[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH] Added timestamps to storage volumes

On 16.07.2012 09:45, Hendrik Schwartke wrote:
On 13.07.2012 17:14, Eric Blake wrote:
On 07/13/2012 08:38 AM, Hendrik Schwartke wrote:
!!! DON'T PUSH until stat-time lgpl 3 issue is fixed
!!! To tests this change lgpl version to 3 in bootstrap.conf:176

The access, birth, modification and change times are added to
storage volumes and corresponding xml representations.
  bootstrap.conf                |    1 +
  docs/formatstorage.html.in    |   13 +++++++++++++
docs/schemas/storagevol.rng | 36 ++++++++++++++++++++++++++++++++++++
  src/conf/storage_conf.c       |   18 ++++++++++++++++++
  src/conf/storage_conf.h       |   13 +++++++++++++
  src/storage/storage_backend.c |    6 ++++++
  6 files changed, 87 insertions(+)

diff --git a/bootstrap.conf b/bootstrap.conf
index 9b42cbf..da0b960 100644
--- a/bootstrap.conf
+++ b/bootstrap.conf
@@ -115,6 +115,7 @@ vc-list-files
Insert in sorted order.

@@ -172,6 +177,14 @@
          contains the MAC (eg SELinux) label string.
<span class="since">Since 0.4.1</span>
+<dd>Provides timing information about the volume. The four sub elements
since btime is omitted on Linux, maybe this would read better as 'Up to
four sub-elements are present, where'

+<code>atime</code>,<code>btime</code>,<code>ctime</code> and<code>mtime</code> + hold the access, birth, change and modification time of the volume, where known. + The used time format is&lt;seconds&gt;.&lt;nanoseconds&gt; since the beginning + of the epoch. This is a readonly attribute and is ignored when creating
+        a volume.<span class="since">Since 0.10.0</span>

+<define name='timestamps'>
+<element name='timestamps'>
+<element name='atime'>
+<data type="string">
+<param name="pattern">[0-9]+\.[0-9]+</param>
It might be worth writing the regex to permit eliding the sub-second
resolution, on file systems that only have 1 second resolution.  Given
Well, the problem here is that stat-time doesn't offer a way to determine if sub-second resolution is available. If the system doesn't support it then tv_nsec is simply zero. So there is always a sub-second part in the timestamp and such an regex could be slightly misleading. I will change it anyway and add a comment to the schema.
that we are repeating this<data>  four times, it might be worth defining
it, for a shorter diff:

<element name='atime'>
<ref name='timestamp'/>

<define name='timestamp'>
<data type='string'>
<param name='pattern'>[0-9]+(\.[0-9]+)?</param>

+++ b/src/conf/storage_conf.c
@@ -1277,6 +1277,24 @@ virStorageVolTargetDefFormat(virStorageVolOptionsPtr options,


+    virBufferAddLit(buf, "<timestamps>\n");
+    virBufferAsprintf(buf, "<atime>%llu.%ld</atime>\n",
+ (unsigned long long) def->timestamps.atime.tv_sec,
+                      def->timestamps.atime.tv_nsec);
Eliding a sub-second suffix when tv_nsec == 0 would be easier with a
helper function:

virStorageVolTimestampFormat(virBufferPtr buf, const char *name,
                              struct timespec *ts)
     if (ts->tv_nsec<  0)
That's never the case. See above.
Yups, wrong line. Of course that could be the case. But again I prefer to check tv_sec also.
     virBufferAsprintf(buf, "<%s>%llu", name,
                       (unsigned long long) ts->tv_sec);
     if (ts->tv_nsec)
That the line I wanted to comment. I'm not sure if it's such a good idea to omit the sub second part. Although it's very unlikely that this happends on systems that support tv_nsec it could be misleading.
         virBufferAsprintf(buf, ".%ld", tv->tv_nsec);
     virBufferAsprintf(buf, "</%s>\n", name);

called as:

virStorageVolTimestampFormat(buf, "atime",&def->timestamps.atime);
virStorageVolTimestampFormat(buf, "atime",&def->timestamps.btime);

and so on.

Actually, I'd list atime, mtime, ctime, btime - in that order - rather
than trying to sort the names alphabetically (that is, match typical
'struct stat' ordering).
Well I thought about that and I think it's better to sort it alphabetically, because everyone who doesn't know 'struct stat' could be very puzzled about atime, mtime, ctime, btime.

+typedef virStorageTimestamps *virStorageTimestampsPtr;
+struct _virStorageTimestamps {
+    struct timespec atime;
+    /* if btime.tv_sec == -1&&  btime.tv_nsec == -1 than
+     * birth time is unknown
Doesn't gnulib guarantee that tv_nsec == -1 in isolation is sufficient
to point out an unknown value? That is, checking tv_sec == -1 is overhead.
Well, actually yes, but the the description on get_stat_birthtime says: "Return *ST's birth time, if available; otherwise return a value with tv_sec and tv_nsec both equal to -1.". So to be sure I prefer to check both.
Looking nicer.  I'll have to ping upstream on gnulib about the last
holdout on the relicensing of stat-time; and I'm also still waiting for
the security fix in updated automake to hit Fedora.

Ok, please let me know if there are some changes here. Meanwhile I will adapt my patch.

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]