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 typedef struct {
  56     GenericCB gencb;
  57     BDRVQEDState *s;
  58     QEDTable *orig_table;
  59     QEDTable *table;
  60     bool flush;             /* flush after write? */
  61
  62     struct iovec iov;
  63     QEMUIOVector qiov;
  64 } QEDWriteTableCB;
  65
  66 static void qed_write_table_cb(void *opaque, int ret)
  67 {
  68     QEDWriteTableCB *write_table_cb = opaque;
  69     BDRVQEDState *s = write_table_cb->s;
  70
  71     trace_qed_write_table_cb(s,
  72                              write_table_cb->orig_table,
  73                              write_table_cb->flush,
  74                              ret);
  75
  76     if (ret) {
  77         goto out;
  78     }
  79
  80     if (write_table_cb->flush) {
  81         /* We still need to flush first */
  82         write_table_cb->flush = false;
  83         qed_acquire(s);
  84         bdrv_aio_flush(write_table_cb->s->bs, qed_write_table_cb,
  85                        write_table_cb);
  86         qed_release(s);
  87         return;
  88     }
  89
  90 out:
  91     qemu_vfree(write_table_cb->table);
  92     gencb_complete(&write_table_cb->gencb, ret);
  93 }
  94
  95 /**
  96  * Write out an updated part or all of a table
  97  *
  98  * @s:          QED state
  99  * @offset:     Offset of table in image file, in bytes
 100  * @table:      Table
 101  * @index:      Index of first element
 102  * @n:          Number of elements
 103  * @flush:      Whether or not to sync to disk
 104  * @cb:         Completion function
 105  * @opaque:     Argument for completion function
 106  */
 107 static void qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
 108                             unsigned int index, unsigned int n, bool flush,
 109                             BlockCompletionFunc *cb, void *opaque)
 110 {
 111     QEDWriteTableCB *write_table_cb;
 112     unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1;
 113     unsigned int start, end, i;
 114     size_t len_bytes;
 115
 116     trace_qed_write_table(s, offset, table, index, n);
 117
 118     /* Calculate indices of the first and one after last elements */
 119     start = index & ~sector_mask;
 120     end = (index + n + sector_mask) & ~sector_mask;
 121
 122     len_bytes = (end - start) * sizeof(uint64_t);
 123
 124     write_table_cb = gencb_alloc(sizeof(*write_table_cb), cb, opaque);
 125     write_table_cb->s = s;
 126     write_table_cb->orig_table = table;
 127     write_table_cb->flush = flush;
 128     write_table_cb->table = qemu_blockalign(s->bs, len_bytes);
 129     write_table_cb->iov.iov_base = write_table_cb->table->offsets;
 130     write_table_cb->iov.iov_len = len_bytes;
 131     qemu_iovec_init_external(&write_table_cb->qiov, &write_table_cb->iov, 1);
 132
 133     /* Byteswap table */
 134     for (i = start; i < end; i++) {
 135         uint64_t le_offset = cpu_to_le64(table->offsets[i]);
 136         write_table_cb->table->offsets[i - start] = le_offset;
 137     }
 138
 139     /* Adjust for offset into table */
 140     offset += start * sizeof(uint64_t);
 141
 142     bdrv_aio_writev(s->bs->file, offset / BDRV_SECTOR_SIZE,
 143                     &write_table_cb->qiov,
 144                     write_table_cb->qiov.size / BDRV_SECTOR_SIZE,
 145                     qed_write_table_cb, write_table_cb);
 146 }
 147
 148 /**
 149  * Propagate return value from async callback
 150  */
 151 static void qed_sync_cb(void *opaque, int ret)
 152 {
 153     *(int *)opaque = ret;
 154 }
 155
 156 int qed_read_l1_table_sync(BDRVQEDState *s)
 157 {
 158     return qed_read_table(s, s->header.l1_table_offset, s->l1_table);
 159 }
 160
 161 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 162                         BlockCompletionFunc *cb, void *opaque)
 163 {
 164     BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
 165     qed_write_table(s, s->header.l1_table_offset,
 166                     s->l1_table, index, n, false, cb, opaque);
 167 }
 168
 169 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 170                             unsigned int n)
 171 {
 172     int ret = -EINPROGRESS;
 173
 174     qed_write_l1_table(s, index, n, qed_sync_cb, &ret);
 175     BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS);
 176
 177     return ret;
 178 }
 179
 180 int qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 181 {
 182     int ret;
 183
 184     qed_unref_l2_cache_entry(request->l2_table);
 185
 186     /* Check for cached L2 entry */
 187     request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 188     if (request->l2_table) {
 189         return 0;
 190     }
 191
 192     request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
 193     request->l2_table->table = qed_alloc_table(s);
 194
 195     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
 196     ret = qed_read_table(s, offset, request->l2_table->table);
 197
 198     qed_acquire(s);
 199     if (ret) {
 200         /* can't trust loaded L2 table anymore */
 201         qed_unref_l2_cache_entry(request->l2_table);
 202         request->l2_table = NULL;
 203     } else {
 204         request->l2_table->offset = offset;
 205
 206         qed_commit_l2_cache_entry(&s->l2_cache, request->l2_table);
 207
 208         /* This is guaranteed to succeed because we just committed the entry
 209          * to the cache.
 210          */
 211         request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 212         assert(request->l2_table != NULL);
 213     }
 214     qed_release(s);
 215
 216     return ret;
 217 }
 218
 219 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 220 {
 221     return qed_read_l2_table(s, request, offset);
 222 }
 223
 224 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 225                         unsigned int index, unsigned int n, bool flush,
 226                         BlockCompletionFunc *cb, void *opaque)
 227 {
 228     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
 229     qed_write_table(s, request->l2_table->offset,
 230                     request->l2_table->table, index, n, flush, cb, opaque);
 231 }
 232
 233 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 234                             unsigned int index, unsigned int n, bool flush)
 235 {
 236     int ret = -EINPROGRESS;
 237
 238     qed_write_l2_table(s, request, index, n, flush, qed_sync_cb, &ret);
 239     BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS);
 240
 241     return ret;
 242 }