On Thu, 16 May 2013 02:46:02 +0900 (JST), Ryusuke Konishi wrote:
> On Wed, 15 May 2013 17:32:10 +0400, Vyacheslav Dubeyko wrote:
>> Hi Ryusuke,
>>
>> This is third version of the patch.
>>
>> v2->v3
>> * Trivial BUG_ON checks were removed.
>> * Whole calculation algorithm was moved into
>> nilfs_palloc_count_max_entries() method.
>> * Improve error processing in the code.
>> * The nilfs_palloc_mdt_file_can_grow() method was simplified.
>> * The nilfs_ifile_count_free_inodes() method was simplified.
>>
>> v1->v2
>> * Change __statfs_word on u64 type.
>> * Rename nilfs_count_free_inodes() into nilfs_ifile_count_free_inodes()
>> method.
>> * Introduce auxiliary functions: nilfs_palloc_count_max_entries(),
>> nilfs_palloc_count_desc_blocks(), nilfs_palloc_mdt_file_can_grow().
>> * Rework processing of returned error from nilfs_ifile_count_free_inodes()
>> in nilfs_statfs().
>>
>> With the best regards,
>> Vyacheslav Dubeyko.
>> ---
>> From: Vyacheslav Dubeyko <slava@xxxxxxxxxxx>
>> Subject: [PATCH v3] nilfs2: implement calculation of free inodes count
>>
>> Currently, NILFS2 returns 0 as free inodes count (f_ffree) and
>> current used inodes count as total file nodes in file system
>> (f_files):
>>
>> df -i
>> Filesystem Inodes IUsed IFree IUse% Mounted on
>> /dev/loop0 2 2 0 100% /mnt/nilfs2
>>
>> This patch implements real calculation of free inodes count.
>> First of all, it is calculated total file nodes in file system
>> as (desc_blocks_count * groups_per_desc_block * entries_per_group).
>> Then, it is calculated free inodes count as difference the total
>> file nodes and used inodes count. As a result, we have such output
>> for NILFS2:
>>
>> df -i
>> Filesystem Inodes IUsed IFree IUse% Mounted on
>> /dev/loop0 4194304 2114701 2079603 51% /mnt/nilfs2
>>
>> Reported-by: Clemens Eisserer <linuxhippy@xxxxxxxxx>
>> Signed-off-by: Vyacheslav Dubeyko <slava@xxxxxxxxxxx>
>> CC: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxxxxxx>
>> Tested-by: Vyacheslav Dubeyko <slava@xxxxxxxxxxx>
>
> This patch still has a potential overflow issue that I pointed out in
> the previous comment; the patch handles the number of descriptor
> blocks in 32-bit wide variables without checking its upper limit.
>
> If you won't to add checks for the number of descriptor blocks, I
> propose you to change the definition of nilfs_palloc_groups_count() instead:
>
> static inline unsigned long
> nilfs_palloc_groups_count(const struct inode *inode)
> {
> - return 1UL << (BITS_PER_LONG - (inode->i_blkbits + 3 /* log2(8) */));
> + return nilfs_palloc_groups_per_desc_block(inode) << 32;
> }
>
> This implies the maximum number of descriptor block is 1 ^ 32.
>
> Because the above diff changes the maximum number of groups, I think
> it should be inserted as a separate patch.
Oops, sorry, this change can overflow in 32-bit architectures
because the type of return value is still unsigned long.
That calculation should be revised more carefully.
Ryusuke Konishi
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
