Blocks are tracked by u32, so the max permitted filesize is
U32_MAX * BLOCK_SIZE. Additionally, in order to support crypto data unit
sizes of 4K with a 16K block size with IV_INO_LBLK_{32,63}, we must
further restrict max filesize to U32_MAX * 4096. This does not affect 4K
blocksize f2fs as the natural limit for files are well below that.
Fixes: d7e9a9037de2 ("f2fs: Support Block Size == Page Size")
Signed-off-by: Daniel Rosenberg <[email protected]>
---
fs/f2fs/super.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 033af907c3b1..18a2189a0dc4 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -3364,6 +3364,18 @@ loff_t max_file_blocks(struct inode *inode)
leaf_count *= NIDS_PER_BLOCK;
result += leaf_count;
+ /*
+ * For compatibility with FSCRYPT_POLICY_IV_INO_LBLK_{64,32} with a
+ * 4K crypto data unit, we must restrict the max filesize to what can
+ * fit within U32_MAX data units.
+ *
+ * Since the blocksize must currently be equal to the page size,
+ * we can use a constant for that. Note if this is not the case
+ * in the future that inode is NULL while setting up the superblock.
+ */
+
+ result = min(result, ((loff_t) U32_MAX * 4096) >> F2FS_BLKSIZE_BITS);
+
return result;
}
base-commit: d346fa09abff46988de9267b67b6900d9913d5a2
--
2.43.0.rc2.451.g8631bc7472-goog
On Mon, Dec 04, 2023 at 03:46:15PM -0800, Daniel Rosenberg via Linux-f2fs-devel wrote:
> Blocks are tracked by u32, so the max permitted filesize is
> U32_MAX * BLOCK_SIZE. Additionally, in order to support crypto data unit
> sizes of 4K with a 16K block size with IV_INO_LBLK_{32,63}, we must
{32,63} should be {32,64}
> + /*
> + * For compatibility with FSCRYPT_POLICY_IV_INO_LBLK_{64,32} with a
> + * 4K crypto data unit, we must restrict the max filesize to what can
> + * fit within U32_MAX data units.
FSCRYPT_POLICY_IV_INO_LBLK_{64,32} should be
FSCRYPT_POLICY_FLAG_IV_INO_LBLK_{64,32}
> + *
> + * Since the blocksize must currently be equal to the page size,
> + * we can use a constant for that. Note if this is not the case
> + * in the future that inode is NULL while setting up the superblock.
I'm not sure what the last sentence is trying to say.
> + */
> +
> + result = min(result, ((loff_t) U32_MAX * 4096) >> F2FS_BLKSIZE_BITS);
Is it intentional that this is off by 1? If indices can be up to U32_MAX, then
the maximum size is U32_MAX + 1. It's not a bad idea to go with the lower size,
so that max_index + 1 does not overflow. But that's not what the explanation
says, so this seems to be accidental.
- Eric