Skip to contentSkip to conversation

qcow2 check improvements part I

PATCH Superseded Vladimir Sementsov-Ogievskiy 3 participants 15 messages Dec 14, 2018 – Feb 27, 2019 Block layer

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover qcow2 check improvements part I

Hi all!

These series are derived from "[PATCH v3 0/8] qcow2 check improvements"
(https://lists.gnu.org/archive/html/qemu-devel/2018-10/msg02284.html)

Some patches are new, some patches are not included and are postponed to part II.

So, here:
01,03 - unchanged
02: handle broken iotest
04-05: new

Vladimir Sementsov-Ogievskiy (5):
qcow2-refcount: fix check_oflag_copied
qcow2-refcount: avoid eating RAM
qcow2-refcount: check_refcounts_l2: reduce ignored overlaps
qcow2-refcount: check_refcounts_l2: don't count fixed cluster as
allocated
qcow2-refcount: don't mask corruptions under internal errors

 block/qcow2-refcount.c     | 87 +++++++++++++++++++++++---------------
 tests/qemu-iotests/138     | 12 +++---
 tests/qemu-iotests/138.out |  5 ++-
 3 files changed, 61 insertions(+), 43 deletions(-)
Signature · 1 line
2.18.0
#1 lore raw

PATCH v4 1/5 qcow2-refcount: fix check_oflag_copied

Increase corruptions_fixed only after successful fix.

---
 block/qcow2-refcount.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
block/qcow2-refcount.c +4 -4
index 1c63ac244a..b717fdecc6 100644
@@ -1822,7 +1822,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
     for (i = 0; i < s->l1_size; i++) {
         uint64_t l1_entry = s->l1_table[i];
         uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
-        bool l2_dirty = false;
+        int l2_dirty = 0;
 
         if (!l2_offset) {
             continue;
@@ -1884,8 +1884,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                         l2_table[j] = cpu_to_be64(refcount == 1
                                     ? l2_entry |  QCOW_OFLAG_COPIED
                                     : l2_entry & ~QCOW_OFLAG_COPIED);
-                        l2_dirty = true;
-                        res->corruptions_fixed++;
+                        l2_dirty++;
                     } else {
                         res->corruptions++;
                     }
@@ -1893,7 +1892,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
             }
         }
 
-        if (l2_dirty) {
+        if (l2_dirty > 0) {
             ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
                                                 l2_offset, s->cluster_size);
             if (ret < 0) {
@@ -1911,6 +1910,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                 res->check_errors++;
                 goto fail;
             }
+            res->corruptions_fixed += l2_dirty;
         }
     }
 
Signature · 1 line
2.18.0
#3 lore raw

PATCH v4 2/5 qcow2-refcount: avoid eating RAM

In reply to Vladimir Sementsov-Ogievskiy

qcow2_inc_refcounts_imrt() (through realloc_refcount_array()) can eat an unpredictable amount of memory on corrupted table entries, which are referencing regions far beyond the end of file.

Prevent this, by skipping such regions from further processing.

Interesting that iotest 138 checks exactly the behavior which we fix here. So, change the test appropriately.

---
 block/qcow2-refcount.c     | 18 ++++++++++++++++++
 tests/qemu-iotests/138     | 12 +++++-------
 tests/qemu-iotests/138.out |  5 ++++-
 3 files changed, 27 insertions(+), 8 deletions(-)
block/qcow2-refcount.c +18 -0
index b717fdecc6..c41c1fbe00 100644
@@ -1505,12 +1505,30 @@ int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
 {
     BDRVQcow2State *s = bs->opaque;
     uint64_t start, last, cluster_offset, k, refcount;
+    int64_t file_len;
     int ret;
 
     if (size <= 0) {
         return 0;
     }
 
+    file_len = bdrv_getlength(bs->file->bs);
+    if (file_len < 0) {
+        return file_len;
+    }
+
+    /* Last cluster of qcow2 image may be semi-allocated, so it's may be OK to
+     * reference some space after file end but it should be less than one
+     * cluster.
+     */
+    if (offset + size - file_len >= s->cluster_size) {
+        fprintf(stderr, "ERROR: counting reference for region exceeding the "
+                "end of the file by one cluster or more: offset 0x%" PRIx64
+                " size 0x%" PRIx64 "\n", offset, size);
+        res->corruptions++;
+        return 0;
+    }
+
     start = start_of_cluster(s, offset);
     last = start_of_cluster(s, offset + size - 1);
     for(cluster_offset = start; cluster_offset <= last;
tests/qemu-iotests/138 +5 -7
index eccbcae3a6..12e20762e5 100755
@@ -54,15 +54,13 @@ $QEMU_IO -c 'write 0 512' "$TEST_IMG" | _filter_qemu_io
 # Put the data cluster at a multiple of 2 TB, resulting in the image apparently
 # having a multiple of 2^32 clusters
 # (To be more specific: It is at 32 PB)
-poke_file "$TEST_IMG" 2048 "\x80\x80\x00\x00\x00\x00\x00\x00"
+poke_file "$TEST_IMG" $((2048 + 8)) "\x00\x80\x00\x00\x00\x00\x00\x00"
 
 # An offset of 32 PB results in qemu-img check having to allocate an in-memory
-# refcount table of 128 TB (16 bit refcounts, 512 byte clusters).
-# This should be generally too much for any system and thus fail.
-# What this test is checking is that the qcow2 driver actually tries to allocate
-# such a large amount of memory (and is consequently aborting) instead of having
-# truncated the cluster count somewhere (which would result in much less memory
-# being allocated and then a segfault occurring).
+# refcount table of 128 TB (16 bit refcounts, 512 byte clusters), if qemu-img
+# don't check that referenced data cluster is far beyond the end of file.
+# But starting from 4.0, qemu-img does this check, and instead of "Cannot
+# allocate memory", we have an error showing that l2 entry is invalid.
 _check_test_img
 
 # success, all done
tests/qemu-iotests/138.out +4 -1
index 3fe911f85a..aca7d47a80 100644
@@ -5,5 +5,8 @@ QA output created by 138
 Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=512
 wrote 512/512 bytes at offset 0
 512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-qemu-img: Check failed: Cannot allocate memory
+ERROR: counting reference for region exceeding the end of the file by one cluster or more: offset 0x80000000000000 size 0x200
+
+1 errors were found on the image.
+Data may be corrupted, or further writes to the image may corrupt it.
 *** done
Signature · 1 line
2.18.0
Reviewed-by on v4 2/5
#4 lore raw

Re: [Qemu-devel] [PATCH v4 2/5] qcow2-refcount: avoid eating RAM

Quoted 5 lines · Vladimir Sementsov-Ogievskiy

qcow2_inc_refcounts_imrt() (through realloc_refcount_array()) can eat an unpredictable amount of memory on corrupted table entries, which are referencing regions far beyond the end of file.

Prevent this, by skipping such regions from further processing.

Is it actually wrong to reference clusters beyond the EOF?

Hmm... The spec says everywhere "offset into the image file" (except for L2 entries pointing to data clusters, but let's say that's just a mistake). This implies that offsets have to be within the confines of the file, OK.

Quoted 16 lines

Interesting that iotest 138 checks exactly the behavior which we fix here. So, change the test appropriately.

---
 block/qcow2-refcount.c     | 18 ++++++++++++++++++
 tests/qemu-iotests/138     | 12 +++++-------
 tests/qemu-iotests/138.out |  5 ++++-
 3 files changed, 27 insertions(+), 8 deletions(-)
block/qcow2-refcount.c +0 -0
index b717fdecc6..c41c1fbe00 100644
@@ -1505,12 +1505,30 @@ int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
 {

block/qcow2-refcount.c int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, · line 1515

Quoted 10 lines
block/qcow2-refcount.c +2 -0
     BDRVQcow2State *s = bs->opaque;
     uint64_t start, last, cluster_offset, k, refcount;
+    int64_t file_len;
     int ret;
 
     if (size <= 0) {
         return 0;
     }
 
+    file_len = bdrv_getlength(bs->file->bs);

I just had to delete a paragraph where I was questioning the use of bdrv_getlength() here again, because I remembered I did that before and decided to give in...

block/qcow2-refcount.c int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, · line 1520

Quoted 5 lines
block/qcow2-refcount.c +5 -0
+    if (file_len < 0) {
+        return file_len;
+    }
+
+    /* Last cluster of qcow2 image may be semi-allocated, so it's may be OK to

With s/it's/it/:

Quoted 54 lines
block/qcow2-refcount.c +11 -0
+     * reference some space after file end but it should be less than one
+     * cluster.
+     */
+    if (offset + size - file_len >= s->cluster_size) {
+        fprintf(stderr, "ERROR: counting reference for region exceeding the "
+                "end of the file by one cluster or more: offset 0x%" PRIx64
+                " size 0x%" PRIx64 "\n", offset, size);
+        res->corruptions++;
+        return 0;
+    }
+
     start = start_of_cluster(s, offset);
     last = start_of_cluster(s, offset + size - 1);
     for(cluster_offset = start; cluster_offset <= last;
tests/qemu-iotests/138 +5 -7
index eccbcae3a6..12e20762e5 100755
@@ -54,15 +54,13 @@ $QEMU_IO -c 'write 0 512' "$TEST_IMG" | _filter_qemu_io
 # Put the data cluster at a multiple of 2 TB, resulting in the image apparently
 # having a multiple of 2^32 clusters
 # (To be more specific: It is at 32 PB)
-poke_file "$TEST_IMG" 2048 "\x80\x80\x00\x00\x00\x00\x00\x00"
+poke_file "$TEST_IMG" $((2048 + 8)) "\x00\x80\x00\x00\x00\x00\x00\x00"
 
 # An offset of 32 PB results in qemu-img check having to allocate an in-memory
-# refcount table of 128 TB (16 bit refcounts, 512 byte clusters).
-# This should be generally too much for any system and thus fail.
-# What this test is checking is that the qcow2 driver actually tries to allocate
-# such a large amount of memory (and is consequently aborting) instead of having
-# truncated the cluster count somewhere (which would result in much less memory
-# being allocated and then a segfault occurring).
+# refcount table of 128 TB (16 bit refcounts, 512 byte clusters), if qemu-img
+# don't check that referenced data cluster is far beyond the end of file.
+# But starting from 4.0, qemu-img does this check, and instead of "Cannot
+# allocate memory", we have an error showing that l2 entry is invalid.
 _check_test_img
 
 # success, all done
tests/qemu-iotests/138.out +4 -1
index 3fe911f85a..aca7d47a80 100644
@@ -5,5 +5,8 @@ QA output created by 138
 Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=512
 wrote 512/512 bytes at offset 0
 512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-qemu-img: Check failed: Cannot allocate memory
+ERROR: counting reference for region exceeding the end of the file by one cluster or more: offset 0x80000000000000 size 0x200
+
+1 errors were found on the image.
+Data may be corrupted, or further writes to the image may corrupt it.
 *** done
#5 lore raw

Re: [Qemu-devel] [PATCH v4 2/5] qcow2-refcount: avoid eating RAM

In reply to Vladimir Sementsov-Ogievskiy
Quoted 12 lines · Vladimir Sementsov-Ogievskiy

qcow2_inc_refcounts_imrt() (through realloc_refcount_array()) can eat an unpredictable amount of memory on corrupted table entries, which are referencing regions far beyond the end of file.

Prevent this, by skipping such regions from further processing.

Interesting that iotest 138 checks exactly the behavior which we fix here. So, change the test appropriately.

---

block/qcow2-refcount.c int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, · line 1522

Quoted 3 lines
block/qcow2-refcount.c +3 -0
+    /* Last cluster of qcow2 image may be semi-allocated, so it's may be OK to
+     * reference some space after file end but it should be less than one
+     * cluster.

Are we guaranteed that this is always the case, even when incrementally building up an image. That is, if we get a power failure in the middle of allocating both a new L2 table and a larger refcount table, can we see an image which temporary references 2 or more clusters beyond the end of the file that it was previously using? Or are we guaranteed that the way we update the refcount table will never see more than one cluster beyond the end of the file for any other reference? I guess where I'm going with this is a question of whether we should permit a finite overrun to account for multiple pieces of metadata all being in flight at once, and only reject the obvious overruns that are definitely beyond that sane limit, and whether 1 cluster is too small for that sane limit.

block/qcow2-refcount.c int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, · line 1527

Quoted 5 lines
block/qcow2-refcount.c +5 -0
+     */
+    if (offset + size - file_len >= s->cluster_size) {
+        fprintf(stderr, "ERROR: counting reference for region exceeding the "
+                "end of the file by one cluster or more: offset 0x%" PRIx64
+                " size 0x%" PRIx64 "\n", offset, size);

Should we be using something smarter than fprintf() here?

Grammar suggestion:

ERROR: reference to a region too far beyond the end of the file:

(which is shorter than what you wrote, and also has the nice property of allowing us to change our mind on whether it is 1 cluster, 2, 10, or some other finite limit as our heuristic of when we consider the image corrupt without having to reword it).

Quoted 10 lines

+++ b/tests/qemu-iotests/138

tests/qemu-iotests/138 +1 -2
@@ -54,15 +54,13 @@ $QEMU_IO -c 'write 0 512' "$TEST_IMG" | _filter_qemu_io
 # Put the data cluster at a multiple of 2 TB, resulting in the image apparently
 # having a multiple of 2^32 clusters
 # (To be more specific: It is at 32 PB)
-poke_file "$TEST_IMG" 2048 "\x80\x80\x00\x00\x00\x00\x00\x00"
+poke_file "$TEST_IMG" $((2048 + 8)) "\x00\x80\x00\x00\x00\x00\x00\x00"
 
 # An offset of 32 PB results in qemu-img check having to allocate an in-memory
-# refcount table of 128 TB (16 bit refcounts, 512 byte clusters).

tests/qemu-iotests/138 $QEMU_IO -c 'write 0 512' "$TEST_IMG" | _filter_qemu_io · line 64

Quoted 10 lines
tests/qemu-iotests/138 +4 -5
-# This should be generally too much for any system and thus fail.
-# What this test is checking is that the qcow2 driver actually tries to allocate
-# such a large amount of memory (and is consequently aborting) instead of having
-# truncated the cluster count somewhere (which would result in much less memory
-# being allocated and then a segfault occurring).
+# refcount table of 128 TB (16 bit refcounts, 512 byte clusters), if qemu-img
+# don't check that referenced data cluster is far beyond the end of file.
+# But starting from 4.0, qemu-img does this check, and instead of "Cannot
+# allocate memory", we have an error showing that l2 entry is invalid.
 _check_test_img

In other words, we are still gracefully invalidating the file, but now because of a heuristic rather than a memory allocation failure.

I might word the comment differently:

An offset of 32 PB would result in qemu-img check having to allocate an in-memory refcount table of 128 TB (16 bit refcounts, 512 byte clusters). It is unlikely that a system has this much memory, so the test ensures that qemu-img either gracefully fails an allocation (prior to 4.0) or flags the image as invalid (the reference points to a cluster too far beyond the actual file), rather than silently allocating a truncated memory amount or dumping core.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#6 lore raw

PATCH v4 3/5 qcow2-refcount: check_refcounts_l2: reduce ignored overlaps

In reply to Vladimir Sementsov-Ogievskiy

Reduce number of structures ignored in overlap check: when checking active table ignore active tables, when checking inactive table ignore inactive ones.

---
 block/qcow2-refcount.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
block/qcow2-refcount.c +9 -7
index c41c1fbe00..e76085d9aa 100644
@@ -1575,7 +1575,7 @@ enum {
 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                               void **refcount_table,
                               int64_t *refcount_table_size, int64_t l2_offset,
-                              int flags, BdrvCheckMode fix)
+                              int flags, BdrvCheckMode fix, bool active)
 {
     BDRVQcow2State *s = bs->opaque;
     uint64_t *l2_table, l2_entry;
@@ -1659,11 +1659,12 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                     if (fix & BDRV_FIX_ERRORS) {
                         uint64_t l2e_offset =
                             l2_offset + (uint64_t)i * sizeof(uint64_t);
+                        int ign = active ? QCOW2_OL_ACTIVE_L2 :
+                                           QCOW2_OL_INACTIVE_L2;
 
                         l2_entry = QCOW_OFLAG_ZERO;
                         l2_table[i] = cpu_to_be64(l2_entry);
-                        ret = qcow2_pre_write_overlap_check(bs,
-                                QCOW2_OL_ACTIVE_L2 | QCOW2_OL_INACTIVE_L2,
+                        ret = qcow2_pre_write_overlap_check(bs, ign,
                                 l2e_offset, sizeof(uint64_t));
                         if (ret < 0) {
                             fprintf(stderr, "ERROR: Overlap check failed\n");
@@ -1737,7 +1738,7 @@ static int check_refcounts_l1(BlockDriverState *bs,
                               void **refcount_table,
                               int64_t *refcount_table_size,
                               int64_t l1_table_offset, int l1_size,
-                              int flags, BdrvCheckMode fix)
+                              int flags, BdrvCheckMode fix, bool active)
 {
     BDRVQcow2State *s = bs->opaque;
     uint64_t *l1_table = NULL, l2_offset, l1_size2;
@@ -1793,7 +1794,7 @@ static int check_refcounts_l1(BlockDriverState *bs,
             /* Process and check L2 entries */
             ret = check_refcounts_l2(bs, res, refcount_table,
                                      refcount_table_size, l2_offset, flags,
-                                     fix);
+                                     fix, active);
             if (ret < 0) {
                 goto fail;
             }
@@ -2079,7 +2080,7 @@ static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
     /* current L1 table */
     ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
                              s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO,
-                             fix);
+                             fix, true);
     if (ret < 0) {
         return ret;
     }
@@ -2102,7 +2103,8 @@ static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
             continue;
         }
         ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
-                                 sn->l1_table_offset, sn->l1_size, 0, fix);
+                                 sn->l1_table_offset, sn->l1_size, 0, fix,
+                                 false);
         if (ret < 0) {
             return ret;
         }
Signature · 1 line
2.18.0
#7 lore raw

PATCH v4 4/5 qcow2-refcount: check_refcounts_l2: don't count fixed cluster as allocated

In reply to Vladimir Sementsov-Ogievskiy

Do not count a cluster which is fixed to be ZERO as allocated.

---
 block/qcow2-refcount.c | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)
block/qcow2-refcount.c +9 -9
index e76085d9aa..8da0e91dd3 100644
@@ -1637,15 +1637,6 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
         {
             uint64_t offset = l2_entry & L2E_OFFSET_MASK;
 
-            if (flags & CHECK_FRAG_INFO) {
-                res->bfi.allocated_clusters++;
-                if (next_contiguous_offset &&
-                    offset != next_contiguous_offset) {
-                    res->bfi.fragmented_clusters++;
-                }
-                next_contiguous_offset = offset + s->cluster_size;
-            }
-
             /* Correct offsets are cluster aligned */
             if (offset_into_cluster(s, offset)) {
                 if (qcow2_get_cluster_type(l2_entry) ==
@@ -1698,6 +1689,15 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                 }
             }
 
+            if (flags & CHECK_FRAG_INFO) {
+                res->bfi.allocated_clusters++;
+                if (next_contiguous_offset &&
+                    offset != next_contiguous_offset) {
+                    res->bfi.fragmented_clusters++;
+                }
+                next_contiguous_offset = offset + s->cluster_size;
+            }
+
             /* Mark cluster as used */
             ret = qcow2_inc_refcounts_imrt(bs, res,
                                            refcount_table, refcount_table_size,
Signature · 1 line
2.18.0
Reviewed-by on v4 4/5
#8 lore raw

Re: [Qemu-devel] [PATCH v4 4/5] qcow2-refcount: check_refcounts_l2: don't count fixed cluster as allocated

Quoted 6 lines · Vladimir Sementsov-Ogievskiy

Do not count a cluster which is fixed to be ZERO as allocated.

---
 block/qcow2-refcount.c | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)
#9 lore raw

PATCH v4 5/5 qcow2-refcount: don't mask corruptions under internal errors

In reply to Vladimir Sementsov-Ogievskiy

No reasons for not reporting found corruptions as corruptions in case of some internal errors, especially in case of just failed to fix l2 entry (and in this case, missed corruptions may influence comparing logic, when we calculate difference between corruptions fields of two results)

---
 block/qcow2-refcount.c | 31 ++++++++++++++-----------------
 1 file changed, 14 insertions(+), 17 deletions(-)
block/qcow2-refcount.c +14 -17
index 8da0e91dd3..b0e006e628 100644
@@ -1639,6 +1639,8 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
 
             /* Correct offsets are cluster aligned */
             if (offset_into_cluster(s, offset)) {
+                res->corruptions++;
+
                 if (qcow2_get_cluster_type(l2_entry) ==
                     QCOW2_CLUSTER_ZERO_ALLOC)
                 {
@@ -1674,18 +1676,16 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                             /* Do not abort, continue checking the rest of this
                              * L2 table's entries */
                         } else {
+                            res->corruptions--;
                             res->corruptions_fixed++;
                             /* Skip marking the cluster as used
                              * (it is unused now) */
                             continue;
                         }
-                    } else {
-                        res->corruptions++;
                     }
                 } else {
                     fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is "
                         "not properly aligned; L2 entry corrupted.\n", offset);
-                    res->corruptions++;
                 }
             }
 
@@ -1854,6 +1854,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
             continue;
         }
         if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
+            res->corruptions++;
             fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
                     "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
                     repair ? "Repairing" : "ERROR", i, l1_entry, refcount);
@@ -1866,9 +1867,8 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                     res->check_errors++;
                     goto fail;
                 }
+                res->corruptions--;
                 res->corruptions_fixed++;
-            } else {
-                res->corruptions++;
             }
         }
 
@@ -1899,19 +1899,16 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                     fprintf(stderr, "%s OFLAG_COPIED data cluster: "
                             "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
                             repair ? "Repairing" : "ERROR", l2_entry, refcount);
-                    if (repair) {
-                        l2_table[j] = cpu_to_be64(refcount == 1
-                                    ? l2_entry |  QCOW_OFLAG_COPIED
-                                    : l2_entry & ~QCOW_OFLAG_COPIED);
-                        l2_dirty++;
-                    } else {
-                        res->corruptions++;
-                    }
+                    l2_table[j] = cpu_to_be64(refcount == 1
+                                ? l2_entry |  QCOW_OFLAG_COPIED
+                                : l2_entry & ~QCOW_OFLAG_COPIED);
+                    l2_dirty++;
                 }
             }
         }
 
-        if (l2_dirty > 0) {
+        res->corruptions += l2_dirty;
+        if (repair && l2_dirty > 0) {
             ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
                                                 l2_offset, s->cluster_size);
             if (ret < 0) {
@@ -1929,6 +1926,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                 res->check_errors++;
                 goto fail;
             }
+            res->corruptions -= l2_dirty;
             res->corruptions_fixed += l2_dirty;
         }
     }
@@ -1967,6 +1965,7 @@ static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
         }
 
         if (cluster >= *nb_clusters) {
+            res->corruptions++;
             fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 
@@ -2006,6 +2005,7 @@ static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
                     goto resize_fail;
                 }
 
+                res->corruptions--;
                 res->corruptions_fixed++;
                 ret = qcow2_inc_refcounts_imrt(bs, res,
                                                refcount_table, nb_clusters,
@@ -2019,12 +2019,9 @@ static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
                 continue;
 
 resize_fail:
-                res->corruptions++;
                 *rebuild = true;
                 fprintf(stderr, "ERROR could not resize image: %s\n",
                         strerror(-ret));
-            } else {
-                res->corruptions++;
             }
             continue;
         }
Signature · 1 line
2.18.0
#10 lore raw

Re: [Qemu-devel] [PATCH v4 5/5] qcow2-refcount: don't mask corruptions under internal errors

Quoted 15 lines · Vladimir Sementsov-Ogievskiy

No reasons for not reporting found corruptions as corruptions in case of some internal errors, especially in case of just failed to fix l2 entry (and in this case, missed corruptions may influence comparing logic, when we calculate difference between corruptions fields of two results)

---
 block/qcow2-refcount.c | 31 ++++++++++++++-----------------
 1 file changed, 14 insertions(+), 17 deletions(-)
block/qcow2-refcount.c +0 -0
index 8da0e91dd3..b0e006e628 100644

[...]

Quoted 6 lines
block/qcow2-refcount.c +0 -2
@@ -1899,19 +1899,16 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                     fprintf(stderr, "%s OFLAG_COPIED data cluster: "
                             "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
                             repair ? "Repairing" : "ERROR", l2_entry, refcount);
-                    if (repair) {
-                        l2_table[j] = cpu_to_be64(refcount == 1

block/qcow2-refcount.c static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, · line 1905

Quoted 10 lines
block/qcow2-refcount.c +4 -6
-                                    ? l2_entry |  QCOW_OFLAG_COPIED
-                                    : l2_entry & ~QCOW_OFLAG_COPIED);
-                        l2_dirty++;
-                    } else {
-                        res->corruptions++;
-                    }
+                    l2_table[j] = cpu_to_be64(refcount == 1
+                                ? l2_entry |  QCOW_OFLAG_COPIED
+                                : l2_entry & ~QCOW_OFLAG_COPIED);
+                    l2_dirty++;

I found the old logic to be easier to understand, actually. It made it clear that if !repair, the L2 table is not going to be touched.

Max

#11 lore raw

Re: [Qemu-devel] [PATCH v4 5/5] qcow2-refcount: don't mask corruptions under internal errors

Quoted 39 lines · Max Reitz
Quoted 15 lines · Vladimir Sementsov-Ogievskiy

No reasons for not reporting found corruptions as corruptions in case of some internal errors, especially in case of just failed to fix l2 entry (and in this case, missed corruptions may influence comparing logic, when we calculate difference between corruptions fields of two results)

---
  block/qcow2-refcount.c | 31 ++++++++++++++-----------------
  1 file changed, 14 insertions(+), 17 deletions(-)
block/qcow2-refcount.c +0 -0
index 8da0e91dd3..b0e006e628 100644

[...]

Quoted 6 lines
block/qcow2-refcount.c +0 -2
@@ -1899,19 +1899,16 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                      fprintf(stderr, "%s OFLAG_COPIED data cluster: "
                              "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
                              repair ? "Repairing" : "ERROR", l2_entry, refcount);
-                    if (repair) {
-                        l2_table[j] = cpu_to_be64(refcount == 1

block/qcow2-refcount.c static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, · line 1905

Quoted 10 lines
block/qcow2-refcount.c +4 -6
-                                    ? l2_entry |  QCOW_OFLAG_COPIED
-                                    : l2_entry & ~QCOW_OFLAG_COPIED);
-                        l2_dirty++;
-                    } else {
-                        res->corruptions++;
-                    }
+                    l2_table[j] = cpu_to_be64(refcount == 1
+                                ? l2_entry |  QCOW_OFLAG_COPIED
+                                : l2_entry & ~QCOW_OFLAG_COPIED);
+                    l2_dirty++;

I found the old logic to be easier to understand, actually. It made it clear that if !repair, the L2 table is not going to be touched.

Quoted 2 lines

Max

so,

   if (repair) {
      keep as is
   }
   res->corruptions++;

ok?

Signature · 2 lines
Best regards,
Vladimir
#12 lore raw

Re: [Qemu-devel] [PATCH v4 5/5] qcow2-refcount: don't mask corruptions under internal errors

Quoted 43 lines · Vladimir Sementsov-Ogievskiy
Quoted 39 lines · Max Reitz
Quoted 15 lines · Vladimir Sementsov-Ogievskiy

No reasons for not reporting found corruptions as corruptions in case of some internal errors, especially in case of just failed to fix l2 entry (and in this case, missed corruptions may influence comparing logic, when we calculate difference between corruptions fields of two results)

---
  block/qcow2-refcount.c | 31 ++++++++++++++-----------------
  1 file changed, 14 insertions(+), 17 deletions(-)
block/qcow2-refcount.c +0 -0
index 8da0e91dd3..b0e006e628 100644

[...]

Quoted 6 lines
block/qcow2-refcount.c +0 -2
@@ -1899,19 +1899,16 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                      fprintf(stderr, "%s OFLAG_COPIED data cluster: "
                              "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
                              repair ? "Repairing" : "ERROR", l2_entry, refcount);
-                    if (repair) {
-                        l2_table[j] = cpu_to_be64(refcount == 1

block/qcow2-refcount.c static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, · line 1905

Quoted 10 lines
block/qcow2-refcount.c +4 -6
-                                    ? l2_entry |  QCOW_OFLAG_COPIED
-                                    : l2_entry & ~QCOW_OFLAG_COPIED);
-                        l2_dirty++;
-                    } else {
-                        res->corruptions++;
-                    }
+                    l2_table[j] = cpu_to_be64(refcount == 1
+                                ? l2_entry |  QCOW_OFLAG_COPIED
+                                : l2_entry & ~QCOW_OFLAG_COPIED);
+                    l2_dirty++;

I found the old logic to be easier to understand, actually. It made it clear that if !repair, the L2 table is not going to be touched.

Quoted 2 lines

Max

Quoted 8 lines

so,

   if (repair) {
      keep as is
   }
   res->corruptions++;

ok?

Yep, that works for me. Or above the fprintf() as you did it in other places in this patch.

Max

#13 lore raw
In reply to Vladimir Sementsov-Ogievskiy

ping

Quoted 26 lines · Vladimir Sementsov-Ogievskiy

Hi all!

These series are derived from "[PATCH v3 0/8] qcow2 check improvements"
(https://lists.gnu.org/archive/html/qemu-devel/2018-10/msg02284.html)

Some patches are new, some patches are not included and are postponed to part II.

So, here:
01,03 - unchanged
02: handle broken iotest
04-05: new

Vladimir Sementsov-Ogievskiy (5):
qcow2-refcount: fix check_oflag_copied
qcow2-refcount: avoid eating RAM
qcow2-refcount: check_refcounts_l2: reduce ignored overlaps
qcow2-refcount: check_refcounts_l2: don't count fixed cluster as
allocated
qcow2-refcount: don't mask corruptions under internal errors

  block/qcow2-refcount.c     | 87 +++++++++++++++++++++++---------------
  tests/qemu-iotests/138     | 12 +++---
  tests/qemu-iotests/138.out |  5 ++-
  3 files changed, 61 insertions(+), 43 deletions(-)
Signature · 2 lines
Best regards,
Vladimir
#14 lore raw

ping

Quoted 31 lines · Vladimir Sementsov-Ogievskiy

ping

Quoted 26 lines · Vladimir Sementsov-Ogievskiy

Hi all!

These series are derived from "[PATCH v3 0/8] qcow2 check improvements"
(https://lists.gnu.org/archive/html/qemu-devel/2018-10/msg02284.html)

Some patches are new, some patches are not included and are postponed to part II.

So, here:
01,03 - unchanged
02: handle broken iotest
04-05: new

Vladimir Sementsov-Ogievskiy (5):
   qcow2-refcount: fix check_oflag_copied
   qcow2-refcount: avoid eating RAM
   qcow2-refcount: check_refcounts_l2: reduce ignored overlaps
   qcow2-refcount: check_refcounts_l2: don't count fixed cluster as allocated
   qcow2-refcount: don't mask corruptions under internal errors

  block/qcow2-refcount.c     | 87 +++++++++++++++++++++++--------------- tests/qemu-iotests/138     | 12 +++--- tests/qemu-iotests/138.out |  5 ++-

  3 files changed, 61 insertions(+), 43 deletions(-)
Signature · 2 lines
Best regards,
Vladimir