Re: [f2fs-dev] [PATCH] f2fs: fix race of pending_pages in decompression

[Daeho Jeong]

STEP_VERITY is enabled by f2fs_need_verity() and the function is like below.
We already know the second condition (idx < DIV_ROUND_UP...) is
satisfied when invoking f2fs_alloc_dic().

static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
{
        return fsverity_active(inode) &&
               idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
}

2020년 12월 4일 (금) 오후 12:28, Eric Biggers <ebiggers@xxxxxxxxxx>님이 작성:
>
> On Fri, Dec 04, 2020 at 09:58:47AM +0900, Daeho Jeong wrote:
> > diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> > index 87090da8693d..cdf72e153da0 100644
> > --- a/fs/f2fs/compress.c
> > +++ b/fs/f2fs/compress.c
> > @@ -803,8 +803,6 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
> >       if (cops->destroy_decompress_ctx)
> >               cops->destroy_decompress_ctx(dic);
> >  out_free_dic:
> > -     if (verity)
> > -             atomic_set(&dic->pending_pages, dic->nr_cpages);
> >       if (!verity)
> >               f2fs_decompress_end_io(dic->rpages, dic->cluster_size,
> >                                                               ret, false);
> > @@ -1498,6 +1496,8 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
> >       dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
> >       dic->inode = cc->inode;
> >       atomic_set(&dic->pending_pages, cc->nr_cpages);
> > +     if (fsverity_active(cc->inode))
> > +             atomic_set(&dic->verity_pages, cc->nr_cpages);
> >       dic->cluster_idx = cc->cluster_idx;
> >       dic->cluster_size = cc->cluster_size;
> >       dic->log_cluster_size = cc->log_cluster_size;
>
> The check for fsverity_active() is wrong.  It looks like you need to know
> whether the bio needs to go through the fs-verity data verification.  The
> correct way to determine that is to check whether STEP_VERITY is enabled in the
> bio's bio_post_read_ctx.  It's set by f2fs_grab_read_bio() when needed.
>
> - Eric