1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * This file and its contents are supplied under the terms of the 4 * Common Development and Distribution License ("CDDL"), version 1.0. 5 * You may only use this file in accordance with the terms of version 6 * 1.0 of the CDDL. 7 * 8 * A full copy of the text of the CDDL should have accompanied this 9 * source. A copy of the CDDL is also available via the Internet at 10 * https://opensource.org/license/CDDL-1.0. 11 */ 12 /* 13 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. 14 * Copyright (c) 2013 by Delphix. All rights reserved. 15 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 16 */ 17 18 #ifndef _SYS_SA_IMPL_H 19 #define _SYS_SA_IMPL_H 20 21 #include <sys/dmu.h> 22 #include <sys/zfs_refcount.h> 23 #include <sys/list.h> 24 25 /* 26 * Array of known attributes and their 27 * various characteristics. 28 */ 29 typedef struct sa_attr_table { 30 sa_attr_type_t sa_attr; 31 uint8_t sa_registered; 32 uint16_t sa_length; 33 sa_bswap_type_t sa_byteswap; 34 char *sa_name; 35 } sa_attr_table_t; 36 37 /* 38 * Zap attribute format for attribute registration 39 * 40 * 64 56 48 40 32 24 16 8 0 41 * +-------+-------+-------+-------+-------+-------+-------+-------+ 42 * | unused | len | bswap | attr num | 43 * +-------+-------+-------+-------+-------+-------+-------+-------+ 44 * 45 * Zap attribute format for layout information. 46 * 47 * layout information is stored as an array of attribute numbers 48 * The name of the attribute is the layout number (0, 1, 2, ...) 49 * 50 * 16 0 51 * +---- ---+ 52 * | attr # | 53 * +--------+ 54 * | attr # | 55 * +--- ----+ 56 * ...... 57 * 58 */ 59 60 #define ATTR_BSWAP(x) BF32_GET(x, 16, 8) 61 #define ATTR_LENGTH(x) BF32_GET(x, 24, 16) 62 #define ATTR_NUM(x) BF32_GET(x, 0, 16) 63 #define ATTR_ENCODE(x, attr, length, bswap) \ 64 { \ 65 BF64_SET(x, 24, 16, length); \ 66 BF64_SET(x, 16, 8, bswap); \ 67 BF64_SET(x, 0, 16, attr); \ 68 } 69 70 #define TOC_OFF(x) BF32_GET(x, 0, 23) 71 #define TOC_ATTR_PRESENT(x) BF32_GET(x, 31, 1) 72 #define TOC_LEN_IDX(x) BF32_GET(x, 24, 4) 73 #define TOC_ATTR_ENCODE(x, len_idx, offset) \ 74 { \ 75 BF32_SET(x, 31, 1, 1); \ 76 BF32_SET(x, 24, 7, len_idx); \ 77 BF32_SET(x, 0, 24, offset); \ 78 } 79 80 #define SA_LAYOUTS "LAYOUTS" 81 #define SA_REGISTRY "REGISTRY" 82 83 /* 84 * Each unique layout will have their own table 85 * sa_lot (layout_table) 86 */ 87 typedef struct sa_lot { 88 avl_node_t lot_num_node; 89 avl_node_t lot_hash_node; 90 uint64_t lot_num; 91 uint64_t lot_hash; 92 sa_attr_type_t *lot_attrs; /* array of attr #'s */ 93 uint32_t lot_var_sizes; /* how many aren't fixed size */ 94 uint32_t lot_attr_count; /* total attr count */ 95 list_t lot_idx_tab; /* should be only a couple of entries */ 96 int lot_instance; /* used with lot_hash to identify entry */ 97 } sa_lot_t; 98 99 /* index table of offsets */ 100 typedef struct sa_idx_tab { 101 list_node_t sa_next; 102 sa_lot_t *sa_layout; 103 uint16_t *sa_variable_lengths; 104 zfs_refcount_t sa_refcount; 105 uint32_t *sa_idx_tab; /* array of offsets */ 106 } sa_idx_tab_t; 107 108 /* 109 * Since the offset/index information into the actual data 110 * will usually be identical we can share that information with 111 * all handles that have the exact same offsets. 112 * 113 * You would typically only have a large number of different table of 114 * contents if you had a several variable sized attributes. 115 * 116 * Two AVL trees are used to track the attribute layout numbers. 117 * one is keyed by number and will be consulted when a DMU_OT_SA 118 * object is first read. The second tree is keyed by the hash signature 119 * of the attributes and will be consulted when an attribute is added 120 * to determine if we already have an instance of that layout. Both 121 * of these tree's are interconnected. The only difference is that 122 * when an entry is found in the "hash" tree the list of attributes will 123 * need to be compared against the list of attributes you have in hand. 124 * The assumption is that typically attributes will just be updated and 125 * adding a completely new attribute is a very rare operation. 126 */ 127 struct sa_os { 128 kmutex_t sa_lock; 129 boolean_t sa_need_attr_registration; 130 boolean_t sa_force_spill; 131 uint64_t sa_master_obj; 132 uint64_t sa_reg_attr_obj; 133 uint64_t sa_layout_attr_obj; 134 int sa_num_attrs; 135 sa_attr_table_t *sa_attr_table; /* private attr table */ 136 sa_update_cb_t *sa_update_cb; 137 avl_tree_t sa_layout_num_tree; /* keyed by layout number */ 138 avl_tree_t sa_layout_hash_tree; /* keyed by layout hash value */ 139 int sa_user_table_sz; 140 sa_attr_type_t *sa_user_table; /* user name->attr mapping table */ 141 }; 142 143 /* 144 * header for all bonus and spill buffers. 145 * 146 * The header has a fixed portion with a variable number 147 * of "lengths" depending on the number of variable sized 148 * attributes which are determined by the "layout number" 149 */ 150 151 #define SA_MAGIC 0x2F505A /* ZFS SA */ 152 typedef struct sa_hdr_phys { 153 uint32_t sa_magic; 154 /* 155 * Encoded with hdrsize and layout number as follows: 156 * 16 10 0 157 * +--------+-------+ 158 * | hdrsz |layout | 159 * +--------+-------+ 160 * 161 * Bits 0-9 (10 bits) are the layout number (0-1023) 162 * Bits 10-15 (6 bits) are the size of the header (0-63) 163 * The hdrsize is the number * 8 164 * 165 * For example. 166 * hdrsz of 1 ==> 8 byte header 167 * 2 ==> 16 byte header 168 * 169 */ 170 uint16_t sa_layout_info; 171 uint16_t sa_lengths[1]; /* optional sizes for variable length attrs */ 172 /* ... Data follows the lengths. */ 173 } sa_hdr_phys_t; 174 175 #define SA_HDR_LAYOUT_NUM(hdr) BF32_GET(hdr->sa_layout_info, 0, 10) 176 #define SA_HDR_SIZE(hdr) BF32_GET_SB(hdr->sa_layout_info, 10, 6, 3, 0) 177 #define SA_HDR_LAYOUT_INFO_ENCODE(x, num, size) \ 178 { \ 179 BF32_SET_SB(x, 10, 6, 3, 0, size); \ 180 BF32_SET(x, 0, 10, num); \ 181 } 182 183 typedef enum sa_buf_type { 184 SA_BONUS = 1, 185 SA_SPILL = 2 186 } sa_buf_type_t; 187 188 typedef enum sa_data_op { 189 SA_LOOKUP, 190 SA_UPDATE, 191 SA_ADD, 192 SA_REPLACE, 193 SA_REMOVE 194 } sa_data_op_t; 195 196 /* 197 * Opaque handle used for most sa functions 198 * 199 * This needs to be kept as small as possible. 200 */ 201 202 struct sa_handle { 203 dmu_buf_user_t sa_dbu; 204 kmutex_t sa_lock; 205 dmu_buf_t *sa_bonus; 206 dmu_buf_t *sa_spill; 207 objset_t *sa_os; 208 void *sa_userp; 209 sa_idx_tab_t *sa_bonus_tab; /* idx of bonus */ 210 sa_idx_tab_t *sa_spill_tab; /* only present if spill activated */ 211 }; 212 213 #define SA_GET_DB(hdl, type) \ 214 (dmu_buf_impl_t *)((type == SA_BONUS) ? hdl->sa_bonus : hdl->sa_spill) 215 216 #define SA_GET_HDR(hdl, type) \ 217 ((sa_hdr_phys_t *)((dmu_buf_impl_t *)(SA_GET_DB(hdl, \ 218 type))->db.db_data)) 219 220 #define SA_IDX_TAB_GET(hdl, type) \ 221 (type == SA_BONUS ? hdl->sa_bonus_tab : hdl->sa_spill_tab) 222 223 #define IS_SA_BONUSTYPE(a) \ 224 ((a == DMU_OT_SA) ? B_TRUE : B_FALSE) 225 226 #define SA_BONUSTYPE_FROM_DB(db) \ 227 (dmu_get_bonustype((dmu_buf_t *)db)) 228 229 #define SA_BLKPTR_SPACE (DN_OLD_MAX_BONUSLEN - sizeof (blkptr_t)) 230 231 #define SA_LAYOUT_NUM(x, type) \ 232 ((!IS_SA_BONUSTYPE(type) ? 0 : (((IS_SA_BONUSTYPE(type)) && \ 233 ((SA_HDR_LAYOUT_NUM(x)) == 0)) ? 1 : SA_HDR_LAYOUT_NUM(x)))) 234 235 236 #define SA_REGISTERED_LEN(sa, attr) sa->sa_attr_table[attr].sa_length 237 238 #define SA_ATTR_LEN(sa, idx, attr, hdr) ((SA_REGISTERED_LEN(sa, attr) == 0) ?\ 239 hdr->sa_lengths[TOC_LEN_IDX(idx->sa_idx_tab[attr])] : \ 240 SA_REGISTERED_LEN(sa, attr)) 241 242 #define SA_SET_HDR(hdr, num, size) \ 243 { \ 244 hdr->sa_magic = SA_MAGIC; \ 245 SA_HDR_LAYOUT_INFO_ENCODE(hdr->sa_layout_info, num, size); \ 246 } 247 248 #define SA_ATTR_INFO(sa, idx, hdr, attr, bulk, type, hdl) \ 249 { \ 250 bulk.sa_size = SA_ATTR_LEN(sa, idx, attr, hdr); \ 251 bulk.sa_buftype = type; \ 252 bulk.sa_addr = \ 253 (void *)((uintptr_t)TOC_OFF(idx->sa_idx_tab[attr]) + \ 254 (uintptr_t)hdr); \ 255 } 256 257 #define SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb) \ 258 (SA_HDR_SIZE(hdr) == (sizeof (sa_hdr_phys_t) + \ 259 (tb->lot_var_sizes > 1 ? P2ROUNDUP((tb->lot_var_sizes - 1) * \ 260 sizeof (uint16_t), 8) : 0))) 261 262 int sa_add_impl(sa_handle_t *, sa_attr_type_t, 263 uint32_t, sa_data_locator_t, void *, dmu_tx_t *); 264 265 void sa_register_update_callback_locked(objset_t *, sa_update_cb_t *); 266 267 void sa_default_locator(void **, uint32_t *, uint32_t, boolean_t, void *); 268 int sa_attr_size(sa_os_t *, sa_idx_tab_t *, sa_attr_type_t, 269 uint16_t *, sa_hdr_phys_t *); 270 271 #ifdef __cplusplus 272 extern "C" { 273 #endif 274 275 #ifdef __cplusplus 276 } 277 #endif 278 279 #endif /* _SYS_SA_IMPL_H */ 280