[v2,0/2] btrfs: dio partial write corruption fix

Message

Boris Burkov Feb. 22, 2023, 12:49 a.m. UTC

If there is a page fault while btrfs reads the write buffer for a dio
write, then iomap will issue a partial bio which ultimately results in
an effective hole in the btrfs representation of the file. If what was
being written was not zeros, this means incorrect file contents.

The patch series consists of a prep patch creating a new ordered extent
allocation function and the business patch, which contains the fix
as well as the gory details of the bug itself.

---
Changelog:
v2:
- rename new ordered extent function
- pull the new function into a prep patch
- reorganize how the ordered_extent is stored/passed around to avoid so
many annoying memsets and exposing it to fs/btrfs/file.c
- lots of small code style improvements
- remove unintentional whitespace changes
- commit message improvements
- various ASSERTs for clarity/debugging

Boris Burkov (2):
  btrfs: btrfs_alloc_ordered_extent
  btrfs: fix dio continue after short write due to buffer page fault

 fs/btrfs/btrfs_inode.h  |  1 +
 fs/btrfs/file.c         | 11 +++++-
 fs/btrfs/inode.c        | 75 ++++++++++++++++++++++++++++++-----------
 fs/btrfs/ordered-data.c | 45 ++++++++++++++++++++-----
 fs/btrfs/ordered-data.h |  7 +++-
 5 files changed, 109 insertions(+), 30 deletions(-)

Comments

David Sterba Feb. 28, 2023, 10:43 p.m. UTC | #1

On Tue, Feb 21, 2023 at 04:49:58PM -0800, Boris Burkov wrote:
> If there is a page fault while btrfs reads the write buffer for a dio
> write, then iomap will issue a partial bio which ultimately results in
> an effective hole in the btrfs representation of the file. If what was
> being written was not zeros, this means incorrect file contents.
> 
> The patch series consists of a prep patch creating a new ordered extent
> allocation function and the business patch, which contains the fix
> as well as the gory details of the bug itself.
> 
> ---
> Changelog:
> v2:
> - rename new ordered extent function
> - pull the new function into a prep patch
> - reorganize how the ordered_extent is stored/passed around to avoid so
> many annoying memsets and exposing it to fs/btrfs/file.c
> - lots of small code style improvements
> - remove unintentional whitespace changes
> - commit message improvements
> - various ASSERTs for clarity/debugging
> 
> Boris Burkov (2):
>   btrfs: btrfs_alloc_ordered_extent
>   btrfs: fix dio continue after short write due to buffer page fault

Added to misc-next with the suggested updates, thanks.