On Wed, Sep 27, 2023 at 10:40:21AM +0300, Dan Carpenter wrote:
> The patch a3c38500d469: "buffer: hoist GFP flags from grow_dev_page()
> to __getblk_gfp()" from Sep 14, 2023 (linux-next), leads to the
> following Smatch static checker warning:
>
> fs/buffer.c:1065 grow_dev_page()
> warn: NEW missing error code 'ret'
Smatch is right.
> fs/buffer.c
> 1037 static int
> 1038 grow_dev_page(struct block_device *bdev, sector_t block,
> 1039 pgoff_t index, int size, int sizebits, gfp_t gfp)
> 1040 {
> 1041 struct inode *inode = bdev->bd_inode;
> 1042 struct folio *folio;
> 1043 struct buffer_head *bh;
> 1044 sector_t end_block;
> 1045 int ret = 0;
> 1046
> 1047 folio = __filemap_get_folio(inode->i_mapping, index,
> 1048 FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
> 1049 if (IS_ERR(folio))
> 1050 return PTR_ERR(folio);
> 1051
> 1052 bh = folio_buffers(folio);
> 1053 if (bh) {
> 1054 if (bh->b_size == size) {
> 1055 end_block = folio_init_buffers(folio, bdev,
> 1056 (sector_t)index << sizebits, size);
> 1057 goto done;
> 1058 }
> 1059 if (!try_to_free_buffers(folio))
> 1060 goto failed;
> 1061 }
> 1062
> 1063 bh = folio_alloc_buffers(folio, size, gfp | __GFP_ACCOUNT);
> 1064 if (!bh)
> --> 1065 goto failed;
>
> Should this be an error code? It's kind of complicated because I think
> the other goto failed path deliberately returns zero?
Yes, it's very confusing. Which is a common refrain in this part of the
VFS. Thanks for the report, I'll sort it out.
