1 /* 2 * QEMU Enhanced Disk Format 3 * 4 * Copyright IBM, Corp. 2010 5 * 6 * Authors: 7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> 8 * Anthony Liguori <aliguori@us.ibm.com> 9 * 10 * This work is licensed under the terms of the GNU LGPL, version 2 or later. 11 * See the COPYING.LIB file in the top-level directory. 12 * 13 */ 14 15 #ifndef BLOCK_QED_H 16 #define BLOCK_QED_H 17 18 #include "block_int.h" 19 20 /* The layout of a QED file is as follows: 21 * 22 * +--------+----------+----------+----------+-----+ 23 * | header | L1 table | cluster0 | cluster1 | ... | 24 * +--------+----------+----------+----------+-----+ 25 * 26 * There is a 2-level pagetable for cluster allocation: 27 * 28 * +----------+ 29 * | L1 table | 30 * +----------+ 31 * ,------' | '------. 32 * +----------+ | +----------+ 33 * | L2 table | ... | L2 table | 34 * +----------+ +----------+ 35 * ,------' | '------. 36 * +----------+ | +----------+ 37 * | Data | ... | Data | 38 * +----------+ +----------+ 39 * 40 * The L1 table is fixed size and always present. L2 tables are allocated on 41 * demand. The L1 table size determines the maximum possible image size; it 42 * can be influenced using the cluster_size and table_size values. 43 * 44 * All fields are little-endian on disk. 45 */ 46 47 enum { 48 QED_MAGIC = 'Q' | 'E' << 8 | 'D' << 16 | '\0' << 24, 49 50 /* The image supports a backing file */ 51 QED_F_BACKING_FILE = 0x01, 52 53 /* The image needs a consistency check before use */ 54 QED_F_NEED_CHECK = 0x02, 55 56 /* The backing file format must not be probed, treat as raw image */ 57 QED_F_BACKING_FORMAT_NO_PROBE = 0x04, 58 59 /* Feature bits must be used when the on-disk format changes */ 60 QED_FEATURE_MASK = QED_F_BACKING_FILE | /* supported feature bits */ 61 QED_F_NEED_CHECK | 62 QED_F_BACKING_FORMAT_NO_PROBE, 63 QED_COMPAT_FEATURE_MASK = 0, /* supported compat feature bits */ 64 QED_AUTOCLEAR_FEATURE_MASK = 0, /* supported autoclear feature bits */ 65 66 /* Data is stored in groups of sectors called clusters. Cluster size must 67 * be large to avoid keeping too much metadata. I/O requests that have 68 * sub-cluster size will require read-modify-write. 69 */ 70 QED_MIN_CLUSTER_SIZE = 4 * 1024, /* in bytes */ 71 QED_MAX_CLUSTER_SIZE = 64 * 1024 * 1024, 72 QED_DEFAULT_CLUSTER_SIZE = 64 * 1024, 73 74 /* Allocated clusters are tracked using a 2-level pagetable. Table size is 75 * a multiple of clusters so large maximum image sizes can be supported 76 * without jacking up the cluster size too much. 77 */ 78 QED_MIN_TABLE_SIZE = 1, /* in clusters */ 79 QED_MAX_TABLE_SIZE = 16, 80 QED_DEFAULT_TABLE_SIZE = 4, 81 82 /* Delay to flush and clean image after last allocating write completes */ 83 QED_NEED_CHECK_TIMEOUT = 5, /* in seconds */ 84 }; 85 86 typedef struct { 87 uint32_t magic; /* QED\0 */ 88 89 uint32_t cluster_size; /* in bytes */ 90 uint32_t table_size; /* for L1 and L2 tables, in clusters */ 91 uint32_t header_size; /* in clusters */ 92 93 uint64_t features; /* format feature bits */ 94 uint64_t compat_features; /* compatible feature bits */ 95 uint64_t autoclear_features; /* self-resetting feature bits */ 96 97 uint64_t l1_table_offset; /* in bytes */ 98 uint64_t image_size; /* total logical image size, in bytes */ 99 100 /* if (features & QED_F_BACKING_FILE) */ 101 uint32_t backing_filename_offset; /* in bytes from start of header */ 102 uint32_t backing_filename_size; /* in bytes */ 103 } QEDHeader; 104 105 typedef struct { 106 uint64_t offsets[0]; /* in bytes */ 107 } QEDTable; 108 109 /* The L2 cache is a simple write-through cache for L2 structures */ 110 typedef struct CachedL2Table { 111 QEDTable *table; 112 uint64_t offset; /* offset=0 indicates an invalidate entry */ 113 QTAILQ_ENTRY(CachedL2Table) node; 114 int ref; 115 } CachedL2Table; 116 117 typedef struct { 118 QTAILQ_HEAD(, CachedL2Table) entries; 119 unsigned int n_entries; 120 } L2TableCache; 121 122 typedef struct QEDRequest { 123 CachedL2Table *l2_table; 124 } QEDRequest; 125 126 typedef struct QEDAIOCB { 127 BlockDriverAIOCB common; 128 QEMUBH *bh; 129 int bh_ret; /* final return status for completion bh */ 130 QSIMPLEQ_ENTRY(QEDAIOCB) next; /* next request */ 131 bool is_write; /* false - read, true - write */ 132 bool *finished; /* signal for cancel completion */ 133 uint64_t end_pos; /* request end on block device, in bytes */ 134 135 /* User scatter-gather list */ 136 QEMUIOVector *qiov; 137 size_t qiov_offset; /* byte count already processed */ 138 139 /* Current cluster scatter-gather list */ 140 QEMUIOVector cur_qiov; 141 uint64_t cur_pos; /* position on block device, in bytes */ 142 uint64_t cur_cluster; /* cluster offset in image file */ 143 unsigned int cur_nclusters; /* number of clusters being accessed */ 144 int find_cluster_ret; /* used for L1/L2 update */ 145 146 QEDRequest request; 147 } QEDAIOCB; 148 149 typedef struct { 150 BlockDriverState *bs; /* device */ 151 uint64_t file_size; /* length of image file, in bytes */ 152 153 QEDHeader header; /* always cpu-endian */ 154 QEDTable *l1_table; 155 L2TableCache l2_cache; /* l2 table cache */ 156 uint32_t table_nelems; 157 uint32_t l1_shift; 158 uint32_t l2_shift; 159 uint32_t l2_mask; 160 161 /* Allocating write request queue */ 162 QSIMPLEQ_HEAD(, QEDAIOCB) allocating_write_reqs; 163 bool allocating_write_reqs_plugged; 164 165 /* Periodic flush and clear need check flag */ 166 QEMUTimer *need_check_timer; 167 } BDRVQEDState; 168 169 enum { 170 QED_CLUSTER_FOUND, /* cluster found */ 171 QED_CLUSTER_ZERO, /* zero cluster found */ 172 QED_CLUSTER_L2, /* cluster missing in L2 */ 173 QED_CLUSTER_L1, /* cluster missing in L1 */ 174 }; 175 176 /** 177 * qed_find_cluster() completion callback 178 * 179 * @opaque: User data for completion callback 180 * @ret: QED_CLUSTER_FOUND Success 181 * QED_CLUSTER_L2 Data cluster unallocated in L2 182 * QED_CLUSTER_L1 L2 unallocated in L1 183 * -errno POSIX error occurred 184 * @offset: Data cluster offset 185 * @len: Contiguous bytes starting from cluster offset 186 * 187 * This function is invoked when qed_find_cluster() completes. 188 * 189 * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range 190 * in the image file. 191 * 192 * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1 193 * table offset, respectively. len is number of contiguous unallocated bytes. 194 */ 195 typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len); 196 197 /** 198 * Generic callback for chaining async callbacks 199 */ 200 typedef struct { 201 BlockDriverCompletionFunc *cb; 202 void *opaque; 203 } GenericCB; 204 205 void *gencb_alloc(size_t len, BlockDriverCompletionFunc *cb, void *opaque); 206 void gencb_complete(void *opaque, int ret); 207 208 /** 209 * L2 cache functions 210 */ 211 void qed_init_l2_cache(L2TableCache *l2_cache); 212 void qed_free_l2_cache(L2TableCache *l2_cache); 213 CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache); 214 void qed_unref_l2_cache_entry(CachedL2Table *entry); 215 CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset); 216 void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table); 217 218 /** 219 * Table I/O functions 220 */ 221 int qed_read_l1_table_sync(BDRVQEDState *s); 222 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n, 223 BlockDriverCompletionFunc *cb, void *opaque); 224 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index, 225 unsigned int n); 226 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, 227 uint64_t offset); 228 void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset, 229 BlockDriverCompletionFunc *cb, void *opaque); 230 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request, 231 unsigned int index, unsigned int n, bool flush, 232 BlockDriverCompletionFunc *cb, void *opaque); 233 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request, 234 unsigned int index, unsigned int n, bool flush); 235 236 /** 237 * Cluster functions 238 */ 239 void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos, 240 size_t len, QEDFindClusterFunc *cb, void *opaque); 241 242 /** 243 * Consistency check 244 */ 245 int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix); 246 247 QEDTable *qed_alloc_table(BDRVQEDState *s); 248 249 /** 250 * Round down to the start of a cluster 251 */ 252 static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset) 253 { 254 return offset & ~(uint64_t)(s->header.cluster_size - 1); 255 } 256 257 static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset) 258 { 259 return offset & (s->header.cluster_size - 1); 260 } 261 262 static inline uint64_t qed_bytes_to_clusters(BDRVQEDState *s, uint64_t bytes) 263 { 264 return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) / 265 (s->header.cluster_size - 1); 266 } 267 268 static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos) 269 { 270 return pos >> s->l1_shift; 271 } 272 273 static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos) 274 { 275 return (pos >> s->l2_shift) & s->l2_mask; 276 } 277 278 /** 279 * Test if a cluster offset is valid 280 */ 281 static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset) 282 { 283 uint64_t header_size = (uint64_t)s->header.header_size * 284 s->header.cluster_size; 285 286 if (offset & (s->header.cluster_size - 1)) { 287 return false; 288 } 289 return offset >= header_size && offset < s->file_size; 290 } 291 292 /** 293 * Test if a table offset is valid 294 */ 295 static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset) 296 { 297 uint64_t end_offset = offset + (s->header.table_size - 1) * 298 s->header.cluster_size; 299 300 /* Overflow check */ 301 if (end_offset <= offset) { 302 return false; 303 } 304 305 return qed_check_cluster_offset(s, offset) && 306 qed_check_cluster_offset(s, end_offset); 307 } 308 309 static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s, 310 uint64_t offset) 311 { 312 if (qed_offset_into_cluster(s, offset)) { 313 return false; 314 } 315 return true; 316 } 317 318 static inline bool qed_offset_is_unalloc_cluster(uint64_t offset) 319 { 320 if (offset == 0) { 321 return true; 322 } 323 return false; 324 } 325 326 static inline bool qed_offset_is_zero_cluster(uint64_t offset) 327 { 328 if (offset == 1) { 329 return true; 330 } 331 return false; 332 } 333 334 #endif /* BLOCK_QED_H */ 335