MADV_DONTNEED historically rejects mlocked ranges, but with
MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
there are valid use cases for depopulating locked ranges as well.
Users mlock memory to protect secrets. There are allocators for secure
buffers that want in-use memory generally mlocked, but cleared and
invalidated memory to give up the physical pages. This could be done
with explicit munlock -> mlock calls on free -> alloc of course, but
that adds two unnecessary syscalls, heavy mmap_sem write locks, vma
splits and re-merges - only to get rid of the backing pages.
Users also mlockall(MCL_ONFAULT) to suppress sustained paging, but are
okay with on-demand initial population. It seems valid to selectively
free some memory during the lifetime of such a process, without having
to mess with its overall policy.
Why add a separate flag? Isn't this a pretty niche usecase?
- MADV_DONTNEED has been bailing on locked vmas forever. It's at least
conceivable that someone, somewhere is relying on mlock to protect
data from perhaps broader invalidation calls. Changing this behavior
now could lead to quiet data corruption.
- It also clarifies expectations around MADV_FREE and maybe
MADV_REMOVE. It avoids the situation where one quietly behaves
different than the others. MADV_FREE_LOCKED can be added later.
- The combination of mlock() and madvise() in the first place is
probably niche. But where it happens, I'd say that dropping pages
from a locked region once they don't contain secrets or won't page
anymore is much saner than relying on mlock to protect memory from
speculative or errant invalidation calls. It's just that we can't
change the default behavior because of the two previous points.
Given that, an explicit new flag seems to make the most sense.
Signed-off-by: Johannes Weiner <[email protected]>
---
include/uapi/asm-generic/mman-common.h | 2 ++
mm/madvise.c | 16 +++++++++++++---
2 files changed, 15 insertions(+), 3 deletions(-)
diff --git a/include/uapi/asm-generic/mman-common.h b/include/uapi/asm-generic/mman-common.h
index 1567a3294c3d..6c1aa92a92e4 100644
--- a/include/uapi/asm-generic/mman-common.h
+++ b/include/uapi/asm-generic/mman-common.h
@@ -75,6 +75,8 @@
#define MADV_POPULATE_READ 22 /* populate (prefault) page tables readable */
#define MADV_POPULATE_WRITE 23 /* populate (prefault) page tables writable */
+#define MADV_DONTNEED_LOCKED 24 /* like DONTNEED, but drop locked pages too */
+
/* compatibility flags */
#define MAP_FILE 0
diff --git a/mm/madvise.c b/mm/madvise.c
index 5604064df464..12dfa14bc985 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -800,6 +800,13 @@ static long madvise_dontneed_single_vma(struct vm_area_struct *vma,
return 0;
}
+static bool can_madv_dontneed_free(struct vm_area_struct *vma, int behavior)
+{
+ if (behavior == MADV_DONTNEED_LOCKED)
+ return !(vma->vm_flags & (VM_HUGETLB|VM_PFNMAP));
+ return can_madv_lru_vma(vma);
+}
+
static long madvise_dontneed_free(struct vm_area_struct *vma,
struct vm_area_struct **prev,
unsigned long start, unsigned long end,
@@ -808,7 +815,8 @@ static long madvise_dontneed_free(struct vm_area_struct *vma,
struct mm_struct *mm = vma->vm_mm;
*prev = vma;
- if (!can_madv_lru_vma(vma))
+
+ if (!can_madv_dontneed_free(vma, behavior))
return -EINVAL;
if (!userfaultfd_remove(vma, start, end)) {
@@ -830,7 +838,7 @@ static long madvise_dontneed_free(struct vm_area_struct *vma,
*/
return -ENOMEM;
}
- if (!can_madv_lru_vma(vma))
+ if (!can_madv_dontneed_free(vma, behavior))
return -EINVAL;
if (end > vma->vm_end) {
/*
@@ -850,7 +858,7 @@ static long madvise_dontneed_free(struct vm_area_struct *vma,
VM_WARN_ON(start >= end);
}
- if (behavior == MADV_DONTNEED)
+ if (behavior == MADV_DONTNEED || behavior == MADV_DONTNEED_LOCKED)
return madvise_dontneed_single_vma(vma, start, end);
else if (behavior == MADV_FREE)
return madvise_free_single_vma(vma, start, end);
@@ -988,6 +996,7 @@ static int madvise_vma_behavior(struct vm_area_struct *vma,
return madvise_pageout(vma, prev, start, end);
case MADV_FREE:
case MADV_DONTNEED:
+ case MADV_DONTNEED_LOCKED:
return madvise_dontneed_free(vma, prev, start, end, behavior);
case MADV_POPULATE_READ:
case MADV_POPULATE_WRITE:
@@ -1113,6 +1122,7 @@ madvise_behavior_valid(int behavior)
case MADV_REMOVE:
case MADV_WILLNEED:
case MADV_DONTNEED:
+ case MADV_DONTNEED_LOCKED:
case MADV_FREE:
case MADV_COLD:
case MADV_PAGEOUT:
--
2.35.1
> On Mar 3, 2022, at 1:29 PM, Johannes Weiner <[email protected]> wrote:
>
> MADV_DONTNEED historically rejects mlocked ranges, but with
> MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
> there are valid use cases for depopulating locked ranges as well.
...
> @@ -850,7 +858,7 @@ static long madvise_dontneed_free(struct vm_area_struct *vma,
> VM_WARN_ON(start >= end);
> }
>
> - if (behavior == MADV_DONTNEED)
> + if (behavior == MADV_DONTNEED || behavior == MADV_DONTNEED_LOCKED)
> return madvise_dontneed_single_vma(vma, start, end);
> else if (behavior == MADV_FREE)
> return madvise_free_single_vma(vma, start, end);
> @@ -988,6 +996,7 @@ static int madvise_vma_behavior(struct vm_area_struct *vma,
> return madvise_pageout(vma, prev, start, end);
> case MADV_FREE:
> case MADV_DONTNEED:
> + case MADV_DONTNEED_LOCKED:
> return madvise_dontneed_free(vma, prev, start, end, behavior);
> case MADV_POPULATE_READ:
> case MADV_POPULATE_WRITE:
> @@ -1113,6 +1122,7 @@ madvise_behavior_valid(int behavior)
> case MADV_REMOVE:
> case MADV_WILLNEED:
> case MADV_DONTNEED:
> + case MADV_DONTNEED_LOCKED:
> case MADV_FREE:
> case MADV_COLD:
> case MADV_PAGEOUT:
Don’t you want to change madvise_need_mmap_write() as well and add
MADV_DONTNEED_LOCKED there too?
On Thu, Mar 03, 2022 at 01:35:56PM -0800, Nadav Amit wrote:
>
>
> > On Mar 3, 2022, at 1:29 PM, Johannes Weiner <[email protected]> wrote:
> >
> > MADV_DONTNEED historically rejects mlocked ranges, but with
> > MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
> > there are valid use cases for depopulating locked ranges as well.
>
> ...
>
> > @@ -850,7 +858,7 @@ static long madvise_dontneed_free(struct vm_area_struct *vma,
> > VM_WARN_ON(start >= end);
> > }
> >
> > - if (behavior == MADV_DONTNEED)
> > + if (behavior == MADV_DONTNEED || behavior == MADV_DONTNEED_LOCKED)
> > return madvise_dontneed_single_vma(vma, start, end);
> > else if (behavior == MADV_FREE)
> > return madvise_free_single_vma(vma, start, end);
> > @@ -988,6 +996,7 @@ static int madvise_vma_behavior(struct vm_area_struct *vma,
> > return madvise_pageout(vma, prev, start, end);
> > case MADV_FREE:
> > case MADV_DONTNEED:
> > + case MADV_DONTNEED_LOCKED:
> > return madvise_dontneed_free(vma, prev, start, end, behavior);
> > case MADV_POPULATE_READ:
> > case MADV_POPULATE_WRITE:
> > @@ -1113,6 +1122,7 @@ madvise_behavior_valid(int behavior)
> > case MADV_REMOVE:
> > case MADV_WILLNEED:
> > case MADV_DONTNEED:
> > + case MADV_DONTNEED_LOCKED:
> > case MADV_FREE:
> > case MADV_COLD:
> > case MADV_PAGEOUT:
>
> Don’t you want to change madvise_need_mmap_write() as well and add
> MADV_DONTNEED_LOCKED there too?
You're absolutely right, thanks Nadav! It'd be fine, but more
expensive than necessary. Here is the fixlet:
diff --git a/mm/madvise.c b/mm/madvise.c
index 12dfa14bc985..7dbfcd6c955a 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -52,6 +52,7 @@ static int madvise_need_mmap_write(int behavior)
case MADV_REMOVE:
case MADV_WILLNEED:
case MADV_DONTNEED:
+ case MADV_DONTNEED_LOCKED:
case MADV_COLD:
case MADV_PAGEOUT:
case MADV_FREE:
On 03.03.22 22:47, Johannes Weiner wrote:
> On Thu, Mar 03, 2022 at 04:29:56PM -0500, Johannes Weiner wrote:
>> MADV_DONTNEED historically rejects mlocked ranges, but with
>> MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
>> there are valid use cases for depopulating locked ranges as well.
Indeed, there are. I'd have use for that in QEMU for virtio-mem (which
uses MADV_POPULATE_WRITE+MADV_DONTNEED to dynamically allocate/discard
memory in a sparse memory mapping to be used by the VM, currently
doesn't support mlock for obvious reasons).
Further, QEMU postcopy live migration handling requires an munlockall()
before registering the uffd handler and discarding all RAM via
MADV_DONTNEED. Once postcopy finished, we have to mlock() again. I
didn't check if there would be more stopping uffd from working on a
mlocked region, but this would be one piece of the puzzle I guess.
--
Thanks,
David / dhildenb
[please CC linux-api if you are going to repost with the fix suggested
by Nadav]
On Thu 03-03-22 16:47:34, Johannes Weiner wrote:
> On Thu, Mar 03, 2022 at 04:29:56PM -0500, Johannes Weiner wrote:
> > MADV_DONTNEED historically rejects mlocked ranges, but with
> > MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
> > there are valid use cases for depopulating locked ranges as well.
> >
> > Users mlock memory to protect secrets. There are allocators for secure
> > buffers that want in-use memory generally mlocked, but cleared and
> > invalidated memory to give up the physical pages. This could be done
> > with explicit munlock -> mlock calls on free -> alloc of course, but
> > that adds two unnecessary syscalls, heavy mmap_sem write locks, vma
> > splits and re-merges - only to get rid of the backing pages.
> >
> > Users also mlockall(MCL_ONFAULT) to suppress sustained paging, but are
> > okay with on-demand initial population. It seems valid to selectively
> > free some memory during the lifetime of such a process, without having
> > to mess with its overall policy.
> >
> > Why add a separate flag? Isn't this a pretty niche usecase?
> >
> > - MADV_DONTNEED has been bailing on locked vmas forever. It's at least
> > conceivable that someone, somewhere is relying on mlock to protect
> > data from perhaps broader invalidation calls. Changing this behavior
> > now could lead to quiet data corruption.
> >
> > - It also clarifies expectations around MADV_FREE and maybe
> > MADV_REMOVE. It avoids the situation where one quietly behaves
> > different than the others. MADV_FREE_LOCKED can be added later.
> >
> > - The combination of mlock() and madvise() in the first place is
> > probably niche. But where it happens, I'd say that dropping pages
> > from a locked region once they don't contain secrets or won't page
> > anymore is much saner than relying on mlock to protect memory from
> > speculative or errant invalidation calls. It's just that we can't
> > change the default behavior because of the two previous points.
> >
> > Given that, an explicit new flag seems to make the most sense.
> >
> > Signed-off-by: Johannes Weiner <[email protected]>
>
> Just for context, I found this discussion back from 2018:
>
> https://lkml.iu.edu/hypermail/linux/kernel/1806.1/00483.html
>
> It seems to me that the usecase wasn't really in question, but people
> weren't sure about the API, and then Jason found a workaround before
> the discussion really concluded. I was asked internally about this
> feature, so I'm submitting another patch in this direction, but with
> more thoughts on why I chose to go with a new flag. Hopefully we can
> work it out this time around :-)
Thanks for the link. The topic sounded familiar but I couldn't really
remember any details anymore. Now I do remember that I wasn't happy
about special casing MLOCK_ONFAULT. A dedicated madvise operation
is definitely safer and I am OK with that. Presented usecases make sense
to me as well.
Btw. I have a recollection that Mike is working on MADV_DONTNEED support
for hugetlb pages. I do not know the current state of that work. Not
that it would make nay impact on your new flag but some minor changes
might be needed.
Anyway, after the madvise_need_mmap_write is addressed, feel free to add
Acked-by: Michal Hocko <[email protected]>
Thanks!
--
Michal Hocko
SUSE Labs
On 3/4/22 09:19, Johannes Weiner wrote:
> MADV_DONTNEED historically rejects mlocked ranges, but with
> MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
> there are valid use cases for depopulating locked ranges as well.
>
> Users mlock memory to protect secrets. There are allocators for secure
> buffers that want in-use memory generally mlocked, but cleared and
> invalidated memory to give up the physical pages. This could be done
> with explicit munlock -> mlock calls on free -> alloc of course, but
> that adds two unnecessary syscalls, heavy mmap_sem write locks, vma
> splits and re-merges - only to get rid of the backing pages.
>
> Users also mlockall(MCL_ONFAULT) to suppress sustained paging, but are
> okay with on-demand initial population. It seems valid to selectively
> free some memory during the lifetime of such a process, without having
> to mess with its overall policy.
>
> Why add a separate flag? Isn't this a pretty niche usecase?
>
> - MADV_DONTNEED has been bailing on locked vmas forever. It's at least
> conceivable that someone, somewhere is relying on mlock to protect
> data from perhaps broader invalidation calls. Changing this behavior
> now could lead to quiet data corruption.
>
> - It also clarifies expectations around MADV_FREE and maybe
> MADV_REMOVE. It avoids the situation where one quietly behaves
> different than the others. MADV_FREE_LOCKED can be added later.
>
> - The combination of mlock() and madvise() in the first place is
> probably niche. But where it happens, I'd say that dropping pages
> from a locked region once they don't contain secrets or won't page
> anymore is much saner than relying on mlock to protect memory from
> speculative or errant invalidation calls. It's just that we can't
> change the default behavior because of the two previous points.
>
> Given that, an explicit new flag seems to make the most sense.
>
> Signed-off-by: Johannes Weiner <[email protected]>
> Acked-by: Michal Hocko <[email protected]>
> ---
> include/uapi/asm-generic/mman-common.h | 2 ++
> mm/madvise.c | 24 ++++++++++++++----------
> 2 files changed, 16 insertions(+), 10 deletions(-)
>
> v2:
> - mmap_sem for read is enough for DONTNEED_LOCKED, thanks Nadav
> - rebased on top of Mike's hugetlb DONTNEED patch in -mm
Thanks for rebasing on top of recent changes.
Reviewed-by: Mike Kravetz <[email protected]>
Looks like we both will be making madvise man page changes soon.
--
Mike Kravetz
On Fri, Mar 04, 2022 at 12:19:12PM -0500, Johannes Weiner wrote:
> MADV_DONTNEED historically rejects mlocked ranges, but with
> MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
> there are valid use cases for depopulating locked ranges as well.
> Users mlock memory to protect secrets. There are allocators for secure
> buffers that want in-use memory generally mlocked, but cleared and
> invalidated memory to give up the physical pages. This could be done
> with explicit munlock -> mlock calls on free -> alloc of course, but
> that adds two unnecessary syscalls, heavy mmap_sem write locks, vma
> splits and re-merges - only to get rid of the backing pages.
> Users also mlockall(MCL_ONFAULT) to suppress sustained paging, but are
> okay with on-demand initial population. It seems valid to selectively
> free some memory during the lifetime of such a process, without having
> to mess with its overall policy.
> Why add a separate flag? Isn't this a pretty niche usecase?
> - MADV_DONTNEED has been bailing on locked vmas forever. It's at least
> conceivable that someone, somewhere is relying on mlock to protect
> data from perhaps broader invalidation calls. Changing this behavior
> now could lead to quiet data corruption.
> - It also clarifies expectations around MADV_FREE and maybe
> MADV_REMOVE. It avoids the situation where one quietly behaves
> different than the others. MADV_FREE_LOCKED can be added later.
> - The combination of mlock() and madvise() in the first place is
> probably niche. But where it happens, I'd say that dropping pages
> from a locked region once they don't contain secrets or won't page
> anymore is much saner than relying on mlock to protect memory from
> speculative or errant invalidation calls. It's just that we can't
> change the default behavior because of the two previous points.
> Given that, an explicit new flag seems to make the most sense.
> Signed-off-by: Johannes Weiner <[email protected]>
> Acked-by: Michal Hocko <[email protected]>
Reviewed-by: Shakeel Butt <[email protected]>
On Thu, Mar 03, 2022 at 04:29:56PM -0500, Johannes Weiner wrote:
> MADV_DONTNEED historically rejects mlocked ranges, but with
> MLOCK_ONFAULT and MCL_ONFAULT allowing to mlock without populating,
> there are valid use cases for depopulating locked ranges as well.
>
> Users mlock memory to protect secrets. There are allocators for secure
> buffers that want in-use memory generally mlocked, but cleared and
> invalidated memory to give up the physical pages. This could be done
> with explicit munlock -> mlock calls on free -> alloc of course, but
> that adds two unnecessary syscalls, heavy mmap_sem write locks, vma
> splits and re-merges - only to get rid of the backing pages.
>
> Users also mlockall(MCL_ONFAULT) to suppress sustained paging, but are
> okay with on-demand initial population. It seems valid to selectively
> free some memory during the lifetime of such a process, without having
> to mess with its overall policy.
>
> Why add a separate flag? Isn't this a pretty niche usecase?
>
> - MADV_DONTNEED has been bailing on locked vmas forever. It's at least
> conceivable that someone, somewhere is relying on mlock to protect
> data from perhaps broader invalidation calls. Changing this behavior
> now could lead to quiet data corruption.
>
> - It also clarifies expectations around MADV_FREE and maybe
> MADV_REMOVE. It avoids the situation where one quietly behaves
> different than the others. MADV_FREE_LOCKED can be added later.
>
> - The combination of mlock() and madvise() in the first place is
> probably niche. But where it happens, I'd say that dropping pages
> from a locked region once they don't contain secrets or won't page
> anymore is much saner than relying on mlock to protect memory from
> speculative or errant invalidation calls. It's just that we can't
> change the default behavior because of the two previous points.
>
> Given that, an explicit new flag seems to make the most sense.
>
> Signed-off-by: Johannes Weiner <[email protected]>
Just for context, I found this discussion back from 2018:
https://lkml.iu.edu/hypermail/linux/kernel/1806.1/00483.html
It seems to me that the usecase wasn't really in question, but people
weren't sure about the API, and then Jason found a workaround before
the discussion really concluded. I was asked internally about this
feature, so I'm submitting another patch in this direction, but with
more thoughts on why I chose to go with a new flag. Hopefully we can
work it out this time around :-)
Thanks