qemu-block
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v2 3/5] parallels: Add checking and repairing duplicate offse


From: Denis V. Lunev
Subject: Re: [PATCH v2 3/5] parallels: Add checking and repairing duplicate offsets in BAT
Date: Tue, 31 Jan 2023 16:45:09 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101 Thunderbird/102.4.2

On 1/12/23 16:01, Alexander Ivanov wrote:
Cluster offsets must be unique among all the BAT entries. Find duplicate
offsets in the BAT and fix it by copying the content of the relevant
cluster to a newly allocated cluster and set the new cluster offset to the
duplicated entry.

Add host_cluster_index() and highest_offset() helpers to deduplicate the
code.

Move parallels_fix_leak() call to parallels_co_check() to fix both types
of leak: real corruption and a leak produced by allocate_clusters()
during deduplication.

Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
---
  block/parallels.c | 168 +++++++++++++++++++++++++++++++++++++++++-----
  1 file changed, 151 insertions(+), 17 deletions(-)

diff --git a/block/parallels.c b/block/parallels.c
index da1e75096c..73e992875a 100644
--- a/block/parallels.c
+++ b/block/parallels.c
@@ -136,6 +136,26 @@ static int cluster_remainder(BDRVParallelsState *s, 
int64_t sector_num,
      return MIN(nb_sectors, ret);
  }
+static uint32_t host_cluster_index(BDRVParallelsState *s, int64_t off)
+{
+    off -= s->header->data_off << BDRV_SECTOR_BITS;
+    return off / s->cluster_size;
+}
+
+static int64_t highest_offset(BDRVParallelsState *s)
+{
+    int64_t off, high_off = 0;
+    int i;
+
+    for (i = 0; i < s->bat_size; i++) {
+        off = bat2sect(s, i) << BDRV_SECTOR_BITS;
+        if (off > high_off) {
+            high_off = off;
+        }
+    }
+    return high_off;
+}
+
  static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
                              int nb_sectors, int *pnum)
  {
@@ -518,17 +538,9 @@ static int parallels_check_leak(BlockDriverState *bs,
                                  BdrvCheckMode fix)
  {
      BDRVParallelsState *s = bs->opaque;
-    int64_t off, high_off, count, leak_size;
-    uint32_t i;
-    int ret;
+    int64_t high_off, count, leak_size;
- high_off = 0;
-    for (i = 0; i < s->bat_size; i++) {
-        off = bat2sect(s, i) << BDRV_SECTOR_BITS;
-        if (off > high_off) {
-            high_off = off;
-        }
-    }
+    high_off = highest_offset(s);
res->image_end_offset = high_off + s->cluster_size; @@ -541,13 +553,6 @@ static int parallels_check_leak(BlockDriverState *bs,
          return 0;
      }
- if (fix & BDRV_FIX_LEAKS) {
-        ret = parallels_fix_leak(bs, res);
-        if (ret < 0) {
-            return ret;
-        }
-    }
-
      count = DIV_ROUND_UP(leak_size, s->cluster_size);
      fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
              fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR", leak_size);
@@ -560,6 +565,122 @@ static int parallels_check_leak(BlockDriverState *bs,
      return 0;
  }
+static int parallels_check_duplicate(BlockDriverState *bs,
+                                     BdrvCheckResult *res,
+                                     BdrvCheckMode *fix)
+{
+    BDRVParallelsState *s = bs->opaque;
+    QEMUIOVector qiov;
+    int64_t off, high_off, sector;
+    unsigned long *bitmap;
+    uint32_t i, bitmap_size, cluster_index;
+    int n, ret = 0;
+    uint64_t *buf = NULL;
+
+    high_off = highest_offset(s);
+    if (high_off == 0) {
+        return 0;
+    }
+
+    /*
+     * Create a bitmap of used clusters.
+     * If a bit is set, there is a BAT entry pointing to this cluster.
+     * Loop through the BAT entries, check bits relevant to an entry offset.
+     * If bit is set, this entry is duplicated. Otherwise set the bit.
+     *
+     * We shouldn't worry about newly allocated clusters outside the image
+     * because they are created higher then any existing cluster pointed by
+     * a BAT entry.
+     */
+    bitmap_size = host_cluster_index(s, high_off) + 1;
+    bitmap = bitmap_new(bitmap_size);
+
+    buf = qemu_memalign(4096, s->cluster_size);
+    qemu_iovec_init(&qiov, 0);
+    qemu_iovec_add(&qiov, buf, s->cluster_size);
+
+    for (i = 0; i < s->bat_size; i++) {
+        off = bat2sect(s, i) << BDRV_SECTOR_BITS;
+        if (off == 0) {
+            continue;
+        }
+
+        cluster_index = host_cluster_index(s, off);
+        if (test_bit(cluster_index, bitmap)) {
+            /* this cluster duplicates another one */
+            fprintf(stderr,
+                    "%s duplicate offset in BAT entry %u\n",
+                    *fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
+
+            res->corruptions++;
+
+            if (*fix & BDRV_FIX_ERRORS) {
+                /*
+                 * Reset the entry and allocate a new cluster
+                 * for the relevant guest offset. In this way we let
+                 * the lower layer to place the new cluster properly.
+                 * Copy the original cluster to the allocated one.
+                 */
+                parallels_set_bat_entry(s, i, 0);
+
+                ret = bdrv_co_pread(bs->file, off, s->cluster_size, buf, 0);
+                if (ret < 0) {
+                    res->check_errors++;
+                    goto out;
+                }
+
+                sector = (i * s->cluster_size) >> BDRV_SECTOR_BITS;
+                sector = allocate_clusters(bs, sector, s->tracks, &n);
+                if (sector < 0) {
+                    res->check_errors++;
+                    ret = sector;
+                    goto out;
+                }
+                off = sector << BDRV_SECTOR_BITS;
+                if (off > high_off) {
+                    high_off = off;
+                }
+
+                ret = bdrv_co_pwritev(bs->file, off, s->cluster_size, &qiov, 
0);
+                if (ret < 0) {
+                    res->check_errors++;
+                    goto out;
+                }
+
+                /*
+                 * In the future allocate_cluster() will reuse holed offsets
+                 * inside the image. Keep the used clusters bitmap content
+                 * consistent for the new allocated clusters too.
+                 *
+                 * Note, clusters allocated outside the current image are not
+                 * considered, and the bitmap size doesn't change.
+                 */
+                cluster_index = host_cluster_index(s, off);
+                if (cluster_index < bitmap_size) {
+                    bitmap_set(bitmap, cluster_index, 1);
+                }
+
+                /*
+                 * When new clusters are allocated, file size increases by
+                 * 128 Mb blocks. We need to truncate the file to the right
+                 * size. Let the leak fix code make its job.
+                 */
+                *fix |= BDRV_FIX_LEAKS;
+                res->corruptions_fixed++;
+            }
+            res->image_end_offset = high_off + s->cluster_size;
+        } else {
+            bitmap_set(bitmap, cluster_index, 1);
+        }
+    }
+
+out:
+    qemu_iovec_destroy(&qiov);
+    g_free(buf);
+    g_free(bitmap);
+    return ret;
+}
+
  static void parallels_collect_statistics(BlockDriverState *bs,
                                           BdrvCheckResult *res,
                                           BdrvCheckMode fix)
@@ -608,7 +729,20 @@ static int coroutine_fn 
parallels_co_check(BlockDriverState *bs,
              return ret;
          }
+ ret = parallels_check_duplicate(bs, res, &fix);
+        if (ret < 0) {
+            return ret;
+        }
+
          parallels_collect_statistics(bs, res, fix);
+
+        if (fix & BDRV_FIX_LEAKS &&
+            (res->corruptions_fixed || res->leaks_fixed)) {
+            ret = parallels_fix_leak(bs, res);
+            if (ret < 0) {
+                return ret;
+            }
+        }
      }
ret = bdrv_co_flush(bs);
I would be more happy if this patch will be split - helpers creation
is better to be separated from functional changes.

Den



reply via email to

[Prev in Thread] Current Thread [Next in Thread]