[v11,0/7] Implement IOCTL to get and optionally clear info about PTEs

Message

Muhammad Usama Anjum March 9, 2023, 1:57 p.m. UTC

These patches are based on next-20230307 and UFFD_FEATURE_WP_UNPOPULATED
patches from Peter.

*Changes in v11*
- Rebase on top of next-20230307
- Base patches on UFFD_FEATURE_WP_UNPOPULATED (https://lore.kernel.org/all/[e-mail])
- Do a lot of cosmetic changes and review updates
- Remove ENGAGE_WP + ! GET operation as it can be performed with UFFDIO_WRITEPROTECT

*Changes in v10*
- Add specific condition to return error if hugetlb is used with wp
  async
- Move changes in tools/include/uapi/linux/fs.h to separate patch
- Add documentation

*Changes in v9:*
- Correct fault resolution for userfaultfd wp async
- Fix build warnings and errors which were happening on some configs
- Simplify pagemap ioctl's code

*Changes in v8:*
- Update uffd async wp implementation
- Improve PAGEMAP_IOCTL implementation

*Changes in v7:*
- Add uffd wp async
- Update the IOCTL to use uffd under the hood instead of soft-dirty
  flags

Hello,

Note:
Soft-dirty pages and pages which have been written-to are synonyms. As
kernel already has soft-dirty feature inside which we have given up to
use, we are using written-to terminology while using UFFD async WP under
the hood.

This IOCTL, PAGEMAP_SCAN on pagemap file can be used to get and/or clear
the info about page table entries. The following operations are
supported in this ioctl:
- Get the information if the pages have been written-to (PAGE_IS_WRITTEN),
  file mapped (PAGE_IS_FILE), present (PAGE_IS_PRESENT) or swapped
  (PAGE_IS_SWAPPED).
- Write-protect the pages (PAGEMAP_WP_ENGAGE) to start finding which
  pages have been written-to.
- Find pages which have been written-to and write protect the pages
  (atomic PAGE_IS_WRITTEN + PAGEMAP_WP_ENGAGE)

It is possible to find and clear soft-dirty pages entirely in userspace.
But it isn't efficient:
- The mprotect and SIGSEGV handler for bookkeeping
- The userfaultfd wp (synchronous) with the handler for bookkeeping

Some benchmarks can be seen here[1]. This series adds features that weren't
present earlier:
- There is no atomic get soft-dirty/Written-to status and clear present in
  the kernel.
- The pages which have been written-to can not be found in accurate way.
  (Kernel's soft-dirty PTE bit + sof_dirty VMA bit shows more soft-dirty
  pages than there actually are.)

Historically, soft-dirty PTE bit tracking has been used in the CRIU
project. The procfs interface is enough for finding the soft-dirty bit
status and clearing the soft-dirty bit of all the pages of a process.
We have the use case where we need to track the soft-dirty PTE bit for
only specific pages on-demand. We need this tracking and clear mechanism
of a region of memory while the process is running to emulate the
getWriteWatch() syscall of Windows.

*(Moved to using UFFD instead of soft-dirtyi feature to find pages which
have been written-to from v7 patch series)*:
Stop using the soft-dirty flags for finding which pages have been
written to. It is too delicate and wrong as it shows more soft-dirty
pages than the actual soft-dirty pages. There is no interest in
correcting it [2][3] as this is how the feature was written years ago.
It shouldn't be updated to changed behaviour. Peter Xu has suggested
using the async version of the UFFD WP [4] as it is based inherently
on the PTEs.

So in this patch series, I've added a new mode to the UFFD which is
asynchronous version of the write protect. When this variant of the
UFFD WP is used, the page faults are resolved automatically by the
kernel. The pages which have been written-to can be found by reading
pagemap file (!PM_UFFD_WP). This feature can be used successfully to
find which pages have been written to from the time the pages were
write protected. This works just like the soft-dirty flag without
showing any extra pages which aren't soft-dirty in reality.

The information related to pages if the page is file mapped, present and
swapped is required for the CRIU project [5][6]. The addition of the
required mask, any mask, excluded mask and return masks are also required
for the CRIU project [5].

The IOCTL returns the addresses of the pages which match the specific
masks. The page addresses are returned in struct page_region in a compact
form. The max_pages is needed to support a use case where user only wants
to get a specific number of pages. So there is no need to find all the
pages of interest in the range when max_pages is specified. The IOCTL
returns when the maximum number of the pages are found. The max_pages is
optional. If max_pages is specified, it must be equal or greater than the
vec_size. This restriction is needed to handle worse case when one
page_region only contains info of one page and it cannot be compacted.
This is needed to emulate the Windows getWriteWatch() syscall.

The patch series include the detailed selftest which can be used as an
example for the uffd async wp test and PAGEMAP_IOCTL. It shows the
interface usages as well.

[1] https://lore.kernel.org/lkml/[e-mail]/
[2] https://lore.kernel.org/all/[e-mail]
[3] https://lore.kernel.org/all/[e-mail]
[4] https://lore.kernel.org/all/Y6Hc2d+7eTKs7AiH@x1n
[5] https://lore.kernel.org/all/[e-mail]/
[6] https://lore.kernel.org/all/[e-mail]/

Regards,
Muhammad Usama Anjum

Muhammad Usama Anjum (7):
  userfaultfd: Add UFFD WP Async support
  userfaultfd: Define dummy uffd_wp_range()
  userfaultfd: update documentation to describe UFFD_FEATURE_WP_ASYNC
  fs/proc/task_mmu: Implement IOCTL to get and optionally clear info
    about PTEs
  tools headers UAPI: Update linux/fs.h with the kernel sources
  mm/pagemap: add documentation of PAGEMAP_SCAN IOCTL
  selftests: mm: add pagemap ioctl tests

 Documentation/admin-guide/mm/pagemap.rst     |  56 ++
 Documentation/admin-guide/mm/userfaultfd.rst |  21 +
 fs/proc/task_mmu.c                           | 366 ++++++++
 fs/userfaultfd.c                             |  25 +-
 include/linux/userfaultfd_k.h                |  14 +
 include/uapi/linux/fs.h                      |  53 ++
 include/uapi/linux/userfaultfd.h             |  11 +-
 mm/memory.c                                  |  27 +-
 tools/include/uapi/linux/fs.h                |  53 ++
 tools/testing/selftests/mm/.gitignore        |   1 +
 tools/testing/selftests/mm/Makefile          |   4 +-
 tools/testing/selftests/mm/config            |   1 +
 tools/testing/selftests/mm/pagemap_ioctl.c   | 920 +++++++++++++++++++
 tools/testing/selftests/mm/run_vmtests.sh    |   4 +
 14 files changed, 1549 insertions(+), 7 deletions(-)
 create mode 100644 tools/testing/selftests/mm/pagemap_ioctl.c
 mode change 100644 => 100755 tools/testing/selftests/mm/run_vmtests.sh

Comments

Andrew Morton March 9, 2023, 7:58 p.m. UTC | #1

On Thu,  9 Mar 2023 18:57:11 +0500 Muhammad Usama Anjum <[e-mail]> wrote:

> The information related to pages if the page is file mapped, present and
> swapped is required for the CRIU project [5][6]. The addition of the
> required mask, any mask, excluded mask and return masks are also required
> for the CRIU project [5].

It's a ton of new code and what I'm not seeing in here (might have
missed it?) is a clear statement of the value of this feature to our
users.

I see hints that CRIU would like it, but no description of how valuable
this is to CRIU's users.

So please spend some time preparing this info.

Also, are any other applications of this feature anticipated?  If so,
what are they?

IOW, please sell this stuff to us!

Muhammad Usama Anjum March 9, 2023, 10:24 p.m. UTC | #2

Hi Andrew,

Thank you for your message.

On 3/10/23 12:58 AM, Andrew Morton wrote:
> On Thu,  9 Mar 2023 18:57:11 +0500 Muhammad Usama Anjum <[e-mail]> wrote:
> 
>> The information related to pages if the page is file mapped, present and
>> swapped is required for the CRIU project [5][6]. The addition of the
>> required mask, any mask, excluded mask and return masks are also required
>> for the CRIU project [5].
> 
> It's a ton of new code and what I'm not seeing in here (might have
> missed it?) is a clear statement of the value of this feature to our
> users.
Sorry, let me give some clear details about its value.

> 
> I see hints that CRIU would like it, but no description of how valuable
> this is to CRIU's users.
> 
> So please spend some time preparing this info.
> 
> Also, are any other applications of this feature anticipated?  If so,
> what are they?
> 
> IOW, please sell this stuff to us!
The real motivation for adding PAGEMAP_SCAN IOCTL is to emulate Windows
GetWriteWatch() syscall [1].
> The GetWriteWatch{} retrieves the addresses of the pages that are written
> to in a region of virtual memory.
This syscall is used in Windows applications and games etc. This syscall is
being emulated in pretty slow manner in userspace. Our purpose is to
enhance the kernel such that we translate it efficiently in a better way.
Currently some out of tree hack patches are being used to efficiently
emulate it in some kernels. We intend to replace those with these patches.
So the whole gaming on Linux can effectively get benefit from this. It
means there would be tons of users of this code.

CRIU use case [2] was mentioned by Andrei and Danylo:
> Use cases for migrating sparse VMAs are binaries sanitized with ASAN,
> MSAN or TSAN [3]. All of these sanitizers produce sparse mappings of
> shadow memory [4]. Being able to migrate such binaries allows to highly
> reduce the amount of work needed to identify and fix post-migration
> crashes, which happen constantly.
At [10]:
> For Andrei's use case is to iteratively dump memory.

@Andrei and Danylo can elaborate more on their use cases.


*Implementation Evolution (Short Summary)*
From the definition of GetWriteWatch(), we feel like kernel's soft-dirty
feature can be used under the hood with some additions like:
* reset soft-dirty flag for only a specific region of memory instead of
clearing the flag for the entire process
* get and clear soft-dirty flag for a specific region atomically

So we decided to use ioctl on pagemap file to read or/and reset soft-dirty
flag. But using soft-dirty flag, sometimes we get extra pages which weren't
even written. They had become soft-dirty because of VMA merging and
VM_SOFTDIRTY flag. This breaks the definition of GetWriteWatch(). We were
able to by-pass this short coming by ignoring VM_SOFTDIRTY until David
reported that mprotect etc messes up the soft-dirty flag while ignoring
VM_SOFTDIRTY [5]. This wasn't happening until [6] got introduced. We
discussed if we can revert these patches. But we could not reach to any
conclusion. So at this point, I made couple of tries to solve this whole
VM_SOFTDIRTY issue by correcting the soft-dirty implementation:
* [7] Correct the bug fixed wrongly back in 2014. It had potential to cause
regression. We left it behind.
* [8] Keep a list of soft-dirty part of a VMA across splits and merges. I
got the reply don't increase the size of the VMA by 8 bytes.

At this point, we left soft-dirty considering it is too much delicate and
userfaultfd [9] seemed like the only way forward. From there onward, we
have been basing soft-dirty emulation on userfaultfd wp feature where
kernel resolves the faults itself when WP_ASYNC feature is used. It was
straight forward to add WP_ASYNC feature in userfautlfd. Now we get only
those pages dirty or written-to which are really written in reality. (PS
There is another WP_UNPOPULATED userfautfd feature is required which is
needed to avoid pre-faulting memory before write-protecting [9].)

All the different masks were added on the request of CRIU devs to create
interface more generic and better.

[1]
https://learn.microsoft.com/en-us/windows/win32/api/memoryapi/nf-memoryapi-getwritewatch
[2] https://lore.kernel.org/all/[e-mail]
[3] https://github.com/google/sanitizers	
[4] https://github.com/google/sanitizers/wiki/AddressSanitizerAlgorithm#64-bit
[5]
https://lore.kernel.org/all/[e-mail]
[6] https://lore.kernel.org/all/[e-mail]/
[7]
https://lore.kernel.org/all/[e-mail]
[8]
https://lore.kernel.org/all/[e-mail]
[9] https://lore.kernel.org/all/[e-mail]
[10] https://lore.kernel.org/all/[e-mail]

Andrei Vagin March 20, 2023, 6:30 p.m. UTC | #3

On Thu, Mar 9, 2023 at 11:58 AM Andrew Morton <[e-mail]> wrote:
>
> On Thu,  9 Mar 2023 18:57:11 +0500 Muhammad Usama Anjum <[e-mail]> wrote:
>
> > The information related to pages if the page is file mapped, present and
> > swapped is required for the CRIU project [5][6]. The addition of the
> > required mask, any mask, excluded mask and return masks are also required
> > for the CRIU project [5].
>
> It's a ton of new code and what I'm not seeing in here (might have
> missed it?) is a clear statement of the value of this feature to our
> users.
>
> I see hints that CRIU would like it, but no description of how valuable
> this is to CRIU's users.


Hi Andrew,

The current interface works for CRIU, and I can't say we have anything
critical with it right now.

On the other hand, the new interface has a number of significant improvements:

* it is more granular and allows us to track changed pages more
  effectively. The current interface can clear dirty bits for the entire
  process only. In addition, reading info about pages is a separate
  operation. It means we must freeze the process to read information
  about all its pages, reset dirty bits, only then we can start dumping
  pages. The information about pages becomes more and more outdated,
  while we are processing pages. The new interface solves both these
  downsides. First, it allows us to read pte bits and clear the
  soft-dirty bit atomically. It means that CRIU will not need to freeze
  processes to pre-dump their memory. Second, it clears soft-dirty bits
  for a specified region of memory. It means CRIU will have actual info
  about pages to the moment of dumping them.

* The new interface has to be much faster because basic page filtering
  is happening in the kernel. With the old interface, we have to read
  pagemap for each page.


Thanks,
Andrei

>
> So please spend some time preparing this info.
>
> Also, are any other applications of this feature anticipated?  If so,
> what are they?
>
> IOW, please sell this stuff to us!

Mike Rapoport March 21, 2023, 12:41 p.m. UTC | #4

On Mon, Mar 20, 2023 at 11:30:00AM -0700, Andrei Vagin wrote:
> On Thu, Mar 9, 2023 at 11:58 AM Andrew Morton <[e-mail]> wrote:
> >
> > On Thu,  9 Mar 2023 18:57:11 +0500 Muhammad Usama Anjum <[e-mail]> wrote:
> >
> > > The information related to pages if the page is file mapped, present and
> > > swapped is required for the CRIU project [5][6]. The addition of the
> > > required mask, any mask, excluded mask and return masks are also required
> > > for the CRIU project [5].
> >
> > It's a ton of new code and what I'm not seeing in here (might have
> > missed it?) is a clear statement of the value of this feature to our
> > users.
> >
> > I see hints that CRIU would like it, but no description of how valuable
> > this is to CRIU's users.
> 
> Hi Andrew,
> 
> The current interface works for CRIU, and I can't say we have anything
> critical with it right now.
> 
> On the other hand, the new interface has a number of significant improvements:
> 
> * it is more granular and allows us to track changed pages more
>   effectively. The current interface can clear dirty bits for the entire
>   process only. In addition, reading info about pages is a separate
>   operation. It means we must freeze the process to read information
>   about all its pages, reset dirty bits, only then we can start dumping
>   pages. The information about pages becomes more and more outdated,
>   while we are processing pages. The new interface solves both these
>   downsides. First, it allows us to read pte bits and clear the
>   soft-dirty bit atomically. It means that CRIU will not need to freeze
>   processes to pre-dump their memory. Second, it clears soft-dirty bits
>   for a specified region of memory. It means CRIU will have actual info
>   about pages to the moment of dumping them.
> 
> * The new interface has to be much faster because basic page filtering
>   is happening in the kernel. With the old interface, we have to read
>   pagemap for each page.

There is still a caveat in using userfaultfd for tracking dirty pages in
CRIU because we still don't support C/R of processes that use uffd. 
 
> Thanks,
> Andrei
> 
> >
> > So please spend some time preparing this info.
> >
> > Also, are any other applications of this feature anticipated?  If so,
> > what are they?
> >
> > IOW, please sell this stuff to us!

Peter Xu March 21, 2023, 3:10 p.m. UTC | #5

On Tue, Mar 21, 2023 at 02:41:53PM +0200, Mike Rapoport wrote:
> On Mon, Mar 20, 2023 at 11:30:00AM -0700, Andrei Vagin wrote:
> > On Thu, Mar 9, 2023 at 11:58 AM Andrew Morton <[e-mail]> wrote:
> > >
> > > On Thu,  9 Mar 2023 18:57:11 +0500 Muhammad Usama Anjum <[e-mail]> wrote:
> > >
> > > > The information related to pages if the page is file mapped, present and
> > > > swapped is required for the CRIU project [5][6]. The addition of the
> > > > required mask, any mask, excluded mask and return masks are also required
> > > > for the CRIU project [5].
> > >
> > > It's a ton of new code and what I'm not seeing in here (might have
> > > missed it?) is a clear statement of the value of this feature to our
> > > users.
> > >
> > > I see hints that CRIU would like it, but no description of how valuable
> > > this is to CRIU's users.
> > 
> > Hi Andrew,
> > 
> > The current interface works for CRIU, and I can't say we have anything
> > critical with it right now.
> > 
> > On the other hand, the new interface has a number of significant improvements:
> > 
> > * it is more granular and allows us to track changed pages more
> >   effectively. The current interface can clear dirty bits for the entire
> >   process only. In addition, reading info about pages is a separate
> >   operation. It means we must freeze the process to read information
> >   about all its pages, reset dirty bits, only then we can start dumping
> >   pages. The information about pages becomes more and more outdated,
> >   while we are processing pages. The new interface solves both these
> >   downsides. First, it allows us to read pte bits and clear the
> >   soft-dirty bit atomically. It means that CRIU will not need to freeze
> >   processes to pre-dump their memory. Second, it clears soft-dirty bits
> >   for a specified region of memory. It means CRIU will have actual info
> >   about pages to the moment of dumping them.
> > 
> > * The new interface has to be much faster because basic page filtering
> >   is happening in the kernel. With the old interface, we have to read
> >   pagemap for each page.
> 
> There is still a caveat in using userfaultfd for tracking dirty pages in
> CRIU because we still don't support C/R of processes that use uffd. 

This reminded me whether the interface can also expose soft-dirty as a
ranged soft-dirty collector too to replace existing pagemap read()s?  Just
in case userfault cannot be used.  The code addition should be trivial IIUC.

Then maybe PAGE_IS_WRITTEN will be a name too generic, it can be two bits
PAGE_IS_UFFD_WP and PAGE_IS_SOFT_DIRTY, having PAGE_IS_UFFD_WP the inverted
meaning of current PAGE_IS_WRITTEN.