[PATCH] erofs: support for mounting a single block device with multiple devices

Jia Zhu zhujia.zj at bytedance.com
Wed Mar 1 18:44:37 AEDT 2023



在 2023/3/1 15:08, Gao Xiang 写道:
> Hi Jia,
> 
> On 2023/3/1 15:04, Jia Zhu wrote:
>> In order to support mounting multi-layer container image as a block
>> device, add single block device with multiple devices feature for EROFS.
> 
> In order to support mounting multi-blob container image as a single
> flattened block device, add flattened block device feature for EROFS.
> 
Thanks, I would revise it.
>>
>> In this mode, all meta/data contents will be mapped into one block 
>> address.
>> User could directly mount the block device by EROFS.
>>
>> Signed-off-by: Jia Zhu <zhujia.zj at bytedance.com>
>> Reviewed-by: Xin Yin <yinxin.x at bytedance.com>
>> ---
>>   fs/erofs/data.c  | 8 ++++++--
>>   fs/erofs/super.c | 5 +++++
>>   2 files changed, 11 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/erofs/data.c b/fs/erofs/data.c
>> index e16545849ea7..870b1f7fe1d4 100644
>> --- a/fs/erofs/data.c
>> +++ b/fs/erofs/data.c
>> @@ -195,9 +195,9 @@ int erofs_map_dev(struct super_block *sb, struct 
>> erofs_map_dev *map)
>>   {
>>       struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
>>       struct erofs_device_info *dif;
>> +    bool flatdev = !!sb->s_bdev;
> 
> I'd like to land it in sbi and set it in advance?
> 
I'll revise that in next version.
> Also, did you test this patch?

I've tested the patch using the following steps mentioned by
https://github.com/dragonflyoss/image-service/pull/1111

1. Compose a (nbd)block device from an EROFS image.
2. mount -t erofs /dev/nbdx /mnt/
3. compare the md5sum between source dir and /mnt dir.

> 
> Thanks,
> Gao Xiang
> 
> 
>>       int id;
>> -    /* primary device by default */
>>       map->m_bdev = sb->s_bdev;
>>       map->m_daxdev = EROFS_SB(sb)->dax_dev;
>>       map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
>> @@ -210,12 +210,16 @@ int erofs_map_dev(struct super_block *sb, struct 
>> erofs_map_dev *map)
>>               up_read(&devs->rwsem);
>>               return -ENODEV;
>>           }
>> +        if (flatdev) {
>> +            map->m_pa += blknr_to_addr(dif->mapped_blkaddr);
>> +            map->m_deviceid = 0;
>> +        }
>>           map->m_bdev = dif->bdev;
>>           map->m_daxdev = dif->dax_dev;
>>           map->m_dax_part_off = dif->dax_part_off;
>>           map->m_fscache = dif->fscache;
>>           up_read(&devs->rwsem);
>> -    } else if (devs->extra_devices) {
>> +    } else if (devs->extra_devices && !flatdev) {
>>           down_read(&devs->rwsem);
>>           idr_for_each_entry(&devs->tree, dif, id) {
>>               erofs_off_t startoff, length;
>> diff --git a/fs/erofs/super.c b/fs/erofs/super.c
>> index 19b1ae79cec4..4f9725b0950c 100644
>> --- a/fs/erofs/super.c
>> +++ b/fs/erofs/super.c
>> @@ -226,6 +226,7 @@ static int erofs_init_device(struct erofs_buf 
>> *buf, struct super_block *sb,
>>       struct erofs_fscache *fscache;
>>       struct erofs_deviceslot *dis;
>>       struct block_device *bdev;
>> +    bool flatdev = !!sb->s_bdev;
>>       void *ptr;
>>       ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*pos), EROFS_KMAP);
>> @@ -248,6 +249,10 @@ static int erofs_init_device(struct erofs_buf 
>> *buf, struct super_block *sb,
>>           if (IS_ERR(fscache))
>>               return PTR_ERR(fscache);
>>           dif->fscache = fscache;
>> +    } else if (flatdev) {
>> +        dif->bdev = sb->s_bdev;
>> +        dif->dax_dev = EROFS_SB(sb)->dax_dev;
>> +        dif->dax_part_off = sbi->dax_part_off;
>>       } else {
>>           bdev = blkdev_get_by_path(dif->path, FMODE_READ | FMODE_EXCL,
>>                         sb->s_type);


More information about the Linux-erofs mailing list