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 static int qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table)
  22 {
  23     QEMUIOVector qiov;
  24     int noffsets;
  25     int i, ret;
  26
  27     struct iovec iov = {
  28         .iov_base = table->offsets,
  29         .iov_len = s->header.cluster_size * s->header.table_size,
  30     };
  31     qemu_iovec_init_external(&qiov, &iov, 1);
  32
  33     trace_qed_read_table(s, offset, table);
  34
  35     ret = bdrv_preadv(s->bs->file, offset, &qiov);
  36     if (ret < 0) {
  37         goto out;
  38     }
  39
  40     /* Byteswap offsets */
  41     qed_acquire(s);
  42     noffsets = qiov.size / sizeof(uint64_t);
  43     for (i = 0; i < noffsets; i++) {
  44         table->offsets[i] = le64_to_cpu(table->offsets[i]);
  45     }
  46     qed_release(s);
  47
  48     ret = 0;
  49 out:
  50     /* Completion */
  51     trace_qed_read_table_cb(s, table, ret);
  52     return ret;
  53 }
  54
  55 /**
  56  * Write out an updated part or all of a table
  57  *
  58  * @s:          QED state
  59  * @offset:     Offset of table in image file, in bytes
  60  * @table:      Table
  61  * @index:      Index of first element
  62  * @n:          Number of elements
  63  * @flush:      Whether or not to sync to disk
  64  * @cb:         Completion function
  65  * @opaque:     Argument for completion function
  66  */
  67 static void qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
  68                             unsigned int index, unsigned int n, bool flush,
  69                             BlockCompletionFunc *cb, void *opaque)
  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     ret = bdrv_pwritev(s->bs->file, offset, &qiov);
 104     trace_qed_write_table_cb(s, table, flush, ret);
 105     if (ret < 0) {
 106         goto out;
 107     }
 108
 109     if (flush) {
 110         qed_acquire(s);
 111         ret = bdrv_flush(s->bs);
 112         qed_release(s);
 113         if (ret < 0) {
 114             goto out;
 115         }
 116     }
 117
 118     ret = 0;
 119 out:
 120     qemu_vfree(new_table);
 121     cb(opaque, ret);
 122 }
 123
 124 /**
 125  * Propagate return value from async callback
 126  */
 127 static void qed_sync_cb(void *opaque, int ret)
 128 {
 129     *(int *)opaque = ret;
 130 }
 131
 132 int qed_read_l1_table_sync(BDRVQEDState *s)
 133 {
 134     return qed_read_table(s, s->header.l1_table_offset, s->l1_table);
 135 }
 136
 137 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 138                         BlockCompletionFunc *cb, void *opaque)
 139 {
 140     BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
 141     qed_write_table(s, s->header.l1_table_offset,
 142                     s->l1_table, index, n, false, cb, opaque);
 143 }
 144
 145 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 146                             unsigned int n)
 147 {
 148     int ret = -EINPROGRESS;
 149
 150     qed_write_l1_table(s, index, n, qed_sync_cb, &ret);
 151     BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS);
 152
 153     return ret;
 154 }
 155
 156 int qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 157 {
 158     int ret;
 159
 160     qed_unref_l2_cache_entry(request->l2_table);
 161
 162     /* Check for cached L2 entry */
 163     request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 164     if (request->l2_table) {
 165         return 0;
 166     }
 167
 168     request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
 169     request->l2_table->table = qed_alloc_table(s);
 170
 171     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
 172     ret = qed_read_table(s, offset, request->l2_table->table);
 173
 174     qed_acquire(s);
 175     if (ret) {
 176         /* can't trust loaded L2 table anymore */
 177         qed_unref_l2_cache_entry(request->l2_table);
 178         request->l2_table = NULL;
 179     } else {
 180         request->l2_table->offset = offset;
 181
 182         qed_commit_l2_cache_entry(&s->l2_cache, request->l2_table);
 183
 184         /* This is guaranteed to succeed because we just committed the entry
 185          * to the cache.
 186          */
 187         request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 188         assert(request->l2_table != NULL);
 189     }
 190     qed_release(s);
 191
 192     return ret;
 193 }
 194
 195 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 196 {
 197     return qed_read_l2_table(s, request, offset);
 198 }
 199
 200 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 201                         unsigned int index, unsigned int n, bool flush,
 202                         BlockCompletionFunc *cb, void *opaque)
 203 {
 204     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
 205     qed_write_table(s, request->l2_table->offset,
 206                     request->l2_table->table, index, n, flush, cb, opaque);
 207 }
 208
 209 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 210                             unsigned int index, unsigned int n, bool flush)
 211 {
 212     int ret = -EINPROGRESS;
 213
 214     qed_write_l2_table(s, request, index, n, flush, qed_sync_cb, &ret);
 215     BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS);
 216
 217     return ret;
 218 }