Skip to contentSkip to conversation

block/file: Show extent size in qemu-img info

PATCH Superseded Hanna Reitz 3 participants 15 messages May 3, 2022 – May 4, 2022 Block layer

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover block/file: Show extent size in qemu-img info

Hi,

This series makes qemu-img info print the extent size of an image file (if available). To do so, we have to do a couple of things:
1. Add a .bdrv_get_specific_info handler to the file-posix driver (patch 4)
2. Have bdrv_query_image_info() collect this driver-specific info not only on the format level, but also on the protocol level (if there is an unambiguous node) so we actually get the information from the file-posix node when querying whatever format node is on top (patch 2)
3. Have bdrv_image_info_dump() print this protocol-level information (patch 3)

Extent size informations seems unavailable on ext4 at least, so if we did just this, you would see the following for an image on ext4:

```
$ qemu-img info -f raw test.img
image: test.img
file format: raw
virtual size: 1 MiB (1048576 bytes)
disk size: 4 KiB
Protocol specific information:
```

That last part looks a bit strange -- it's a heading without a section. So patch 1 makes bdrv_image_info_specific_dump() omit that heading if there is no information that can be printed.

Hanna Reitz (4):
block: Improve empty format-specific info dump
block: Add protocol-specific image info
block: Print protocol-specific information
block/file: Add file-specific image info

 qapi/block-core.json | 32 ++++++++++++++++++++--
 include/block/qapi.h |  3 +-
 block/file-posix.c   | 30 ++++++++++++++++++++
 block/qapi.c         | 65 +++++++++++++++++++++++++++++++++++++++++---
 qemu-io-cmds.c       |  4 +--
 5 files changed, 124 insertions(+), 10 deletions(-)
Signature · 1 line
2.35.1
#1 lore raw

PATCH 1/4 block: Improve empty format-specific info dump

When a block driver supports obtaining format-specific information, but that object only contains optional fields, it is possible that none of them are present, so that dump_qobject() (called by bdrv_image_info_specific_dump()) will not print anything.

The callers of bdrv_image_info_specific_dump() put a header above this information ("Format specific information:\n"), which will look strange when there is nothing below. Modify bdrv_image_info_specific_dump() to print this header instead of its callers, and only if there is indeed something to be printed.

---
 include/block/qapi.h |  3 ++-
 block/qapi.c         | 41 +++++++++++++++++++++++++++++++++++++----
 qemu-io-cmds.c       |  4 ++--
 3 files changed, 41 insertions(+), 7 deletions(-)
include/block/qapi.h +2 -1
index 22c7807c89..c09859ea78 100644
@@ -40,6 +40,7 @@ void bdrv_query_image_info(BlockDriverState *bs,
                            Error **errp);
 
 void bdrv_snapshot_dump(QEMUSnapshotInfo *sn);
-void bdrv_image_info_specific_dump(ImageInfoSpecific *info_spec);
+void bdrv_image_info_specific_dump(ImageInfoSpecific *info_spec,
+                                   const char *prefix);
 void bdrv_image_info_dump(ImageInfo *info);
 #endif
block/qapi.c +37 -4

58 lines not on this page · raw message

qemu-io-cmds.c +2 -2
index 2f0d8ac25a..084ec44d3b 100644
@@ -1819,8 +1819,8 @@ static int info_f(BlockBackend *blk, int argc, char **argv)
         return -EIO;
     }
     if (spec_info) {
-        printf("Format specific information:\n");
-        bdrv_image_info_specific_dump(spec_info);
+        bdrv_image_info_specific_dump(spec_info,
+                                      "Format specific information:\n");
         qapi_free_ImageInfoSpecific(spec_info);
     }
 
Signature · 1 line
2.35.1
Reviewed-by on v1 1/4
#2 lore raw

Re: [PATCH 1/4] block: Improve empty format-specific info dump

Quoted 11 lines · Hanna Reitz

When a block driver supports obtaining format-specific information, but that object only contains optional fields, it is possible that none of them are present, so that dump_qobject() (called by bdrv_image_info_specific_dump()) will not print anything.

The callers of bdrv_image_info_specific_dump() put a header above this information ("Format specific information:\n"), which will look strange when there is nothing below. Modify bdrv_image_info_specific_dump() to print this header instead of its callers, and only if there is indeed something to be printed.

Quoted 7 lines
---
 include/block/qapi.h |  3 ++-
 block/qapi.c         | 41 +++++++++++++++++++++++++++++++++++++----
 qemu-io-cmds.c       |  4 ++--
 3 files changed, 41 insertions(+), 7 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
Reviewed-by on v1 1/4
#3 lore raw

Re: [PATCH 1/4] block: Improve empty format-specific info dump

In reply to Hanna Reitz

Am 03.05.2022 um 16:55 hat Hanna Reitz geschrieben:

Quoted 5 lines

When a block driver supports obtaining format-specific information, but that object only contains optional fields, it is possible that none of them are present, so that dump_qobject() (called by bdrv_image_info_specific_dump()) will not print anything.

Quoted 7 lines

The callers of bdrv_image_info_specific_dump() put a header above this information ("Format specific information:\n"), which will look strange when there is nothing below. Modify bdrv_image_info_specific_dump() to print this header instead of its callers, and only if there is indeed something to be printed.

#4 lore raw

PATCH 2/4 block: Add protocol-specific image info

In reply to Hanna Reitz

The ImageInfo object currently only contains (optional) format-specific image information. However, perhaps the protocol node can provide some additional information, so add a new field presenting it.

---
 qapi/block-core.json |  6 +++++-
 block/qapi.c         | 19 +++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)
qapi/block-core.json +5 -1
index beeb91952a..e7d6c2e0cc 100644
@@ -236,6 +236,9 @@
 # @format-specific: structure supplying additional format-specific
 #                   information (since 1.7)
 #
+# @protocol-specific: structure supplying additional protocol-specific
+#                     information (since 7.1)
+#
 # Since: 1.3
 #
 ##
@@ -246,7 +249,8 @@
            '*backing-filename': 'str', '*full-backing-filename': 'str',
            '*backing-filename-format': 'str', '*snapshots': ['SnapshotInfo'],
            '*backing-image': 'ImageInfo',
-           '*format-specific': 'ImageInfoSpecific' } }
+           '*format-specific': 'ImageInfoSpecific',
+           '*protocol-specific': 'ImageInfoSpecific' } }
 
 ##
 # @ImageCheck:
block/qapi.c +19 -0
index 51202b470a..293983cf82 100644
@@ -262,6 +262,7 @@ void bdrv_query_image_info(BlockDriverState *bs,
     int64_t size;
     const char *backing_filename;
     BlockDriverInfo bdi;
+    BlockDriverState *protocol_bs;
     int ret;
     Error *err = NULL;
     ImageInfo *info;
@@ -303,6 +304,24 @@ void bdrv_query_image_info(BlockDriverState *bs,
     }
     info->has_format_specific = info->format_specific != NULL;
 
+    /* Try to look for an unambiguous protocol node */
+    protocol_bs = bs;
+    while (protocol_bs && !QLIST_EMPTY(&protocol_bs->children)) {
+        protocol_bs = bdrv_primary_bs(protocol_bs);
+    }
+    if (protocol_bs) {
+        /* Assert that this is a protocol node */
+        assert(QLIST_EMPTY(&protocol_bs->children));
+
+        info->protocol_specific = bdrv_get_specific_info(protocol_bs, &err);
+        if (err) {
+            error_propagate(errp, err);
+            qapi_free_ImageInfo(info);
+            goto out;
+        }
+        info->has_protocol_specific = info->protocol_specific != NULL;
+    }
+
     backing_filename = bs->backing_file;
     if (backing_filename[0] != '\0') {
         char *backing_filename2;
Signature · 1 line
2.35.1
Reviewed-by on v1 2/4
#5 lore raw

Re: [PATCH 2/4] block: Add protocol-specific image info

Quoted 9 lines · Hanna Reitz

The ImageInfo object currently only contains (optional) format-specific image information. However, perhaps the protocol node can provide some additional information, so add a new field presenting it.

---
 qapi/block-core.json |  6 +++++-
 block/qapi.c         | 19 +++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#6 lore raw

Re: [PATCH 2/4] block: Add protocol-specific image info

In reply to Hanna Reitz

Am 03.05.2022 um 16:55 hat Hanna Reitz geschrieben:

Quoted 25 lines

The ImageInfo object currently only contains (optional) format-specific image information. However, perhaps the protocol node can provide some additional information, so add a new field presenting it.

---
 qapi/block-core.json |  6 +++++-
 block/qapi.c         | 19 +++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)
qapi/block-core.json +3 -0
index beeb91952a..e7d6c2e0cc 100644
@@ -236,6 +236,9 @@
 # @format-specific: structure supplying additional format-specific
 #                   information (since 1.7)
 #
+# @protocol-specific: structure supplying additional protocol-specific
+#                     information (since 7.1)
+#
 # Since: 1.3
 #
 ##

qapi/block-core.json line 253

Quoted 6 lines
qapi/block-core.json +2 -1
@@ -246,7 +249,8 @@
            '*backing-filename': 'str', '*full-backing-filename': 'str',
            '*backing-filename-format': 'str', '*snapshots': ['SnapshotInfo'],
            '*backing-image': 'ImageInfo',
-           '*format-specific': 'ImageInfoSpecific' } }
+           '*format-specific': 'ImageInfoSpecific',
+           '*protocol-specific': 'ImageInfoSpecific' } }

I'm not a fan of this one. It solves the problem for exactly one special case (even if admittedly a common one) and leaves everything else as it is. It is unclear what it produces in configurations that aren't the simple one format node on top of one protocol node layout.

I would rather interpret 'format-specific' as 'driver-specific' and make the ImageInfo for any child node accessible.

With rbd we already interpret it like a generic driver thing that is not just for formats that because it implements .bdrv_get_specific_info even though we didn't have a 'protocol-specific' yet.

Making other nodes has precedence, too. 'backing-image' is even in the context of this hunk. VMDK exposes its extents the same way. So maybe what we really want is a 'children' list with the ImageInfo of every child node. And then qemu-img could go through all children and print headings like "Driver specific information for file (#block123)". Then filters like blkdebug could add their information and it would be printed, too.

Quoted 15 lines
qapi/block-core.json +0 -0
 ##
 # @ImageCheck:
block/qapi.c +1 -0
index 51202b470a..293983cf82 100644
@@ -262,6 +262,7 @@ void bdrv_query_image_info(BlockDriverState *bs,
     int64_t size;
     const char *backing_filename;
     BlockDriverInfo bdi;
+    BlockDriverState *protocol_bs;
     int ret;
     Error *err = NULL;
     ImageInfo *info;

block/qapi.c void bdrv_query_image_info(BlockDriverState *bs, · line 311

Quoted 8 lines
block/qapi.c +5 -0
@@ -303,6 +304,24 @@ void bdrv_query_image_info(BlockDriverState *bs,
     }
     info->has_format_specific = info->format_specific != NULL;
 
+    /* Try to look for an unambiguous protocol node */
+    protocol_bs = bs;
+    while (protocol_bs && !QLIST_EMPTY(&protocol_bs->children)) {
+        protocol_bs = bdrv_primary_bs(protocol_bs);
+    }

If bs is already a leaf node, this duplicates the information, which looks weird:

    $ build/qemu-img info -f file ~/tmp/test.raw
    image: /home/kwolf/tmp/test.raw
    file format: file
    virtual size: 10 GiB (10737418240 bytes)
    disk size: 7.63 GiB
    Format specific information:
        extent size: 1048576
    Protocol specific information:
        extent size: 1048576
Quoted 7 lines
block/qapi.c +6 -0
+    if (protocol_bs) {
+        /* Assert that this is a protocol node */
+        assert(QLIST_EMPTY(&protocol_bs->children));
+
+        info->protocol_specific = bdrv_get_specific_info(protocol_bs, &err);
+        if (err) {

block/qapi.c void bdrv_query_image_info(BlockDriverState *bs, · line 327

Quoted 10 lines
block/qapi.c +7 -0
+            error_propagate(errp, err);
+            qapi_free_ImageInfo(info);
+            goto out;
+        }
+        info->has_protocol_specific = info->protocol_specific != NULL;
+    }
+
     backing_filename = bs->backing_file;
     if (backing_filename[0] != '\0') {
         char *backing_filename2;

Kevin

#7 lore raw

Re: [PATCH 2/4] block: Add protocol-specific image info

Quoted 28 lines · Kevin Wolf

Am 03.05.2022 um 16:55 hat Hanna Reitz geschrieben:

Quoted 27 lines

The ImageInfo object currently only contains (optional) format-specific image information. However, perhaps the protocol node can provide some additional information, so add a new field presenting it.

---
  qapi/block-core.json |  6 +++++-
  block/qapi.c         | 19 +++++++++++++++++++
  2 files changed, 24 insertions(+), 1 deletion(-)
qapi/block-core.json +3 -0
index beeb91952a..e7d6c2e0cc 100644
@@ -236,6 +236,9 @@
  # @format-specific: structure supplying additional format-specific
  #                   information (since 1.7)
  #
+# @protocol-specific: structure supplying additional protocol-specific
+#                     information (since 7.1)
+#
  # Since: 1.3
  #
  ##
@@ -246,7 +249,8 @@
             '*backing-filename': 'str', '*full-backing-filename': 'str',
             '*backing-filename-format': 'str', '*snapshots': ['SnapshotInfo'],
Quoted 8 lines

qapi/block-core.json line 253

Quoted 4 lines
qapi/block-core.json +2 -1
             '*backing-image': 'ImageInfo',
-           '*format-specific': 'ImageInfoSpecific' } }
+           '*format-specific': 'ImageInfoSpecific',
+           '*protocol-specific': 'ImageInfoSpecific' } }

I'm not a fan of this one. It solves the problem for exactly one special case (even if admittedly a common one) and leaves everything else as it is. It is unclear what it produces in configurations that aren't the simple one format node on top of one protocol node layout.

I don’t disagree, but I do wonder how often this structure is used outside of qemu-img info, where filter nodes and more complex configurations are very rare.  I understand wanting to support complex block graph configurations everywhere, I’m just wondering whether there is actually much of a use for that here.

Quoted 2 lines

I would rather interpret 'format-specific' as 'driver-specific' and make the ImageInfo for any child node accessible.

Again, I don’t disagree, but I have reservations about that.  I don’t think this is a trivial approach to take.

First, that will be kind of bad for VMDK files, which already have all of their extent children in their driver-specific info, so we’d duplicate that info.

Second, same for the backing child, generally.  Do we want to exclude specifically the backing child from that list of ImageInfos for all children, because we already have an entry for it in ImageInfo itself?  That wouldn’t make much sense.  Deprecate backing-child?  Works for the future, weird in the meantime.

Third, the implementation would not be trivial. bdrv_query_image_info() specifically says to return "flat" image information, i.e. not to query the backing child information. Currently, its callers fill that blank some way or another, with qemu-img info creating a list of files (i.e. the backing chain) instead of using that backing-image field.  I actually have no idea how we should bring that together.  Should bdrv_query_image_info() also not collect that ImageInfo list of all children, and then collect_image_info_list() will put those into its list, too, making it recursive?  Then we have the problem of describing nodes in this graph, and as written below I wouldn’t be happy to use auto-generated node names for this.  Or should bdrv_query_image_info() collect all those children, and then collect_image_info_list() will just drop the backing child from them, so that it still gets a flat backing chain list, but the other children will be nested, allowing users to identify which nodes those are based on nesting?  (And nesting would require adding indentation support to bdrv_image_info_dump(), and bdrv_snapshot_dump().)

Fourth, precisely for the common case of not having filters or other more complex configurations, the additional info provided by the protocol node’s ImageInfo is limited.  Most of it just duplicates information from the format node, the really interesting bit is just the ImageInfoSpecific, so for qemu-img info it’ll mostly just clutter the output.  Many fields are also named on the assumption that this information is about a format node ("file format", "virtual size"), and so I personally find it confusing to see those things in the information about a protocol node when using qemu-img info.

Quoted 3 lines

With rbd we already interpret it like a generic driver thing that is not just for formats that because it implements .bdrv_get_specific_info even though we didn't have a 'protocol-specific' yet.

On one hand, that’s the same thing I’m doing in this series.  On the other, I think the rbd implementation as a whole has not been well thought out, because it must have faced exactly the same problem that I’m trying to solve in this patch here, but obviously it hasn’t been addressed yet.

(Instead, it probably relied on users calling qemu-img info -f rbd, which is just cheating.  I mean, I could do that, too, and just drop anything but patch 4.)

Quoted 2 lines

Making other nodes has precedence, too. 'backing-image' is even in the context of this hunk. VMDK exposes its extents the same way.

Both of which now make the solution to include the list of all children’s ImageInfos just more complicated, yes. O:)

(I know that me saying that simply means that these were probably bad solutions then, and that maybe we should’ve had a list of all children’s ImageInfos from the start.  Which means dancing around the issue even more won’t make it better, I know.  O:)  I’m just trying to say that simply adding this list isn’t an ideal solution now, under the current circumstances; but I’m not saying there is any ideal solution.)

Quoted 4 lines

So maybe
what we really want is a 'children' list with the ImageInfo of every child node. And then qemu-img could go through all children and print headings like "Driver specific information for file (#block123)".

I would very much rather drop auto-generated node names, and instead just print the child name and rely on indentation.  I have an example below.

Quoted 3 lines

Then
filters like blkdebug could add their information and it would be printed, too.

Is this really something that would ever be useful in practice?

I understand your concern (and share it to a degree), but I feel like allowing for this ImageInfo struct to represent and encompass a complex block graph comes at the detriment of readability and understandability of qemu-img info output for plain images.

For example, this is how I’d imagine the output for a raw image:

image: test.raw
file format: raw
virtual size: 10 GiB (10737418240 bytes)
disk size: 1 MiB
child 'file':
    image: test.raw
    file format: file
    virtual size: 10 GiB (10737418240 bytes)
    disk size: 1 MiB
    Driver specific information:
        extent size: 1048576

Personally, I like that less than what this series’s v1 produces.  I understand it represents the modular nature of the block graph, but that’s generally not something I want to see when I run qemu-img info on a plain image (which is 98 % of the use I have for qemu-img info).

Quoted 26 lines
Quoted 15 lines
qapi/block-core.json +0 -0
  ##
  # @ImageCheck:
block/qapi.c +1 -0
index 51202b470a..293983cf82 100644
@@ -262,6 +262,7 @@ void bdrv_query_image_info(BlockDriverState *bs,
      int64_t size;
      const char *backing_filename;
      BlockDriverInfo bdi;
+    BlockDriverState *protocol_bs;
      int ret;
      Error *err = NULL;
      ImageInfo *info;

block/qapi.c void bdrv_query_image_info(BlockDriverState *bs, · line 311

Quoted 8 lines
block/qapi.c +5 -0
@@ -303,6 +304,24 @@ void bdrv_query_image_info(BlockDriverState *bs,
      }
      info->has_format_specific = info->format_specific != NULL;
 
+    /* Try to look for an unambiguous protocol node */
+    protocol_bs = bs;
+    while (protocol_bs && !QLIST_EMPTY(&protocol_bs->children)) {
+        protocol_bs = bdrv_primary_bs(protocol_bs);
+    }

If bs is already a leaf node, this duplicates the information, which looks weird:

Quoted 9 lines
     $ build/qemu-img info -f file ~/tmp/test.raw
     image: /home/kwolf/tmp/test.raw
     file format: file
     virtual size: 10 GiB (10737418240 bytes)
     disk size: 7.63 GiB
     Format specific information:
         extent size: 1048576
     Protocol specific information:
         extent size: 1048576

I mean, that isn’t wrong, but also fixable if need be.

Quoted 18 lines
Quoted 6 lines
block/qapi.c +6 -0
+    if (protocol_bs) {
+        /* Assert that this is a protocol node */
+        assert(QLIST_EMPTY(&protocol_bs->children));
+
+        info->protocol_specific = bdrv_get_specific_info(protocol_bs, &err);
+        if (err) {

block/qapi.c void bdrv_query_image_info(BlockDriverState *bs, · line 327

Quoted 10 lines
block/qapi.c +7 -0
+            error_propagate(errp, err);
+            qapi_free_ImageInfo(info);
+            goto out;
+        }
+        info->has_protocol_specific = info->protocol_specific != NULL;
+    }
+
      backing_filename = bs->backing_file;
      if (backing_filename[0] != '\0') {
          char *backing_filename2;

Kevin

#8 lore raw

PATCH 3/4 block: Print protocol-specific information

In reply to Hanna Reitz

Make bdrv_image_info_dump() print protocol-specific information.

---
 block/qapi.c | 5 +++++
 1 file changed, 5 insertions(+)
block/qapi.c +5 -0
index 293983cf82..169ea08f70 100644
@@ -921,4 +921,9 @@ void bdrv_image_info_dump(ImageInfo *info)
         bdrv_image_info_specific_dump(info->format_specific,
                                       "Format specific information:\n");
     }
+
+    if (info->has_protocol_specific) {
+        bdrv_image_info_specific_dump(info->protocol_specific,
+                                      "Protocol specific information:\n");
+    }
 }
Signature · 1 line
2.35.1
Reviewed-by on v1 3/4
#9 lore raw

Re: [PATCH 3/4] block: Print protocol-specific information

Quoted 6 lines · Hanna Reitz

Make bdrv_image_info_dump() print protocol-specific information.

---
 block/qapi.c | 5 +++++
 1 file changed, 5 insertions(+)

(and now I should probably go write a patch to expose protocol information for NBD...)

Quoted 18 lines
block/qapi.c +5 -0
index 293983cf82..169ea08f70 100644
@@ -921,4 +921,9 @@ void bdrv_image_info_dump(ImageInfo *info)
         bdrv_image_info_specific_dump(info->format_specific,
                                       "Format specific information:\n");
     }
+
+    if (info->has_protocol_specific) {
+        bdrv_image_info_specific_dump(info->protocol_specific,
+                                      "Protocol specific information:\n");
+    }
 }
Signature · 1 line
2.35.1
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#10 lore raw

PATCH 4/4 block/file: Add file-specific image info

In reply to Hanna Reitz

Add some (optional) information that the file driver can provide for image files, namely the extent size.

---
 qapi/block-core.json | 26 ++++++++++++++++++++++++--
 block/file-posix.c   | 30 ++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+), 2 deletions(-)
qapi/block-core.json +24 -2
index e7d6c2e0cc..728da051ae 100644
@@ -139,16 +139,29 @@
       '*encryption-format': 'RbdImageEncryptionFormat'
   } }
 
+##
+# @ImageInfoSpecificFile:
+#
+# @extent-size: Extent size (if available)
+#
+# Since: 7.1
+##
+{ 'struct': 'ImageInfoSpecificFile',
+  'data': {
+      '*extent-size': 'size'
+  } }
+
 ##
 # @ImageInfoSpecificKind:
 #
 # @luks: Since 2.7
 # @rbd: Since 6.1
+# @file: Since 7.1
 #
 # Since: 1.7
 ##
 { 'enum': 'ImageInfoSpecificKind',
-  'data': [ 'qcow2', 'vmdk', 'luks', 'rbd' ] }
+  'data': [ 'qcow2', 'vmdk', 'luks', 'rbd', 'file' ] }
 
 ##
 # @ImageInfoSpecificQCow2Wrapper:
@@ -185,6 +198,14 @@
 { 'struct': 'ImageInfoSpecificRbdWrapper',
   'data': { 'data': 'ImageInfoSpecificRbd' } }
 
+##
+# @ImageInfoSpecificFileWrapper:
+#
+# Since: 7.1
+##
+{ 'struct': 'ImageInfoSpecificFileWrapper',
+  'data': { 'data': 'ImageInfoSpecificFile' } }
+
 ##
 # @ImageInfoSpecific:
 #
@@ -199,7 +220,8 @@
       'qcow2': 'ImageInfoSpecificQCow2Wrapper',
       'vmdk': 'ImageInfoSpecificVmdkWrapper',
       'luks': 'ImageInfoSpecificLUKSWrapper',
-      'rbd': 'ImageInfoSpecificRbdWrapper'
+      'rbd': 'ImageInfoSpecificRbdWrapper',
+      'file': 'ImageInfoSpecificFileWrapper'
   } }
 
 ##
block/file-posix.c +30 -0
index bfd9b21111..4323345c99 100644
@@ -3068,6 +3068,34 @@ static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
     return 0;
 }
 
+static ImageInfoSpecific *raw_get_specific_info(BlockDriverState *bs,
+                                                Error **errp)
+{
+    BDRVRawState *s = bs->opaque;
+    ImageInfoSpecificFile *file_info = g_new0(ImageInfoSpecificFile, 1);
+    ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
+
+    *spec_info = (ImageInfoSpecific){
+        .type = IMAGE_INFO_SPECIFIC_KIND_FILE,
+        .u.file.data = file_info,
+    };
+
+#ifdef FS_IOC_FSGETXATTR
+    {
+        struct fsxattr attr;
+        int ret;
+
+        ret = ioctl(s->fd, FS_IOC_FSGETXATTR, &attr);
+        if (!ret && attr.fsx_extsize != 0) {
+            file_info->has_extent_size = true;
+            file_info->extent_size = attr.fsx_extsize;
+        }
+    }
+#endif
+
+    return spec_info;
+}
+
 static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs)
 {
     BDRVRawState *s = bs->opaque;
@@ -3301,6 +3329,7 @@ BlockDriver bdrv_file = {
     .bdrv_co_truncate = raw_co_truncate,
     .bdrv_getlength = raw_getlength,
     .bdrv_get_info = raw_get_info,
+    .bdrv_get_specific_info = raw_get_specific_info,
     .bdrv_get_allocated_file_size
                         = raw_get_allocated_file_size,
     .bdrv_get_specific_stats = raw_get_specific_stats,
@@ -3673,6 +3702,7 @@ static BlockDriver bdrv_host_device = {
     .bdrv_co_truncate       = raw_co_truncate,
     .bdrv_getlength	= raw_getlength,
     .bdrv_get_info = raw_get_info,
+    .bdrv_get_specific_info = raw_get_specific_info,
     .bdrv_get_allocated_file_size
                         = raw_get_allocated_file_size,
     .bdrv_get_specific_stats = hdev_get_specific_stats,
Signature · 1 line
2.35.1
#11 lore raw

Re: [PATCH 4/4] block/file: Add file-specific image info

Quoted 29 lines · Hanna Reitz

Add some (optional) information that the file driver can provide for image files, namely the extent size.

---
 qapi/block-core.json | 26 ++++++++++++++++++++++++--
 block/file-posix.c   | 30 ++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+), 2 deletions(-)

+++ b/block/file-posix.c

block/file-posix.c +14 -0
@@ -3068,6 +3068,34 @@ static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
     return 0;
 }
 
+static ImageInfoSpecific *raw_get_specific_info(BlockDriverState *bs,
+                                                Error **errp)
+{
+    BDRVRawState *s = bs->opaque;
+    ImageInfoSpecificFile *file_info = g_new0(ImageInfoSpecificFile, 1);
+    ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
+
+    *spec_info = (ImageInfoSpecific){
+        .type = IMAGE_INFO_SPECIFIC_KIND_FILE,
+        .u.file.data = file_info,
+    };
+
+#ifdef FS_IOC_FSGETXATTR
+    {

block/file-posix.c static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) · line 3094

Quoted 10 lines
block/file-posix.c +10 -0
+        struct fsxattr attr;
+        int ret;
+
+        ret = ioctl(s->fd, FS_IOC_FSGETXATTR, &attr);
+        if (!ret && attr.fsx_extsize != 0) {
+            file_info->has_extent_size = true;
+            file_info->extent_size = attr.fsx_extsize;
+        }
+    }
+#endif

Can/should we fall back to stat's st_blksize when the ioctl produces nothing?

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#12 lore raw

Re: [PATCH 4/4] block/file: Add file-specific image info

Quoted 33 lines · Eric Blake
Quoted 32 lines · Hanna Reitz

Add some (optional) information that the file driver can provide for image files, namely the extent size.

---
  qapi/block-core.json | 26 ++++++++++++++++++++++++--
  block/file-posix.c   | 30 ++++++++++++++++++++++++++++++
  2 files changed, 54 insertions(+), 2 deletions(-)

+++ b/block/file-posix.c

block/file-posix.c +18 -0
@@ -3068,6 +3068,34 @@ static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
      return 0;
  }
 
+static ImageInfoSpecific *raw_get_specific_info(BlockDriverState *bs,
+                                                Error **errp)
+{
+    BDRVRawState *s = bs->opaque;
+    ImageInfoSpecificFile *file_info = g_new0(ImageInfoSpecificFile, 1);
+    ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
+
+    *spec_info = (ImageInfoSpecific){
+        .type = IMAGE_INFO_SPECIFIC_KIND_FILE,
+        .u.file.data = file_info,
+    };
+
+#ifdef FS_IOC_FSGETXATTR
+    {
+        struct fsxattr attr;
+        int ret;
+
+        ret = ioctl(s->fd, FS_IOC_FSGETXATTR, &attr);
Quoted 8 lines

block/file-posix.c static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) · line 3094

Quoted 6 lines
block/file-posix.c +6 -0
+        if (!ret && attr.fsx_extsize != 0) {
+            file_info->has_extent_size = true;
+            file_info->extent_size = attr.fsx_extsize;
+        }
+    }
+#endif

Can/should we fall back to stat's st_blksize when the ioctl produces nothing?

I don’t think so, that’s a different value.  For example, by default, we use an extent-size-hint of 1 MB for new images (which is applied at least on XFS), but that doesn’t change st_blksize (which is 4096 here).  So I’d only report the extent-size for filesystems that actually differentiate between the two.

Hanna

#13 lore raw

Re: [PATCH 4/4] block/file: Add file-specific image info

In reply to Hanna Reitz

Am 03.05.2022 um 16:55 hat Hanna Reitz geschrieben:

Quoted 18 lines

Add some (optional) information that the file driver can provide for image files, namely the extent size.

---
 qapi/block-core.json | 26 ++++++++++++++++++++++++--
 block/file-posix.c   | 30 ++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+), 2 deletions(-)
qapi/block-core.json +1 -0
index e7d6c2e0cc..728da051ae 100644
@@ -139,16 +139,29 @@
       '*encryption-format': 'RbdImageEncryptionFormat'
   } }
 
+##

qapi/block-core.json line 152

Quoted 10 lines
qapi/block-core.json +10 -0
+# @ImageInfoSpecificFile:
+#
+# @extent-size: Extent size (if available)
+#
+# Since: 7.1
+##
+{ 'struct': 'ImageInfoSpecificFile',
+  'data': {
+      '*extent-size': 'size'
+  } }

It's not "the extent size" (the whole point of extents is that they don't have a fixed size like blocks), but an extent size *hint* that tells the filesystem the minimum size to allocate for an extent. The xfs_io man page calls it the preferred extent size for allocatino, which works for the documentation if you prefer, but BlockdevCreateOptionsFile has 'extent-size-hint', so I'd prefer consistency on the wire at least.

Kevin

#14 lore raw

Re: [PATCH 4/4] block/file: Add file-specific image info

Quoted 27 lines · Kevin Wolf

Am 03.05.2022 um 16:55 hat Hanna Reitz geschrieben:

Quoted 26 lines

Add some (optional) information that the file driver can provide for image files, namely the extent size.

---
  qapi/block-core.json | 26 ++++++++++++++++++++++++--
  block/file-posix.c   | 30 ++++++++++++++++++++++++++++++
  2 files changed, 54 insertions(+), 2 deletions(-)
qapi/block-core.json +9 -0
index e7d6c2e0cc..728da051ae 100644
@@ -139,16 +139,29 @@
        '*encryption-format': 'RbdImageEncryptionFormat'
    } }
 
+##
+# @ImageInfoSpecificFile:
+#
+# @extent-size: Extent size (if available)
+#
+# Since: 7.1
+##
+{ 'struct': 'ImageInfoSpecificFile',
+  'data': {
Quoted 8 lines

qapi/block-core.json line 152

Quoted 2 lines
qapi/block-core.json +2 -0
+      '*extent-size': 'size'
+  } }

It's not "the extent size" (the whole point of extents is that they don't have a fixed size like blocks), but an extent size *hint* that tells the filesystem the minimum size to allocate for an extent. The xfs_io man page calls it the preferred extent size for allocatino, which works for the documentation if you prefer, but BlockdevCreateOptionsFile has 'extent-size-hint', so I'd prefer consistency on the wire at least.

Got it.