva[1]) / 2) \ 370 10922 Jeff copies--; \ 371 10922 Jeff len += func(buf + len, size - len, \ 372 10922 Jeff "[L%llu %s] %s %s %s %s %s %s%c" \ 373 10922 Jeff "size=%llxL/%llxP birth=%lluL/%lluP fill=%llu%c" \ 374 10922 Jeff "cksum=%llx:%llx:%llx:%llx", \ 375 10922 Jeff (u_longlong_t)BP_GET_LEVEL(bp), \ 376 10922 Jeff type, \ 377 10922 Jeff checksum, \ 378 10922 Jeff compress, \ 379 10922 Jeff BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE", \ 380 10922 Jeff BP_IS_GANG(bp) ? "gang" : "contiguous", \ 381 10922 Jeff BP_GET_DEDUP(bp) ? "dedup" : "unique", \ 382 10922 Jeff copyname[copies], \ 383 10922 Jeff ws, \ 384 10922 Jeff (u_longlong_t)BP_GET_LSIZE(bp), \ 385 10922 Jeff (u_longlong_t)BP_GET_PSIZE(bp), \ 386 10922 Jeff (u_longlong_t)bp->blk_birth, \ 387 10922 Jeff (u_longlong_t)BP_PHYSICAL_BIRTH(bp), \ 388 10922 Jeff (u_longlong_t)bp->blk_fill, \ 389 10922 Jeff ws, \ 390 10922 Jeff (u_longlong_t)bp->blk_cksum.zc_word[0], \ 391 10922 Jeff (u_longlong_t)bp->blk_cksum.zc_word[1], \ 392 10922 Jeff (u_longlong_t)bp->blk_cksum.zc_word[2], \ 393 10922 Jeff (u_longlong_t)bp->blk_cksum.zc_word[3]); \ 394 10922 Jeff } \ 395 10922 Jeff ASSERT(len < size); \ 396 10922 Jeff } 397 10922 Jeff 398 789 ahrens #include <sys/dmu.h> 399 789 ahrens 400 3290 johansen #define BP_GET_BUFC_TYPE(bp) \ 401 3290 johansen (((BP_GET_LEVEL(bp) > 0) || (dmu_ot[BP_GET_TYPE(bp)].ot_metadata)) ? \ 402 3290 johansen ARC_BUFC_METADATA : ARC_BUFC_DATA); 403 789 ahrens 404 789 ahrens /* state manipulation functions */ 405 789 ahrens extern int spa_open(const char *pool, spa_t **, void *tag); 406 10921 Tim extern int spa_open_rewind(const char *pool, spa_t **, void *tag, 407 10921 Tim nvlist_t *policy, nvlist_t **config); 408 1544 eschrock extern int spa_get_stats(const char *pool, nvlist_t **config, 409 1544 eschrock char *altroot, size_t buflen); 410 5094 lling extern int spa_create(const char *pool, nvlist_t *config, nvlist_t *props, 411 7184 timh const char *history_str, nvlist_t *zplprops); 412 7147 taylor extern int spa_import_rootpool(char *devpath, char *devid); 413 5094 lling extern int spa_import(const char *pool, nvlist_t *config, nvlist_t *props); 414 9425 Eric extern int spa_import_verbatim(const char *, nvlist_t *, nvlist_t *); 415 789 ahrens extern nvlist_t *spa_tryimport(nvlist_t *tryconfig); 416 789 ahrens extern int spa_destroy(char *pool); 417 8211 George extern int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 418 8211 George boolean_t hardforce); 419 1544 eschrock extern int spa_reset(char *pool); 420 1544 eschrock extern void spa_async_request(spa_t *spa, int flag); 421 7046 ahrens extern void spa_async_unrequest(spa_t *spa, int flag); 422 1544 eschrock extern void spa_async_suspend(spa_t *spa); 423 1544 eschrock extern void spa_async_resume(spa_t *spa); 424 1544 eschrock extern spa_t *spa_inject_addref(char *pool); 425 1544 eschrock extern void spa_inject_delref(spa_t *spa); 426 1544 eschrock 427 7754 Jeff #define SPA_ASYNC_CONFIG_UPDATE 0x01 428 7754 Jeff #define SPA_ASYNC_REMOVE 0x02 429 7754 Jeff #define SPA_ASYNC_PROBE 0x04 430 7754 Jeff #define SPA_ASYNC_RESILVER_DONE 0x08 431 7754 Jeff #define SPA_ASYNC_RESILVER 0x10 432 9816 George #define SPA_ASYNC_AUTOEXPAND 0x20 433 789 ahrens 434 789 ahrens /* device manipulation */ 435 789 ahrens extern int spa_vdev_add(spa_t *spa, nvlist_t *nvroot); 436 1544 eschrock extern int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, 437 789 ahrens int replacing); 438 8241 Jeff extern int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, 439 8241 Jeff int replace_done); 440 2082 eschrock extern int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare); 441 1354 eschrock extern int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath); 442 9425 Eric extern int spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru); 443 2082 eschrock 444 2082 eschrock /* spare state (which is global across all pools) */ 445 3377 eschrock extern void spa_spare_add(vdev_t *vd); 446 3377 eschrock extern void spa_spare_remove(vdev_t *vd); 447 7214 lling extern boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt); 448 3377 eschrock extern void spa_spare_activate(vdev_t *vd); 449 789 ahrens 450 5450 brendan /* L2ARC state (which is global across all pools) */ 451 5450 brendan extern void spa_l2cache_add(vdev_t *vd); 452 5450 brendan extern void spa_l2cache_remove(vdev_t *vd); 453 5450 brendan extern boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool); 454 5450 brendan extern void spa_l2cache_activate(vdev_t *vd); 455 5450 brendan extern void spa_l2cache_drop(spa_t *spa); 456 5450 brendan 457 789 ahrens /* scrubbing */ 458 7046 ahrens extern int spa_scrub(spa_t *spa, pool_scrub_type_t type); 459 789 ahrens 460 789 ahrens /* spa syncing */ 461 789 ahrens extern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */ 462 789 ahrens extern void spa_sync_allpools(void); 463 10922 Jeff 464 10922 Jeff #define SYNC_PASS_DEFERRED_FREE 1 /* defer frees after this pass */ 465 10922 Jeff #define SYNC_PASS_DONT_COMPRESS 4 /* don't compress after this pass */ 466 10922 Jeff #define SYNC_PASS_REWRITE 1 /* rewrite new bps after this pass */ 467 789 ahrens 468 6987 brendan /* spa namespace global mutex */ 469 6987 brendan extern kmutex_t spa_namespace_lock; 470 6987 brendan 471 789 ahrens /* 472 789 ahrens * SPA configuration functions in spa_config.c 473 789 ahrens */ 474 1635 bonwick 475 1635 bonwick #define SPA_CONFIG_UPDATE_POOL 0 476 1635 bonwick #define SPA_CONFIG_UPDATE_VDEVS 1 477 1635 bonwick 478 6643 eschrock extern void spa_config_sync(spa_t *, boolean_t, boolean_t); 479 789 ahrens extern void spa_config_load(void); 480 789 ahrens extern nvlist_t *spa_all_configs(uint64_t *); 481 789 ahrens extern void spa_config_set(spa_t *spa, nvlist_t *config); 482 789 ahrens extern nvlist_t *spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, 483 789 ahrens int getstats); 484 1635 bonwick extern void spa_config_update(spa_t *spa, int what); 485 789 ahrens 486 789 ahrens /* 487 789 ahrens * Miscellaneous SPA routines in spa_misc.c 488 789 ahrens */ 489 789 ahrens 490 789 ahrens /* Namespace manipulation */ 491 789 ahrens extern spa_t *spa_lookup(const char *name); 492 10921 Tim extern spa_t *spa_add(const char *name, nvlist_t *config, const char *altroot); 493 789 ahrens extern void spa_remove(spa_t *spa); 494 789 ahrens extern spa_t *spa_next(spa_t *prev); 495 789 ahrens 496 789 ahrens /* Refcount functions */ 497 789 ahrens extern void spa_open_ref(spa_t *spa, void *tag); 498 789 ahrens extern void spa_close(spa_t *spa, void *tag); 499 789 ahrens extern boolean_t spa_refcount_zero(spa_t *spa); 500 789 ahrens 501 10685 George #define SCL_NONE 0x00 502 7754 Jeff #define SCL_CONFIG 0x01 503 7754 Jeff #define SCL_STATE 0x02 504 7754 Jeff #define SCL_L2ARC 0x04 /* hack until L2ARC 2.0 */ 505 7754 Jeff #define SCL_ALLOC 0x08 506 7754 Jeff #define SCL_ZIO 0x10 507 7754 Jeff #define SCL_FREE 0x20 508 7754 Jeff #define SCL_VDEV 0x40 509 7754 Jeff #define SCL_LOCKS 7 510 7754 Jeff #define SCL_ALL ((1 << SCL_LOCKS) - 1) 511 7754 Jeff #define SCL_STATE_ALL (SCL_STATE | SCL_L2ARC | SCL_ZIO) 512 7754 Jeff 513 7754 Jeff /* Pool configuration locks */ 514 7754 Jeff extern int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw); 515 7754 Jeff extern void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw); 516 7754 Jeff extern void spa_config_exit(spa_t *spa, int locks, void *tag); 517 7754 Jeff extern int spa_config_held(spa_t *spa, int locks, krw_t rw); 518 789 ahrens 519 789 ahrens /* Pool vdev add/remove lock */ 520 789 ahrens extern uint64_t spa_vdev_enter(spa_t *spa); 521 10594 George extern uint64_t spa_vdev_config_enter(spa_t *spa); 522 10594 George extern void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, 523 10594 George int error, char *tag); 524 789 ahrens extern int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error); 525 7754 Jeff 526 7754 Jeff /* Pool vdev state change lock */ 527 10685 George extern void spa_vdev_state_enter(spa_t *spa, int oplock); 528 7754 Jeff extern int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error); 529 789 ahrens 530 10922 Jeff /* Log state */ 531 10922 Jeff typedef enum spa_log_state { 532 10922 Jeff SPA_LOG_UNKNOWN = 0, /* unknown log state */ 533 10922 Jeff SPA_LOG_MISSING, /* missing log(s) */ 534 10922 Jeff SPA_LOG_CLEAR, /* clear the log(s) */ 535 10922 Jeff SPA_LOG_GOOD, /* log(s) are good */ 536 10922 Jeff } spa_log_state_t; 537 10922 Jeff 538 10922 Jeff extern spa_log_state_t spa_get_log_state(spa_t *spa); 539 10922 Jeff extern void spa_set_log_state(spa_t *spa, spa_log_state_t state); 540 10922 Jeff 541 10922 Jeff /* Log claim callback */ 542 10922 Jeff extern void spa_claim_notify(zio_t *zio); 543 10922 Jeff 544 789 ahrens /* Accessor functions */ 545 7837 Matthew extern boolean_t spa_shutting_down(spa_t *spa); 546 789 ahrens extern struct dsl_pool *spa_get_dsl(spa_t *spa); 547 789 ahrens extern blkptr_t *spa_get_rootblkptr(spa_t *spa); 548 789 ahrens extern void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp); 549 789 ahrens extern void spa_altroot(spa_t *, char *, size_t); 550 789 ahrens extern int spa_sync_pass(spa_t *spa); 551 789 ahrens extern char *spa_name(spa_t *spa); 552 789 ahrens extern uint64_t spa_guid(spa_t *spa); 553 789 ahrens extern uint64_t spa_last_synced_txg(spa_t *spa); 554 789 ahrens extern uint64_t spa_first_txg(spa_t *spa); 555 10922 Jeff extern uint64_t spa_syncing_txg(spa_t *spa); 556 2082 eschrock extern uint64_t spa_version(spa_t *spa); 557 7837 Matthew extern pool_state_t spa_state(spa_t *spa); 558 11147 George extern spa_load_state_t spa_load_state(spa_t *spa); 559 789 ahrens extern uint64_t spa_freeze_txg(spa_t *spa); 560 789 ahrens extern uint64_t spa_get_asize(spa_t *spa, uint64_t lsize); 561 10956 George extern uint64_t spa_get_dspace(spa_t *spa); 562 10956 George extern void spa_update_dspace(spa_t *spa); 563 1775 billm extern uint64_t spa_version(spa_t *spa); 564 10922 Jeff extern boolean_t spa_deflate(spa_t *spa); 565 10922 Jeff extern metaslab_class_t *spa_normal_class(spa_t *spa); 566 10922 Jeff extern metaslab_class_t *spa_log_class(spa_t *spa); 567 1775 billm extern int spa_max_replication(spa_t *spa); 568 789 ahrens extern int spa_busy(void); 569 5329 gw25295 extern uint8_t spa_get_failmode(spa_t *spa); 570 7754 Jeff extern boolean_t spa_suspended(spa_t *spa); 571 10922 Jeff extern uint64_t spa_bootfs(spa_t *spa); 572 10922 Jeff extern uint64_t spa_delegation(spa_t *spa); 573 10922 Jeff extern objset_t *spa_meta_objset(spa_t *spa); 574 10922 Jeff extern enum zio_checksum spa_dedup_checksum(spa_t *spa); 575 789 ahrens 576 789 ahrens /* Miscellaneous support routines */ 577 789 ahrens extern int spa_rename(const char *oldname, const char *newname); 578 789 ahrens extern boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid); 579 789 ahrens extern char *spa_strdup(const char *); 580 789 ahrens extern void spa_strfree(char *); 581 789 ahrens extern uint64_t spa_get_random(uint64_t range); 582 10922 Jeff extern void sprintf_blkptr(char *buf, const blkptr_t *bp); 583 789 ahrens extern void spa_freeze(spa_t *spa); 584 5094 lling extern void spa_upgrade(spa_t *spa, uint64_t version); 585 789 ahrens extern void spa_evict_all(void); 586 6643 eschrock extern vdev_t *spa_lookup_by_guid(spa_t *spa, uint64_t guid, 587 6643 eschrock boolean_t l2cache); 588 2082 eschrock extern boolean_t spa_has_spare(spa_t *, uint64_t guid); 589 10922 Jeff extern uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva); 590 10922 Jeff extern uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp); 591 10922 Jeff extern uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp); 592 4620 perrin extern boolean_t spa_has_slogs(spa_t *spa); 593 6673 eschrock extern boolean_t spa_is_root(spa_t *spa); 594 8241 Jeff extern boolean_t spa_writeable(spa_t *spa); 595 10921 Tim extern void spa_rewind_data_to_nvlist(spa_t *spa, nvlist_t *to); 596 10921 Tim 597 8241 Jeff extern int spa_mode(spa_t *spa); 598 10342 chris extern uint64_t strtonum(const char *str, char **nptr); 599 1544 eschrock 600 2926 ek110237 /* history logging */ 601 4543 marks typedef enum history_log_type { 602 4543 marks LOG_CMD_POOL_CREATE, 603 4543 marks LOG_CMD_NORMAL, 604 4543 marks LOG_INTERNAL 605 4543 marks } history_log_type_t; 606 4543 marks 607 4543 marks typedef struct history_arg { 608 4543 marks const char *ha_history_str; 609 4543 marks history_log_type_t ha_log_type; 610 4543 marks history_internal_events_t ha_event; 611 4543 marks char ha_zone[MAXPATHLEN]; 612 4543 marks } history_arg_t; 613 4543 marks 614 4543 marks extern char *spa_his_ievent_table[]; 615 4543 marks 616 2926 ek110237 extern void spa_history_create_obj(spa_t *spa, dmu_tx_t *tx); 617 2926 ek110237 extern int spa_history_get(spa_t *spa, uint64_t *offset, uint64_t *len_read, 618 2926 ek110237 char *his_buf); 619 2926 ek110237 extern int spa_history_log(spa_t *spa, const char *his_buf, 620 4543 marks history_log_type_t what); 621 9946 Mark extern void spa_history_internal_log(history_internal_events_t event, 622 9946 Mark spa_t *spa, dmu_tx_t *tx, cred_t *cr, const char *fmt, ...); 623 9946 Mark extern void spa_history_log_version(spa_t *spa, history_internal_events_t evt); 624 2926 ek110237 625 1544 eschrock /* error handling */ 626 1544 eschrock struct zbookmark; 627 10922 Jeff extern void spa_log_error(spa_t *spa, zio_t *zio); 628 1544 eschrock extern void zfs_ereport_post(const char *class, spa_t *spa, vdev_t *vd, 629 10922 Jeff zio_t *zio, uint64_t stateoroffset, uint64_t length); 630 4451 eschrock extern void zfs_post_remove(spa_t *spa, vdev_t *vd); 631 10817 Eric extern void zfs_post_state_change(spa_t *spa, vdev_t *vd); 632 4451 eschrock extern void zfs_post_autoreplace(spa_t *spa, vdev_t *vd); 633 1544 eschrock extern uint64_t spa_get_errlog_size(spa_t *spa); 634 1544 eschrock extern int spa_get_errlog(spa_t *spa, void *uaddr, size_t *count); 635 1544 eschrock extern void spa_errlog_rotate(spa_t *spa); 636 1544 eschrock extern void spa_errlog_drain(spa_t *spa); 637 1544 eschrock extern void spa_errlog_sync(spa_t *spa, uint64_t txg); 638 1544 eschrock extern void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub); 639 789 ahrens 640 5810 ek110237 /* vdev cache */ 641 5810 ek110237 extern void vdev_cache_stat_init(void); 642 5810 ek110237 extern void vdev_cache_stat_fini(void); 643 5810 ek110237 644 789 ahrens /* Initialization and termination */ 645 789 ahrens extern void spa_init(int flags); 646 789 ahrens extern void spa_fini(void); 647 6423 gw25295 extern void spa_boot_init(); 648 789 ahrens 649 3912 lling /* properties */ 650 5094 lling extern int spa_prop_set(spa_t *spa, nvlist_t *nvp); 651 5094 lling extern int spa_prop_get(spa_t *spa, nvlist_t **nvp); 652 5094 lling extern void spa_prop_clear_bootfs(spa_t *spa, uint64_t obj, dmu_tx_t *tx); 653 8525 Eric extern void spa_configfile_set(spa_t *, nvlist_t *, boolean_t); 654 4451 eschrock 655 4451 eschrock /* asynchronous event notification */ 656 4451 eschrock extern void spa_event_notify(spa_t *spa, vdev_t *vdev, const char *name); 657 3912 lling 658 789 ahrens #ifdef ZFS_DEBUG 659 3700 ek110237 #define dprintf_bp(bp, fmt, ...) do { \ 660 3700 ek110237 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 661 3700 ek110237 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ 662 10922 Jeff sprintf_blkptr(__blkbuf, (bp)); \ 663 3700 ek110237 dprintf(fmt " %s\n", __VA_ARGS__, __blkbuf); \ 664 3700 ek110237 kmem_free(__blkbuf, BP_SPRINTF_LEN); \ 665 789 ahrens } \ 666 789 ahrens _NOTE(CONSTCOND) } while (0) 667 789 ahrens #else 668 789 ahrens #define dprintf_bp(bp, fmt, ...) 669 789 ahrens #endif 670 789 ahrens 671 8241 Jeff extern int spa_mode_global; /* mode, e.g. FREAD | FWRITE */ 672 789 ahrens 673 789 ahrens #ifdef __cplusplus 674 789 ahrens } 675 789 ahrens #endif 676 789 ahrens 677 789 ahrens #endif /* _SYS_SPA_H */ 678
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.