Skip to contentSkip to conversation

ring-buffer: serialize read-page order with subbuffer resize

PATCH In review Yousef Alhouseen 2 participants 8 messages Jun 28 – Jul 18

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ring-buffer: serialize read-page order with subbuffer resize

ring_buffer_read_page() checks that its spare page has the current subbuffer order before taking cpu_buffer->reader_lock. A concurrent ring_buffer_subbuf_order_set() can change the order and replace the reader page after that check. The reader then copies a larger subbuffer into the old allocation, causing an out-of-bounds write.

Keep spare-page allocation and release under buffer->mutex, which already serializes order changes. Move the read-side order check under reader_lock, the lock used by resize when replacing per-CPU pages.

Fixes f9b94daa542a ("ring-buffer: Set new size of the ring buffer sub page")
Reported-by [email protected]
---
 kernel/trace/ring_buffer.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
kernel/trace/ring_buffer.c +6 -3
index 56a328e94395..eed5d7cffdee 100644
@@ -6950,6 +6950,8 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
 		return ERR_PTR(-ENODEV);
 
+	guard(mutex)(&buffer->mutex);
+
 	bpage = kzalloc_obj(*bpage);
 	if (!bpage)
 		return ERR_PTR(-ENOMEM);
@@ -7000,6 +7002,8 @@ void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu,
 	if (!buffer || !buffer->buffers || !buffer->buffers[cpu])
 		return;
 
+	guard(mutex)(&buffer->mutex);
+
 	cpu_buffer = buffer->buffers[cpu];
 
 	/*
@@ -7091,14 +7095,13 @@ int ring_buffer_read_page(struct trace_buffer *buffer,
 	if (!data_page || !data_page->data)
 		return -1;
 
-	if (data_page->order != buffer->subbuf_order)
-		return -1;
-
 	dpage = data_page->data;
 	if (!dpage)
 		return -1;
 
 	guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
+	if (data_page->order != buffer->subbuf_order)
+		return -1;
 
 	reader = rb_get_reader_page(cpu_buffer);
 	if (!reader)
Signature · 1 line
2.54.0
maintainer
#1 lore raw
Quoted 24 lines · On Sun, 28 Jun 2026 02:46:53 +0200 Yousef Alhouseen…

ring_buffer_read_page() checks that its spare page has the current subbuffer order before taking cpu_buffer->reader_lock. A concurrent ring_buffer_subbuf_order_set() can change the order and replace the reader page after that check. The reader then copies a larger subbuffer into the old allocation, causing an out-of-bounds write.

Keep spare-page allocation and release under buffer->mutex, which already serializes order changes. Move the read-side order check under reader_lock, the lock used by resize when replacing per-CPU pages.

Fixes f9b94daa542a ("ring-buffer: Set new size of the ring buffer sub page")
Reported-by [email protected]
---
 kernel/trace/ring_buffer.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
kernel/trace/ring_buffer.c +0 -0
index 56a328e94395..eed5d7cffdee 100644

kernel/trace/ring_buffer.c ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) · line 6955

Quoted 6 lines
kernel/trace/ring_buffer.c +2 -0
@@ -6950,6 +6950,8 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
 		return ERR_PTR(-ENODEV);
 
+	guard(mutex)(&buffer->mutex);
+
 	bpage = kzalloc_obj(*bpage);

First, do not grab locks around allocations unless the are really needed. This is bad practice, as it extends the critical section and may even add the allocation locking to the lock chain.

That said, just moving things around the current locks should work.

Like this (not compiled nor tested):

kernel/trace/ring_buffer.c +6 -6
index 56a328e94395..8352f935a223 100644
@@ -6954,11 +6954,11 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 	if (!bpage)
 		return ERR_PTR(-ENOMEM);
 
-	bpage->order = buffer->subbuf_order;
 	cpu_buffer = buffer->buffers[cpu];
 	local_irq_save(flags);
 	arch_spin_lock(&cpu_buffer->lock);
 
+	bpage->order = buffer->subbuf_order;
 	if (cpu_buffer->free_page) {
 		bpage->data = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
@@ -7007,13 +7007,13 @@ void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu,
 	 * is different from the subbuffer order of the buffer -
 	 * we can't reuse it
 	 */
-	if (page_ref_count(page) > 1 || data_page->order != buffer->subbuf_order)
+	if (page_ref_count(page) > 1)
 		goto out;
 
 	local_irq_save(flags);
 	arch_spin_lock(&cpu_buffer->lock);
 
-	if (!cpu_buffer->free_page) {
+	if (!cpu_buffer->free_page && data_page->order == buffer->subbuf_order)
 		cpu_buffer->free_page = dpage;
 		dpage = NULL;
 	}
@@ -7091,15 +7091,15 @@ int ring_buffer_read_page(struct trace_buffer *buffer,
 	if (!data_page || !data_page->data)
 		return -1;
 
-	if (data_page->order != buffer->subbuf_order)
-		return -1;
-
 	dpage = data_page->data;
 	if (!dpage)
 		return -1;
 
 	guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
 
+	if (data_page->order != buffer->subbuf_order)
+		return -1;
+
 	reader = rb_get_reader_page(cpu_buffer);
 	if (!reader)
 		return -1;

-- Steve

#2 lore raw

Thanks, Steve. Moving the order reads under the existing per-CPU locks is much tighter and avoids extending the allocation critical path. I’ll compile and test this direction and send v2.

Quoted 96 lines · Steven Rostedt
Quoted 24 lines · On Sun, 28 Jun 2026 02:46:53 +0200 Yousef Alhouseen…

ring_buffer_read_page() checks that its spare page has the current subbuffer order before taking cpu_buffer->reader_lock. A concurrent ring_buffer_subbuf_order_set() can change the order and replace the reader page after that check. The reader then copies a larger subbuffer into the old allocation, causing an out-of-bounds write.

Keep spare-page allocation and release under buffer->mutex, which already serializes order changes. Move the read-side order check under reader_lock, the lock used by resize when replacing per-CPU pages.

Fixes f9b94daa542a ("ring-buffer: Set new size of the ring buffer sub page")
Reported-by [email protected]

---
kernel/trace/ring_buffer.c | 9 ++++++---

1 file changed, 6 insertions(+), 3 deletions(-)
kernel/trace/ring_buffer.c +0 -0
index 56a328e94395..eed5d7cffdee 100644

kernel/trace/ring_buffer.c ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) · line 6955

Quoted 6 lines
kernel/trace/ring_buffer.c +2 -0
@@ -6950,6 +6950,8 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 if (!cpumask_test_cpu(cpu, buffer->cpumask))
 return ERR_PTR(-ENODEV);
 
+ guard(mutex)(&buffer->mutex);
+
 bpage = kzalloc_obj(*bpage);

First, do not grab locks around allocations unless the are really needed. This is bad practice, as it extends the critical section and may even add the allocation locking to the lock chain.

That said, just moving things around the current locks should work.

Like this (not compiled nor tested):

kernel/trace/ring_buffer.c +6 -7
index 56a328e94395..8352f935a223 100644
@@ -6954,11 +6954,11 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 if (!bpage)
 return ERR_PTR(-ENOMEM);
 
- bpage->order = buffer->subbuf_order;
 cpu_buffer = buffer->buffers[cpu];
 local_irq_save(flags);
 arch_spin_lock(&cpu_buffer->lock);
 
+ bpage->order = buffer->subbuf_order;
 if (cpu_buffer->free_page) {
 bpage->data = cpu_buffer->free_page;
 cpu_buffer->free_page = NULL;
@@ -7007,13 +7007,13 @@ void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu,
 * is different from the subbuffer order of the buffer -
 * we can't reuse it
 */
- if (page_ref_count(page) > 1 || data_page->order != buffer->subbuf_order)
+ if (page_ref_count(page) > 1)
 goto out;
 
 local_irq_save(flags);
 arch_spin_lock(&cpu_buffer->lock);
 
- if (!cpu_buffer->free_page) {
+ if (!cpu_buffer->free_page && data_page->order == buffer->subbuf_order)
 cpu_buffer->free_page = dpage;
 dpage = NULL;
 }
@@ -7091,15 +7091,15 @@ int ring_buffer_read_page(struct trace_buffer *buffer,
 if (!data_page || !data_page->data)
 return -1;
 
- if (data_page->order != buffer->subbuf_order)
- return -1;
-
 dpage = data_page->data;
 if (!dpage)
 return -1;
 
 guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
 
+ if (data_page->order != buffer->subbuf_order)
+ return -1;
+
 reader = rb_get_reader_page(cpu_buffer);
 if (!reader)
 return -1;
 
-- Steve
#3 lore raw
In reply to Steven Rostedt

One issue turned up while checking the suggested locking:
ring_buffer_subbuf_order_set() writes buffer->subbuf_order before taking reader_lock and never takes cpu_buffer->lock. An allocator can therefore take cpu_buffer->lock after the new order is published but before resize clears the old-order free_page, tag that old page with the new order, and return it.

I can keep the allocations outside buffer->mutex and hold the mutex only while snapshotting subbuf_order and taking or returning free_page. That removes allocation from the critical section and serializes the order/free-page pair with resize. Would you prefer that, or should free_page and the order transition be synchronized another way?

Quoted 96 lines · Steven Rostedt
Quoted 24 lines · On Sun, 28 Jun 2026 02:46:53 +0200 Yousef Alhouseen…

ring_buffer_read_page() checks that its spare page has the current subbuffer order before taking cpu_buffer->reader_lock. A concurrent ring_buffer_subbuf_order_set() can change the order and replace the reader page after that check. The reader then copies a larger subbuffer into the old allocation, causing an out-of-bounds write.

Keep spare-page allocation and release under buffer->mutex, which already serializes order changes. Move the read-side order check under reader_lock, the lock used by resize when replacing per-CPU pages.

Fixes f9b94daa542a ("ring-buffer: Set new size of the ring buffer sub page")
Reported-by [email protected]

---
kernel/trace/ring_buffer.c | 9 ++++++---

1 file changed, 6 insertions(+), 3 deletions(-)
kernel/trace/ring_buffer.c +0 -0
index 56a328e94395..eed5d7cffdee 100644

kernel/trace/ring_buffer.c ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) · line 6955

Quoted 6 lines
kernel/trace/ring_buffer.c +2 -0
@@ -6950,6 +6950,8 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 if (!cpumask_test_cpu(cpu, buffer->cpumask))
 return ERR_PTR(-ENODEV);
 
+ guard(mutex)(&buffer->mutex);
+
 bpage = kzalloc_obj(*bpage);

First, do not grab locks around allocations unless the are really needed. This is bad practice, as it extends the critical section and may even add the allocation locking to the lock chain.

That said, just moving things around the current locks should work.

Like this (not compiled nor tested):

kernel/trace/ring_buffer.c +6 -7
index 56a328e94395..8352f935a223 100644
@@ -6954,11 +6954,11 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 if (!bpage)
 return ERR_PTR(-ENOMEM);
 
- bpage->order = buffer->subbuf_order;
 cpu_buffer = buffer->buffers[cpu];
 local_irq_save(flags);
 arch_spin_lock(&cpu_buffer->lock);
 
+ bpage->order = buffer->subbuf_order;
 if (cpu_buffer->free_page) {
 bpage->data = cpu_buffer->free_page;
 cpu_buffer->free_page = NULL;
@@ -7007,13 +7007,13 @@ void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu,
 * is different from the subbuffer order of the buffer -
 * we can't reuse it
 */
- if (page_ref_count(page) > 1 || data_page->order != buffer->subbuf_order)
+ if (page_ref_count(page) > 1)
 goto out;
 
 local_irq_save(flags);
 arch_spin_lock(&cpu_buffer->lock);
 
- if (!cpu_buffer->free_page) {
+ if (!cpu_buffer->free_page && data_page->order == buffer->subbuf_order)
 cpu_buffer->free_page = dpage;
 dpage = NULL;
 }
@@ -7091,15 +7091,15 @@ int ring_buffer_read_page(struct trace_buffer *buffer,
 if (!data_page || !data_page->data)
 return -1;
 
- if (data_page->order != buffer->subbuf_order)
- return -1;
-
 dpage = data_page->data;
 if (!dpage)
 return -1;
 
 guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
 
+ if (data_page->order != buffer->subbuf_order)
+ return -1;
+
 reader = rb_get_reader_page(cpu_buffer);
 if (!reader)
 return -1;
 
-- Steve
maintainer
#4 lore raw
Quoted 7 lines · On Tue, 30 Jun 2026 13:45:05 -0700 Yousef Alhouseen…

One issue turned up while checking the suggested locking:
ring_buffer_subbuf_order_set() writes buffer->subbuf_order before taking reader_lock and never takes cpu_buffer->lock. An allocator can therefore take cpu_buffer->lock after the new order is published but before resize clears the old-order free_page, tag that old page with the new order, and return it.

Quoted 6 lines

I can keep the allocations outside buffer->mutex and hold the mutex only while snapshotting subbuf_order and taking or returning free_page. That removes allocation from the critical section and serializes the order/free-page pair with resize. Would you prefer that, or should free_page and the order transition be synchronized another way?

Nothing should be reading when the subbuf_order is being updated. Let's add a flag to state that it's being updated, and make all reads simply fail during that time.

-- Steve

#5 lore raw

Agreed. I’ll add an explicit resize-in-progress flag, set it around the order transition, and make the external read-page allocation/free/read paths reject work while it is set. I’ll check the flag under the locks that serialize each path so it cannot race the transition, then compile and test the resulting v2.

Quoted 22 lines · Steven Rostedt
Quoted 7 lines · On Tue, 30 Jun 2026 13:45:05 -0700 Yousef Alhouseen…

One issue turned up while checking the suggested locking:
ring_buffer_subbuf_order_set() writes buffer->subbuf_order before taking reader_lock and never takes cpu_buffer->lock. An allocator can therefore take cpu_buffer->lock after the new order is published but before resize clears the old-order free_page, tag that old page with the new order, and return it.

Quoted 6 lines

I can keep the allocations outside buffer->mutex and hold the mutex only while snapshotting subbuf_order and taking or returning free_page. That removes allocation from the critical section and serializes the order/free-page pair with resize. Would you prefer that, or should free_page and the order transition be synchronized another way?

Nothing should be reading when the subbuf_order is being updated. Let's add a flag to state that it's being updated, and make all reads simply fail during that time.

-- Steve

maintainer
#6 lore raw
Quoted 5 lines · On Tue, 30 Jun 2026 14:16:48 -0700 Yousef Alhouseen…

Agreed. I’ll add an explicit resize-in-progress flag, set it around the order transition, and make the external read-page allocation/free/read paths reject work while it is set. I’ll check the flag under the locks that serialize each path so it cannot race the transition, then compile and test the resulting v2.

Did you ever get around to sending a v2?

-- Steve

#7 lore raw

Not yet. The mutex-based draft isn't suitable; I still need to implement and test the resize-in-progress flag design you requested. I'll send v2 only after that is done.

Thanks,
Yousef

Quoted 12 lines · Steven Rostedt
Quoted 5 lines · On Tue, 30 Jun 2026 14:16:48 -0700 Yousef Alhouseen…

Agreed. I’ll add an explicit resize-in-progress flag, set it around the order transition, and make the external read-page allocation/free/read paths reject work while it is set. I’ll check the flag under the locks that serialize each path so it cannot race the transition, then compile and test the resulting v2.

Did you ever get around to sending a v2?

-- Steve