minor compression improvements

In v4:
- patch 2 has even more wording tweaks [Kevin]
- patch 3 is new [Berto]
- patch 5 fixes an off-by-one [Berto]

Eric Blake (5):
qcow2: Prefer byte-based calls into bs->file
qcow2: Document some maximum size constraints
qcow2: Reduce REFT_OFFSET_MASK
qcow2: Don't allow overflow during cluster allocation
qcow2: Avoid memory over-allocation on compressed images

 docs/interop/qcow2.txt | 38 +++++++++++++++++++++++++++++++++++---
 block/qcow2.h          |  8 +++++++-
 block/qcow2-cluster.c  | 32 ++++++++++++++++++++------------
 block/qcow2-refcount.c | 27 +++++++++++++++++----------
 block/qcow2.c          |  2 +-
 5 files changed, 80 insertions(+), 27 deletions(-)

2.14.3

We had only three sector-based stragglers left; convert them to use our preferred byte-based accesses.

---
v2: indentation fix

---
 block/qcow2-cluster.c  | 5 ++---
 block/qcow2-refcount.c | 6 +++---
 2 files changed, 5 insertions(+), 6 deletions(-)

index e406b0f3b9e..85be7d5e340 100644
@@ -1615,13 +1615,12 @@ int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
         }
 
         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
-        ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data,
-                        nb_csectors);
+        ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
         if (ret < 0) {
             return ret;
         }
         if (decompress_buffer(s->cluster_cache, s->cluster_size,
-                              s->cluster_data + sector_offset, csize) < 0) {
+                              s->cluster_data, csize) < 0) {
             return -EIO;
         }
         s->cluster_cache_offset = coffset;

index d46b69d7f34..28afbb1b5ea 100644
@@ -2310,8 +2310,8 @@ write_refblocks:
         on_disk_refblock = (void *)((char *) *refcount_table +
                                     refblock_index * s->cluster_size);
 
-        ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE,
-                         on_disk_refblock, s->cluster_sectors);
+        ret = bdrv_pwrite(bs->file, refblock_offset, on_disk_refblock,
+                          s->cluster_size);
         if (ret < 0) {
             fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
             goto fail;

@@ -2533,7 +2533,7 @@ fail:
  * - 0 if writing to this offset will not affect the mentioned metadata
  * - a positive QCow2MetadataOverlap value indicating one overlapping section
  * - a negative value (-errno) indicating an error while performing a check,
- *   e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
+ *   e.g. when bdrv_pread failed on QCOW2_OL_INACTIVE_L2
  */
 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
                                  int64_t size)

2.14.3

Although off_t permits up to 63 bits (8EB) of file offsets, in practice, we're going to hit other limits first. Document some of those limits in the qcow2 spec, and how choice of cluster size can influence some of the limits.

While at it, notice that since we cannot map any virtual cluster to any address higher than 64 PB (56 bits) (due to the current L1/L2 field encoding stopping at bit 55), it makes little sense to require the refcount table to access host offsets beyond that point. Mark the upper bits of the refcount table entries as reserved to match the L1/L2 table, with no ill effects, since it is unlikely that there are any existing images larger than 64PB in the first place, and thus all existing images already have those bits as 0. If 64PB proves to be too small in the future, we could enlarge all three uses of bit 55 into the reserved bits at that time.

However, there is one limit that reserved bits don't help with: for compressed clusters, the L2 layout requires an ever-smaller maximum host offset as cluster size gets larger, down to a 512 TB maximum with 2M clusters.

--
v4: more wording tweaks
v3: new patch

---
 docs/interop/qcow2.txt | 38 +++++++++++++++++++++++++++++++++++---
 1 file changed, 35 insertions(+), 3 deletions(-)

index feb711fb6a8..e32d391e66b 100644
@@ -40,7 +40,17 @@ The first cluster of a qcow2 image contains the file header:
                     with larger cluster sizes.
 
          24 - 31:   size
-                    Virtual disk size in bytes
+                    Virtual disk size in bytes.
+
+                    Note: with a 2 MB cluster size, the maximum
+                    virtual size is 2 EB (61 bits) for a sparse file,
+                    but other sizing limitations in refcount and L1/L2
+                    tables mean that an image cannot have more than 64
+                    PB of populated clusters (and an image may hit
+                    other sizing limitations as well, such as
+                    underlying protocol limits).  With a 512 byte
+                    cluster size, the maximum virtual size drops to
+                    128 GB (37 bits).
 
          32 - 35:   crypt_method
                     0 for no encryption

@@ -318,6 +328,13 @@ for each host cluster. A refcount of 0 means that the cluster is free, 1 means
 that it is used, and >= 2 means that it is used and any write access must
 perform a COW (copy on write) operation.
 
+The refcount table has implications on the maximum host file size; a
+larger cluster size is required for the refcount table to cover larger
+offsets.  Furthermore, all qcow2 metadata must currently reside at
+offsets below 64 PB (56 bits) (this limit could be enlarged by putting
+reserved bits into use, but only if a similar limit on L1/L2 tables is
+revisited at the same time).
+
 The refcounts are managed in a two-level table. The first level is called
 refcount table and has a variable size (which is stored in the header). The
 refcount table can cover multiple clusters, however it needs to be contiguous

@@ -341,7 +358,7 @@ Refcount table entry:
 
     Bit  0 -  8:    Reserved (set to 0)
 
-         9 - 63:    Bits 9-63 of the offset into the image file at which the
+         9 - 55:    Bits 9-55 of the offset into the image file at which the
                     refcount block starts. Must be aligned to a cluster
                     boundary.
 

@@ -349,6 +366,8 @@ Refcount table entry:
                     been allocated. All refcounts managed by this refcount block
                     are 0.
 
+        56 - 63:    Reserved (set to 0)
+
 Refcount block entry (x = refcount_bits - 1):
 
     Bit  0 -  x:    Reference count of the cluster. If refcount_bits implies a

@@ -365,6 +384,17 @@ The L1 table has a variable size (stored in the header) and may use multiple
 clusters, however it must be contiguous in the image file. L2 tables are
 exactly one cluster in size.
 
+The L1 and L2 tables have implications on the maximum virtual file
+size; a larger cluster size is required for the guest to have access
+to more space.  Furthermore, a virtual cluster must currently map to a
+host offset below 64 PB (56 bits) (this limit could be enlarged by
+putting reserved bits into use, but only if a similar limit on
+refcount tables is revisited at the same time).  Additionally, with
+larger cluster sizes, compressed clusters have a smaller limit on host
+cluster mappings (a 2M cluster size requires compressed clusters to
+reside below 512 TB (49 bits), where enlarging this would require an
+incompatible layout change).
+
 Given a offset into the virtual disk, the offset into the image file can be
 obtained as follows:
 

@@ -427,7 +457,9 @@ Standard Cluster Descriptor:
 Compressed Clusters Descriptor (x = 62 - (cluster_bits - 8)):
 
     Bit  0 - x-1:   Host cluster offset. This is usually _not_ aligned to a
-                    cluster or sector boundary!
+                    cluster or sector boundary!  If cluster_bits is
+                    small enough that this field includes bits beyond
+                    55, those upper bits must be set to 0.
 
          x - 61:    Number of additional 512-byte sectors used for the
                     compressed data, beyond the sector containing the offset

2.14.3

Why is this? Because of the refcount_table_clusters field ?

I think the maximum offset allowed by that is ridiculously high, exceeding any other limit imposed by the L1/L2 tables.

If my numbers are right, with the default values that's 64 ZB.

In addition to that, the size that can be covered by the refcount table also depends on the size of refcount entries (refcount_order).

With 512 byte clusters and 64 bit refcount entries I still get 8 PB, way over what's limited by the L1/L2 tables (128 GB).

Berto

Good point. I was basing my comment off of qcow2.h:

/* 8 MB refcount table is enough for 2 PB images at 64k cluster size
* (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
#define QCOW_MAX_REFTABLE_SIZE 0x800000

But that's our implementation choice (we put a maximum amount of memory on the size of the refcount table we are willing to support, while the qcow2 spec would allow an implementation willing to reserve more memory to access even larger sizing).

True.

Do I need to make any modifications to the sentence, then? Or is it still accurate, if vague, to leave the sentence as is because there IS an impact to consider, even if the impact is unlikely to matter in relation to other sizing impacts?

Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

[...]

I guess what surprised me the first time that I read it was that it suggests that this has to be taken into account when calculating the physical limits of an image, while in practice it can be ignored.

You could say something like

Although the larger the cluster size, the larger the offsets that can be covered by the refcount table, in practice these limits cannot be reached because they are larger than the ones imposed by other data structures.

although I'm sure that you can come up with a better wording than mine :)

Berto

Are there any updates here? I guess I personally would just drop the whole paragraph, because I think it really doesn't matter...

Also note that the maximum file size of ext4 is 16 PB (for 4 kB blocks).
OK, it's bigger for XFS, but that still gives some perspective.

Also, long before anyone is going to complain about the specification failing to mention that limit, they are going to complain that qemu refuses to open their image (because of its limit on the reftable size).

Max

Yeah, I need to post a v5 of this series now that 2.13 is nearly open.

Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

Match our code to the spec change in the previous patch - there's no reason for the refcount table to allow larger offsets than the L1/L2 tables. In practice, no image has more than 64PB of allocated clusters anyways, as anything beyond that can't be expressed via L2 mappings to host offsets.

---
v4: new patch

---
 block/qcow2.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index 883802241fb..afc524ead54 100644
@@ -442,7 +442,7 @@ typedef enum QCow2MetadataOverlap {
 #define L2E_OFFSET_MASK 0x00fffffffffffe00ULL
 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL
 
-#define REFT_OFFSET_MASK 0xfffffffffffffe00ULL
+#define REFT_OFFSET_MASK 0x00fffffffffffe00ULL
 
 static inline int64_t start_of_cluster(BDRVQcow2State *s, int64_t offset)
 {

2.14.3

Berto

Our code was already checking that we did not attempt to allocate more clusters than what would fit in an INT64 (the physical maximimum if we can access a full off_t's worth of data). But this does not catch smaller limits enforced by various spots in the qcow2 image description: L1 and normal clusters of L2 are documented as having bits 63-56 reserved for other purposes, capping our maximum offset at 64PB (bit 55 is the maximum bit set). And for compressed images with 2M clusters, the cap drops the maximum offset to bit 48, or a maximum offset of 512TB. If we overflow that offset, we would write compressed data into one place, but try to decompress from another, which won't work.

I don't have 512TB handy to prove whether things break if we compress so much data that we overflow that limit, and don't think that iotests can (quickly) test it either. Test 138 comes close (it corrupts an image into thinking something lives at 32PB, which is half the maximum for L1 sizing - although it relies on 512-byte clusters). But that test points out that we will generally hit other limits first (such as running out of memory for the refcount table, or exceeding file system limits like 16TB on ext4, etc), so this is more a theoretical safety valve than something likely to be hit.

---
v3: use s->cluster_offset_mask instead of open-coding it [Berto], use MIN() to clamp offset on small cluster size, add spec patch first to justify clamping even on refcount allocations

---
 block/qcow2.h          |  6 ++++++
 block/qcow2-refcount.c | 21 ++++++++++++++-------
 2 files changed, 20 insertions(+), 7 deletions(-)

index afc524ead54..785c8d6e222 100644
@@ -41,6 +41,12 @@
 #define QCOW_MAX_CRYPT_CLUSTERS 32
 #define QCOW_MAX_SNAPSHOTS 65536
 
+/* Field widths in qcow2 mean normal cluster offsets cannot reach
+ * 64PB; depending on cluster size, compressed clusters can have a
+ * smaller limit (64PB for up to 16k clusters, then ramps down to
+ * 512TB for 2M clusters).  */
+#define QCOW_MAX_CLUSTER_OFFSET ((1ULL << 56) - 1)
+
 /* 8 MB refcount table is enough for 2 PB images at 64k cluster size
  * (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
 #define QCOW_MAX_REFTABLE_SIZE 0x800000

index 28afbb1b5ea..c46a2257a9e 100644
@@ -31,7 +31,8 @@
 #include "qemu/bswap.h"
 #include "qemu/cutils.h"
 
-static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size);
+static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size,
+                                    uint64_t max);
 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
                             int64_t offset, int64_t length, uint64_t addend,
                             bool decrease, enum qcow2_discard_type type);

@@ -362,7 +363,8 @@ static int alloc_refcount_block(BlockDriverState *bs,
     }
 
     /* Allocate the refcount block itself and mark it as used */
-    int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
+    int64_t new_block = alloc_clusters_noref(bs, s->cluster_size,
+                                             QCOW_MAX_CLUSTER_OFFSET);
     if (new_block < 0) {
         return new_block;
     }

@@ -947,7 +949,8 @@ int qcow2_update_cluster_refcount(BlockDriverState *bs,
 
 
 /* return < 0 if error */
-static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size)
+static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size,
+                                    uint64_t max)
 {
     BDRVQcow2State *s = bs->opaque;
     uint64_t i, nb_clusters, refcount;

@@ -972,9 +975,9 @@ retry:
     }
 
     /* Make sure that all offsets in the "allocated" range are representable
-     * in an int64_t */
+     * in the requested max */
     if (s->free_cluster_index > 0 &&
-        s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits))
+        s->free_cluster_index - 1 > (max >> s->cluster_bits))
     {
         return -EFBIG;
     }

@@ -994,7 +997,7 @@ int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
 
     BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
     do {
-        offset = alloc_clusters_noref(bs, size);
+        offset = alloc_clusters_noref(bs, size, QCOW_MAX_CLUSTER_OFFSET);
         if (offset < 0) {
             return offset;
         }

@@ -1076,7 +1079,11 @@ int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
     free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
     do {
         if (!offset || free_in_cluster < size) {
-            int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size);
+            int64_t new_cluster;
+
+            new_cluster = alloc_clusters_noref(bs, s->cluster_size,
+                                               MIN(s->cluster_offset_mask,
+                                                   QCOW_MAX_CLUSTER_OFFSET));
             if (new_cluster < 0) {
                 return new_cluster;
             }

2.14.3

When reading a compressed image, we were allocating s->cluster_data to 32*cluster_size + 512 (possibly over 64 megabytes, for an image with 2M clusters). Let's check out the history:

Back when qcow2 was first written, we used s->cluster_data for everything, including copy_sectors() and encryption, where we want to operate on more than one cluster at once. Obviously, at that point, the buffer had to be aligned for other users, even though compression itself doesn't require any alignment (the fact that the compressed data generally starts mid-sector means that aligning our buffer buys us nothing - either the protocol already supports byte-based access into whatever offset we want, or we are already using a bounce buffer to read a full sector, and copying into our destination no longer requires alignment).

But commit 1b9f1491 (v1.1!) changed things to allocate parallel buffers on demand rather than sharing a single buffer, for encryption and COW, leaving compression as the final client of s->cluster_data. That use was still preserved, because if a single compressed cluster is read more than once, we reuse the cache instead of decompressing it a second time (someday, we may come up with better caching to avoid wasting repeated decompressions while still being more parallel, but that is a task for another patch; the XXX comment in qcow2_co_preadv for QCOW2_CLUSTER_COMPRESSED is telling).

Much later, in commit de82815d (v2.2), we noticed that a 64M allocation is prone to failure, so we switched over to a graceful memory allocation error message. Elsewhere in the code, we do g_malloc(2 * cluster_size) without ever checking for failure, but even 4M starts to be large enough that trying to be nice is worth the effort, so we want to keep that aspect.

Then even later, in 3e4c7052 (2.11), we realized that allocating a large buffer up front for every qcow2 image is expensive, and switched to lazy allocation only for images that actually had compressed clusters. But in the process, we never even bothered to check whether what we were allocating still made sense in its new context!

So, it's time to cut back on the waste. A compressed cluster written by qemu will NEVER occupy more than an uncompressed cluster, but based on mid-sector alignment, we may still need to read 1 cluster + 1 sector in order to recover enough bytes for the decompression. But third-party producers of qcow2 may not be as smart, and gzip DOES document that because the compression stream adds metadata, and because of the pigeonhole principle, there are worst case scenarios where attempts to compress will actually inflate an image, by up to 0.015% (or 62 sectors larger for an unfortunate 2M compression). In fact, the qcow2 spec permits an all-ones sector count, plus a full sector containing the initial offset, for a maximum read of 2 full clusters; and thanks to the way decompression works, even though such a value is probably too large for the actual compressed data, it really doesn't matter if we read too much (gzip ignores slop, once it has decoded a full cluster). So it's easier to just allocate cluster_data to be as large as we can ever possibly see; even if it still wastes up to 2M on any image created by qcow2, that's still an improvment of 60M less waste than pre-patch.

---
v4: fix off-by-one in assertion and commit message [Berto]
v3: tighten allocation [Berto]
v2: actually check allocation failure (previous version meant to use g_malloc, but ended up posted with g_try_malloc without checking); add assertions outside of conditional, improve commit message to better match reality now that qcow2 spec bug has been fixed

---
 block/qcow2-cluster.c | 27 ++++++++++++++++++---------
 block/qcow2.c         |  2 +-
 2 files changed, 19 insertions(+), 10 deletions(-)

index 85be7d5e340..1f7f00cbea2 100644
@@ -1598,20 +1598,29 @@ int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
         sector_offset = coffset & 511;
         csize = nb_csectors * 512 - sector_offset;
 
-        /* Allocate buffers on first decompress operation, most images are
-         * uncompressed and the memory overhead can be avoided.  The buffers
-         * are freed in .bdrv_close().
+        /* Allocate buffers on the first decompress operation; most
+         * images are uncompressed and the memory overhead can be
+         * avoided.  The buffers are freed in .bdrv_close().  qemu
+         * never writes an inflated cluster, and gzip itself never
+         * inflates a problematic cluster by more than 0.015%, but the
+         * qcow2 format allows up to 1 byte short of 2 full clusters
+         * when including the sector containing offset.  gzip ignores
+         * trailing slop, so just allocate that much up front rather
+         * than reject third-party images with overlarge csize.
          */
+        assert(!!s->cluster_data == !!s->cluster_cache);
+        assert(csize <= 2 * s->cluster_size);
         if (!s->cluster_data) {
-            /* one more sector for decompressed data alignment */
-            s->cluster_data = qemu_try_blockalign(bs->file->bs,
-                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512);
+            s->cluster_data = g_try_malloc(2 * s->cluster_size);
             if (!s->cluster_data) {
                 return -ENOMEM;
             }
-        }
-        if (!s->cluster_cache) {
-            s->cluster_cache = g_malloc(s->cluster_size);
+            s->cluster_cache = g_try_malloc(s->cluster_size);
+            if (!s->cluster_cache) {
+                g_free(s->cluster_data);
+                s->cluster_data = NULL;
+                return -ENOMEM;
+            }
         }
 
         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);

index 288b5299d80..6ad3436e0e5 100644
@@ -2103,7 +2103,7 @@ static void qcow2_close(BlockDriverState *bs)
     g_free(s->image_backing_format);
 
     g_free(s->cluster_cache);
-    qemu_vfree(s->cluster_data);
+    g_free(s->cluster_data);
     qcow2_refcount_close(bs);
     qcow2_free_snapshots(bs);
 }

2.14.3

Berto