[Devel] [PATCH VZ9 3/3] block/blk-cbt: add BLKCBTLIST ioctl
Andrey Zhadchenko
andrey.zhadchenko at virtuozzo.com
Thu Jan 23 13:47:33 MSK 2025
On 1/23/25 10:53, Pavel Tikhomirov wrote:
>
>
> On 1/23/25 17:21, Andrey Zhadchenko wrote:
>>
>>
>> On 1/23/25 09:15, Pavel Tikhomirov wrote:
>>>
>>>
>>> On 1/20/25 16:34, Andrey Zhadchenko wrote:
>>>> so user can query the list of active bitmaps.
>>>>
>>>> https://virtuozzo.atlassian.net/browse/VSTOR-96269
>>>> Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko at virtuozzo.com>
>>>> ---
>>>> block/blk-cbt.c | 38 ++++++++++++++++++++++++++++++++++++++
>>>> block/ioctl.c | 1 +
>>>> include/uapi/linux/fs.h | 9 +++++++++
>>>> 3 files changed, 48 insertions(+)
>>>>
>>>> diff --git a/block/blk-cbt.c b/block/blk-cbt.c
>>>> index 8b0d07ee757c..af242633acc9 100644
>>>> --- a/block/blk-cbt.c
>>>> +++ b/block/blk-cbt.c
>>>> @@ -979,6 +979,42 @@ static int cbt_ioc_set(struct block_device
>>>> *bdev, struct blk_user_cbt_info __use
>>>> return ret;
>>>> }
>>>> +static int cbt_ioc_list(struct block_device *bdev, void __user *arg)
>>>> +{
>>>> + void __user *name_ptr = (void *)((size_t)arg + sizeof(struct
>>>> blk_user_cbt_list));
>>>> + struct request_queue *q = bdev_get_queue(bdev);
>>>> + struct blk_user_cbt_list lst;
>>>> + __u32 total = 0, copied = 0;
>>>> + struct cbt_info *cbt;
>>>> +
>>>> + if (copy_from_user(&lst, arg, sizeof(lst)))
>>>> + return -EFAULT;
>>>> +
>>>> + if (!access_ok(name_ptr,
>>>> + lst.count * CBT_NAME_LENGTH))
>>>> + return -EFAULT;
>>>> +
>>>> + rcu_read_lock();
>>>> + list_for_each_entry_rcu(cbt, &q->cbt_list, list) {
>>>> + if (copied < lst.count) {
>>>> + if (copy_to_user(name_ptr, cbt->name, CBT_NAME_LENGTH))
>>>> + return -EFAULT;
>>>> + copied++;
>>>> + name_ptr = (void *)((size_t)name_ptr + CBT_NAME_LENGTH);
>>>> + }
>>>> + total++;
>>>> + }
>>>> + rcu_read_unlock();
>>>> +
>>>> + lst.count = copied;
>>>> + lst.total_count = total;
>>>> +
>>>> + if (copy_to_user(arg, &lst, sizeof(lst)))
>>>> + return -EFAULT;
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> static int cbt_ioc_misc(struct block_device *bdev, void __user *arg)
>>>> {
>>>> struct request_queue *q = bdev_get_queue(bdev);
>>>> @@ -1024,6 +1060,8 @@ int blk_cbt_ioctl(struct block_device *bdev,
>>>> unsigned cmd, char __user *arg)
>>>> return cbt_ioc_set(bdev, ucbt_ioc, 0);
>>>> case BLKCBTMISC:
>>>> return cbt_ioc_misc(bdev, arg);
>>>> + case BLKCBTLIST:
>>>> + return cbt_ioc_list(bdev, arg);
>>>> default:
>>>> BUG();
>>>> }
>>>> diff --git a/block/ioctl.c b/block/ioctl.c
>>>> index b2d3dbe5bef5..6e76d55790ea 100644
>>>> --- a/block/ioctl.c
>>>> +++ b/block/ioctl.c
>>>> @@ -569,6 +569,7 @@ static int blkdev_common_ioctl(struct
>>>> block_device *bdev, blk_mode_t mode,
>>>> case BLKCBTSET:
>>>> case BLKCBTCLR:
>>>> case BLKCBTMISC:
>>>> + case BLKCBTLIST:
>>>> return blk_cbt_ioctl(bdev, cmd, (char __user *)arg);
>>>> default:
>>>> return -ENOIOCTLCMD;
>>>> diff --git a/include/uapi/linux/fs.h b/include/uapi/linux/fs.h
>>>> index 667157fe864a..d4641a8a70a7 100644
>>>> --- a/include/uapi/linux/fs.h
>>>> +++ b/include/uapi/linux/fs.h
>>>> @@ -169,12 +169,21 @@ struct blk_user_cbt_snap_create {
>>>> __u64 size;
>>>> };
>>>> +struct blk_user_cbt_list {
>>>> + __u32 count; /* Amount of name entries in names
>>>> + * in - available, out - filled
>>>> + */
>>>> + __u32 total_count; /* Total bitmap count of this device
>>>> (out) */
>>>
>>> Why do we need to know how many cbts were filled? It would always be
>>> min(available, total). So it seems we can use just one "count"
>>> variable with the same effect.
>>
>> The only important case is when the amount of cbts is bigger then the
>> provided buffer. e.g. someone added a cbt in-between calls or someone
>> supplied static size buffer with the hopes that it will be enough
>
> Still, my point is that you don't need two variables to handle it.
Why not? If you want to list every bitmap and someone does create a new
one between first and second ioctls, you won't know.
But I do not check it after second ioctl anyway, so probably we should
reduce it to a single field
>
>>
>>>
>>> Also it might be a good idea to consider actual name lengths, so that
>>> buffer size can be decreased, to save (CBT_NAME_LENGTH - strlen(name)
>>> - 1) memory for each name (e.g. separating names with spaces).
>>
>> Well user does not know the size of names beforehand anyway, so I
>> think this is excess complexity.
>
> First user calls with zero buffer, gets the size, allocates buffer of
> the size and does real call (same as in your example in other thread,
> but names are put in a more compact manner). But yes, I don't insist,
> probably for now simpler is better and if we face size problems in
> future we can compact it (though we might need to put version to the
> structure to have smooth switch to different format).
>
>>
>>>
>>> Another idea is to make this ioctl iterative, remembering position,
>>> so that we can read the list one by one using relatively small buffer.
>>
>> I think userspace memory if rather "free"
>> If we make it iterative, we will have to solve a few problems like
>> - Holding cbt pointer so we can get next cbt (what if it someone
>> deletes previous)
>> - When to reset
>> - How to distinguish users
>
> yes, going in this direction is hard
>
> Thus I think restricting maximum number of cbts is the best option.
So, do you want to have some inner hard cap, or should it be a new IOCTL
to set/get bitmap maximum count?
>
>>
>>>
>>>> + __u8 names[0]; /* Array of CBT_NAME_LENGTH long names */
>>>> +};
>>>> +
>>>> #define BLKCBTSTART _IOR(0x12,200, struct blk_user_cbt_info)
>>>> #define BLKCBTSTOP _IOR(0x12, 201, struct blk_user_cbt_info)
>>>> #define BLKCBTGET _IOWR(0x12,202,struct blk_user_cbt_info)
>>>> #define BLKCBTSET _IOR(0x12,203,struct blk_user_cbt_info)
>>>> #define BLKCBTCLR _IOR(0x12,204,struct blk_user_cbt_info)
>>>> #define BLKCBTMISC _IOWR(0x12,205,struct blk_user_cbt_misc_info)
>>>> +#define BLKCBTLIST _IOWR(0x12, 206, struct blk_user_cbt_list)
>>>> /*
>>>> * Flags for the fsx_xflags field
>>>
>
More information about the Devel
mailing list