LXR qemu/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/block_int.h"
  19#include "qemu/cutils.h"
  20
  21/* The layout of a QED file is as follows:
  22 *
  23 * +--------+----------+----------+----------+-----+
  24 * | header | L1 table | cluster0 | cluster1 | ... |
  25 * +--------+----------+----------+----------+-----+
  26 *
  27 * There is a 2-level pagetable for cluster allocation:
  28 *
  29 *                     +----------+
  30 *                     | L1 table |
  31 *                     +----------+
  32 *                ,------'  |  '------.
  33 *           +----------+   |    +----------+
  34 *           | L2 table |  ...   | L2 table |
  35 *           +----------+        +----------+
  36 *       ,------'  |  '------.
  37 *  +----------+   |    +----------+
  38 *  |   Data   |  ...   |   Data   |
  39 *  +----------+        +----------+
  40 *
  41 * The L1 table is fixed size and always present.  L2 tables are allocated on
  42 * demand.  The L1 table size determines the maximum possible image size; it
  43 * can be influenced using the cluster_size and table_size values.
  44 *
  45 * All fields are little-endian on disk.
  46 */
  47#define  QED_DEFAULT_CLUSTER_SIZE  65536
  48enum {
  49    QED_MAGIC = 'Q' | 'E' << 8 | 'D' << 16 | '\0' << 24,
  50
  51    /* The image supports a backing file */
  52    QED_F_BACKING_FILE = 0x01,
  53
  54    /* The image needs a consistency check before use */
  55    QED_F_NEED_CHECK = 0x02,
  56
  57    /* The backing file format must not be probed, treat as raw image */
  58    QED_F_BACKING_FORMAT_NO_PROBE = 0x04,
  59
  60    /* Feature bits must be used when the on-disk format changes */
  61    QED_FEATURE_MASK = QED_F_BACKING_FILE | /* supported feature bits */
  62                       QED_F_NEED_CHECK |
  63                       QED_F_BACKING_FORMAT_NO_PROBE,
  64    QED_COMPAT_FEATURE_MASK = 0,            /* supported compat feature bits */
  65    QED_AUTOCLEAR_FEATURE_MASK = 0,         /* supported autoclear feature bits */
  66
  67    /* Data is stored in groups of sectors called clusters.  Cluster size must
  68     * be large to avoid keeping too much metadata.  I/O requests that have
  69     * sub-cluster size will require read-modify-write.
  70     */
  71    QED_MIN_CLUSTER_SIZE = 4 * 1024, /* in bytes */
  72    QED_MAX_CLUSTER_SIZE = 64 * 1024 * 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
  86typedef 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} QEMU_PACKED QEDHeader;
 104
 105typedef struct {
 106    uint64_t offsets[0];            /* in bytes */
 107} QEDTable;
 108
 109/* The L2 cache is a simple write-through cache for L2 structures */
 110typedef 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
 117typedef struct {
 118    QTAILQ_HEAD(, CachedL2Table) entries;
 119    unsigned int n_entries;
 120} L2TableCache;
 121
 122typedef struct QEDRequest {
 123    CachedL2Table *l2_table;
 124} QEDRequest;
 125
 126enum {
 127    QED_AIOCB_WRITE = 0x0001,       /* read or write? */
 128    QED_AIOCB_ZERO  = 0x0002,       /* zero write, used with QED_AIOCB_WRITE */
 129};
 130
 131typedef struct QEDAIOCB {
 132    BlockAIOCB common;
 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    uint64_t end_pos;               /* request end on block device, in bytes */
 137
 138    /* User scatter-gather list */
 139    QEMUIOVector *qiov;
 140    size_t qiov_offset;             /* byte count already processed */
 141
 142    /* Current cluster scatter-gather list */
 143    QEMUIOVector cur_qiov;
 144    QEMUIOVector *backing_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
 153typedef 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} BDRVQEDState;
 172
 173enum {
 174    QED_CLUSTER_FOUND,         /* cluster found */
 175    QED_CLUSTER_ZERO,          /* zero cluster found */
 176    QED_CLUSTER_L2,            /* cluster missing in L2 */
 177    QED_CLUSTER_L1,            /* cluster missing in L1 */
 178};
 179
 180/**
 181 * qed_find_cluster() completion callback
 182 *
 183 * @opaque:     User data for completion callback
 184 * @ret:        QED_CLUSTER_FOUND   Success
 185 *              QED_CLUSTER_L2      Data cluster unallocated in L2
 186 *              QED_CLUSTER_L1      L2 unallocated in L1
 187 *              -errno              POSIX error occurred
 188 * @offset:     Data cluster offset
 189 * @len:        Contiguous bytes starting from cluster offset
 190 *
 191 * This function is invoked when qed_find_cluster() completes.
 192 *
 193 * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range
 194 * in the image file.
 195 *
 196 * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1
 197 * table offset, respectively.  len is number of contiguous unallocated bytes.
 198 */
 199typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
 200
 201/**
 202 * Generic callback for chaining async callbacks
 203 */
 204typedef struct {
 205    BlockCompletionFunc *cb;
 206    void *opaque;
 207} GenericCB;
 208
 209void *gencb_alloc(size_t len, BlockCompletionFunc *cb, void *opaque);
 210void gencb_complete(void *opaque, int ret);
 211
 212/**
 213 * Header functions
 214 */
 215int qed_write_header_sync(BDRVQEDState *s);
 216
 217/**
 218 * L2 cache functions
 219 */
 220void qed_init_l2_cache(L2TableCache *l2_cache);
 221void qed_free_l2_cache(L2TableCache *l2_cache);
 222CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
 223void qed_unref_l2_cache_entry(CachedL2Table *entry);
 224CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
 225void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
 226
 227/**
 228 * Table I/O functions
 229 */
 230int qed_read_l1_table_sync(BDRVQEDState *s);
 231void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 232                        BlockCompletionFunc *cb, void *opaque);
 233int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 234                            unsigned int n);
 235int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 236                           uint64_t offset);
 237void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
 238                       BlockCompletionFunc *cb, void *opaque);
 239void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 240                        unsigned int index, unsigned int n, bool flush,
 241                        BlockCompletionFunc *cb, void *opaque);
 242int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 243                            unsigned int index, unsigned int n, bool flush);
 244
 245/**
 246 * Cluster functions
 247 */
 248void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos,
 249                      size_t len, QEDFindClusterFunc *cb, void *opaque);
 250
 251/**
 252 * Consistency check
 253 */
 254int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
 255
 256QEDTable *qed_alloc_table(BDRVQEDState *s);
 257
 258/**
 259 * Round down to the start of a cluster
 260 */
 261static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
 262{
 263    return offset & ~(uint64_t)(s->header.cluster_size - 1);
 264}
 265
 266static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
 267{
 268    return offset & (s->header.cluster_size - 1);
 269}
 270
 271static inline uint64_t qed_bytes_to_clusters(BDRVQEDState *s, uint64_t bytes)
 272{
 273    return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
 274           (s->header.cluster_size - 1);
 275}
 276
 277static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
 278{
 279    return pos >> s->l1_shift;
 280}
 281
 282static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
 283{
 284    return (pos >> s->l2_shift) & s->l2_mask;
 285}
 286
 287/**
 288 * Test if a cluster offset is valid
 289 */
 290static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
 291{
 292    uint64_t header_size = (uint64_t)s->header.header_size *
 293                           s->header.cluster_size;
 294
 295    if (offset & (s->header.cluster_size - 1)) {
 296        return false;
 297    }
 298    return offset >= header_size && offset < s->file_size;
 299}
 300
 301/**
 302 * Test if a table offset is valid
 303 */
 304static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
 305{
 306    uint64_t end_offset = offset + (s->header.table_size - 1) *
 307                          s->header.cluster_size;
 308
 309    /* Overflow check */
 310    if (end_offset <= offset) {
 311        return false;
 312    }
 313
 314    return qed_check_cluster_offset(s, offset) &&
 315           qed_check_cluster_offset(s, end_offset);
 316}
 317
 318static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s,
 319                                                 uint64_t offset)
 320{
 321    if (qed_offset_into_cluster(s, offset)) {
 322        return false;
 323    }
 324    return true;
 325}
 326
 327static inline bool qed_offset_is_unalloc_cluster(uint64_t offset)
 328{
 329    if (offset == 0) {
 330        return true;
 331    }
 332    return false;
 333}
 334
 335static inline bool qed_offset_is_zero_cluster(uint64_t offset)
 336{
 337    if (offset == 1) {
 338        return true;
 339    }
 340    return false;
 341}
 342
 343#endif /* BLOCK_QED_H */
 344