[SRU][F][PATCH 1/1] RDMA/odp: Ensure the mm is still alive before creating an implicit child
Chengen Du
chengen.du at canonical.com
Fri Dec 15 10:03:43 UTC 2023
From: Jason Gunthorpe <jgg at mellanox.com>
BugLink: https://bugs.launchpad.net/bugs/2046534
Registration of a mmu_notifier requires the caller to hold a mmget() on
the mm as registration is not permitted to race with exit_mmap(). There is
a BUG_ON inside the mmu_notifier to guard against this.
Normally creating a umem is done against current which implicitly holds
the mmget(), however an implicit ODP child is created from a pagefault
work queue and is not guaranteed to have a mmget().
Call mmget() around this registration and abort faulting if the MM has
gone to exit_mmap().
Before the patch below the notifier was registered when the implicit ODP
parent was created, so there was no chance to register a notifier outside
of current.
Fixes: c571feca2dc9 ("RDMA/odp: use mmu_notifier_get/put for 'struct ib_ucontext_per_mm'")
Link: https://lore.kernel.org/r/20200227114118.94736-1-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro at mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg at mellanox.com>
(backported from commit a4e63bce1414df7ab6eb82ca9feb8494ce13e554)
[chengen - remove tgid-related code and adjust jump label from out_tgid
to out_mmput]
Signed-off-by: Chengen Du <chengen.du at canonical.com>
---
drivers/infiniband/core/umem_odp.c | 22 ++++++++++++++++++----
1 file changed, 18 insertions(+), 4 deletions(-)
diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c
index 615e4ce0343b..0e9445cb5627 100644
--- a/drivers/infiniband/core/umem_odp.c
+++ b/drivers/infiniband/core/umem_odp.c
@@ -356,12 +356,26 @@ struct ib_umem_odp *ib_umem_odp_alloc_child(struct ib_umem_odp *root,
umem->owning_mm = root->umem.owning_mm;
odp_data->page_shift = PAGE_SHIFT;
- ret = ib_init_umem_odp(odp_data);
- if (ret) {
- kfree(odp_data);
- return ERR_PTR(ret);
+ /*
+ * A mmget must be held when registering a notifier, the owming_mm only
+ * has a mm_grab at this point.
+ */
+ if (!mmget_not_zero(umem->owning_mm)) {
+ ret = -EFAULT;
+ goto out_free;
}
+
+ ret = ib_init_umem_odp(odp_data);
+ if (ret)
+ goto out_mmput;
+ mmput(umem->owning_mm);
return odp_data;
+
+out_mmput:
+ mmput(umem->owning_mm);
+out_free:
+ kfree(odp_data);
+ return ERR_PTR(ret);
}
EXPORT_SYMBOL(ib_umem_odp_alloc_child);
--
2.40.1
More information about the kernel-team
mailing list