[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v8 03/36] block: Introduce image file locking
From: |
Max Reitz |
Subject: |
Re: [Qemu-devel] [PATCH v8 03/36] block: Introduce image file locking |
Date: |
Fri, 21 Oct 2016 23:04:13 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.4.0 |
On 30.09.2016 14:09, Fam Zheng wrote:
> Block drivers can implement this new operation .bdrv_lockf to actually lock
> the
> image in the protocol specific way.
>
> Signed-off-by: Fam Zheng <address@hidden>
> ---
> block.c | 52
> +++++++++++++++++++++++++++++++++++++++++++++++
> include/block/block.h | 4 +++-
> include/block/block_int.h | 5 +++++
> include/hw/block/block.h | 2 ++
> 4 files changed, 62 insertions(+), 1 deletion(-)
>
> diff --git a/block.c b/block.c
> index 493ecf3..9d600df 100644
> --- a/block.c
> +++ b/block.c
> @@ -235,6 +235,7 @@ BlockDriverState *bdrv_new(void)
> notifier_with_return_list_init(&bs->before_write_notifiers);
> bs->refcnt = 1;
> bs->aio_context = qemu_get_aio_context();
> + bs->cur_lock = IMAGE_LOCK_MODE__MAX;
(Yes, I know, I'm supposed to write a high-level review, but...)
I don't really like using values for enums that are not actually
supposed to be part of the enum. Maybe nolock would be a reasonable choice?
> qemu_co_queue_init(&bs->flush_queue);
>
> @@ -925,6 +926,48 @@ out:
> g_free(gen_node_name);
> }
>
> +ImageLockMode bdrv_lock_mode_from_flags(int flags)
> +{
> + if (flags & BDRV_O_NO_LOCK) {
> + return IMAGE_LOCK_MODE_NOLOCK;
> + } else if (flags & BDRV_O_SHARED_LOCK) {
> + return IMAGE_LOCK_MODE_SHARED;
> + } else if (flags & BDRV_O_EXCLUSIVE_LOCK) {
> + return IMAGE_LOCK_MODE_EXCLUSIVE;
> + } else {
> + return IMAGE_LOCK_MODE_AUTO;
> + }
> +}
I don't know if there's been any discussion about the order of the flags
here, but I personally would order them exactly the other way around:
Asking for exclusive locking should override nolock, in my opinion.
> +
> +ImageLockMode bdrv_get_lock_mode(BlockDriverState *bs)
> +{
> + return bs->cur_lock;
> +}
> +
> +int bdrv_set_lock_mode(BlockDriverState *bs, ImageLockMode mode)
> +{
> + int ret;
> +
> + if (bs->cur_lock == mode) {
> + return 0;
> + } else if (!bs->drv) {
> + return -ENOMEDIUM;
> + } else if (!bs->drv->bdrv_lockf) {
> + if (bs->file) {
> + return bdrv_set_lock_mode(bs->file->bs, mode);
> + }
> + return 0;
> + }
> + ret = bs->drv->bdrv_lockf(bs, mode);
> + if (ret == -ENOTSUP) {
> + /* Handle it the same way as !bs->drv->bdrv_lockf */
> + ret = 0;
Yes, well, why do you handle both as success? Wouldn't returning
-ENOTSUP make more sense?
I guess the caller can find out itself by checking whether bs->cur_lock
has changed, but...
> + } else if (ret == 0) {
> + bs->cur_lock = mode;
> + }
> + return ret;
> +}
> +
> static QemuOptsList bdrv_runtime_opts = {
> .name = "bdrv_common",
> .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
> @@ -1076,6 +1119,10 @@ static int bdrv_open_common(BlockDriverState *bs,
> BdrvChild *file,
> goto free_and_fail;
> }
>
> + if (open_flags & BDRV_O_INACTIVE) {
> + open_flags = (open_flags & ~BDRV_O_LOCK_MASK) & BDRV_O_NO_LOCK;
I suppose the second & is supposed to be a |?
> + }
> +
> ret = refresh_total_sectors(bs, bs->total_sectors);
> if (ret < 0) {
> error_setg_errno(errp, -ret, "Could not refresh total sector count");
> @@ -2273,6 +2320,7 @@ static void bdrv_close(BlockDriverState *bs)
> if (bs->drv) {
> BdrvChild *child, *next;
>
> + bdrv_set_lock_mode(bs, IMAGE_LOCK_MODE_NOLOCK);
> bs->drv->bdrv_close(bs);
> bs->drv = NULL;
>
> @@ -3188,6 +3236,9 @@ void bdrv_invalidate_cache(BlockDriverState *bs, Error
> **errp)
This function's name is pretty weird... Maybe it would be better to
rename it to "bdrv_complete_incoming" or something. (Unrelated to this
series, of course.)
> error_setg_errno(errp, -ret, "Could not refresh total sector count");
> return;
> }
> + if (bs->cur_lock != IMAGE_LOCK_MODE__MAX) {
> + bdrv_set_lock_mode(bs, bs->cur_lock);
> + }
> }
>
> void bdrv_invalidate_cache_all(Error **errp)
> @@ -3230,6 +3281,7 @@ static int bdrv_inactivate_recurse(BlockDriverState *bs,
> }
>
> if (setting_flag) {
> + ret = bdrv_set_lock_mode(bs, IMAGE_LOCK_MODE_NOLOCK);
Maybe it would make sense to do something with the return value...? :-)
At least you should probably check whether bdrv_get_lock_mode(bs) ==
IMAGE_LOCK_MODE_NOLOCK.
Max
> bs->open_flags |= BDRV_O_INACTIVE;
> }
> return 0;
signature.asc
Description: OpenPGP digital signature
- Re: [Qemu-devel] [PATCH v8 03/36] block: Introduce image file locking,
Max Reitz <=