block/qed.h

   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
  83 typedef struct {
  84     uint32_t magic;                 /* QED\0 */
  85
  86     uint32_t cluster_size;          /* in bytes */
  87     uint32_t table_size;            /* for L1 and L2 tables, in clusters */
  88     uint32_t header_size;           /* in clusters */
  89
  90     uint64_t features;              /* format feature bits */
  91     uint64_t compat_features;       /* compatible feature bits */
  92     uint64_t autoclear_features;    /* self-resetting feature bits */
  93
  94     uint64_t l1_table_offset;       /* in bytes */
  95     uint64_t image_size;            /* total logical image size, in bytes */
  96
  97     /* if (features & QED_F_BACKING_FILE) */
  98     uint32_t backing_filename_offset; /* in bytes from start of header */
  99     uint32_t backing_filename_size;   /* in bytes */
 100 } QEDHeader;
 101
 102 typedef struct {
 103     uint64_t offsets[0];            /* in bytes */
 104 } QEDTable;
 105
 106 /* The L2 cache is a simple write-through cache for L2 structures */
 107 typedef struct CachedL2Table {
 108     QEDTable *table;
 109     uint64_t offset;    /* offset=0 indicates an invalidate entry */
 110     QTAILQ_ENTRY(CachedL2Table) node;
 111     int ref;
 112 } CachedL2Table;
 113
 114 typedef struct {
 115     QTAILQ_HEAD(, CachedL2Table) entries;
 116     unsigned int n_entries;
 117 } L2TableCache;
 118
 119 typedef struct QEDRequest {
 120     CachedL2Table *l2_table;
 121 } QEDRequest;
 122
 123 typedef struct QEDAIOCB {
 124     BlockDriverAIOCB common;
 125     QEMUBH *bh;
 126     int bh_ret;                     /* final return status for completion bh */
 127     QSIMPLEQ_ENTRY(QEDAIOCB) next;  /* next request */
 128     bool is_write;                  /* false - read, true - write */
 129     bool *finished;                 /* signal for cancel completion */
 130     uint64_t end_pos;               /* request end on block device, in bytes */
 131
 132     /* User scatter-gather list */
 133     QEMUIOVector *qiov;
 134     size_t qiov_offset;             /* byte count already processed */
 135
 136     /* Current cluster scatter-gather list */
 137     QEMUIOVector cur_qiov;
 138     uint64_t cur_pos;               /* position on block device, in bytes */
 139     uint64_t cur_cluster;           /* cluster offset in image file */
 140     unsigned int cur_nclusters;     /* number of clusters being accessed */
 141     int find_cluster_ret;           /* used for L1/L2 update */
 142
 143     QEDRequest request;
 144 } QEDAIOCB;
 145
 146 typedef struct {
 147     BlockDriverState *bs;           /* device */
 148     uint64_t file_size;             /* length of image file, in bytes */
 149
 150     QEDHeader header;               /* always cpu-endian */
 151     QEDTable *l1_table;
 152     L2TableCache l2_cache;          /* l2 table cache */
 153     uint32_t table_nelems;
 154     uint32_t l1_shift;
 155     uint32_t l2_shift;
 156     uint32_t l2_mask;
 157
 158     /* Allocating write request queue */
 159     QSIMPLEQ_HEAD(, QEDAIOCB) allocating_write_reqs;
 160 } BDRVQEDState;
 161
 162 enum {
 163     QED_CLUSTER_FOUND,         /* cluster found */
 164     QED_CLUSTER_ZERO,          /* zero cluster found */
 165     QED_CLUSTER_L2,            /* cluster missing in L2 */
 166     QED_CLUSTER_L1,            /* cluster missing in L1 */
 167 };
 168
 169 /**
 170  * qed_find_cluster() completion callback
 171  *
 172  * @opaque:     User data for completion callback
 173  * @ret:        QED_CLUSTER_FOUND   Success
 174  *              QED_CLUSTER_L2      Data cluster unallocated in L2
 175  *              QED_CLUSTER_L1      L2 unallocated in L1
 176  *              -errno              POSIX error occurred
 177  * @offset:     Data cluster offset
 178  * @len:        Contiguous bytes starting from cluster offset
 179  *
 180  * This function is invoked when qed_find_cluster() completes.
 181  *
 182  * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range
 183  * in the image file.
 184  *
 185  * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1
 186  * table offset, respectively.  len is number of contiguous unallocated bytes.
 187  */
 188 typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
 189
 190 /**
 191  * Generic callback for chaining async callbacks
 192  */
 193 typedef struct {
 194     BlockDriverCompletionFunc *cb;
 195     void *opaque;
 196 } GenericCB;
 197
 198 void *gencb_alloc(size_t len, BlockDriverCompletionFunc *cb, void *opaque);
 199 void gencb_complete(void *opaque, int ret);
 200
 201 /**
 202  * L2 cache functions
 203  */
 204 void qed_init_l2_cache(L2TableCache *l2_cache);
 205 void qed_free_l2_cache(L2TableCache *l2_cache);
 206 CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
 207 void qed_unref_l2_cache_entry(CachedL2Table *entry);
 208 CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
 209 void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
 210
 211 /**
 212  * Table I/O functions
 213  */
 214 int qed_read_l1_table_sync(BDRVQEDState *s);
 215 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 216                         BlockDriverCompletionFunc *cb, void *opaque);
 217 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 218                             unsigned int n);
 219 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 220                            uint64_t offset);
 221 void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
 222                        BlockDriverCompletionFunc *cb, void *opaque);
 223 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 224                         unsigned int index, unsigned int n, bool flush,
 225                         BlockDriverCompletionFunc *cb, void *opaque);
 226 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 227                             unsigned int index, unsigned int n, bool flush);
 228
 229 /**
 230  * Cluster functions
 231  */
 232 void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos,
 233                       size_t len, QEDFindClusterFunc *cb, void *opaque);
 234
 235 /**
 236  * Consistency check
 237  */
 238 int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
 239
 240 QEDTable *qed_alloc_table(BDRVQEDState *s);
 241
 242 /**
 243  * Round down to the start of a cluster
 244  */
 245 static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
 246 {
 247     return offset & ~(uint64_t)(s->header.cluster_size - 1);
 248 }
 249
 250 static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
 251 {
 252     return offset & (s->header.cluster_size - 1);
 253 }
 254
 255 static inline unsigned int qed_bytes_to_clusters(BDRVQEDState *s, size_t bytes)
 256 {
 257     return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
 258            (s->header.cluster_size - 1);
 259 }
 260
 261 static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
 262 {
 263     return pos >> s->l1_shift;
 264 }
 265
 266 static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
 267 {
 268     return (pos >> s->l2_shift) & s->l2_mask;
 269 }
 270
 271 /**
 272  * Test if a cluster offset is valid
 273  */
 274 static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
 275 {
 276     uint64_t header_size = (uint64_t)s->header.header_size *
 277                            s->header.cluster_size;
 278
 279     if (offset & (s->header.cluster_size - 1)) {
 280         return false;
 281     }
 282     return offset >= header_size && offset < s->file_size;
 283 }
 284
 285 /**
 286  * Test if a table offset is valid
 287  */
 288 static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
 289 {
 290     uint64_t end_offset = offset + (s->header.table_size - 1) *
 291                           s->header.cluster_size;
 292
 293     /* Overflow check */
 294     if (end_offset <= offset) {
 295         return false;
 296     }
 297
 298     return qed_check_cluster_offset(s, offset) &&
 299            qed_check_cluster_offset(s, end_offset);
 300 }
 301
 302 static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s,
 303                                                  uint64_t offset)
 304 {
 305     if (qed_offset_into_cluster(s, offset)) {
 306         return false;
 307     }
 308     return true;
 309 }
 310
 311 static inline bool qed_offset_is_unalloc_cluster(uint64_t offset)
 312 {
 313     if (offset == 0) {
 314         return true;
 315     }
 316     return false;
 317 }
 318
 319 static inline bool qed_offset_is_zero_cluster(uint64_t offset)
 320 {
 321     if (offset == 1) {
 322         return true;
 323     }
 324     return false;
 325 }
 326
 327 #endif /* BLOCK_QED_H */