On 10.01.2014 18:55, Kevin Wolf wrote:
Am 10.01.2014 um 18:27 hat Max Reitz geschrieben:
On 09.01.2014 11:59, Kevin Wolf wrote:
[ CCing Max, who was recently active in this area, for another opinion ]
Am 08.01.2014 um 20:43 hat Peter Feiner geschrieben:
When a backing file is opened such that (1) a protocol is directly
used as the block driver and (2) the block driver has bdrv_file_open,
bdrv_open_backing_file segfaults. The problem arises because
bdrv_open_common returns without setting bd->backing_hd->file.
To effect (1), you seem to have to use the -F flag in qemu-img. There
are several block drivers that satisfy (2), such as "file" and "nbd".
Here are some concrete examples:
#!/bin/bash
echo Test file format
./qemu-img create -f file base.file 1m
./qemu-img create -f qcow2 -F file -o backing_file=base.file\
file-overlay.qcow2
./qemu-img convert -O raw file-overlay.qcow2 file-convert.raw
echo Test nbd format
SOCK=$PWD/nbd.sock
./qemu-img create -f raw base.raw 1m
./qemu-nbd -t -k $SOCK base.raw &
trap "kill $!" EXIT
while ! test -e $SOCK; do sleep 1; done
./qemu-img create -f qcow2 -F nbd -o backing_file=nbd:unix:$SOCK\
nbd-overlay.qcow2
./qemu-img convert -O raw nbd-overlay.qcow2 nbd-convert.raw
Without this patch, the two qemu-img convert commands segfault.
This is a regression that was introduced in v1.7 by
dbecebddfa4932d1c83915bcb9b5ba5984eb91be.
Signed-off-by: Peter Feiner <pe...@gridcentric.ca>
---
block.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/block.c b/block.c
index 64e7d22..a4a172d 100644
--- a/block.c
+++ b/block.c
@@ -1016,8 +1016,9 @@ int bdrv_open_backing_file(BlockDriverState *bs, QDict
*options, Error **errp)
error_free(local_err);
return ret;
}
- pstrcpy(bs->backing_file, sizeof(bs->backing_file),
- bs->backing_hd->file->filename);
+ if (bs->backing_hd->file)
+ pstrcpy(bs->backing_file, sizeof(bs->backing_file),
+ bs->backing_hd->file->filename);
return 0;
}
I think if there is no bs->backing_hd->file, we should get the filename
>from bs->backing_hd->filename instead of leaving it empty.
In fact, can we always do that or does bs->backing_hd normally lack the
filename? If so, perhaps that is what we need to fix, so we can always
directly use bs->backing_hd->filename here.
bs->backing_hd->filename would be set by the bdrv_open_common() in
bdrv_open(), the filename is read from file->filename (if file !=
NULL; in this case, that would be bs->backing_hd->file->filename) or
from the configuration option "filename".
The latter configuration option is not used by
bdrv_open_backing_file(), as far as I can see. However,
bs->backing_hd->file->filename is exactly the field the old code
uses, therefore, using bs->backing_hd->filename directly should not
break anything.
However, the patch does something different: If file is NULL, it
leaves bs->backing_file unchanged; whereas using
bs->backing_hd->filename would in this case result in the value of
the "filename" option. I think leaving bs->backing_file unchanged is
probably better, unless it is "" and the "filename" option is set.
If we want bs->backing_hd->filename to always point to a valid
filename, we'd probably have to copy to contents of bs->backing_file
there at some point in time, if it is not valid. But this is exactly
a point in code where bs->backing_file is updated, so there'd be no
gain if we instead updated bs->backing_hd->filename if necessary and
then copied that to bs->backing_file, as long as there is no other
place in the code where bs->backing_hd->filename always has to be a
valid filename.
Thus, I think the patch is okay, but I'd probably prefer "if
(bs->backing_hd->filename[0]) pstrcpy(...,
bs->backing_hd->filename);" - although that should not differ from
the given patch, unless the "filename" option is set for the
backing_hd.
Ok, if you're happy with it, I'll apply it. Can I put your Reviewed-by
there?
Yes, feel free to.
In the long run, we need to get rid of all this copying anyway. I'm
imagining a BlockDriver function that returns a file name to reproduce
the same setup, and a removal of bs->backing_file and bs->file_name.
For some drivers, the returned filename would be a URL or some other
string that that particular driver can parse.
While doing that, we might also consider a fake protocol that handles
filenames like 'json:{"driver":"qcow2","lazy-refcounts":"on",...}',
because for some drivers this might be the only thing that comes close
to a filename as it is a single string at least...
Urgh. *g*
I'm not sure if we should force every BDS to have a clearly defining
file name. If there are options, which completely change the behavior of
the block driver (I wouldn't consider lazy-refcounts one of them since
it doesn't change the contents of the block device), I'd rather return
NULL when asked for a file name. But then again, maybe an ugly filename
is better than none at all…
In general, I'd prefer abandoning filenames* (especially protocol
filenames) altogether. The set of options with which to recreate the
same BDS is already available.
Max
*Of course, we need filenames for, well, opening files, but I'm
referring to have an explicit string "filename" in addition to the
option dicts (nearly) everywhere. And as far as I see it, we don't
really want the user to specify a filename outside of the options
anymore anyway, therefore we should probably not encourage him/her to do
so by introducing filename abominations which may contain all options. ;-)