Re: [PATCH 08/19] pramfs: file operations for dirs

[Marco Stornelli <marco.stornelli@gmail.com>]

Il 07/09/2013 17:01, Al Viro ha scritto:
> On Sat, Sep 07, 2013 at 10:22:36AM +0200, Marco Stornelli wrote:
>> +int pram_add_link(struct dentry *dentry, struct inode *inode)
>> +{
>> +	struct inode *dir = dentry->d_parent->d_inode;
>> +	struct pram_inode *pidir, *pi, *pitail = NULL;
>> +	u64 tail_ino, prev_ino;
>> +
>> +	const char *name = dentry->d_name.name;
>> +
>> +	int namelen = min_t(unsigned int, dentry->d_name.len, PRAM_NAME_LEN);
>
> Whatever the hell for?  Your ->lookup() rejects dentries with names longer
> than PRAM_NAME_LEN with an error, so they won't reach this function at all.
>

Ok. I'll remove it.

>> +int pram_remove_link(struct inode *inode)
>
> Umm...  That's called on rename (for old one) *and* inode eviction when link
> count goes to zero.  What's the point of keeping unlinked ones (unlink/rmdir/
> rename victims) on those lists?  Sure, you skip them on lookups, but why
> delay link removal until eviction?  You pay for that with extra locking,
> BTW - if not for that, you wouldn't need your i_link_mutex at all.
>

Good question. The only answer I've got now is for "historical" reason, 
I can't see at the moment why we can remove the link information in case 
of opened-but-unlinked, instead of delay the operation until evict.

>> +	pi = pram_get_inode(sb, inode->i_ino);
>> +
>> +	switch ((u32)file->f_pos) {
>> +	case 0:
>> +		ret = dir_emit_dot(file, ctx);
>> +		ctx->pos = 1;
>> +		return ret;
>
> Really?  So on the first call of ->iterate() you just generate one
> entry and don't even try to produce more?  And it looks like the
> rest is no nicer...
>

I'll try to improve the behavior here.

Marco