[pve-devel] [PATCH qemu] fix #1534: vma: Add extract filter for disk images

[Fiona Ebner]

Am 07.08.23 um 12:08 schrieb Filip Schauer:
> Add a filter to the "vma extract" command. A wildcard can be passed with
> "-f" to match disk images that should be extracted.

The bug report doesn't mention wildcards, but a list of image names.
Restoring drive-ide2 and drive-scsi0 becomes awkward with a single
pattern and I don't see why a list of just the names shouldn't suffice.
Cases with many disks are rather rare and even in extreme cases it won't
be more than ~30 disks or so. And if users request patterns, we can
still add it on top.

> @@ -2064,12 +2069,12 @@ index 0000000000..304f02bc84
>  +
>  +    int i;
>  +    int vmstate_fd = -1;
> -+    guint8 vmstate_stream = 0;
> ++    uint8_t filter_bitmap[256 / 8];

I'd rather use a bool array (if the compiler optimizes it to a bitmap
great, otherwise it's still easier to read), but...

> ++    memset(filter_bitmap, 0, sizeof(filter_bitmap));
>  +
>  +    for (i = 1; i < 255; i++) {
>  +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
>  +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
> -+            vmstate_stream = i;
>  +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
>  +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
>  +            if (vmstate_fd < 0) {
> @@ -2078,6 +2083,16 @@ index 0000000000..304f02bc84
>  +            }
>  +            g_free(statefn);
>  +        } else if (di) {
> ++            if (filter && (fnmatch(filter, di->devname, 0) != 0)) {
> ++                if (vma_reader_register_bs(vmar, i, NULL, true, true, &errp) < 0) {
> ++                    g_error("%s", error_get_pretty(errp));
> ++                }
> ++
> ++                continue;
> ++            }

...instead of doing this, you could also just make sure not to allocate
an image when the name is not included and set the skip flag when
calling vma_reader_register_bs(), re-using the existing mechanism for
partial restore.

Does not registering all even work? I thought vma was written in such a
way that it would read everything, which is why the skip logic was
necessary IIRC. Peeking at the code, doesn't restore_write_data() call
restore_write_data() with a dev_id from the backup we didn't register
the stream for at some point otherwise?