block/qed-table.c

   1 /*
   2  * QEMU Enhanced Disk Format Table I/O
   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 #include "qemu/osdep.h"
  16 #include "trace.h"
  17 #include "qemu/sockets.h" /* for EINPROGRESS on Windows */
  18 #include "qed.h"
  19 #include "qemu/bswap.h"
  20
  21 /* Called with table_lock held.  */
  22 static int qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table)
  23 {
  24     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(
  25         qiov, table->offsets, s->header.cluster_size * s->header.table_size);
  26     int noffsets;
  27     int i, ret;
  28
  29     trace_qed_read_table(s, offset, table);
  30
  31     qemu_co_mutex_unlock(&s->table_lock);
  32     ret = bdrv_preadv(s->bs->file, offset, &qiov);
  33     qemu_co_mutex_lock(&s->table_lock);
  34     if (ret < 0) {
  35         goto out;
  36     }
  37
  38     /* Byteswap offsets */
  39     noffsets = qiov.size / sizeof(uint64_t);
  40     for (i = 0; i < noffsets; i++) {
  41         table->offsets[i] = le64_to_cpu(table->offsets[i]);
  42     }
  43
  44     ret = 0;
  45 out:
  46     /* Completion */
  47     trace_qed_read_table_cb(s, table, ret);
  48     return ret;
  49 }
  50
  51 /**
  52  * Write out an updated part or all of a table
  53  *
  54  * @s:          QED state
  55  * @offset:     Offset of table in image file, in bytes
  56  * @table:      Table
  57  * @index:      Index of first element
  58  * @n:          Number of elements
  59  * @flush:      Whether or not to sync to disk
  60  *
  61  * Called with table_lock held.
  62  */
  63 static int qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
  64                            unsigned int index, unsigned int n, bool flush)
  65 {
  66     unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1;
  67     unsigned int start, end, i;
  68     QEDTable *new_table;
  69     QEMUIOVector qiov;
  70     size_t len_bytes;
  71     int ret;
  72
  73     trace_qed_write_table(s, offset, table, index, n);
  74
  75     /* Calculate indices of the first and one after last elements */
  76     start = index & ~sector_mask;
  77     end = (index + n + sector_mask) & ~sector_mask;
  78
  79     len_bytes = (end - start) * sizeof(uint64_t);
  80
  81     new_table = qemu_blockalign(s->bs, len_bytes);
  82     qemu_iovec_init_buf(&qiov, new_table->offsets, len_bytes);
  83
  84     /* Byteswap table */
  85     for (i = start; i < end; i++) {
  86         uint64_t le_offset = cpu_to_le64(table->offsets[i]);
  87         new_table->offsets[i - start] = le_offset;
  88     }
  89
  90     /* Adjust for offset into table */
  91     offset += start * sizeof(uint64_t);
  92
  93     qemu_co_mutex_unlock(&s->table_lock);
  94     ret = bdrv_pwritev(s->bs->file, offset, &qiov);
  95     qemu_co_mutex_lock(&s->table_lock);
  96     trace_qed_write_table_cb(s, table, flush, ret);
  97     if (ret < 0) {
  98         goto out;
  99     }
 100
 101     if (flush) {
 102         ret = bdrv_flush(s->bs);
 103         if (ret < 0) {
 104             goto out;
 105         }
 106     }
 107
 108     ret = 0;
 109 out:
 110     qemu_vfree(new_table);
 111     return ret;
 112 }
 113
 114 int qed_read_l1_table_sync(BDRVQEDState *s)
 115 {
 116     return qed_read_table(s, s->header.l1_table_offset, s->l1_table);
 117 }
 118
 119 /* Called with table_lock held.  */
 120 int qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n)
 121 {
 122     BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
 123     return qed_write_table(s, s->header.l1_table_offset,
 124                            s->l1_table, index, n, false);
 125 }
 126
 127 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 128                             unsigned int n)
 129 {
 130     return qed_write_l1_table(s, index, n);
 131 }
 132
 133 /* Called with table_lock held.  */
 134 int qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 135 {
 136     int ret;
 137
 138     qed_unref_l2_cache_entry(request->l2_table);
 139
 140     /* Check for cached L2 entry */
 141     request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 142     if (request->l2_table) {
 143         return 0;
 144     }
 145
 146     request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
 147     request->l2_table->table = qed_alloc_table(s);
 148
 149     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
 150     ret = qed_read_table(s, offset, request->l2_table->table);
 151
 152     if (ret) {
 153         /* can't trust loaded L2 table anymore */
 154         qed_unref_l2_cache_entry(request->l2_table);
 155         request->l2_table = NULL;
 156     } else {
 157         request->l2_table->offset = offset;
 158
 159         qed_commit_l2_cache_entry(&s->l2_cache, request->l2_table);
 160
 161         /* This is guaranteed to succeed because we just committed the entry
 162          * to the cache.
 163          */
 164         request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 165         assert(request->l2_table != NULL);
 166     }
 167
 168     return ret;
 169 }
 170
 171 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 172 {
 173     return qed_read_l2_table(s, request, offset);
 174 }
 175
 176 /* Called with table_lock held.  */
 177 int qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 178                        unsigned int index, unsigned int n, bool flush)
 179 {
 180     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
 181     return qed_write_table(s, request->l2_table->offset,
 182                            request->l2_table->table, index, n, flush);
 183 }
 184
 185 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 186                             unsigned int index, unsigned int n, bool flush)
 187 {
 188     return qed_write_l2_table(s, request, index, n, flush);
 189 }