On Thursday, 5 October 2017 17:36:09 CEST Richard W.M. Jones wrote:
--- inspector/inspector.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/inspector/inspector.c b/inspector/inspector.c index 3583c61df..30d279987 100644 --- a/inspector/inspector.c +++ b/inspector/inspector.c @@ -347,6 +347,7 @@ output_root (xmlTextWriterPtr xo, char *root) char buf[32]; char *canonical_root; size_t size; + int is_bsd;
XMLERROR (-1, xmlTextWriterStartElement (xo, BAD_CAST "operatingsystem"));
@@ -362,6 +363,10 @@ output_root (xmlTextWriterPtr xo, char *root) if (STRNEQ (str, "unknown")) XMLERROR (-1, xmlTextWriterWriteElement (xo, BAD_CAST "name", BAD_CAST str)); + is_bsd = + STREQ (str, "freebsd") || + STREQ (str, "netbsd") || + STREQ (str, "openbsd"); free (str);
str = guestfs_inspect_get_arch (g, root); @@ -451,8 +456,13 @@ output_root (xmlTextWriterPtr xo, char *root)
/* We need to mount everything up in order to read out the list of * applications and the icon, ie. everything below this point. + * + * XXX As a workaround for BSD guests, because the Linux kernel + * driver cannot just mount a UFS filesystem, we must disable this + * for all *BSD operating systems. We cannot read the apps or icon + * from *BSD anyway. */
This is not true, libguestfs can actually read those properties. The proof of that is running virt-inspector with --no-applications, and --no-icon shows the proper details (such as product name, mount points, host name, etc) of the guest. The trick here is what check_for_filesystem_on in daemon/inspect_fs.ml does, by trying few mount options for ufs filesystems: usually one of those attempts succeeds, so we extract the information from that partition. Later on in virt-inspector, though, these extra mount options are not retained, so it fails. The two possible options I had in mind were: a) duplicate the same mount logic also in virt-inspector, which is ugly, duplicates code, etc b) save these extra mount options, exposing them via a new inspect_* API: IMHO it is ugly as well, since it would "make official" the extra mount attempts hack tried in inspect_fs.ml but both are ugly, in one way or another, so I did not implement either of them. Anyway, IMHO this patch is not the correct approach. -- Pino Toscano