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 backing file format must not be probed, treat as raw image */
  54     QED_F_BACKING_FORMAT_NO_PROBE = 0x04,
  55
  56     /* Feature bits must be used when the on-disk format changes */
  57     QED_FEATURE_MASK = QED_F_BACKING_FILE | /* supported feature bits */
  58                        QED_F_BACKING_FORMAT_NO_PROBE,
  59     QED_COMPAT_FEATURE_MASK = 0,            /* supported compat feature bits */
  60     QED_AUTOCLEAR_FEATURE_MASK = 0,         /* supported autoclear feature bits */
  61
  62     /* Data is stored in groups of sectors called clusters.  Cluster size must
  63      * be large to avoid keeping too much metadata.  I/O requests that have
  64      * sub-cluster size will require read-modify-write.
  65      */
  66     QED_MIN_CLUSTER_SIZE = 4 * 1024, /* in bytes */
  67     QED_MAX_CLUSTER_SIZE = 64 * 1024 * 1024,
  68     QED_DEFAULT_CLUSTER_SIZE = 64 * 1024,
  69
  70     /* Allocated clusters are tracked using a 2-level pagetable.  Table size is
  71      * a multiple of clusters so large maximum image sizes can be supported
  72      * without jacking up the cluster size too much.
  73      */
  74     QED_MIN_TABLE_SIZE = 1,        /* in clusters */
  75     QED_MAX_TABLE_SIZE = 16,
  76     QED_DEFAULT_TABLE_SIZE = 4,
  77 };
  78
  79 typedef struct {
  80     uint32_t magic;                 /* QED\0 */
  81
  82     uint32_t cluster_size;          /* in bytes */
  83     uint32_t table_size;            /* for L1 and L2 tables, in clusters */
  84     uint32_t header_size;           /* in clusters */
  85
  86     uint64_t features;              /* format feature bits */
  87     uint64_t compat_features;       /* compatible feature bits */
  88     uint64_t autoclear_features;    /* self-resetting feature bits */
  89
  90     uint64_t l1_table_offset;       /* in bytes */
  91     uint64_t image_size;            /* total logical image size, in bytes */
  92
  93     /* if (features & QED_F_BACKING_FILE) */
  94     uint32_t backing_filename_offset; /* in bytes from start of header */
  95     uint32_t backing_filename_size;   /* in bytes */
  96 } QEDHeader;
  97
  98 typedef struct {
  99     uint64_t offsets[0];            /* in bytes */
 100 } QEDTable;
 101
 102 /* The L2 cache is a simple write-through cache for L2 structures */
 103 typedef struct CachedL2Table {
 104     QEDTable *table;
 105     uint64_t offset;    /* offset=0 indicates an invalidate entry */
 106     QTAILQ_ENTRY(CachedL2Table) node;
 107     int ref;
 108 } CachedL2Table;
 109
 110 typedef struct {
 111     QTAILQ_HEAD(, CachedL2Table) entries;
 112     unsigned int n_entries;
 113 } L2TableCache;
 114
 115 typedef struct QEDRequest {
 116     CachedL2Table *l2_table;
 117 } QEDRequest;
 118
 119 typedef struct QEDAIOCB {
 120     BlockDriverAIOCB common;
 121     QEMUBH *bh;
 122     int bh_ret;                     /* final return status for completion bh */
 123     QSIMPLEQ_ENTRY(QEDAIOCB) next;  /* next request */
 124     bool is_write;                  /* false - read, true - write */
 125     bool *finished;                 /* signal for cancel completion */
 126     uint64_t end_pos;               /* request end on block device, in bytes */
 127
 128     /* User scatter-gather list */
 129     QEMUIOVector *qiov;
 130     size_t qiov_offset;             /* byte count already processed */
 131
 132     /* Current cluster scatter-gather list */
 133     QEMUIOVector cur_qiov;
 134     uint64_t cur_pos;               /* position on block device, in bytes */
 135     uint64_t cur_cluster;           /* cluster offset in image file */
 136     unsigned int cur_nclusters;     /* number of clusters being accessed */
 137     int find_cluster_ret;           /* used for L1/L2 update */
 138
 139     QEDRequest request;
 140 } QEDAIOCB;
 141
 142 typedef struct {
 143     BlockDriverState *bs;           /* device */
 144     uint64_t file_size;             /* length of image file, in bytes */
 145
 146     QEDHeader header;               /* always cpu-endian */
 147     QEDTable *l1_table;
 148     L2TableCache l2_cache;          /* l2 table cache */
 149     uint32_t table_nelems;
 150     uint32_t l1_shift;
 151     uint32_t l2_shift;
 152     uint32_t l2_mask;
 153
 154     /* Allocating write request queue */
 155     QSIMPLEQ_HEAD(, QEDAIOCB) allocating_write_reqs;
 156 } BDRVQEDState;
 157
 158 enum {
 159     QED_CLUSTER_FOUND,         /* cluster found */
 160     QED_CLUSTER_L2,            /* cluster missing in L2 */
 161     QED_CLUSTER_L1,            /* cluster missing in L1 */
 162 };
 163
 164 /**
 165  * qed_find_cluster() completion callback
 166  *
 167  * @opaque:     User data for completion callback
 168  * @ret:        QED_CLUSTER_FOUND   Success
 169  *              QED_CLUSTER_L2      Data cluster unallocated in L2
 170  *              QED_CLUSTER_L1      L2 unallocated in L1
 171  *              -errno              POSIX error occurred
 172  * @offset:     Data cluster offset
 173  * @len:        Contiguous bytes starting from cluster offset
 174  *
 175  * This function is invoked when qed_find_cluster() completes.
 176  *
 177  * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range
 178  * in the image file.
 179  *
 180  * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1
 181  * table offset, respectively.  len is number of contiguous unallocated bytes.
 182  */
 183 typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
 184
 185 /**
 186  * Generic callback for chaining async callbacks
 187  */
 188 typedef struct {
 189     BlockDriverCompletionFunc *cb;
 190     void *opaque;
 191 } GenericCB;
 192
 193 void *gencb_alloc(size_t len, BlockDriverCompletionFunc *cb, void *opaque);
 194 void gencb_complete(void *opaque, int ret);
 195
 196 /**
 197  * L2 cache functions
 198  */
 199 void qed_init_l2_cache(L2TableCache *l2_cache);
 200 void qed_free_l2_cache(L2TableCache *l2_cache);
 201 CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
 202 void qed_unref_l2_cache_entry(CachedL2Table *entry);
 203 CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
 204 void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
 205
 206 /**
 207  * Table I/O functions
 208  */
 209 int qed_read_l1_table_sync(BDRVQEDState *s);
 210 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 211                         BlockDriverCompletionFunc *cb, void *opaque);
 212 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 213                             unsigned int n);
 214 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 215                            uint64_t offset);
 216 void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
 217                        BlockDriverCompletionFunc *cb, void *opaque);
 218 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 219                         unsigned int index, unsigned int n, bool flush,
 220                         BlockDriverCompletionFunc *cb, void *opaque);
 221 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 222                             unsigned int index, unsigned int n, bool flush);
 223
 224 /**
 225  * Cluster functions
 226  */
 227 void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos,
 228                       size_t len, QEDFindClusterFunc *cb, void *opaque);
 229
 230 /**
 231  * Consistency check
 232  */
 233 int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
 234
 235 QEDTable *qed_alloc_table(BDRVQEDState *s);
 236
 237 /**
 238  * Round down to the start of a cluster
 239  */
 240 static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
 241 {
 242     return offset & ~(uint64_t)(s->header.cluster_size - 1);
 243 }
 244
 245 static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
 246 {
 247     return offset & (s->header.cluster_size - 1);
 248 }
 249
 250 static inline unsigned int qed_bytes_to_clusters(BDRVQEDState *s, size_t bytes)
 251 {
 252     return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
 253            (s->header.cluster_size - 1);
 254 }
 255
 256 static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
 257 {
 258     return pos >> s->l1_shift;
 259 }
 260
 261 static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
 262 {
 263     return (pos >> s->l2_shift) & s->l2_mask;
 264 }
 265
 266 /**
 267  * Test if a cluster offset is valid
 268  */
 269 static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
 270 {
 271     uint64_t header_size = (uint64_t)s->header.header_size *
 272                            s->header.cluster_size;
 273
 274     if (offset & (s->header.cluster_size - 1)) {
 275         return false;
 276     }
 277     return offset >= header_size && offset < s->file_size;
 278 }
 279
 280 /**
 281  * Test if a table offset is valid
 282  */
 283 static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
 284 {
 285     uint64_t end_offset = offset + (s->header.table_size - 1) *
 286                           s->header.cluster_size;
 287
 288     /* Overflow check */
 289     if (end_offset <= offset) {
 290         return false;
 291     }
 292
 293     return qed_check_cluster_offset(s, offset) &&
 294            qed_check_cluster_offset(s, end_offset);
 295 }
 296
 297 #endif /* BLOCK_QED_H */