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     /* 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 enum {
 127     QED_AIOCB_WRITE = 0x0001,       /* read or write? */
 128 };
 129
 130 typedef struct QEDAIOCB {
 131     BlockDriverAIOCB common;
 132     QEMUBH *bh;
 133     int bh_ret;                     /* final return status for completion bh */
 134     QSIMPLEQ_ENTRY(QEDAIOCB) next;  /* next request */
 135     int flags;                      /* QED_AIOCB_* bits ORed together */
 136     bool *finished;                 /* signal for cancel completion */
 137     uint64_t end_pos;               /* request end on block device, in bytes */
 138
 139     /* User scatter-gather list */
 140     QEMUIOVector *qiov;
 141     size_t qiov_offset;             /* byte count already processed */
 142
 143     /* Current cluster scatter-gather list */
 144     QEMUIOVector cur_qiov;
 145     uint64_t cur_pos;               /* position on block device, in bytes */
 146     uint64_t cur_cluster;           /* cluster offset in image file */
 147     unsigned int cur_nclusters;     /* number of clusters being accessed */
 148     int find_cluster_ret;           /* used for L1/L2 update */
 149
 150     QEDRequest request;
 151 } QEDAIOCB;
 152
 153 typedef struct {
 154     BlockDriverState *bs;           /* device */
 155     uint64_t file_size;             /* length of image file, in bytes */
 156
 157     QEDHeader header;               /* always cpu-endian */
 158     QEDTable *l1_table;
 159     L2TableCache l2_cache;          /* l2 table cache */
 160     uint32_t table_nelems;
 161     uint32_t l1_shift;
 162     uint32_t l2_shift;
 163     uint32_t l2_mask;
 164
 165     /* Allocating write request queue */
 166     QSIMPLEQ_HEAD(, QEDAIOCB) allocating_write_reqs;
 167     bool allocating_write_reqs_plugged;
 168
 169     /* Periodic flush and clear need check flag */
 170     QEMUTimer *need_check_timer;
 171
 172     Error *migration_blocker;
 173 } BDRVQEDState;
 174
 175 enum {
 176     QED_CLUSTER_FOUND,         /* cluster found */
 177     QED_CLUSTER_ZERO,          /* zero cluster found */
 178     QED_CLUSTER_L2,            /* cluster missing in L2 */
 179     QED_CLUSTER_L1,            /* cluster missing in L1 */
 180 };
 181
 182 /**
 183  * qed_find_cluster() completion callback
 184  *
 185  * @opaque:     User data for completion callback
 186  * @ret:        QED_CLUSTER_FOUND   Success
 187  *              QED_CLUSTER_L2      Data cluster unallocated in L2
 188  *              QED_CLUSTER_L1      L2 unallocated in L1
 189  *              -errno              POSIX error occurred
 190  * @offset:     Data cluster offset
 191  * @len:        Contiguous bytes starting from cluster offset
 192  *
 193  * This function is invoked when qed_find_cluster() completes.
 194  *
 195  * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range
 196  * in the image file.
 197  *
 198  * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1
 199  * table offset, respectively.  len is number of contiguous unallocated bytes.
 200  */
 201 typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
 202
 203 /**
 204  * Generic callback for chaining async callbacks
 205  */
 206 typedef struct {
 207     BlockDriverCompletionFunc *cb;
 208     void *opaque;
 209 } GenericCB;
 210
 211 void *gencb_alloc(size_t len, BlockDriverCompletionFunc *cb, void *opaque);
 212 void gencb_complete(void *opaque, int ret);
 213
 214 /**
 215  * L2 cache functions
 216  */
 217 void qed_init_l2_cache(L2TableCache *l2_cache);
 218 void qed_free_l2_cache(L2TableCache *l2_cache);
 219 CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
 220 void qed_unref_l2_cache_entry(CachedL2Table *entry);
 221 CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
 222 void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
 223
 224 /**
 225  * Table I/O functions
 226  */
 227 int qed_read_l1_table_sync(BDRVQEDState *s);
 228 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 229                         BlockDriverCompletionFunc *cb, void *opaque);
 230 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 231                             unsigned int n);
 232 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 233                            uint64_t offset);
 234 void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
 235                        BlockDriverCompletionFunc *cb, void *opaque);
 236 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 237                         unsigned int index, unsigned int n, bool flush,
 238                         BlockDriverCompletionFunc *cb, void *opaque);
 239 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 240                             unsigned int index, unsigned int n, bool flush);
 241
 242 /**
 243  * Cluster functions
 244  */
 245 void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos,
 246                       size_t len, QEDFindClusterFunc *cb, void *opaque);
 247
 248 /**
 249  * Consistency check
 250  */
 251 int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
 252
 253 QEDTable *qed_alloc_table(BDRVQEDState *s);
 254
 255 /**
 256  * Round down to the start of a cluster
 257  */
 258 static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
 259 {
 260     return offset & ~(uint64_t)(s->header.cluster_size - 1);
 261 }
 262
 263 static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
 264 {
 265     return offset & (s->header.cluster_size - 1);
 266 }
 267
 268 static inline uint64_t qed_bytes_to_clusters(BDRVQEDState *s, uint64_t bytes)
 269 {
 270     return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
 271            (s->header.cluster_size - 1);
 272 }
 273
 274 static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
 275 {
 276     return pos >> s->l1_shift;
 277 }
 278
 279 static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
 280 {
 281     return (pos >> s->l2_shift) & s->l2_mask;
 282 }
 283
 284 /**
 285  * Test if a cluster offset is valid
 286  */
 287 static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
 288 {
 289     uint64_t header_size = (uint64_t)s->header.header_size *
 290                            s->header.cluster_size;
 291
 292     if (offset & (s->header.cluster_size - 1)) {
 293         return false;
 294     }
 295     return offset >= header_size && offset < s->file_size;
 296 }
 297
 298 /**
 299  * Test if a table offset is valid
 300  */
 301 static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
 302 {
 303     uint64_t end_offset = offset + (s->header.table_size - 1) *
 304                           s->header.cluster_size;
 305
 306     /* Overflow check */
 307     if (end_offset <= offset) {
 308         return false;
 309     }
 310
 311     return qed_check_cluster_offset(s, offset) &&
 312            qed_check_cluster_offset(s, end_offset);
 313 }
 314
 315 static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s,
 316                                                  uint64_t offset)
 317 {
 318     if (qed_offset_into_cluster(s, offset)) {
 319         return false;
 320     }
 321     return true;
 322 }
 323
 324 static inline bool qed_offset_is_unalloc_cluster(uint64_t offset)
 325 {
 326     if (offset == 0) {
 327         return true;
 328     }
 329     return false;
 330 }
 331
 332 static inline bool qed_offset_is_zero_cluster(uint64_t offset)
 333 {
 334     if (offset == 1) {
 335         return true;
 336     }
 337     return false;
 338 }
 339
 340 #endif /* BLOCK_QED_H */