Fix race condition in Devres
Subsystems (3)
Danilo Krummrich
maintainerv2 cover Fix race condition in Devres
This patch series fixes a race condition in Devres.
Please see patch 3 "rust: devres: fix race in Devres::drop()" for a detailed description of the race and how it is fixed.
None of the upstream users of Devres is prone to this race, hence we do not necessarily have to backport this -- yet, it must be fixed.
Thanks to Alice for catching and reporting this!
A branch containing those patches can be found in [1].
[1] https://git.kernel.org/pub/scm/linux/kernel/git/dakr/linux.git/log/?h=rust/devres-race
Changes in v2:
- Completion
- s/can queue themselves up/have been queued up/
- move up Send / Sync impl
- refer to complete_all() from wait_for_completion()
- clarify "permanentely done" with "i.e. signals all current and future waiters"
- Devres
- pick more appropriate 'Fixes' tag
- extend commit message with an illustration of the race
Danilo Krummrich (3):
rust: completion: implement initial abstraction
rust: revocable: indicate whether data has been revoked already
rust: devres: fix race in Devres::drop()
rust/bindings/bindings_helper.h | 1 + rust/helpers/completion.c | 8 +++ rust/helpers/helpers.c | 1 + rust/kernel/devres.rs | 33 ++++++++-- rust/kernel/revocable.rs | 18 +++-- rust/kernel/sync.rs | 2 + rust/kernel/sync/completion.rs | 112 ++++++++++++++++++++++++++++++++ 7 files changed, 164 insertions(+), 11 deletions(-) create mode 100644 rust/helpers/completion.c create mode 100644 rust/kernel/sync/completion.rs
base-commit: 19272b37aa4f83ca52bdf9c16d5d81bdd1354494
Signature · 1 line
2.49.0
Danilo Krummrich
maintainerPATCH v2 1/3 rust: completion: implement initial abstraction
Implement a minimal abstraction for the completion synchronization primitive.
This initial abstraction only adds complete_all() and wait_for_completion(), since that is what is required for the subsequent Devres patch.
--- rust/bindings/bindings_helper.h | 1 + rust/helpers/completion.c | 8 +++ rust/helpers/helpers.c | 1 + rust/kernel/sync.rs | 2 + rust/kernel/sync/completion.rs | 112 ++++++++++++++++++++++++++++++++ 5 files changed, 124 insertions(+) create mode 100644 rust/helpers/completion.c create mode 100644 rust/kernel/sync/completion.rs
rust/bindings/bindings_helper.h +1 -0
@@ -39,6 +39,7 @@#include <linux/blk_types.h> #include <linux/blkdev.h> #include <linux/clk.h> +#include <linux/completion.h> #include <linux/configfs.h> #include <linux/cpu.h> #include <linux/cpufreq.h>
rust/helpers/completion.c new file +8 -0
@@ -0,0 +1,8 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <linux/completion.h> + +void rust_helper_init_completion (struct completion *x) +{ +init_completion (x); +}
rust/helpers/helpers.c +1 -0
@@ -13,6 +13,7 @@#include "build_assert.c" #include "build_bug.c" #include "clk.c" +#include "completion.c" #include "cpufreq.c" #include "cpumask.c" #include "cred.c"
rust/kernel/sync.rs +2 -0
@@ -10,6 +10,7 @@use pin_init;mod arc; +pub mod completion;mod condvar;pub mod lock;mod locked_by;
@@ -17,6 +18,7 @@pub mod rcu;pub use arc::{Arc ,ArcBorrow ,UniqueArc }; +pub use completion::Completion ;pub use condvar::{new_condvar,CondVar ,CondVarTimeoutResult };pub use lock::global::{global_lock,GlobalGuard ,GlobalLock ,GlobalLockBackend ,GlobalLockedBy };pub use lock::mutex::{new_mutex,Mutex ,MutexGuard };
rust/kernel/sync/completion.rs new file +112 -0
112 lines not on this page · raw message
Signature · 1 line
2.49.0
Re: [PATCH v2 1/3] rust: completion: implement initial abstraction
Quoted 8 lines · Danilo Krummrich
Implement a minimal abstraction for the completion synchronization primitive.
This initial abstraction only adds
complete_all()andwait_for_completion(), since that is what is required for the subsequent Devres patch.
Quoted 13 lines
Cc Ingo Molnar <[email protected]>Cc Peter Zijlstra <[email protected]>Cc Juri Lelli <[email protected]>Cc Vincent Guittot <[email protected]>Cc Dietmar Eggemann <[email protected]>Cc Steven Rostedt <[email protected]>Cc Ben Segall <[email protected]>Cc Mel Gorman <[email protected]>Cc Valentin Schneider <[email protected]>Reviewed-by Alice Ryhl <[email protected]>Reviewed-by Boqun Feng <[email protected]>Reviewed-by Benno Lossin <[email protected]>Signed-off-by Danilo Krummrich <[email protected]>
This is part of a fix, so it is a bit of an edge case, but it is just the bits Danilo needs, so as long as nobody in the Cc list above complains:
Quoted 1 line
+/// The [
Completion] synchronization primitive signales when a certain task has been completed by
signals
Cheers,
Miguel
Danilo Krummrich
maintainerPATCH v2 2/3 rust: revocable: indicate whether `data` has been revoked already
In reply to Danilo KrummrichReturn a boolean from Revocable::revoke() and Revocable::revoke_nosync() to indicate whether the data has been revoked already.
Return true if the data hasn't been revoked yet (i.e. this call revoked the data), false otherwise.
--- rust/kernel/revocable.rs | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-)
rust/kernel/revocable.rs +14 -4
@@ -154,8 +154,10 @@ pub unsafe fn access(&self) -> &T {/// # Safety /// /// Callers must ensure that there are no more concurrent users of the revocable object. -unsafe fn revoke_internal <const SYNC :bool >(&self) { -if self.is_available.swap (false ,Ordering ::Relaxed ) { +unsafe fn revoke_internal <const SYNC :bool >(&self) ->bool { +let revoke = self.is_available.swap (false ,Ordering ::Relaxed ); + +if revoke {if SYNC {// SAFETY: Just an FFI call, there are no further requirements. unsafe { bindings::synchronize_rcu () };
@@ -165,6 +167,8 @@ unsafe fn revoke_internal<const SYNC: bool>(&self) {// `compare_exchange` above that takes `is_available` from `true` to `false`. unsafe {drop_in_place (self.data.get ()) }; } + + revoke }/// Revokes access to and drops the wrapped object.
@@ -172,10 +176,13 @@ unsafe fn revoke_internal<const SYNC: bool>(&self) {/// Access to the object is revoked immediately to new callers of [`Revocable::try_access`], /// expecting that there are no concurrent users of the object. /// +/// Returns `true` if `&self` has been revoked with this call, `false` if it was revoked +/// already. +/// /// # Safety /// /// Callers must ensure that there are no more concurrent users of the revocable object. -pub unsafe fn revoke_nosync (&self) { +pub unsafe fn revoke_nosync (&self) ->bool {// SAFETY: By the safety requirement of this function, the caller ensures that nobody is // accessing the data anymore and hence we don't have to wait for the grace period to // finish.
@@ -189,7 +196,10 @@ pub unsafe fn revoke_nosync(&self) {/// If there are concurrent users of the object (i.e., ones that called /// [`Revocable::try_access`] beforehand and still haven't dropped the returned guard), this /// function waits for the concurrent access to complete before dropping the wrapped object. -pub fn revoke (&self) { +/// +/// Returns `true` if `&self` has been revoked with this call, `false` if it was revoked +/// already. +pub fn revoke (&self) ->bool {// SAFETY: By passing `true` we ask `revoke_internal` to wait for the grace period to // finish. unsafe { self.revoke_internal ::<true >() }
Signature · 1 line
2.49.0
Re: [PATCH v2 2/3] rust: revocable: indicate whether `data` has been revoked already
Quoted 9 lines · Danilo Krummrich
Return a boolean from
Revocable::revoke()andRevocable::revoke_nosync()to indicate whether the data has been revoked already.Return true if the data hasn't been revoked yet (i.e. this call revoked the data), false otherwise.
Reviewed-by Benno Lossin <[email protected]>Signed-off-by Danilo Krummrich <[email protected]>
It could help those reading the Git log later on to mention the use case comes with the next commit (i.e. the "why").
Cheers,
Miguel
Danilo Krummrich
maintainerPATCH v2 3/3 rust: devres: fix race in Devres::drop()
In reply to Danilo KrummrichIn Devres::drop() we first remove the devres action and then drop the wrapped device resource.
The design goal is to give the owner of a Devres object control over when the device resource is dropped, but limit the overall scope to the corresponding device being bound to a driver.
However, there's a race that was introduced with commit 8ff656643d30 ("rust: devres: remove action in Devres::drop"), but also has been (partially) present from the initial version on.
In Devres::drop(), the devres action is removed successfully and subsequently the destructor of the wrapped device resource runs. However, there is no guarantee that the destructor of the wrapped device resource completes before the driver core is done unbinding the corresponding device.
If in Devres::drop(), the devres action can't be removed, it means that the devres callback has been executed already, or is still running concurrently. In case of the latter, either Devres::drop() wins revoking the Revocable or the devres callback wins revoking the Revocable. If Devres::drop() wins, we (again) have no guarantee that the destructor of the wrapped device resource completes before the driver core is done unbinding the corresponding device.
CPU0 CPU1
------------------------------------------------------------------------Devres::drop() { Devres::devres_callback() {
self.data.revoke() { this.data.revoke() {
is_available.swap() == true
is_available.swap == false
}
}
// [...]
// device fully unbound
drop_in_place() {
// release device resource
} }
}
Depending on the specific device resource, this can potentially lead to user-after-free bugs.
In order to fix this, implement the following logic.
In the devres callback, we're always good when we get to revoke the device resource ourselves, i.e. Revocable::revoke() returns true.
If Revocable::revoke() returns false, it means that Devres::drop(), concurrently, already drops the device resource and we have to wait for Devres::drop() to signal that it finished dropping the device resource.
Note that if we hit the case where we need to wait for the completion of Devres::drop() in the devres callback, it means that we're actually racing with a concurrent Devres::drop() call, which already started revoking the device resource for us. This is rather unlikely and means that the concurrent Devres::drop() already started doing our work and we just need to wait for it to complete it for us. Hence, there should not be any additional overhead from that.
(Actually, for now it's even better if Devres::drop() does the work for us, since it can bypass the synchronize_rcu() call implied by Revocable::revoke(), but this goes away anyways once I get to implement the split devres callback approach, which allows us to first flip the atomics of all registered Devres objects of a certain device, execute a single synchronize_rcu() and then drop all revocable objects.)
In Devres::drop() we try to revoke the device resource. If that is *not* successful, it means that the devres callback already did and we're good.
Otherwise, we try to remove the devres action, which, if successful, means that we're good, since the device resource has just been revoked by us *before* we removed the devres action successfully.
If the devres action could not be removed, it means that the devres callback must be running concurrently, hence we signal that the device resource has been revoked by us, using the completion.
This makes it safe to drop a Devres object from any task and at any point of time, which is one of the design goals.
--- rust/kernel/devres.rs | 33 ++++++++++++++++++++++++++------- 1 file changed, 26 insertions(+), 7 deletions(-)
rust/kernel/devres.rs +26 -7
@@ -13,7 +13,7 @@ ffi::c_void, prelude::*, revocable::Revocable, - sync::Arc, + sync::{Arc, Completion}, types::ARef, };
@@ -25,6 +25,8 @@ struct DevresInner<T> { callback: unsafe extern "C" fn(*mut c_void),#[pin] data: Revocable<T>, +#[pin] + revoke: Completion, }/// This abstraction is meant to be used by subsystems to containerize [`Device`] bound resources to
@@ -102,6 +104,7 @@ fn new(dev: &Device<Bound>, data: T, flags: Flags) -> Result<Arc<DevresInner<T>> dev: dev.into(), callback: Self::devres_callback, data <- Revocable::new(data), + revoke <- Completion::new(), }), flags, )?;
@@ -130,26 +133,28 @@ fn as_ptr(&self) -> *const Self { selfas _ } -fn remove_action (this: &Arc <Self >) { +fn remove_action (this: &Arc <Self >) ->bool {// SAFETY: // - `self.inner.dev` is a valid `Device`, // - the `action` and `data` pointers are the exact same ones as given to devm_add_action() // previously, // - `self` is always valid, even if the action has been released already. -let ret =unsafe { +let success =unsafe { bindings::devm_remove_action_nowarn ( this.dev.as_raw (),Some (this.callback), this.as_ptr ()as _ , ) - }; + } ==0 ; -if ret ==0 { +if success {// SAFETY: We leaked an `Arc` reference to devm_add_action() in `DevresInner::new`; if // devm_remove_action_nowarn() was successful we can (and have to) claim back ownership // of this reference. let _ =unsafe {Arc ::from_raw (this.as_ptr ()) }; } + + success }#[allow(clippy::missing_safety_doc)]
@@ -161,7 +166,12 @@ fn remove_action(this: &Arc<Self>) {// `DevresInner::new`. let inner =unsafe {Arc ::from_raw (ptr) }; - inner.data.revoke (); +if !inner.data.revoke () { +// If `revoke()` returns false, it means that `Devres::drop` already started revoking +// `inner.data` for us. Hence we have to wait until `Devres::drop()` signals that it +// completed revoking `inner.data`. + inner.revoke.wait_for_completion (); + } } }
@@ -232,6 +242,15 @@ fn deref(&self) -> &Self::Target {impl <T >Drop for Devres <T > {fn drop (&mut self) { -DevresInner ::remove_action (&self.0 ); +// SAFETY: When `drop` runs, it is guaranteed that nobody is accessing the revocable data +// anymore, hence it is safe not to wait for the grace period to finish. +if unsafe { self.revoke_nosync () } { +// We revoked `self.0.data` before the devres action did, hence try to remove it. +if !DevresInner ::remove_action (&self.0 ) { +// We could not remove the devres action, which means that it now runs concurrently, +// hence signal that `self.0.data` has been revoked successfully. + self.0 .revoke.complete_all (); + } + } } }
Signature · 1 line
2.49.0
Re: [PATCH v2 3/3] rust: devres: fix race in Devres::drop()
Quoted 27 lines · Danilo Krummrich
In
Devres::drop()we first remove the devres action and then drop the wrapped device resource.The design goal is to give the owner of a Devres object control over when the device resource is dropped, but limit the overall scope to the corresponding device being bound to a driver.
However, there's a race that was introduced with commit
8ff656643d30("rust: devres: remove action inDevres::drop"), but also has been (partially) present from the initial version on.In
Devres::drop(), the devres action is removed successfully and subsequently the destructor of the wrapped device resource runs. However, there is no guarantee that the destructor of the wrapped device resource completes before the driver core is done unbinding the corresponding device.If in
Devres::drop(), the devres action can't be removed, it means that the devres callback has been executed already, or is still running concurrently. In case of the latter, eitherDevres::drop()wins revoking the Revocable or the devres callback wins revoking the Revocable. IfDevres::drop()wins, we (again) have no guarantee that the destructor of the wrapped device resource completes before the driver core is done unbinding the corresponding device.CPU0 CPU1
------------------------------------------------------------------------
Quoted 14 lines
Devres::drop() { Devres::devres_callback() {
self.data.revoke() { this.data.revoke() {
is_available.swap() == true
is_available.swap == false
}
}
// [...]
// device fully unbound
drop_in_place() {
// release device resource
}
}
}I forgot to mention: you used tabs, which breaks when tabstop is not set to 8 (such as in my editor. This is how it looks for me ):
------------------------------------------------------------------------
Devres::drop() { Devres::devres_callback() {
self.data.revoke() { this.data.revoke() {
is_available.swap() == true
is_available.swap == false
}
}
// [...]
// device fully unbound
drop_in_place() {
// release device resource
}
}
}I personally would have used spaces for this, but it looks fine in my gitlog, so feel free to keep it this way.
Quoted 38 lines
Depending on the specific device resource, this can potentially lead to user-after-free bugs.
In order to fix this, implement the following logic.
In the devres callback, we're always good when we get to revoke the device resource ourselves, i.e.
Revocable::revoke()returns true.If
Revocable::revoke()returns false, it means thatDevres::drop(), concurrently, already drops the device resource and we have to wait forDevres::drop()to signal that it finished dropping the device resource.Note that if we hit the case where we need to wait for the completion of
Devres::drop()in the devres callback, it means that we're actually racing with a concurrentDevres::drop()call, which already started revoking the device resource for us. This is rather unlikely and means that the concurrentDevres::drop()already started doing our work and we just need to wait for it to complete it for us. Hence, there should not be any additional overhead from that.(Actually, for now it's even better if
Devres::drop()does the work for us, since it can bypass thesynchronize_rcu()call implied byRevocable::revoke(), but this goes away anyways once I get to implement the split devres callback approach, which allows us to first flip the atomics of all registered Devres objects of a certain device, execute a singlesynchronize_rcu()and then drop all revocable objects.)In
Devres::drop()we try to revoke the device resource. If that is *not* successful, it means that the devres callback already did and we're good.Otherwise, we try to remove the devres action, which, if successful, means that we're good, since the device resource has just been revoked by us *before* we removed the devres action successfully.
If the devres action could not be removed, it means that the devres callback must be running concurrently, hence we signal that the device resource has been revoked by us, using the completion.
Quoted 7 lines
This makes it safe to drop a Devres object from any task and at any point of time, which is one of the design goals.
Fixes 76c01ded724b ("rust: add devres abstraction")Reported-by Alice Ryhl <[email protected]>Signed-off-by Danilo Krummrich <[email protected]>
---
Cheers,
Benno
Quoted 3 lines
--- rust/kernel/devres.rs | 33 ++++++++++++++++++++++++++------- 1 file changed, 26 insertions(+), 7 deletions(-)
Quoted 1 line · Danilo Krummrich
This patch series fixes a race condition in Devres.
Applied to driver-core-linus, thanks!