Re: [PATCH 2/2] fs/quota: Fix spectre gadget in do_quotactl

From: Jan Kara
Date: Wed Aug 22 2018 - 12:08:38 EST


On Tue 31-07-18 01:37:31, Jeremy Cline wrote:
> 'type' is user-controlled, so sanitize it after the bounds check to
> avoid using it in speculative execution. This covers the following
> potential gadgets detected with the help of smatch:
>
> * fs/ext4/super.c:5741 ext4_quota_read() warn: potential spectre issue
> 'sb_dqopt(sb)->files' [r]
> * fs/ext4/super.c:5778 ext4_quota_write() warn: potential spectre issue
> 'sb_dqopt(sb)->files' [r]
> * fs/f2fs/super.c:1552 f2fs_quota_read() warn: potential spectre issue
> 'sb_dqopt(sb)->files' [r]
> * fs/f2fs/super.c:1608 f2fs_quota_write() warn: potential spectre issue
> 'sb_dqopt(sb)->files' [r]
> * fs/quota/dquot.c:412 mark_info_dirty() warn: potential spectre issue
> 'sb_dqopt(sb)->info' [w]
> * fs/quota/dquot.c:933 dqinit_needed() warn: potential spectre issue
> 'dquots' [r]
> * fs/quota/dquot.c:2112 dquot_commit_info() warn: potential spectre
> issue 'dqopt->ops' [r]
> * fs/quota/dquot.c:2362 vfs_load_quota_inode() warn: potential spectre
> issue 'dqopt->files' [w] (local cap)
> * fs/quota/dquot.c:2369 vfs_load_quota_inode() warn: potential spectre
> issue 'dqopt->ops' [w] (local cap)
> * fs/quota/dquot.c:2370 vfs_load_quota_inode() warn: potential spectre
> issue 'dqopt->info' [w] (local cap)
> * fs/quota/quota.c:110 quota_getfmt() warn: potential spectre issue
> 'sb_dqopt(sb)->info' [r]
> * fs/quota/quota_v2.c:84 v2_check_quota_file() warn: potential spectre
> issue 'quota_magics' [w]
> * fs/quota/quota_v2.c:85 v2_check_quota_file() warn: potential spectre
> issue 'quota_versions' [w]
> * fs/quota/quota_v2.c:96 v2_read_file_info() warn: potential spectre
> issue 'dqopt->info' [r]
> * fs/quota/quota_v2.c:172 v2_write_file_info() warn: potential spectre
> issue 'dqopt->info' [r]
>
> Additionally, a quick inspection indicates there are array accesses with
> 'type' in quota_on() and quota_off() functions which are also addressed
> by this.
>
> Cc: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Jeremy Cline <jcline@xxxxxxxxxx>

OK, this looks good. I'll take both patches through my tree (I'll fixup the
first patch myself).

Honza

> ---
>
> This patch isn't going to cleanly apply to stable without the "fs/quota:
> Replace XQM_MAXQUOTAS usage with MAXQUOTAS" patch, but I'm not sure that
> patch is really stable material and XQM_MAXQUOTAS has been 3 since
> pre-v4.4 so the end result will be the same even if that patch isn't
> backported.
>
> fs/quota/quota.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/fs/quota/quota.c b/fs/quota/quota.c
> index d403392d8a0f..f0cbf58ad4da 100644
> --- a/fs/quota/quota.c
> +++ b/fs/quota/quota.c
> @@ -18,6 +18,7 @@
> #include <linux/quotaops.h>
> #include <linux/types.h>
> #include <linux/writeback.h>
> +#include <linux/nospec.h>
>
> static int check_quotactl_permission(struct super_block *sb, int type, int cmd,
> qid_t id)
> @@ -701,6 +702,7 @@ static int do_quotactl(struct super_block *sb, int type, int cmd, qid_t id,
>
> if (type >= MAXQUOTAS)
> return -EINVAL;
> + type = array_index_nospec(type, MAXQUOTAS);
> /*
> * Quota not supported on this fs? Check this before s_quota_types
> * since they needn't be set if quota is not supported at all.
> --
> 2.17.1
>
>
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR