If mlockall() is called with only MCL_ONFAULT as flag,
it removes any previously applied lockings and does
nothing else.
This behavior is counter-intuitive and doesn't match the
Linux man page.
Consequently, return the error EINVAL, if only MCL_ONFAULT
is passed. That way, applications will at least detect that
they are calling mlockall() incorrectly.
Fixes: b0f205c2a308 ("mm: mlock: add mlock flags to enable VM_LOCKONFAULT usage")
Signed-off-by: Stefan Potyra <[email protected]>
---
Sparse shows a warning for mlock.c, but it is not related to
this patch.
mm/mlock.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mm/mlock.c b/mm/mlock.c
index e492a155c51a..03f39cbdd4c4 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -797,7 +797,8 @@ SYSCALL_DEFINE1(mlockall, int, flags)
unsigned long lock_limit;
int ret;
- if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
+ if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
+ flags == MCL_ONFAULT)
return -EINVAL;
if (!can_do_mlock())
--
2.20.1
[ Adding linux-api and some of the people who were involved in the
MCL_ONFAULT/mlock2/etc discussions. Author of the Fixes patch appears to
have moved on. ]
On Wed, May 22, 2019 at 11:23:37AM +0000, Potyra, Stefan wrote:
> If mlockall() is called with only MCL_ONFAULT as flag,
> it removes any previously applied lockings and does
> nothing else.
The change looks reasonable. Hard to imagine any application relies on it, and
they really shouldn't be if they are. Debian codesearch turned up only a few
cases where stress-ng was doing this for unknown reasons[1] and this change
isn't gonna break those. In this case I think changing the syscall's behavior
is justified.
> This behavior is counter-intuitive and doesn't match the
> Linux man page.
I'd quote it for the changelog:
For mlockall():
EINVAL Unknown flags were specified or MCL_ONFAULT was specified with‐
out either MCL_FUTURE or MCL_CURRENT.
With that you can add
Reviewed-by: Daniel Jordan <[email protected]>
[1] https://sources.debian.org/src/stress-ng/0.09.50-1/stress-mlock.c/?hl=203#L203
On Fri 24-05-19 17:43:04, Daniel Jordan wrote:
> [ Adding linux-api and some of the people who were involved in the
> MCL_ONFAULT/mlock2/etc discussions. Author of the Fixes patch appears to
> have moved on. ]
>
> On Wed, May 22, 2019 at 11:23:37AM +0000, Potyra, Stefan wrote:
> > If mlockall() is called with only MCL_ONFAULT as flag,
> > it removes any previously applied lockings and does
> > nothing else.
>
> The change looks reasonable. Hard to imagine any application relies on it, and
> they really shouldn't be if they are. Debian codesearch turned up only a few
> cases where stress-ng was doing this for unknown reasons[1] and this change
> isn't gonna break those. In this case I think changing the syscall's behavior
> is justified.
>
> > This behavior is counter-intuitive and doesn't match the
> > Linux man page.
>
> I'd quote it for the changelog:
>
> For mlockall():
>
> EINVAL Unknown flags were specified or MCL_ONFAULT was specified with‐
> out either MCL_FUTURE or MCL_CURRENT.
>
> With that you can add
>
> Reviewed-by: Daniel Jordan <[email protected]>
>
> [1] https://sources.debian.org/src/stress-ng/0.09.50-1/stress-mlock.c/?hl=203#L203
Well spotted and the fix looks reasonable as well. Quoting the man page
seems useful as well.
Acked-by: Michal Hocko <[email protected]>
Thanks!
--
Michal Hocko
SUSE Labs
If mlockall() is called with only MCL_ONFAULT as flag,
it removes any previously applied lockings and does
nothing else.
This behavior is counter-intuitive and doesn't match the
Linux man page.
For mlockall():
EINVAL Unknown flags were specified or MCL_ONFAULT was specified with‐
out either MCL_FUTURE or MCL_CURRENT.
Consequently, return the error EINVAL, if only MCL_ONFAULT
is passed. That way, applications will at least detect that
they are calling mlockall() incorrectly.
Fixes: b0f205c2a308 ("mm: mlock: add mlock flags to enable VM_LOCKONFAULT usage")
Signed-off-by: Stefan Potyra <[email protected]>
Reviewed-by: Daniel Jordan <[email protected]>
Acked-by: Michal Hocko <[email protected]>
---
mm/mlock.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mm/mlock.c b/mm/mlock.c
index e492a155c51a..03f39cbdd4c4 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -797,7 +797,8 @@ SYSCALL_DEFINE1(mlockall, int, flags)
unsigned long lock_limit;
int ret;
- if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
+ if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
+ flags == MCL_ONFAULT)
return -EINVAL;
if (!can_do_mlock())
--
2.20.1
On 5/27/19 9:53 AM, Potyra, Stefan wrote:
> If mlockall() is called with only MCL_ONFAULT as flag,
> it removes any previously applied lockings and does
> nothing else.
>
> This behavior is counter-intuitive and doesn't match the
> Linux man page.
>
> For mlockall():
>
> EINVAL Unknown flags were specified or MCL_ONFAULT was specified with‐
> out either MCL_FUTURE or MCL_CURRENT.
>
> Consequently, return the error EINVAL, if only MCL_ONFAULT
> is passed. That way, applications will at least detect that
> they are calling mlockall() incorrectly.
>
> Fixes: b0f205c2a308 ("mm: mlock: add mlock flags to enable VM_LOCKONFAULT usage")
> Signed-off-by: Stefan Potyra <[email protected]>
> Reviewed-by: Daniel Jordan <[email protected]>
> Acked-by: Michal Hocko <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Thanks, shame we didn't catch it during review. Hope nobody will report
a regression.
> ---
> mm/mlock.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/mlock.c b/mm/mlock.c
> index e492a155c51a..03f39cbdd4c4 100644
> --- a/mm/mlock.c
> +++ b/mm/mlock.c
> @@ -797,7 +797,8 @@ SYSCALL_DEFINE1(mlockall, int, flags)
> unsigned long lock_limit;
> int ret;
>
> - if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
> + if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
> + flags == MCL_ONFAULT)
> return -EINVAL;
>
> if (!can_do_mlock())
>