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;
  25     int noffsets;
  26     int i, ret;
  27
  28     struct iovec iov = {
  29         .iov_base = table->offsets,
  30         .iov_len = s->header.cluster_size * s->header.table_size,
  31     };
  32     qemu_iovec_init_external(&qiov, &iov, 1);
  33
  34     trace_qed_read_table(s, offset, table);
  35
  36     qemu_co_mutex_unlock(&s->table_lock);
  37     ret = bdrv_preadv(s->bs->file, offset, &qiov);
  38     qemu_co_mutex_lock(&s->table_lock);
  39     if (ret < 0) {
  40         goto out;
  41     }
  42
  43     /* Byteswap offsets */
  44     noffsets = qiov.size / sizeof(uint64_t);
  45     for (i = 0; i < noffsets; i++) {
  46         table->offsets[i] = le64_to_cpu(table->offsets[i]);
  47     }
  48
  49     ret = 0;
  50 out:
  51     /* Completion */
  52     trace_qed_read_table_cb(s, table, ret);
  53     return ret;
  54 }
  55
  56 /**
  57  * Write out an updated part or all of a table
  58  *
  59  * @s:          QED state
  60  * @offset:     Offset of table in image file, in bytes
  61  * @table:      Table
  62  * @index:      Index of first element
  63  * @n:          Number of elements
  64  * @flush:      Whether or not to sync to disk
  65  *
  66  * Called with table_lock held.
  67  */
  68 static int qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
  69                            unsigned int index, unsigned int n, bool flush)
  70 {
  71     unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1;
  72     unsigned int start, end, i;
  73     QEDTable *new_table;
  74     struct iovec iov;
  75     QEMUIOVector qiov;
  76     size_t len_bytes;
  77     int ret;
  78
  79     trace_qed_write_table(s, offset, table, index, n);
  80
  81     /* Calculate indices of the first and one after last elements */
  82     start = index & ~sector_mask;
  83     end = (index + n + sector_mask) & ~sector_mask;
  84
  85     len_bytes = (end - start) * sizeof(uint64_t);
  86
  87     new_table = qemu_blockalign(s->bs, len_bytes);
  88     iov = (struct iovec) {
  89         .iov_base = new_table->offsets,
  90         .iov_len = len_bytes,
  91     };
  92     qemu_iovec_init_external(&qiov, &iov, 1);
  93
  94     /* Byteswap table */
  95     for (i = start; i < end; i++) {
  96         uint64_t le_offset = cpu_to_le64(table->offsets[i]);
  97         new_table->offsets[i - start] = le_offset;
  98     }
  99
 100     /* Adjust for offset into table */
 101     offset += start * sizeof(uint64_t);
 102
 103     qemu_co_mutex_unlock(&s->table_lock);
 104     ret = bdrv_pwritev(s->bs->file, offset, &qiov);
 105     qemu_co_mutex_lock(&s->table_lock);
 106     trace_qed_write_table_cb(s, table, flush, ret);
 107     if (ret < 0) {
 108         goto out;
 109     }
 110
 111     if (flush) {
 112         ret = bdrv_flush(s->bs);
 113         if (ret < 0) {
 114             goto out;
 115         }
 116     }
 117
 118     ret = 0;
 119 out:
 120     qemu_vfree(new_table);
 121     return ret;
 122 }
 123
 124 int qed_read_l1_table_sync(BDRVQEDState *s)
 125 {
 126     return qed_read_table(s, s->header.l1_table_offset, s->l1_table);
 127 }
 128
 129 /* Called with table_lock held.  */
 130 int qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n)
 131 {
 132     BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
 133     return qed_write_table(s, s->header.l1_table_offset,
 134                            s->l1_table, index, n, false);
 135 }
 136
 137 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 138                             unsigned int n)
 139 {
 140     return qed_write_l1_table(s, index, n);
 141 }
 142
 143 /* Called with table_lock held.  */
 144 int qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 145 {
 146     int ret;
 147
 148     qed_unref_l2_cache_entry(request->l2_table);
 149
 150     /* Check for cached L2 entry */
 151     request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 152     if (request->l2_table) {
 153         return 0;
 154     }
 155
 156     request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
 157     request->l2_table->table = qed_alloc_table(s);
 158
 159     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
 160     ret = qed_read_table(s, offset, request->l2_table->table);
 161
 162     if (ret) {
 163         /* can't trust loaded L2 table anymore */
 164         qed_unref_l2_cache_entry(request->l2_table);
 165         request->l2_table = NULL;
 166     } else {
 167         request->l2_table->offset = offset;
 168
 169         qed_commit_l2_cache_entry(&s->l2_cache, request->l2_table);
 170
 171         /* This is guaranteed to succeed because we just committed the entry
 172          * to the cache.
 173          */
 174         request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 175         assert(request->l2_table != NULL);
 176     }
 177
 178     return ret;
 179 }
 180
 181 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 182 {
 183     return qed_read_l2_table(s, request, offset);
 184 }
 185
 186 /* Called with table_lock held.  */
 187 int qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 188                        unsigned int index, unsigned int n, bool flush)
 189 {
 190     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
 191     return qed_write_table(s, request->l2_table->offset,
 192                            request->l2_table->table, index, n, flush);
 193 }
 194
 195 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 196                             unsigned int index, unsigned int n, bool flush)
 197 {
 198     return qed_write_l2_table(s, request, index, n, flush);
 199 }