Re: [PATCH] isofs: limit empty zisofs blocks to their uncompressed size
From: Jan Kara
Date: Fri Nov 13 2026 - 13:59:57 EST
On Thu 12-11-26 17:35:40, Mikhail Dmitrichenko wrote:
> When a zisofs block is smaller than a page, the empty-block path in
> zisofs_uncompress_block() zeroes the rest of the page and returns that
> length. zisofs_fill_pages() then advances past the page without reading
> the remaining blocks, returning zeros instead of their contents.
>
> For example, with 64 KiB pages and 32 KiB zisofs blocks, an empty first
> block should zero only the first half of the page. Instead, it zeroes
> the entire page and returns 64 KiB, so the second block is never read.
>
> Commit 68d4d3e78150 ("isofs: fix out-of-bounds page array access on empty
> zisofs block") accounted for poffset but left this older problem in
> place. It was pointed out by Sashiko while reviewing a 5.10 backport.
>
> Limit the zeroed range and returned length to one uncompressed block
> and the available output space. Only mark a folio uptodate when the
> range reaches its end. zisofs_fill_pages() already zeroes the tail and
> marks the partially filled folio uptodate at EOF.
>
> Fixes: 59bc055211b8 ("zisofs: Implement reading of compressed files when PAGE_CACHE_SIZE > compress block size")
> Cc: stable@xxxxxxxxxxxxxxx
> Link: https://sashiko.dev/#/patchset/20261105083209.922-1-mdmitrichenko%40astralinux.ru
> Signed-off-by: Mikhail Dmitrichenko <mdmitrichenko@xxxxxxxxxxxxx>
Thanks. I've added this to my tree.
Honza
> ---
> fs/isofs/compress.c | 16 ++++++++++++----
> 1 file changed, 12 insertions(+), 4 deletions(-)
>
> diff --git a/fs/isofs/compress.c b/fs/isofs/compress.c
> index 2d23abaeb874..c471007a76ed 100644
> --- a/fs/isofs/compress.c
> +++ b/fs/isofs/compress.c
> @@ -64,16 +64,24 @@ static loff_t zisofs_uncompress_block(struct inode *inode, loff_t block_start,
> }
> /* Empty block? */
> if (block_size == 0) {
> - for ( i = 0 ; i < pcount ; i++ ) {
> + unsigned int bytes = min_t(loff_t, 1U << zisofs_block_shift,
> + (((loff_t)pcount) << PAGE_SHIFT) - poffset);
> + unsigned int remaining = bytes;
> +
> + for (i = 0; i < pcount; i++) {
> unsigned int off = i ? 0 : poffset;
> + unsigned int len = min_t(unsigned int, remaining,
> + PAGE_SIZE - off);
> struct folio *folio = folios[i];
>
> + remaining -= len;
> if (!folio)
> continue;
> - folio_zero_range(folio, off, folio_size(folio) - off);
> - folio_mark_uptodate(folio);
> + folio_zero_range(folio, off, len);
> + if (off + len == folio_size(folio))
> + folio_mark_uptodate(folio);
> }
> - return (((loff_t)pcount) << PAGE_SHIFT) - poffset;
> + return bytes;
> }
>
> /* Because zlib is not thread-safe, do all the I/O at the top. */
> --
> 2.43.0
>
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR