[net-next,v8,0/2] Fix late DMA unmap crash for page pool

Message

Toke Høiland-Jørgensen April 7, 2025, 4:53 p.m. UTC

This series fixes the late dma_unmap crash for page pool first reported
by Yonglong Liu in [0]. It is an alternative approach to the one
submitted by Yunsheng Lin, most recently in [1]. The first commit just
wraps some tests in a helper function, in preparation of the main change
in patch 2. See the commit message of patch 2 for the details.

-Toke

[0] https://lore.kernel.org/lkml/[e-mail]/T/
[1] https://lore.kernel.org/r/[e-mail]

Signed-off-by: Toke Høiland-Jørgensen <[e-mail]>
---
Changes in v8:
- Move defines to mm.h
- Keep pp->dma_sync as 1-bit wide
- Unset pp->dma_addr on id alloc failure
- Rebase on -rc1
- Link to v7: https://lore.kernel.org/r/[e-mail]

Changes in v7:
- Change WARN_ON logic if xarray alloc fails
- Don't leak xarray ID if page_pool_set_dma_addr_netmem() fails
- Unconditionally init xarray in page_pool_init()
- Rebase on current net-next
- Link to v6: https://lore.kernel.org/r/[e-mail]

Changes in v6:
- Add READ_ONCE() around both reads of pp->dma_sync
- Link to v5: https://lore.kernel.org/r/[e-mail]

Changes in v5:
- Dereferencing pp->p.dev if pp->pma_sync is unset could lead to a
  crash, so make sure we don't do that.
- With the change above, patch 2 was just changing a single field, so
  squash it into patch 3
- Link to v4: https://lore.kernel.org/r/[e-mail]

Changes in v4:
- Rebase on net-next
- Collect tags
- Link to v3: https://lore.kernel.org/r/[e-mail]

Changes in v3:
- Use a full-width bool for pp->dma_sync instead of a full unsigned
  long (in patch 2), and leave pp->dma_sync_cpu alone.

- Link to v2: https://lore.kernel.org/r/[e-mail]

Changes in v2:
- Always leave two bits at the top of pp_magic as zero, instead of one

- Add an rcu_read_lock() around __page_pool_dma_sync_for_device()

- Add a comment in poison.h with a reference to the bitmask definition

- Add a longer description of the logic of the bitmask definitions to
  the comment in types.h, and a summary of the security implications of
  using the pp_magic field to the commit message of patch 3

- Collect Mina's Reviewed-by and Yonglong's Tested-by tags

- Link to v1: https://lore.kernel.org/r/[e-mail]

---
Toke Høiland-Jørgensen (2):
      page_pool: Move pp_magic check into helper functions
      page_pool: Track DMA-mapped pages and unmap them when destroying the pool

 drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c |  4 +-
 include/linux/mm.h                               | 59 +++++++++++++++++
 include/linux/poison.h                           |  4 ++
 include/net/page_pool/types.h                    |  7 ++
 mm/page_alloc.c                                  |  8 +--
 net/core/netmem_priv.h                           | 33 +++++++++-
 net/core/page_pool.c                             | 81 ++++++++++++++++++++----
 net/core/skbuff.c                                | 16 +----
 net/core/xdp.c                                   |  4 +-
 9 files changed, 178 insertions(+), 38 deletions(-)
---
base-commit: 61f96e684edd28ca40555ec49ea1555df31ba619
change-id: 20250310-page-pool-track-dma-0332343a460e