| Message ID | 20200204170848.614480-23-mreitz@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | block: Introduce real BdrvChildRole | expand |
On 2/4/20 11:08 AM, Max Reitz wrote: > Signed-off-by: Max Reitz <mreitz@redhat.com> Another sparse commit message (a recurring theme of this series). The subject line says 'what', and the patch appears to be faithful to that, but if a future bisection lands here, even a one-sentence 'why' would be handy; maybe: This is part of a larger series of unifying block device relationships via child_of_bds. to at least hint that searching nearby commits gives a better why. > --- > block.c | 26 ++++++++++++++++++++------ > block/backup-top.c | 2 +- > block/vvfat.c | 3 ++- > tests/test-bdrv-drain.c | 13 +++++++------ > 4 files changed, 30 insertions(+), 14 deletions(-) > > diff --git a/block.c b/block.c > index 77755f0c6c..6b705ee23a 100644 > --- a/block.c > +++ b/block.c > @@ -2748,6 +2748,20 @@ static bool bdrv_inherits_from_recursive(BlockDriverState *child, > return child != NULL; > } > > +/* > + * Return the BdrvChildRole for @bs's backing child. bs->backing is > + * mostly used for COW backing children (role = COW), but also for > + * filtered children (role = FILTERED | PRIMARY). > + */ > +static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) > +{ > + if (bs->drv && bs->drv->is_filter) { > + return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; And here's the first point (that I've spotted at least) in this series where you are definitely returning a non-BdrvChildRole through a return type of BdrvChildRole, rather than me just guessing you might (the integer formed by bitwise-or of two enum values is not itself an enum value). Repeating what I said earlier, the C language is loose enough to allow your usage, and your usage is somewhat better self-documenting than using an unsigned int; but it would not fly in other languages. So I won't insist you change it, but at least think about it. (And the latter has already happened if you read my paragraph - so can we call that enough thought on the matter? ;) > + } else { > + return BDRV_CHILD_COW; > + } > +} > + Reviewed-by: Eric Blake <eblake@redhat.com>
On 05.02.20 23:45, Eric Blake wrote: > On 2/4/20 11:08 AM, Max Reitz wrote: >> Signed-off-by: Max Reitz <mreitz@redhat.com> > > Another sparse commit message (a recurring theme of this series). The > subject line says 'what', and the patch appears to be faithful to that, > but if a future bisection lands here, even a one-sentence 'why' would be > handy; maybe: > > This is part of a larger series of unifying block device relationships > via child_of_bds. Sure, works for me. Or maybe: Make all parents of backing files pass the appropriate BdrvChildRole. By doing so, we can switch their BdrvChildClass over to the generic child_of_bds, which will do the right thing when given a correct BdrvChildRole. (Because actually the point of this series is not child_of_bds, but the BdrvChildRole, which allows the “Deal with filters” series to implement the child access functions in a more obvious way. I hope.) Max
On 2/6/20 5:33 AM, Max Reitz wrote: > On 05.02.20 23:45, Eric Blake wrote: >> On 2/4/20 11:08 AM, Max Reitz wrote: >>> Signed-off-by: Max Reitz <mreitz@redhat.com> >> >> Another sparse commit message (a recurring theme of this series). The >> subject line says 'what', and the patch appears to be faithful to that, >> but if a future bisection lands here, even a one-sentence 'why' would be >> handy; maybe: >> >> This is part of a larger series of unifying block device relationships >> via child_of_bds. > > Sure, works for me. Or maybe: > > Make all parents of backing files pass the appropriate BdrvChildRole. > By doing so, we can switch their BdrvChildClass over to the generic > child_of_bds, which will do the right thing when given a correct > BdrvChildRole. Sounds good. > > (Because actually the point of this series is not child_of_bds, but the > BdrvChildRole, which allows the “Deal with filters” series to implement > the child access functions in a more obvious way. I hope.) And hopefully I finish my review of the rest of the series today, to see if we met that goal.
diff --git a/block.c b/block.c index 77755f0c6c..6b705ee23a 100644 --- a/block.c +++ b/block.c @@ -2748,6 +2748,20 @@ static bool bdrv_inherits_from_recursive(BlockDriverState *child, return child != NULL; } +/* + * Return the BdrvChildRole for @bs's backing child. bs->backing is + * mostly used for COW backing children (role = COW), but also for + * filtered children (role = FILTERED | PRIMARY). + */ +static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) +{ + if (bs->drv && bs->drv->is_filter) { + return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; + } else { + return BDRV_CHILD_COW; + } +} + /* * Sets the backing file link of a BDS. A new reference is created; callers * which don't need their own reference any more must call bdrv_unref(). @@ -2775,8 +2789,8 @@ void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, goto out; } - bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing, - 0, errp); + bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds, + bdrv_backing_role(bs), errp); /* If backing_hd was already part of bs's backing chain, and * inherits_from pointed recursively to bs then let's update it to * point directly to bs (else it will become NULL). */ @@ -2873,7 +2887,7 @@ int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, } backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, - &child_backing, 0, errp); + &child_of_bds, BDRV_CHILD_COW, errp); if (!backing_hd) { bs->open_flags |= BDRV_O_NO_BACKING; error_prepend(errp, "Could not open backing file: "); @@ -3709,8 +3723,8 @@ int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) if (state->replace_backing_bs && state->new_backing_bs) { uint64_t nperm, nshared; bdrv_child_perm(state->bs, state->new_backing_bs, - NULL, &child_backing, 0, bs_queue, - state->perm, state->shared_perm, + NULL, &child_of_bds, bdrv_backing_role(state->bs), + bs_queue, state->perm, state->shared_perm, &nperm, &nshared); ret = bdrv_check_update_perm(state->new_backing_bs, NULL, nperm, nshared, NULL, NULL, errp); @@ -6656,7 +6670,7 @@ void bdrv_refresh_filename(BlockDriverState *bs) drv->bdrv_gather_child_options(bs, opts, backing_overridden); } else { QLIST_FOREACH(child, &bs->children, next) { - if (child->klass == &child_backing && !backing_overridden) { + if (child == bs->backing && !backing_overridden) { /* We can skip the backing BDS if it has not been overridden */ continue; } diff --git a/block/backup-top.c b/block/backup-top.c index 36ed71413b..8ac7009852 100644 --- a/block/backup-top.c +++ b/block/backup-top.c @@ -138,7 +138,7 @@ static void backup_top_child_perm(BlockDriverState *bs, BdrvChild *c, return; } - if (child_class == &child_file) { + if (!(role & BDRV_CHILD_FILTERED)) { /* * Target child * diff --git a/block/vvfat.c b/block/vvfat.c index 3bcf860af7..2bc5812dc5 100644 --- a/block/vvfat.c +++ b/block/vvfat.c @@ -3228,7 +3228,8 @@ static void vvfat_child_perm(BlockDriverState *bs, BdrvChild *c, { BDRVVVFATState *s = bs->opaque; - assert(c == s->qcow || child_class == &child_backing); + assert(c == s->qcow || + (child_class == &child_of_bds && (role & BDRV_CHILD_COW))); if (c == s->qcow) { /* This is a private node, nobody should try to attach to it */ diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index b3d7960bd0..15393a0140 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -96,7 +96,7 @@ static void bdrv_test_child_perm(BlockDriverState *bs, BdrvChild *c, * bdrv_format_default_perms() accepts only these two, so disguise * detach_by_driver_cb_parent as one of them. */ - if (child_class != &child_file && child_class != &child_backing) { + if (child_class != &child_file && child_class != &child_of_bds) { child_class = &child_file; } @@ -1399,8 +1399,8 @@ static void test_detach_indirect(bool by_parent_cb) bdrv_ref(a); child_b = bdrv_attach_child(parent_b, b, "PB-B", &child_file, 0, &error_abort); - child_a = bdrv_attach_child(parent_b, a, "PB-A", &child_backing, 0, - &error_abort); + child_a = bdrv_attach_child(parent_b, a, "PB-A", &child_of_bds, + BDRV_CHILD_COW, &error_abort); bdrv_ref(a); bdrv_attach_child(parent_a, a, "PA-A", @@ -1793,7 +1793,7 @@ static void test_drop_intermediate_poll(void) int i; int ret; - chain_child_class = child_backing; + chain_child_class = child_of_bds; chain_child_class.update_filename = drop_intermediate_poll_update_filename; for (i = 0; i < 3; i++) { @@ -1816,7 +1816,7 @@ static void test_drop_intermediate_poll(void) /* Takes the reference to chain[i - 1] */ chain[i]->backing = bdrv_attach_child(chain[i], chain[i - 1], "chain", &chain_child_class, - 0, &error_abort); + BDRV_CHILD_COW, &error_abort); } } @@ -2034,7 +2034,8 @@ static void do_test_replace_child_mid_drain(int old_drain_count, bdrv_ref(old_child_bs); parent_bs->backing = bdrv_attach_child(parent_bs, old_child_bs, "child", - &child_backing, 0, &error_abort); + &child_of_bds, BDRV_CHILD_COW, + &error_abort); for (i = 0; i < old_drain_count; i++) { bdrv_drained_begin(old_child_bs);
Signed-off-by: Max Reitz <mreitz@redhat.com> --- block.c | 26 ++++++++++++++++++++------ block/backup-top.c | 2 +- block/vvfat.c | 3 ++- tests/test-bdrv-drain.c | 13 +++++++------ 4 files changed, 30 insertions(+), 14 deletions(-)