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
  20 typedef struct {
  21     GenericCB gencb;
  22     BDRVQEDState *s;
  23     QEDTable *table;
  24
  25     struct iovec iov;
  26     QEMUIOVector qiov;
  27 } QEDReadTableCB;
  28
  29 static void qed_read_table_cb(void *opaque, int ret)
  30 {
  31     QEDReadTableCB *read_table_cb = opaque;
  32     QEDTable *table = read_table_cb->table;
  33     int noffsets = read_table_cb->qiov.size / sizeof(uint64_t);
  34     int i;
  35
  36     /* Handle I/O error */
  37     if (ret) {
  38         goto out;
  39     }
  40
  41     /* Byteswap offsets */
  42     for (i = 0; i < noffsets; i++) {
  43         table->offsets[i] = le64_to_cpu(table->offsets[i]);
  44     }
  45
  46 out:
  47     /* Completion */
  48     trace_qed_read_table_cb(read_table_cb->s, read_table_cb->table, ret);
  49     gencb_complete(&read_table_cb->gencb, ret);
  50 }
  51
  52 static void qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
  53                            BlockCompletionFunc *cb, void *opaque)
  54 {
  55     QEDReadTableCB *read_table_cb = gencb_alloc(sizeof(*read_table_cb),
  56                                                 cb, opaque);
  57     QEMUIOVector *qiov = &read_table_cb->qiov;
  58
  59     trace_qed_read_table(s, offset, table);
  60
  61     read_table_cb->s = s;
  62     read_table_cb->table = table;
  63     read_table_cb->iov.iov_base = table->offsets,
  64     read_table_cb->iov.iov_len = s->header.cluster_size * s->header.table_size,
  65
  66     qemu_iovec_init_external(qiov, &read_table_cb->iov, 1);
  67     bdrv_aio_readv(s->bs->file->bs, offset / BDRV_SECTOR_SIZE, qiov,
  68                    qiov->size / BDRV_SECTOR_SIZE,
  69                    qed_read_table_cb, read_table_cb);
  70 }
  71
  72 typedef struct {
  73     GenericCB gencb;
  74     BDRVQEDState *s;
  75     QEDTable *orig_table;
  76     QEDTable *table;
  77     bool flush;             /* flush after write? */
  78
  79     struct iovec iov;
  80     QEMUIOVector qiov;
  81 } QEDWriteTableCB;
  82
  83 static void qed_write_table_cb(void *opaque, int ret)
  84 {
  85     QEDWriteTableCB *write_table_cb = opaque;
  86
  87     trace_qed_write_table_cb(write_table_cb->s,
  88                              write_table_cb->orig_table,
  89                              write_table_cb->flush,
  90                              ret);
  91
  92     if (ret) {
  93         goto out;
  94     }
  95
  96     if (write_table_cb->flush) {
  97         /* We still need to flush first */
  98         write_table_cb->flush = false;
  99         bdrv_aio_flush(write_table_cb->s->bs, qed_write_table_cb,
 100                        write_table_cb);
 101         return;
 102     }
 103
 104 out:
 105     qemu_vfree(write_table_cb->table);
 106     gencb_complete(&write_table_cb->gencb, ret);
 107 }
 108
 109 /**
 110  * Write out an updated part or all of a table
 111  *
 112  * @s:          QED state
 113  * @offset:     Offset of table in image file, in bytes
 114  * @table:      Table
 115  * @index:      Index of first element
 116  * @n:          Number of elements
 117  * @flush:      Whether or not to sync to disk
 118  * @cb:         Completion function
 119  * @opaque:     Argument for completion function
 120  */
 121 static void qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
 122                             unsigned int index, unsigned int n, bool flush,
 123                             BlockCompletionFunc *cb, void *opaque)
 124 {
 125     QEDWriteTableCB *write_table_cb;
 126     unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1;
 127     unsigned int start, end, i;
 128     size_t len_bytes;
 129
 130     trace_qed_write_table(s, offset, table, index, n);
 131
 132     /* Calculate indices of the first and one after last elements */
 133     start = index & ~sector_mask;
 134     end = (index + n + sector_mask) & ~sector_mask;
 135
 136     len_bytes = (end - start) * sizeof(uint64_t);
 137
 138     write_table_cb = gencb_alloc(sizeof(*write_table_cb), cb, opaque);
 139     write_table_cb->s = s;
 140     write_table_cb->orig_table = table;
 141     write_table_cb->flush = flush;
 142     write_table_cb->table = qemu_blockalign(s->bs, len_bytes);
 143     write_table_cb->iov.iov_base = write_table_cb->table->offsets;
 144     write_table_cb->iov.iov_len = len_bytes;
 145     qemu_iovec_init_external(&write_table_cb->qiov, &write_table_cb->iov, 1);
 146
 147     /* Byteswap table */
 148     for (i = start; i < end; i++) {
 149         uint64_t le_offset = cpu_to_le64(table->offsets[i]);
 150         write_table_cb->table->offsets[i - start] = le_offset;
 151     }
 152
 153     /* Adjust for offset into table */
 154     offset += start * sizeof(uint64_t);
 155
 156     bdrv_aio_writev(s->bs->file->bs, offset / BDRV_SECTOR_SIZE,
 157                     &write_table_cb->qiov,
 158                     write_table_cb->qiov.size / BDRV_SECTOR_SIZE,
 159                     qed_write_table_cb, write_table_cb);
 160 }
 161
 162 /**
 163  * Propagate return value from async callback
 164  */
 165 static void qed_sync_cb(void *opaque, int ret)
 166 {
 167     *(int *)opaque = ret;
 168 }
 169
 170 int qed_read_l1_table_sync(BDRVQEDState *s)
 171 {
 172     int ret = -EINPROGRESS;
 173
 174     qed_read_table(s, s->header.l1_table_offset,
 175                    s->l1_table, qed_sync_cb, &ret);
 176     while (ret == -EINPROGRESS) {
 177         aio_poll(bdrv_get_aio_context(s->bs), true);
 178     }
 179
 180     return ret;
 181 }
 182
 183 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
 184                         BlockCompletionFunc *cb, void *opaque)
 185 {
 186     BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
 187     qed_write_table(s, s->header.l1_table_offset,
 188                     s->l1_table, index, n, false, cb, opaque);
 189 }
 190
 191 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
 192                             unsigned int n)
 193 {
 194     int ret = -EINPROGRESS;
 195
 196     qed_write_l1_table(s, index, n, qed_sync_cb, &ret);
 197     while (ret == -EINPROGRESS) {
 198         aio_poll(bdrv_get_aio_context(s->bs), true);
 199     }
 200
 201     return ret;
 202 }
 203
 204 typedef struct {
 205     GenericCB gencb;
 206     BDRVQEDState *s;
 207     uint64_t l2_offset;
 208     QEDRequest *request;
 209 } QEDReadL2TableCB;
 210
 211 static void qed_read_l2_table_cb(void *opaque, int ret)
 212 {
 213     QEDReadL2TableCB *read_l2_table_cb = opaque;
 214     QEDRequest *request = read_l2_table_cb->request;
 215     BDRVQEDState *s = read_l2_table_cb->s;
 216     CachedL2Table *l2_table = request->l2_table;
 217     uint64_t l2_offset = read_l2_table_cb->l2_offset;
 218
 219     if (ret) {
 220         /* can't trust loaded L2 table anymore */
 221         qed_unref_l2_cache_entry(l2_table);
 222         request->l2_table = NULL;
 223     } else {
 224         l2_table->offset = l2_offset;
 225
 226         qed_commit_l2_cache_entry(&s->l2_cache, l2_table);
 227
 228         /* This is guaranteed to succeed because we just committed the entry
 229          * to the cache.
 230          */
 231         request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
 232         assert(request->l2_table != NULL);
 233     }
 234
 235     gencb_complete(&read_l2_table_cb->gencb, ret);
 236 }
 237
 238 void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
 239                        BlockCompletionFunc *cb, void *opaque)
 240 {
 241     QEDReadL2TableCB *read_l2_table_cb;
 242
 243     qed_unref_l2_cache_entry(request->l2_table);
 244
 245     /* Check for cached L2 entry */
 246     request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
 247     if (request->l2_table) {
 248         cb(opaque, 0);
 249         return;
 250     }
 251
 252     request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
 253     request->l2_table->table = qed_alloc_table(s);
 254
 255     read_l2_table_cb = gencb_alloc(sizeof(*read_l2_table_cb), cb, opaque);
 256     read_l2_table_cb->s = s;
 257     read_l2_table_cb->l2_offset = offset;
 258     read_l2_table_cb->request = request;
 259
 260     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
 261     qed_read_table(s, offset, request->l2_table->table,
 262                    qed_read_l2_table_cb, read_l2_table_cb);
 263 }
 264
 265 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
 266 {
 267     int ret = -EINPROGRESS;
 268
 269     qed_read_l2_table(s, request, offset, qed_sync_cb, &ret);
 270     while (ret == -EINPROGRESS) {
 271         aio_poll(bdrv_get_aio_context(s->bs), true);
 272     }
 273
 274     return ret;
 275 }
 276
 277 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
 278                         unsigned int index, unsigned int n, bool flush,
 279                         BlockCompletionFunc *cb, void *opaque)
 280 {
 281     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
 282     qed_write_table(s, request->l2_table->offset,
 283                     request->l2_table->table, index, n, flush, cb, opaque);
 284 }
 285
 286 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
 287                             unsigned int index, unsigned int n, bool flush)
 288 {
 289     int ret = -EINPROGRESS;
 290
 291     qed_write_l2_table(s, request, index, n, flush, qed_sync_cb, &ret);
 292     while (ret == -EINPROGRESS) {
 293         aio_poll(bdrv_get_aio_context(s->bs), true);
 294     }
 295
 296     return ret;
 297 }