Le Monday 03 Feb 2014 à 21:22:45 (+0100), Max Reitz a écrit :
> On 03.02.2014 20:11, Benoît Canet wrote:
> >From: Benoît Canet <ben...@irqsave.net>
> >
> >Example of command line:
> >-drive if=virtio,file.driver=quorum,\
> >file.children.0.file.filename=1.raw,\
> >file.children.0.node-name=1.raw,\
> >file.children.0.driver=raw,\
> >file.children.1.file.filename=2.raw,\
> >file.children.1.node-name=2.raw,\
> >file.children.1.driver=raw,\
> >file.children.2.file.filename=3.raw,\
> >file.children.2.node-name=3.raw,\
> >file.children.2.driver=raw,\
> >file.vote_threshold=2
> >
> >file.blkverify=on with file.vote_threshold=2 and two files can be passed to
> >emulated blkverify.
> >
> >Signed-off-by: Benoit Canet <ben...@irqsave.net>
> >---
> > block/quorum.c | 169
> > +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > qapi-schema.json | 21 ++++++-
> > 2 files changed, 189 insertions(+), 1 deletion(-)
> >
> >diff --git a/block/quorum.c b/block/quorum.c
> >index a4716b3..582bf2c 100644
> >--- a/block/quorum.c
> >+++ b/block/quorum.c
> >@@ -17,8 +17,12 @@
> > #include <gnutls/crypto.h>
> > #include "block/block_int.h"
> > #include "qapi/qmp/qjson.h"
> >+#include "qapi/qmp/types.h"
> >+#include "qemu-common.h"
> > #define HASH_LENGTH 32
> >+#define KEY_PREFIX "children."
> >+#define KEY_FILENAME_SUFFIX ".file.filename"
> > /* This union holds a vote hash value */
> > typedef union QuorumVoteValue {
> >@@ -720,12 +724,177 @@ static bool
> >quorum_recurse_is_first_non_filter(BlockDriverState *bs,
> > return false;
> > }
> >+static int quorum_valid_threshold(int threshold,
> >+ int total,
> >+ Error **errp)
> >+{
> >+
> >+ if (threshold < 1) {
> >+ error_set(errp, QERR_INVALID_PARAMETER_VALUE,
> >+ "vote-threshold", "value >= 1");
> >+ return -ERANGE;
> >+ }
> >+
> >+ if (threshold > total) {
> >+ error_setg(errp, "threshold may not exceed children count");
> >+ return -ERANGE;
> >+ }
> >+
> >+ return 0;
> >+}
> >+
> >+static int quorum_open(BlockDriverState *bs,
> >+ QDict *options,
> >+ int flags,
> >+ Error **errp)
> >+{
> >+ BDRVQuorumState *s = bs->opaque;
> >+ Error *local_err = NULL;
> >+ bool *opened;
> >+ QDict *sub = NULL;
> >+ QList *list = NULL;
> >+ const QListEntry *lentry;
> >+ const QDictEntry *dentry;
> >+ const char *value;
> >+ char *next;
> >+ int i;
> >+ int ret = 0;
> >+ unsigned long long threshold = 0;
> >+
> >+ qdict_extract_subqdict(options, &sub, "children.");
> >+ qdict_array_split(sub, &list);
> >+
> >+ /* count how many different children are present and validate */
> >+ s->total = !qlist_size(list) ? qdict_size(sub) : qlist_size(list);
> >+ if (s->total < 2) {
> >+ error_setg(&local_err,
> >+ "Number of provided children must be greater than 1");
> >+ ret = -EINVAL;
> >+ goto exit;
> >+ }
> >+
> >+ ret = qdict_get_try_int(options, "vote-threshold", -1);
> >+ /* from QMP */
> >+ if (ret != -1) {
> >+ qdict_del(options, "vote-threshold");
> >+ s->threshold = ret;
> >+ /* from command line */
> >+ } else {
> >+ /* retrieve the threshold option from the command line */
> >+ value = qdict_get_try_str(options, "vote_threshold");
> >+ if (!value) {
> >+ error_setg(&local_err,
> >+ "vote_threshold must be provided");
> >+ ret = -EINVAL;
> >+ goto exit;
> >+ }
> >+ qdict_del(options, "vote_threshold");
> >+
> >+ ret = parse_uint(value, &threshold, &next, 10);
> >+
> >+ /* no int found -> scan fail */
> >+ if (ret < 0) {
> >+ error_setg(&local_err,
> >+ "invalid vote_threshold specified");
> >+ ret = -EINVAL;
> >+ goto exit;
> >+ }
> >+ s->threshold = threshold;
> >+ }
> >+
> >+ /* and validate it againt s->total */
>
> One step closer, but "against" is still one letter away. ;-)
>
> >+ ret = quorum_valid_threshold(s->threshold, s->total, &local_err);
> >+ if (ret < 0) {
> >+ goto exit;
> >+ }
> >+
> >+ /* is the driver in blkverify mode */
> >+ value = qdict_get_try_str(options, "blkverify");
> >+ if (value && !strcmp(value, "on") &&
> >+ s->total == 2 && s->threshold == 2) {
> >+ s->is_blkverify = true;
> >+ } else if (value && strcmp(value, "off")) {
> >+ fprintf(stderr, "blkverify mode is set by setting blkverify=on "
> >+ "and using two files with vote_threshold=2");
>
> This probably needs a \n at the end of the format string.
>
> >+ }
> >+ qdict_del(options, "blkverify");
> >+
> >+ /* allocate the children BlockDriverState array */
> >+ s->bs = g_new0(BlockDriverState *, s->total);
> >+ opened = g_new0(bool, s->total);
> >+
> >+ /* Opening by file name or options */
> >+ for (i = 0, lentry = qlist_first(list);
> >+ s->total == qlist_size(list) && lentry;
>
> I know I said I find it clever to put such a condition into the loop
> condition instead of wrapping an if around it, but this time, I'd
> prefer the wrapping if for s->total == qlist_size(size) instead. ;-)
>
> This is not a trivial loop and the compiler may be unable to verify
> that this condition doesn't change on each iteration. Also, an if
> with this loop in the if branch and the following loop in the else
> branch would be much more evident in what it does, at least to me.
>
> >+ lentry = qlist_next(lentry), i++) {
> >+ ret = bdrv_open(&s->bs[i], NULL, NULL,
> >qobject_to_qdict(lentry->value),
> >+ flags, NULL, &local_err);
> >+ if (ret < 0) {
> >+ goto close_exit;
> >+ }
> >+ opened[i] = true;
> >+ }
> >+
> >+ /* Opening by reference */
> >+ for (i = 0, dentry = qdict_first(sub);
> >+ s->total == qdict_size(sub) && dentry;
> >+ dentry = qdict_next(sub, dentry), i++) {
> >+ ret = bdrv_open(&s->bs[i], NULL,
> >+ qstring_get_str(qobject_to_qstring(dentry->value)),
> >+ NULL, flags, NULL, &local_err);
> >+ if (ret < 0) {
> >+ goto close_exit;
> >+ }
> >+ opened[i] = true;
> >+ }
>
> Okay, I see… Using qdict_array_split() doesn't help you with QMP
> invocation, so now you're not verifying the indices in that case.
> Well, we have three options:
> 1) Leave it like this.
> 2) Check whether the QDict keys are actually contiguous (this
> shouldn't be too much effort, I hope…) and pull the index here ("i")
> from the QDict key.
> 3) Use qdict_flatten(options) before everything and thus make the
> difference between QMP and command-line invocation disappear.
>
> I'd go with 3, especially considering that it should have been
> already called in qmp_blockdev_add() (and therefore we don't even
> have to call it here).
>
> I don't know why exactly you have this special code for QMP in the
> first place; perhaps it is because I extended qdict_flatten() to
> cover QLists not until November or December last year. Until then,
> qdict_flatten() stopped at QLists and only flattened QDicts; but
> now, it will flatten QLists as well, in a manner consistent with
> qdict_array_split().
This special code is not for QMP but for the QMP by references string.
dict.children will be a list of qstring and qdict_array_split will leave them
in the original
qdict.
qdict.children is not a dict of string list.
Moving the qstring in the qlist will require special case handling of the
correct QTYPE in the bdrv_open loop.
I don't know what is the best way to handle this.
Best regards
Benoît
>
> Thus, I think the extra handling for QMP should be unnecessary by
> now. If you have anything to test this hypothesis, by any chance,
> could you try it out?
>
> Max
