Hi Eric,
On 2020/6/19 2:13, Eric Biggers wrote:
> Hi Chao,
>
> On Thu, Jun 18, 2020 at 06:06:02PM +0800, Chao Yu wrote:
>>> @@ -936,8 +972,11 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio)
>>>
>>> inc_page_count(sbi, WB_DATA_TYPE(bio_page));
>>>
>>> - if (io->bio && !io_is_mergeable(sbi, io->bio, io, fio,
>>> - io->last_block_in_bio, fio->new_blkaddr))
>>> + if (io->bio &&
>>> + (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
>>> + fio->new_blkaddr) ||
>>> + !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
>>> + fio->page->index, fio)))
>>
>> bio_page->index, fio)))
>>
>>> __submit_merged_bio(io);
>>> alloc_new:
>>> if (io->bio == NULL) {
>>> @@ -949,6 +988,8 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio)
>>> goto skip;
>>> }
>>> io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
>>> + f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
>>> + fio->page->index, fio, GFP_NOIO);
>>
>> bio_page->index, fio, GFP_NOIO);
>>
>
> We're using ->mapping->host and ->index. Ordinarily that would mean the page
> needs to be a pagecache page. But bio_page can also be a compressed page or a
> bounce page containing fs-layer encrypted contents.
I'm concerning about compression + inlinecrypt case.
>
> Is your suggestion to keep using fio->page->mapping->host (since encrypted pages
Yup,
> don't have a mapping), but start using bio_page->index (since f2fs apparently
I meant that we need to use bio_page->index as tweak value in write path to
keep consistent as we did in read path, otherwise we may read the wrong
decrypted data later to incorrect tweak value.
- f2fs_read_multi_pages (only comes from compression inode)
- f2fs_alloc_dic
- f2fs_set_compressed_page(page, cc->inode,
start_idx + i + 1, dic);
^^^^^^^^^^^^^^^^^
- dic->cpages[i] = page;
- for ()
struct page *page = dic->cpages[i];
if (!bio)
- f2fs_grab_read_bio(..., page->index,..)
- f2fs_set_bio_crypt_ctx(..., first_idx, ..) /* first_idx == cpage->index */
You can see that cpage->index was set to page->index + 1, that's why we need
to use one of cpage->index/page->index as tweak value all the time rather than
using both index mixed in read/write path.
But note that for fs-layer encryption, we have used cpage->index as tweak value,
so here I suggest we can keep consistent to use cpage->index in inlinecrypt case.
> *does* set ->index for compressed pages, and if the file uses fs-layer
> encryption then f2fs_set_bio_crypt_ctx() won't use the index anyway)?
>
> Does this mean the code is currently broken for compression + inline encryption
> because it's using the wrong ->index? I think the answer is no, since
I guess it's broken now for compression + inlinecrypt case.
> f2fs_write_compressed_pages() will still pass the first 'nr_cpages' pagecache
> pages along with the compressed pages. In that case, your suggestion would be a
> cleanup rather than a fix?
That's a fix.
>
> It would be helpful if there was an f2fs mount option to auto-enable compression
> on all files (similar to how test_dummy_encryption auto-enables encryption on
> all files) so that it could be tested more easily.
Agreed.
Previously I changed mkfs to allow to add compression flag to root inode for
compression test. :P
Thanks,
>
> - Eric
> .
>
