Change order of metadata update to prevent loosing guest data because of unexpec…

…ted exit (Gleb Natapov) Currently the order is this (during cow since it's the interesting case): 1. Decrement refcount of old clusters 2. Increment refcount for newly allocated clusters 3. Copy content of old sectors that will not be rewritten 4. Update L2 table with pointers to new clusters 5. Write guest data into new clusters (asynchronously) There are several problems with this order. The first one is that if qemu crashes (or killed or host reboots) after new clusters are linked into L2 table but before user data is written there, then on the next reboot guest will find neither old data nor new one in those sectors and this is not what gust expects even when journaling file system is in use. The other problem is that if qemu is killed between steps 1 and 4 then refcount of old cluster will be incorrect and may cause snapshot corruption. The patch change the order to be like this: 1. Increment refcount for newly allocated clusters 2. Write guest data into new clusters (asynchronously) 3. Copy content of old sectors that were not rewritten 4. Update L2 table with pointers to new clusters 5. Decrement refcount of old clusters Unexpected crash may cause cluster leakage, but guest data should be safe. Signed-off-by: Gleb Natapov <gleb@redhat.com> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@5861 c046a42c-6fe2-441c-8c8c-71466251a162

Change order of metadata update to prevent loosing guest data because of unexpec…
…ted exit (Gleb Natapov) Currently the order is this (during cow since it's the interesting case): 1. Decrement refcount of old clusters 2. Increment refcount for newly allocated clusters 3. Copy content of old sectors that will not be rewritten 4. Update L2 table with pointers to new clusters 5. Write guest data into new clusters (asynchronously) There are several problems with this order. The first one is that if qemu crashes (or killed or host reboots) after new clusters are linked into L2 table but before user data is written there, then on the next reboot guest will find neither old data nor new one in those sectors and this is not what gust expects even when journaling file system is in use. The other problem is that if qemu is killed between steps 1 and 4 then refcount of old cluster will be incorrect and may cause snapshot corruption. The patch change the order to be like this: 1. Increment refcount for newly allocated clusters 2. Write guest data into new clusters (asynchronously) 3. Copy content of old sectors that were not rewritten 4. Update L2 table with pointers to new clusters 5. Decrement refcount of old clusters Unexpected crash may cause cluster leakage, but guest data should be safe. Signed-off-by: Gleb Natapov <gleb@redhat.com> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@5861 c046a42c-6fe2-441c-8c8c-71466251a162
aliguori
1 parent 6db6c638
Showing 1 changed file with 92 additions and 63 deletions
block-qcow2.c
@@ -856,6 +856,70 @@ static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
     return cluster_offset;
 }
+typedef struct QCowL2Meta
+{
+    uint64_t offset;
+    int n_start;
+    int nb_available;
+    int nb_clusters;
+} QCowL2Meta;
+
+static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
+        QCowL2Meta *m)
+{
+    BDRVQcowState *s = bs->opaque;
+    int i, j = 0, l2_index, ret;
+    uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
+
+    if (m->nb_clusters == 0)
+        return 0;
+
+    if (!(old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t))))
+        return -ENOMEM;
+
+    /* copy content of unmodified sectors */
+    start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
+    if (m->n_start) {
+        ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
+        if (ret < 0)
+            goto err;
+    }
+
+    if (m->nb_available & (s->cluster_sectors - 1)) {
+        uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
+        ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
+                m->nb_available - end, s->cluster_sectors);
+        if (ret < 0)
+            goto err;
+    }
+
+    ret = -EIO;
+    /* update L2 table */
+    if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
+        goto err;
+
+    for (i = 0; i < m->nb_clusters; i++) {
+        if(l2_table[l2_index + i] != 0)
+            old_cluster[j++] = l2_table[l2_index + i];
+
+        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
+                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
+     }
+
+    if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
+                l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
+            m->nb_clusters * sizeof(uint64_t))
+        goto err;
+
+    for (i = 0; i < j; i++)
+        free_any_clusters(bs, old_cluster[i], 1);
+
+    ret = 0;
+err:
+    qemu_free(old_cluster);
+    return ret;
+ }
+
 /*
  * alloc_cluster_offset
  *
@@ -872,13 +936,12 @@ static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
                                      uint64_t offset,
                                      int n_start, int n_end,
-                                     int *num)
+                                     int *num, QCowL2Meta *m)
 {
     BDRVQcowState *s = bs->opaque;
     int l2_index, ret;
     uint64_t l2_offset, *l2_table, cluster_offset;
-    int nb_available, nb_clusters, i = 0;
-    uint64_t start_sect;
+    int nb_clusters, i = 0;
     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
     if (ret == 0)
@@ -886,8 +949,7 @@ static uint64_t alloc_cluster_offset(BlockDriverState *bs,
     nb_clusters = size_to_clusters(s, n_end << 9);
-    if (nb_clusters > s->l2_size - l2_index)
-            nb_clusters = s->l2_size - l2_index;
+    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
     cluster_offset = be64_to_cpu(l2_table[l2_index]);
@@ -897,11 +959,8 @@ static uint64_t alloc_cluster_offset(BlockDriverState *bs,
         nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
                 &l2_table[l2_index], 0);
-        nb_available = nb_clusters << (s->cluster_bits - 9);
-        if (nb_available > n_end)
-            nb_available = n_end;
-
         cluster_offset &= ~QCOW_OFLAG_COPIED;
+        m->nb_clusters = 0;
         goto out;
     }
@@ -914,7 +973,6 @@ static uint64_t alloc_cluster_offset(BlockDriverState *bs,
     /* how many available clusters ? */
     while (i < nb_clusters) {
-        int j;
         i += count_contiguous_free_clusters(nb_clusters - i,
                 &l2_table[l2_index + i]);
@@ -924,14 +982,9 @@ static uint64_t alloc_cluster_offset(BlockDriverState *bs,
                 (cluster_offset & QCOW_OFLAG_COMPRESSED))
             break;
-        j = count_contiguous_clusters(nb_clusters - i, s->cluster_size,
+        i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
                 &l2_table[l2_index + i], 0);
-        if (j)
-            free_any_clusters(bs, cluster_offset, j);
-
-        i += j;
-
         if(be64_to_cpu(l2_table[l2_index + i]))
             break;
     }
@@ -941,48 +994,15 @@ static uint64_t alloc_cluster_offset(BlockDriverState *bs,
     cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
-    /* we must initialize the cluster content which won't be
-       written */
-
-    nb_available = nb_clusters << (s->cluster_bits - 9);
-    if (nb_available > n_end)
-        nb_available = n_end;
-
-    /* copy content of unmodified sectors */
-
-    start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
-    if (n_start) {
-        ret = copy_sectors(bs, start_sect, cluster_offset, 0, n_start);
-        if (ret < 0)
-            return 0;
-    }
-
-    if (nb_available & (s->cluster_sectors - 1)) {
-        uint64_t end = nb_available & ~(uint64_t)(s->cluster_sectors - 1);
-        ret = copy_sectors(bs, start_sect + end,
-                           cluster_offset + (end << 9),
-                           nb_available - end,
-                           s->cluster_sectors);
-        if (ret < 0)
-            return 0;
-    }
-
-    /* update L2 table */
-
-    for (i = 0; i < nb_clusters; i++)
-        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
-                                             (i << s->cluster_bits)) |
-                                             QCOW_OFLAG_COPIED);
-
-    if (bdrv_pwrite(s->hd,
-                    l2_offset + l2_index * sizeof(uint64_t),
-                    l2_table + l2_index,
-                    nb_clusters * sizeof(uint64_t)) !=
-                    nb_clusters * sizeof(uint64_t))
-        return 0;
+    /* save info needed for meta data update */
+    m->offset = offset;
+    m->n_start = n_start;
+    m->nb_clusters = nb_clusters;
 out:
-    *num = nb_available - n_start;
+    m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
+
+    *num = m->nb_available - n_start;
     return cluster_offset;
 }
@@ -1113,6 +1133,7 @@ static int qcow_write(BlockDriverState *bs, int64_t sector_num,
     int ret, index_in_cluster, n;
     uint64_t cluster_offset;
     int n_end;
+    QCowL2Meta l2meta;
     while (nb_sectors > 0) {
         index_in_cluster = sector_num & (s->cluster_sectors - 1);
@@ -1122,7 +1143,7 @@ static int qcow_write(BlockDriverState *bs, int64_t sector_num,
             n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
         cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
                                               index_in_cluster,
-                                              n_end, &n);
+                                              n_end, &n, &l2meta);
         if (!cluster_offset)
             return -1;
         if (s->crypt_method) {
@@ -1133,8 +1154,10 @@ static int qcow_write(BlockDriverState *bs, int64_t sector_num,
         } else {
             ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
         }
-        if (ret != n * 512)
+        if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
+            free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
             return -1;
+        }
         nb_sectors -= n;
         sector_num += n;
         buf += n * 512;
@@ -1153,6 +1176,7 @@ typedef struct QCowAIOCB {
     uint8_t *cluster_data;
     BlockDriverAIOCB *hd_aiocb;
     QEMUBH *bh;
+    QCowL2Meta l2meta;
 } QCowAIOCB;
 static void qcow_aio_read_cb(void *opaque, int ret);
@@ -1281,6 +1305,7 @@ static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
     acb->nb_sectors = nb_sectors;
     acb->n = 0;
     acb->cluster_offset = 0;
+    acb->l2meta.nb_clusters = 0;
     return acb;
 }
@@ -1304,7 +1329,6 @@ static void qcow_aio_write_cb(void *opaque, int ret)
     BlockDriverState *bs = acb->common.bs;
     BDRVQcowState *s = bs->opaque;
     int index_in_cluster;
-    uint64_t cluster_offset;
     const uint8_t *src_buf;
     int n_end;
@@ -1317,6 +1341,11 @@ static void qcow_aio_write_cb(void *opaque, int ret)
         return;
     }
+    if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
+        free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
+        goto fail;
+    }
+
     acb->nb_sectors -= acb->n;
     acb->sector_num += acb->n;
     acb->buf += acb->n * 512;
@@ -1334,10 +1363,10 @@ static void qcow_aio_write_cb(void *opaque, int ret)
         n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
         n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
-    cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
+    acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
                                           index_in_cluster,
-                                          n_end, &acb->n);
-    if (!cluster_offset || (cluster_offset & 511) != 0) {
+                                          n_end, &acb->n, &acb->l2meta);
+    if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
         ret = -EIO;
         goto fail;
     }
@@ -1357,7 +1386,7 @@ static void qcow_aio_write_cb(void *opaque, int ret)
         src_buf = acb->buf;
     }
     acb->hd_aiocb = bdrv_aio_write(s->hd,
-                                   (cluster_offset >> 9) + index_in_cluster,
+                                   (acb->cluster_offset >> 9) + index_in_cluster,
                                    src_buf, acb->n,
                                    qcow_aio_write_cb, acb);
     if (acb->hd_aiocb == NULL)