Re: [PATCH 2/2] f2fs: support data compression

[Jaegeuk Kim]

On 10/30, Eric Biggers wrote:
> On Wed, Oct 30, 2019 at 04:43:52PM +0800, Chao Yu wrote:
> > On 2019/10/30 10:55, Eric Biggers wrote:
> > > On Tue, Oct 29, 2019 at 04:33:36PM +0800, Chao Yu wrote:
> > >> On 2019/10/28 6:50, Eric Biggers wrote:
> > >>>> +bool f2fs_is_compressed_page(struct page *page)
> > >>>> +{
> > >>>> +	if (!page_private(page))
> > >>>> +		return false;
> > >>>> +	if (IS_ATOMIC_WRITTEN_PAGE(page) || IS_DUMMY_WRITTEN_PAGE(page))
> > >>>> +		return false;
> > >>>> +	return *((u32 *)page_private(page)) == F2FS_COMPRESSED_PAGE_MAGIC;
> > >>>> +}
> > >>>
> > >>> This code implies that there can be multiple page private structures each of
> > >>> which has a different magic number.  But I only see F2FS_COMPRESSED_PAGE_MAGIC.
> > >>> Where in the code is the other one(s)?
> > >>
> > >> I'm not sure I understood you correctly, did you mean it needs to introduce
> > >> f2fs_is_atomic_written_page() and f2fs_is_dummy_written_page() like
> > >> f2fs_is_compressed_page()?
> > >>
> > > 
> > > No, I'm asking what is the case where the line
> > > 
> > > 	*((u32 *)page_private(page)) == F2FS_COMPRESSED_PAGE_MAGIC
> > > 
> > > returns false?
> > 
> > Should be this?
> > 
> > if (!page_private(page))
> > 	return false;
> > f2fs_bug_on(*((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC)
> > return true;
> 
> Yes, that makes more sense, unless there are other cases.
> 
> > 
> > > 
> > >>>
> > >>>> +
> > >>>> +static void f2fs_set_compressed_page(struct page *page,
> > >>>> +		struct inode *inode, pgoff_t index, void *data, refcount_t *r)
> > >>>> +{
> > >>>> +	SetPagePrivate(page);
> > >>>> +	set_page_private(page, (unsigned long)data);
> > >>>> +
> > >>>> +	/* i_crypto_info and iv index */
> > >>>> +	page->index = index;
> > >>>> +	page->mapping = inode->i_mapping;
> > >>>> +	if (r)
> > >>>> +		refcount_inc(r);
> > >>>> +}
> > >>>
> > >>> It isn't really appropriate to create fake pagecache pages like this.  Did you
> > >>> consider changing f2fs to use fscrypt_decrypt_block_inplace() instead?
> > >>
> > >> We need to store i_crypto_info and iv index somewhere, in order to pass them to
> > >> fscrypt_decrypt_block_inplace(), where did you suggest to store them?
> > >>
> > > 
> > > The same place where the pages are stored.
> > 
> > Still we need allocate space for those fields, any strong reason to do so?
> > 
> 
> page->mapping set implies that the page is a pagecache page.  Faking it could
> cause problems with code elsewhere.

I've checked it with minchan, and it seems to be fine that filesystem uses
this page internally only, not in pagecache.

> 
> > > 
> > >>>> +
> > >>>> +void f2fs_destroy_compress_ctx(struct compress_ctx *cc)
> > >>>> +{
> > >>>> +	kvfree(cc->rpages);
> > >>>> +}
> > >>>
> > >>> The memory is allocated with kzalloc(), so why is it freed with kvfree() and not
> > >>> just kfree()?
> > >>
> > >> It was allocated by f2fs_*alloc() which will fallback to kvmalloc() once
> > >> kmalloc() failed.
> > > 
> > > This seems to be a bug in f2fs_kmalloc() -- it inappropriately falls back to
> > > kvmalloc().  As per its name, it should only use kmalloc().  f2fs_kvmalloc()
> > > already exists, so it can be used when the fallback is wanted.
> > 
> > We can introduce f2fs_memalloc() to wrap f2fs_kmalloc() and f2fs_kvmalloc() as
> > below:
> > 
> > f2fs_memalloc()
> > {
> > 	mem = f2fs_kmalloc();
> > 	if (mem)
> > 		return mem;
> > 	return f2fs_kvmalloc();
> > }
> > 
> > It can be used in specified place where we really need it, like the place
> > descirbied in 5222595d093e ("f2fs: use kvmalloc, if kmalloc is failed") in where
> > we introduced original logic.
> 
> No, just use kvmalloc().  The whole point of kvmalloc() is that it tries
> kmalloc() and then falls back to vmalloc() if it fails.
> 
> - Eric