dma-buf/sw_sync: Avoid recursive lock during fence signal
Subsystems (3)
PATCH 1/1 dma-buf/sw_sync: Avoid recursive lock during fence signal
From: Rob Clark <[email protected]>
If a signal callback releases the sw_sync fence, that will trigger a deadlock as the timeline_fence_release recurses onto the fence->lock (used both for signaling and the the timeline tree).
To avoid that, temporarily hold an extra reference to the signalled fences until after we drop the lock.
(This is an alternative implementation of https://patchwork.kernel.org/patch/11664717/
which avoids some potential UAF issues with the original patch.)
--- drivers/dma-buf/sw_sync.c | 10 ++++++++++ 1 file changed, 10 insertions(+)
drivers/dma-buf/sw_sync.c +10 -0
@@ -191,6 +191,7 @@ static const struct dma_fence_ops timeline_fence_ops = {*/ static void sync_timeline_signal (struct sync_timeline *obj,unsigned int inc) { +LIST_HEAD (signalled);struct sync_pt *pt, *next;trace_sync_timeline (obj);
@@ -203,9 +204,13 @@ static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc)if (!timeline_fence_signaled (&pt->base))break ; +dma_fence_get (&pt->base); +list_del_init (&pt->link);rb_erase (&pt->node, &obj->pt_tree); +list_add_tail (&pt->link, &signalled); +/* * A signal callback may release the last reference to this * fence, causing it to be freed. That operation has to be
@@ -218,6 +223,11 @@ static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) }spin_unlock_irq (&obj->lock); + +list_for_each_entry_safe (pt, next, &signalled, link) { +list_del (&pt->link); +dma_fence_put (&pt->base); + } }/**
Signature · 1 line
2.41.0
Christian König
maintainerAm 17.08.23 um 23:37 schrieb Rob Clark:
Quoted 32 lines
From: Rob Clark <[email protected]>
If a signal callback releases the
sw_syncfence, that will trigger a deadlock as thetimeline_fence_releaserecurses onto thefence->lock(used both for signaling and the the timeline tree).To avoid that, temporarily hold an extra reference to the signalled fences until after we drop the lock.
(This is an alternative implementation of https://patchwork.kernel.org/patch/11664717/
which avoids some potential UAF issues with the original patch.)Reported-by Bas Nieuwenhuizen <[email protected]>Fixes d3c6dd1fb30d ("dma-buf/sw_sync: Synchronize signal vs syncpt free")Signed-off-by Rob Clark <[email protected]>--- drivers/dma-buf/sw_sync.c | 10 ++++++++++ 1 file changed, 10 insertions(+)drivers/dma-buf/sw_sync.c +1
-0@@ -191,6 +191,7 @@ static const struct dma_fence_ops timeline_fence_ops = {*/ static void sync_timeline_signal (struct sync_timeline *obj,unsigned int inc) { +LIST_HEAD (signalled);struct sync_pt *pt, *next;trace_sync_timeline (obj);
drivers/dma-buf/sw_sync.c static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) · line 207
Quoted 4 lines
drivers/dma-buf/sw_sync.c +1
-0@@ -203,9 +204,13 @@ static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc)if (!timeline_fence_signaled (&pt->base))break ; +dma_fence_get (&pt->base);
Question is why don't have the fences a reference on the list in the first place?
drivers/dma-buf/sw_sync.c static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) · line 212
Quoted 5 lines
drivers/dma-buf/sw_sync.c +2
-0+list_del_init (&pt->link);rb_erase (&pt->node, &obj->pt_tree); +list_add_tail (&pt->link, &signalled);
Instead of list_del()/list_add_tail() you could also use list_move_tail() here.
drivers/dma-buf/sw_sync.c static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) · line 228
Quoted 11 lines
drivers/dma-buf/sw_sync.c +4
-0+/* * A signal callback may release the last reference to this * fence, causing it to be freed. That operation has to be @@ -218,6 +223,11 @@ static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) }spin_unlock_irq (&obj->lock); + +list_for_each_entry_safe (pt, next, &signalled, link) { +list_del (&pt->link);
You must use list_del_init() here or otherwise the pt->link will keep pointing to the prev/next entries and the list_empty() check in timeline_fence_release() will fail and potentially corrupt things.
Regards,
Christian.
Quoted 5 lines
drivers/dma-buf/sw_sync.c +2
-0+dma_fence_put (&pt->base); + } }/**
Quoted 37 lines · Christian König
Am 17.08.23 um 23:37 schrieb Rob Clark:
Quoted 35 lines
From: Rob Clark <[email protected]>
If a signal callback releases the
sw_syncfence, that will trigger a deadlock as thetimeline_fence_releaserecurses onto thefence->lock(used both for signaling and the the timeline tree).To avoid that, temporarily hold an extra reference to the signalled fences until after we drop the lock.
(This is an alternative implementation of https://patchwork.kernel.org/patch/11664717/
which avoids some potential UAF issues with the original patch.)Reported-by Bas Nieuwenhuizen <[email protected]>Fixes d3c6dd1fb30d ("dma-buf/sw_sync: Synchronize signal vs syncpt free")Signed-off-by Rob Clark <[email protected]>--- drivers/dma-buf/sw_sync.c | 10 ++++++++++ 1 file changed, 10 insertions(+)drivers/dma-buf/sw_sync.c +1
-0@@ -191,6 +191,7 @@ static const struct dma_fence_ops timeline_fence_ops = {*/ static void sync_timeline_signal (struct sync_timeline *obj,unsigned int inc) { +LIST_HEAD (signalled);struct sync_pt *pt, *next;trace_sync_timeline (obj);@@ -203,9 +204,13 @@ static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc)if (!timeline_fence_signaled (&pt->base))break ;
Quoted 4 lines
drivers/dma-buf/sw_sync.c static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) · line 207
Quoted 1 line
drivers/dma-buf/sw_sync.c +1
-0+dma_fence_get (&pt->base);Question is why don't have the fences a reference on the list in the first place?
As best I can tell, it is because the fences hold a reference to the timeline, so a reference in the other direction would cause a loop.
BR,
-R
Quoted 34 lines
drivers/dma-buf/sw_sync.c static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) · line 212
Quoted 5 lines
drivers/dma-buf/sw_sync.c +2
-0+list_del_init (&pt->link);rb_erase (&pt->node, &obj->pt_tree); +list_add_tail (&pt->link, &signalled);Instead of list_del()/list_add_tail() you could also use
list_move_tail()here.drivers/dma-buf/sw_sync.c static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) · line 228
Quoted 11 lines
drivers/dma-buf/sw_sync.c +4
-0+/* * A signal callback may release the last reference to this * fence, causing it to be freed. That operation has to be @@ -218,6 +223,11 @@ static void sync_timeline_signal(struct sync_timeline *obj, unsigned int inc) }spin_unlock_irq (&obj->lock); + +list_for_each_entry_safe (pt, next, &signalled, link) { +list_del (&pt->link);You must use
list_del_init()here or otherwise thept->linkwill keep pointing to the prev/next entries and thelist_empty()check intimeline_fence_release()will fail and potentially corrupt things.Regards,
Christian.Quoted 5 lines
drivers/dma-buf/sw_sync.c +2
-0+dma_fence_put (&pt->base); + } }/**