[PATCH 6/9] erofs-utils: add inode hash helper

[Gao Xiang]

On 2023/7/27 14:55, Jingbo Xu wrote:
> 
> 
> On 7/27/23 2:12 PM, Gao Xiang wrote:
>>
>>
>> On 2023/7/27 12:57, Jingbo Xu wrote:
>>> Add erofs_insert_ihash() helper inserting inode into inode hash table,
>>> and erofs_cleanup_ihash() helper cleaning up inode hash table.
>>>
>>> Also add prototypes of erofs_iget() and erofs_iget_by_nid() in the
>>> header file.
>>>
>>> Signed-off-by: Jingbo Xu <jefflexu at linux.alibaba.com>
>>> ---
>>>    include/erofs/inode.h |  4 ++++
>>>    lib/inode.c           | 22 +++++++++++++++++++---
>>>    2 files changed, 23 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/include/erofs/inode.h b/include/erofs/inode.h
>>> index e8a5670..aba2a94 100644
>>> --- a/include/erofs/inode.h
>>> +++ b/include/erofs/inode.h
>>> @@ -25,6 +25,10 @@ u32 erofs_new_encode_dev(dev_t dev);
>>>    unsigned char erofs_mode_to_ftype(umode_t mode);
>>>    unsigned char erofs_ftype_to_dtype(unsigned int filetype);
>>>    void erofs_inode_manager_init(void);
>>> +void erofs_insert_ihash(struct erofs_inode *inode, dev_t dev, ino_t
>>> ino);
>>> +void erofs_cleanup_ihash(void);
>>> +struct erofs_inode *erofs_iget(dev_t dev, ino_t ino);
>>> +struct erofs_inode *erofs_iget_by_nid(erofs_nid_t nid);
>>>    unsigned int erofs_iput(struct erofs_inode *inode);
>>>    erofs_nid_t erofs_lookupnid(struct erofs_inode *inode);
>>>    struct erofs_dentry *erofs_d_alloc(struct erofs_inode *parent,
>>> diff --git a/lib/inode.c b/lib/inode.c
>>> index d54f84f..d82ea95 100644
>>> --- a/lib/inode.c
>>> +++ b/lib/inode.c
>>> @@ -75,6 +75,24 @@ void erofs_inode_manager_init(void)
>>>            init_list_head(&inode_hashtable[i]);
>>>    }
>>>    +void erofs_insert_ihash(struct erofs_inode *inode, dev_t dev, ino_t
>>> ino)
>>> +{
>>> +    list_add(&inode->i_hash,
>>> +         &inode_hashtable[(ino ^ dev) % NR_INODE_HASHTABLE]);
>>> +}
>>> +
>>> +void erofs_cleanup_ihash(void)
>>> +{
>>> +    unsigned int i;
>>> +    struct erofs_inode *inode, *n;
>>> +
>>> +    for (i = 0; i < NR_INODE_HASHTABLE; ++i) {
>>> +        list_for_each_entry_safe(inode, n, &inode_hashtable[i], i_hash)
>>
>> Why are there still inodes here?
> 
> The inode hash table is used to detect hardlink file.  In union mode,
> only ino (dev is always 0) comes to play when calculating the hash.
> Thus the inode hash table is built on a per-image basis.  When we have
> finished parsing one input image, erofs_cleanup_ihash() is called to
> clean up the whole inode hash table, otherwise inode (with the same ino)
> from next input image may conflict when hashing.

I guess we'd better to inline the code directly and add comments for
this usage (also you'd better to ensure that hashtable is empty in
the beginning) since it's not a safe way for other use cases.

Thanks,
Gao Xiang