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    BlockDriverState *bs;
 133    QSIMPLEQ_ENTRY(QEDAIOCB) next;  /* next request */
 134    int flags;                      /* QED_AIOCB_* bits ORed together */
 135    uint64_t end_pos;               /* request end on block device, in bytes */
 136
 137    /* User scatter-gather list */
 138    QEMUIOVector *qiov;
 139    size_t qiov_offset;             /* byte count already processed */
 140
 141    /* Current cluster scatter-gather list */
 142    QEMUIOVector cur_qiov;
 143    uint64_t cur_pos;               /* position on block device, in bytes */
 144    uint64_t cur_cluster;           /* cluster offset in image file */
 145    unsigned int cur_nclusters;     /* number of clusters being accessed */
 146    int find_cluster_ret;           /* used for L1/L2 update */
 147
 148    QEDRequest request;
 149} QEDAIOCB;
 150
 151typedef struct {
 152    BlockDriverState *bs;           /* device */
 153
 154    /* Written only by an allocating write or the timer handler (the latter
 155     * while allocating reqs are plugged).
 156     */
 157    QEDHeader header;               /* always cpu-endian */
 158
 159    /* Protected by table_lock.  */
 160    CoMutex table_lock;
 161    QEDTable *l1_table;
 162    L2TableCache l2_cache;          /* l2 table cache */
 163    uint32_t table_nelems;
 164    uint32_t l1_shift;
 165    uint32_t l2_shift;
 166    uint32_t l2_mask;
 167    uint64_t file_size;             /* length of image file, in bytes */
 168
 169    /* Allocating write request queue */
 170    QEDAIOCB *allocating_acb;
 171    CoQueue allocating_write_reqs;
 172    bool allocating_write_reqs_plugged;
 173
 174    /* Periodic flush and clear need check flag */
 175    QEMUTimer *need_check_timer;
 176} BDRVQEDState;
 177
 178enum {
 179    QED_CLUSTER_FOUND,         /* cluster found */
 180    QED_CLUSTER_ZERO,          /* zero cluster found */
 181    QED_CLUSTER_L2,            /* cluster missing in L2 */
 182    QED_CLUSTER_L1,            /* cluster missing in L1 */
 183};
 184
 185/**
 186 * Header functions
 187 */
 188int qed_write_header_sync(BDRVQEDState *s);
 189
 190/**
 191 * L2 cache functions
 192 */
 193void qed_init_l2_cache(L2TableCache *l2_cache);
 194void qed_free_l2_cache(L2TableCache *l2_cache);
 195CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
 196void qed_unref_l2_cache_entry(CachedL2Table *entry);
 197CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
 198void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
 199
 200/**
 201 * Table I/O functions
 202 */
 203int coroutine_fn qed_read_l1_table_sync(BDRVQEDState *s);
 204int coroutine_fn qed_write_l1_table(BDRVQEDState *s, unsigned int index,
 205                                    unsigned int n);
 206int coroutine_fn qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 207                                         unsigned int n);
 208int coroutine_fn qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 209                                        uint64_t offset);
 210int coroutine_fn qed_read_l2_table(BDRVQEDState *s, QEDRequest *request,
 211                                   uint64_t offset);
 212int coroutine_fn qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 213                                    unsigned int index, unsigned int n,
 214                                    bool flush);
 215int coroutine_fn qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 216                                         unsigned int index, unsigned int n,
 217                                         bool flush);
 218
 219/**
 220 * Cluster functions
 221 */
 222int coroutine_fn qed_find_cluster(BDRVQEDState *s, QEDRequest *request,
 223                                  uint64_t pos, size_t *len,
 224                                  uint64_t *img_offset);
 225
 226/**
 227 * Consistency check
 228 */
 229int coroutine_fn qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
 230
 231QEDTable *qed_alloc_table(BDRVQEDState *s);
 232
 233/**
 234 * Round down to the start of a cluster
 235 */
 236static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
 237{
 238    return offset & ~(uint64_t)(s->header.cluster_size - 1);
 239}
 240
 241static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
 242{
 243    return offset & (s->header.cluster_size - 1);
 244}
 245
 246static inline uint64_t qed_bytes_to_clusters(BDRVQEDState *s, uint64_t bytes)
 247{
 248    return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
 249           (s->header.cluster_size - 1);
 250}
 251
 252static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
 253{
 254    return pos >> s->l1_shift;
 255}
 256
 257static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
 258{
 259    return (pos >> s->l2_shift) & s->l2_mask;
 260}
 261
 262/**
 263 * Test if a cluster offset is valid
 264 */
 265static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
 266{
 267    uint64_t header_size = (uint64_t)s->header.header_size *
 268                           s->header.cluster_size;
 269
 270    if (offset & (s->header.cluster_size - 1)) {
 271        return false;
 272    }
 273    return offset >= header_size && offset < s->file_size;
 274}
 275
 276/**
 277 * Test if a table offset is valid
 278 */
 279static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
 280{
 281    uint64_t end_offset = offset + (s->header.table_size - 1) *
 282                          s->header.cluster_size;
 283
 284    /* Overflow check */
 285    if (end_offset <= offset) {
 286        return false;
 287    }
 288
 289    return qed_check_cluster_offset(s, offset) &&
 290           qed_check_cluster_offset(s, end_offset);
 291}
 292
 293static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s,
 294                                                 uint64_t offset)
 295{
 296    if (qed_offset_into_cluster(s, offset)) {
 297        return false;
 298    }
 299    return true;
 300}
 301
 302static inline bool qed_offset_is_unalloc_cluster(uint64_t offset)
 303{
 304    if (offset == 0) {
 305        return true;
 306    }
 307    return false;
 308}
 309
 310static inline bool qed_offset_is_zero_cluster(uint64_t offset)
 311{
 312    if (offset == 1) {
 313        return true;
 314    }
 315    return false;
 316}
 317
 318#endif /* BLOCK_QED_H */
 319