Skip to contentSkip to conversation

ring-buffer: Fix sub-buffer order updates and free_page leaks

PATCH In review Masami Hiramatsu (Google) 4 participants 7 messages Aug 13 – Aug 14

Subsystems (2)
lore
Share & tools
lore

Masami Hiramatsu (Google)

author maintainer
#0 lore raw

cover ring-buffer: Fix sub-buffer order updates and free_page leaks

Hi Vincent,

Here is my patch to fix the race and free_page problem. Let's see what Sashiko says for this.

Thank you,

---
base-commit: 3d6d817622b0a9721e3cc404df3469171582be13

Masami Hiramatsu (Google) (1):
ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

 kernel/trace/ring_buffer.c |   35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)

--
Masami Hiramatsu (Google) <[email protected]>

author maintainer
#1 lore raw

PATCH 1/1 ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

From: Masami Hiramatsu (Google) <[email protected]>

When ring_buffer_subbuf_order_set() updates buffer->subbuf_order, it previously modified buffer->subbuf_order before clearing the cached per-CPU free_page entries. Furthermore, clearing cpu_buffer->free_page was done under cpu_buffer->reader_lock, whereas ring_buffer_alloc_read_page() protects cpu_buffer->free_page using arch_spin_lock(&cpu_buffer->lock).

Because ring_buffer_alloc_read_page(), ring_buffer_free_read_page(), and ring_buffer_read_page() checked buffer->subbuf_order locklessly before accessing reader resources, a TOCTOU race allowed a concurrent reader to obtain, cache, or swap a page allocated under an outdated order while tagging bpage->order with the new order. This allowed undersized pages to be swapped into the ring buffer, leading to heap buffer overflows, or caused free_pages() to be called with an invalid order.

Fix this by:
1. Flushing and freeing all per-CPU cached free_page entries under arch_spin_lock(&cpu_buffer->lock) using old_order before modifying buffer->subbuf_order.
2. Protecting bpage->order assignment under arch_spin_lock(&cpu_buffer->lock) in ring_buffer_alloc_read_page().
3. Moving the buffer->subbuf_order validation inside arch_spin_lock(&cpu_buffer->lock) in ring_buffer_free_read_page().
4. Re-validating buffer->subbuf_order inside reader_lock in ring_buffer_read_page().
5. Protecting cpu_buffer->free_page extraction with arch_spin_lock(&cpu_buffer->lock) in ring_buffer_subbuf_order_set().

Fixes 2808e31ec12e ("ring-buffer: Add interface for configuring trace sub buffer size")
Assisted-by Antigravity:gemini-3.6-flash
---
 kernel/trace/ring_buffer.c |   35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)
kernel/trace/ring_buffer.c +29 -6
index 2667992f0aa2..6d180689ad59 100644
@@ -6957,11 +6957,11 @@ 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;
@@ -7010,13 +7010,13 @@ 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 (data_page->order == buffer->subbuf_order && !cpu_buffer->free_page) {
 		cpu_buffer->free_page = dpage;
 		dpage = NULL;
 	}
@@ -7094,15 +7094,15 @@ 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;
@@ -7350,6 +7350,27 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 	/* Make sure all commits have finished */
 	synchronize_rcu();
 
+	/* Flush any cached free_page allocated with old_order */
+	for_each_buffer_cpu(buffer, cpu) {
+		struct buffer_data_page *old_free;
+		unsigned long flags;
+
+		if (!cpumask_test_cpu(cpu, buffer->cpumask))
+			continue;
+
+		cpu_buffer = buffer->buffers[cpu];
+
+		local_irq_save(flags);
+		arch_spin_lock(&cpu_buffer->lock);
+		old_free = cpu_buffer->free_page;
+		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
+		local_irq_restore(flags);
+
+		if (old_free)
+			free_pages((unsigned long)old_free, old_order);
+	}
+
 	buffer->subbuf_order = order;
 	buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE;
 
@@ -7431,8 +7452,10 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 		cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update;
 		cpu_buffer->nr_pages_to_update = 0;
 
+		arch_spin_lock(&cpu_buffer->lock);
 		old_free_data_page = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
 
 		rb_head_page_activate(cpu_buffer);
 
#2 lore raw

Re: [PATCH] ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Structural TOCTOU race: global order and size are updated before replacing per-CPU buffer pages, leading to Heap Buffer Overflow.
- [High] Incomplete TOCTOU fix: alloc_cpu_data() evaluates buffer->subbuf_order locklessly, causing bpage->order mismatch.
- [High] Race window between pre-flush loop and buffer->subbuf_order update allows caching undersized pages.
- [High] Post-update flush loop unconditionally frees newly cached pages using old_order.
- [High] Error path fails to flush cached pages when rolling back to old_order.
--

--- Patch [1]: [PATCH] ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers --- commit d1c3f71c852aab912bda60e5894b681e43da7c7b
Author: Masami Hiramatsu (Google) <[email protected]>

ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

This commit attempts to fix a TOCTOU race condition in the ring buffer code by ensuring that the per-CPU free_page cache is flushed and protected by the cpu_buffer lock during order changes. It also updates reader functions to validate the page order under the same lock to prevent using mismatched page sizes.

Quoted 17 lines
kernel/trace/ring_buffer.c +1 -1
index c2c57fbf6f448..9a11c663cff5b 100644
@@ -6957,11 +6957,11 @@ 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;
Quoted 1 line

}

[Severity: High]
This isn't a bug introduced by this patch, but does this locklessly evaluate buffer->subbuf_order after dropping the lock?

Looking at the code just below this hunk in ring_buffer_alloc_read_page():

	arch_spin_unlock(&cpu_buffer->lock);
	local_irq_restore(flags);

	if (bpage->data) {
		rb_init_data_page(bpage->data);
	} else {
		bpage->data = alloc_cpu_data(cpu, cpu_buffer->buffer->subbuf_order);
		...

If a concurrent writer changes the global order after the lock is dropped but before alloc_cpu_data() evaluates the argument, the allocated page size will mismatch the stored bpage->order.

When ring_buffer_free_read_page() later frees this page using bpage->order, it could free a mismatched page size and corrupt the buddy allocator.

[ ... ]

Quoted 17 lines
kernel/trace/ring_buffer.c +13 -0
@@ -7350,6 +7350,27 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 	/* Make sure all commits have finished */
 	synchronize_rcu();
 
+	/* Flush any cached free_page allocated with old_order */
+	for_each_buffer_cpu(buffer, cpu) {
+		struct buffer_data_page *old_free;
+		unsigned long flags;
+
+		if (!cpumask_test_cpu(cpu, buffer->cpumask))
+			continue;
+
+		cpu_buffer = buffer->buffers[cpu];
+
+		local_irq_save(flags);
+		arch_spin_lock(&cpu_buffer->lock);
+		old_free = cpu_buffer->free_page;

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7375

Quoted 10 lines
kernel/trace/ring_buffer.c +8 -0
+		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
+		local_irq_restore(flags);
+
+		if (old_free)
+			free_pages((unsigned long)old_free, old_order);
+	}
+
 	buffer->subbuf_order = order;
 	buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE;

[Severity: High]
This isn't a bug introduced by this patch, but does updating the global order after the flush loop leave a race window?

During the unprotected window between the flush loop completing and updating buffer->subbuf_order, the global order is still old_order.

A concurrent reader could call ring_buffer_free_read_page(), see that data_page->order matches buffer->subbuf_order (both old), and cache an old_order page back into cpu_buffer->free_page.

Once the order is updated below, subsequent readers might extract this cached old_order page, tag it with the new order, and eventually free it with the new order, causing buddy allocator corruption.

[Severity: Critical]
This isn't a bug introduced by this patch, but does updating the global order and size here, before replacing per-CPU buffer pages, create a structural race?

If a concurrent reader calls ring_buffer_read_page(), it validates its new-order data_page against the new global order and swaps it into the ring buffer. It extracts the old, unreplaced reader->page into dpage.

If the buffer is expanding, dpage is an undersized old-order page.

However, the reader executes memset() using the prematurely updated global buffer->subbuf_size:

ring_buffer_read_page():

	if (size < buffer->subbuf_size)
		memset(&dpage->data[size], 0, buffer->subbuf_size - size);

This would blindly write zeroes far beyond the bounds of the extracted old-order dpage, resulting in a heap buffer overflow reachable via unprivileged tracefs reads.

[ ... ]

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7460

Quoted 10 lines
kernel/trace/ring_buffer.c +2 -0
@@ -7431,8 +7452,10 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 		cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update;
 		cpu_buffer->nr_pages_to_update = 0;
 
+		arch_spin_lock(&cpu_buffer->lock);
 		old_free_data_page = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
 
 		rb_head_page_activate(cpu_buffer);

[Severity: High]
This isn't a bug introduced by this patch, but does this second flush loop unconditionally free potentially new-order pages with old_order?

After the global order is set to new_order, this function performs a sleepable allocation loop (__rb_allocate_pages). During this time, concurrent readers can allocate and free new_order pages, successfully caching them into cpu_buffer->free_page.

When this code extracts cpu_buffer->free_page and subsequently frees it down below:

	free_pages((unsigned long)old_free_data_page, old_order);

It blindly forces what could be a new_order page to be freed with old_order, corrupting the buddy allocator.

[Severity: High]
This isn't a bug introduced by this patch, but does the error path fail to flush cached pages when rolling back?

If __rb_allocate_pages() fails, the code jumps to the error label and restores buffer->subbuf_order = old_order:

ring_buffer_subbuf_order_set():
error:

	buffer->subbuf_order = old_order;
	buffer->subbuf_size = old_size;

During the window where it was new_order, concurrent readers may have cached new_order pages into cpu_buffer->free_page. The error path returns without flushing cpu_buffer->free_page.

Subsequent readers would extract these new_order pages, tag them as old_order, and free them with old_order, causing a memory leak (if expanding) or allocator corruption (if shrinking).

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/178663777320.475864.4716637934003507750.stgit@devnote2?part=1
#6 lore raw
In reply to Masami Hiramatsu (Google)
Quoted 14 lines · Masami Hiramatsu (Google)

Hi Vincent,

Here is my patch to fix the race and free_page problem. Let's see what Sashiko says for this.

Thank you,

---
base-commit: 3d6d817622b0a9721e3cc404df3469171582be13

Masami Hiramatsu (Google) (1):
ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

Quoted 5 lines
 kernel/trace/ring_buffer.c |   35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)

--
Masami Hiramatsu (Google) <[email protected]>

Sorry I am not sure to understand what is wrong with that series I made here?

https://lore.kernel.org/all/[email protected]/

Signature · 1 line
Vincent
maintainer
#3 lore raw

Re: [PATCH] ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

In reply to Masami Hiramatsu (Google)
Quoted 30 lines · On Fri, 14 Aug 2026 01:16:13 +0900 "Masami Hiramatsu (Google)"…

From: Masami Hiramatsu (Google) <[email protected]>

When ring_buffer_subbuf_order_set() updates buffer->subbuf_order, it previously modified buffer->subbuf_order before clearing the cached per-CPU free_page entries. Furthermore, clearing cpu_buffer->free_page was done under cpu_buffer->reader_lock, whereas ring_buffer_alloc_read_page() protects cpu_buffer->free_page using arch_spin_lock(&cpu_buffer->lock).

Because ring_buffer_alloc_read_page(), ring_buffer_free_read_page(), and ring_buffer_read_page() checked buffer->subbuf_order locklessly before accessing reader resources, a TOCTOU race allowed a concurrent reader to obtain, cache, or swap a page allocated under an outdated order while tagging bpage->order with the new order. This allowed undersized pages to be swapped into the ring buffer, leading to heap buffer overflows, or caused free_pages() to be called with an invalid order.

Fix this by:
1. Flushing and freeing all per-CPU cached free_page entries under arch_spin_lock(&cpu_buffer->lock) using old_order before modifying buffer->subbuf_order.
2. Protecting bpage->order assignment under arch_spin_lock(&cpu_buffer->lock) in ring_buffer_alloc_read_page().
3. Moving the buffer->subbuf_order validation inside arch_spin_lock(&cpu_buffer->lock) in ring_buffer_free_read_page().
4. Re-validating buffer->subbuf_order inside reader_lock in ring_buffer_read_page().
5. Protecting cpu_buffer->free_page extraction with arch_spin_lock(&cpu_buffer->lock) in ring_buffer_subbuf_order_set().

Quoted 4 lines
Fixes 2808e31ec12e ("ring-buffer: Add interface for configuring trace sub buffer size")
Assisted-by Antigravity:gemini-3.6-flash

---

Please rebase on top of ring-buffer/for-next, as I added Vincent's patches to that.

Quoted 8 lines
 kernel/trace/ring_buffer.c |   35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)
kernel/trace/ring_buffer.c +0 -0
index 2667992f0aa2..6d180689ad59 100644

kernel/trace/ring_buffer.c ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) · line 6964

Quoted 9 lines
kernel/trace/ring_buffer.c +1 -1
@@ -6957,11 +6957,11 @@ 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;

This is still needed.

Quoted 6 lines
kernel/trace/ring_buffer.c +0 -0
 	if (cpu_buffer->free_page) {
 		bpage->data = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
@@ -7010,13 +7010,13 @@ 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

kernel/trace/ring_buffer.c void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu, · line 7019

Quoted 10 lines
kernel/trace/ring_buffer.c +2 -2
 	 */
-	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 (data_page->order == buffer->subbuf_order && !cpu_buffer->free_page) {

Swap the order please. It has to check both to continue and if one fails it will not continue. Checking for cpu_buffer->free_page to be NULL first is the quicker check. And also the more likely one to fail.

Quoted 36 lines
kernel/trace/ring_buffer.c +13 -3
 		cpu_buffer->free_page = dpage;
 		dpage = NULL;
 	}
@@ -7094,15 +7094,15 @@ 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;
@@ -7350,6 +7350,27 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 	/* Make sure all commits have finished */
 	synchronize_rcu();
 
+	/* Flush any cached free_page allocated with old_order */
+	for_each_buffer_cpu(buffer, cpu) {
+		struct buffer_data_page *old_free;
+		unsigned long flags;
+
+		if (!cpumask_test_cpu(cpu, buffer->cpumask))
+			continue;
+
+		cpu_buffer = buffer->buffers[cpu];
+

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7372

Quoted 10 lines
kernel/trace/ring_buffer.c +10 -0
+		local_irq_save(flags);
+		arch_spin_lock(&cpu_buffer->lock);
+		old_free = cpu_buffer->free_page;
+		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
+		local_irq_restore(flags);
+
+		if (old_free)
+			free_pages((unsigned long)old_free, old_order);
+	}

Honestly, this should be a separate patch. The first part of this patch is data races with adding and freeing, but this is about changes to the size.

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7458

Quoted 12 lines
kernel/trace/ring_buffer.c +3 -0
+
 	buffer->subbuf_order = order;
 	buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE;
 
@@ -7431,8 +7452,10 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 		cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update;
 		cpu_buffer->nr_pages_to_update = 0;
 
+		arch_spin_lock(&cpu_buffer->lock);
 		old_free_data_page = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);

This is already fixed by Vincent (and I would have asked this to be a separate patch too if it hadn't).

-- Steve

Quoted 3 lines
kernel/trace/ring_buffer.c +0 -0
 		rb_head_page_activate(cpu_buffer);
 
#4 lore raw

Re: [PATCH] ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

Quoted 52 lines · Steven Rostedt
Quoted 30 lines · On Fri, 14 Aug 2026 01:16:13 +0900 "Masami Hiramatsu (Google)"…

From: Masami Hiramatsu (Google) <[email protected]>

When ring_buffer_subbuf_order_set() updates buffer->subbuf_order, it previously modified buffer->subbuf_order before clearing the cached per-CPU free_page entries. Furthermore, clearing cpu_buffer->free_page was done under cpu_buffer->reader_lock, whereas ring_buffer_alloc_read_page() protects cpu_buffer->free_page using arch_spin_lock(&cpu_buffer->lock).

Because ring_buffer_alloc_read_page(), ring_buffer_free_read_page(), and ring_buffer_read_page() checked buffer->subbuf_order locklessly before accessing reader resources, a TOCTOU race allowed a concurrent reader to obtain, cache, or swap a page allocated under an outdated order while tagging bpage->order with the new order. This allowed undersized pages to be swapped into the ring buffer, leading to heap buffer overflows, or caused free_pages() to be called with an invalid order.

Fix this by:
1. Flushing and freeing all per-CPU cached free_page entries under arch_spin_lock(&cpu_buffer->lock) using old_order before modifying buffer->subbuf_order.
2. Protecting bpage->order assignment under arch_spin_lock(&cpu_buffer->lock) in ring_buffer_alloc_read_page().
3. Moving the buffer->subbuf_order validation inside arch_spin_lock(&cpu_buffer->lock) in ring_buffer_free_read_page().
4. Re-validating buffer->subbuf_order inside reader_lock in ring_buffer_read_page().
5. Protecting cpu_buffer->free_page extraction with arch_spin_lock(&cpu_buffer->lock) in ring_buffer_subbuf_order_set().

Quoted 4 lines
Fixes 2808e31ec12e ("ring-buffer: Add interface for configuring trace sub buffer size")
Assisted-by Antigravity:gemini-3.6-flash

---

Please rebase on top of ring-buffer/for-next, as I added Vincent's patches to that.

Quoted 11 lines
 kernel/trace/ring_buffer.c |   35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)
kernel/trace/ring_buffer.c +0 -0
index 2667992f0aa2..6d180689ad59 100644
@@ -6957,11 +6957,11 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
 	if (!bpage)
 		return ERR_PTR(-ENOMEM);
 
Quoted 8 lines

kernel/trace/ring_buffer.c ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) · line 6964

Quoted 6 lines
kernel/trace/ring_buffer.c +1 -1
-	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;

This is still needed.

I am not sure, the lock is per-cpu_buffer but subbuf_order. is global to trace_buffer?

Quoted 36 lines
Quoted 7 lines
kernel/trace/ring_buffer.c +0 -0
 	if (cpu_buffer->free_page) {
 		bpage->data = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
@@ -7010,13 +7010,13 @@ 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

kernel/trace/ring_buffer.c void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu, · line 7019

Quoted 10 lines
kernel/trace/ring_buffer.c +2 -2
 	 */
-	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 (data_page->order == buffer->subbuf_order && !cpu_buffer->free_page) {

Swap the order please. It has to check both to continue and if one fails it will not continue. Checking for cpu_buffer->free_page to be NULL first is the quicker check. And also the more likely one to fail.

Quoted 14 lines
kernel/trace/ring_buffer.c +0 -3
 		cpu_buffer->free_page = dpage;
 		dpage = NULL;
 	}
@@ -7094,15 +7094,15 @@ 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;
 
Quoted 5 lines

kernel/trace/ring_buffer.c int ring_buffer_read_page(struct trace_buffer *buffer, · line 7105

Quoted 5 lines
kernel/trace/ring_buffer.c +3 -0
 	guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
 
+	if (data_page->order != buffer->subbuf_order)
+		return -1;
+

I have modified this as part of

tracing: Fix subbuf resize races with trace_pipe_raw readers

Quoted 53 lines
Quoted 17 lines
kernel/trace/ring_buffer.c +10 -0
 	reader = rb_get_reader_page(cpu_buffer);
 	if (!reader)
 		return -1;
@@ -7350,6 +7350,27 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 	/* Make sure all commits have finished */
 	synchronize_rcu();
 
+	/* Flush any cached free_page allocated with old_order */
+	for_each_buffer_cpu(buffer, cpu) {
+		struct buffer_data_page *old_free;
+		unsigned long flags;
+
+		if (!cpumask_test_cpu(cpu, buffer->cpumask))
+			continue;
+
+		cpu_buffer = buffer->buffers[cpu];
+

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7372

Quoted 10 lines
kernel/trace/ring_buffer.c +10 -0
+		local_irq_save(flags);
+		arch_spin_lock(&cpu_buffer->lock);
+		old_free = cpu_buffer->free_page;
+		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
+		local_irq_restore(flags);
+
+		if (old_free)
+			free_pages((unsigned long)old_free, old_order);
+	}

Honestly, this should be a separate patch. The first part of this patch is data races with adding and freeing, but this is about changes to the size.

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7458

Quoted 12 lines
kernel/trace/ring_buffer.c +3 -0
+
 	buffer->subbuf_order = order;
 	buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE;
 
@@ -7431,8 +7452,10 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 		cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update;
 		cpu_buffer->nr_pages_to_update = 0;
 
+		arch_spin_lock(&cpu_buffer->lock);
 		old_free_data_page = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);

This is already fixed by Vincent (and I would have asked this to be a separate patch too if it hadn't).

-- Steve

Quoted 3 lines
kernel/trace/ring_buffer.c +0 -0
 		rb_head_page_activate(cpu_buffer);
 
Signature · 1 line
Vincent
author maintainer
#5 lore raw

Re: [PATCH] ring-buffer: Fix race between ring_buffer_subbuf_order_set() and readers

Quoted 58 lines · On Fri, 14 Aug 2026 15:53:43 +0100 Vincent Donnefort <[email protected]>…
Quoted 57 lines · Steven Rostedt
Quoted 30 lines · On Fri, 14 Aug 2026 01:16:13 +0900 "Masami Hiramatsu (Google)"…

From: Masami Hiramatsu (Google) <[email protected]>

When ring_buffer_subbuf_order_set() updates buffer->subbuf_order, it previously modified buffer->subbuf_order before clearing the cached per-CPU free_page entries. Furthermore, clearing cpu_buffer->free_page was done under cpu_buffer->reader_lock, whereas ring_buffer_alloc_read_page() protects cpu_buffer->free_page using arch_spin_lock(&cpu_buffer->lock).

Because ring_buffer_alloc_read_page(), ring_buffer_free_read_page(), and ring_buffer_read_page() checked buffer->subbuf_order locklessly before accessing reader resources, a TOCTOU race allowed a concurrent reader to obtain, cache, or swap a page allocated under an outdated order while tagging bpage->order with the new order. This allowed undersized pages to be swapped into the ring buffer, leading to heap buffer overflows, or caused free_pages() to be called with an invalid order.

Fix this by:
1. Flushing and freeing all per-CPU cached free_page entries under arch_spin_lock(&cpu_buffer->lock) using old_order before modifying buffer->subbuf_order.
2. Protecting bpage->order assignment under arch_spin_lock(&cpu_buffer->lock) in ring_buffer_alloc_read_page().
3. Moving the buffer->subbuf_order validation inside arch_spin_lock(&cpu_buffer->lock) in ring_buffer_free_read_page().
4. Re-validating buffer->subbuf_order inside reader_lock in ring_buffer_read_page().
5. Protecting cpu_buffer->free_page extraction with arch_spin_lock(&cpu_buffer->lock) in ring_buffer_subbuf_order_set().

Quoted 4 lines
Fixes 2808e31ec12e ("ring-buffer: Add interface for configuring trace sub buffer size")
Assisted-by Antigravity:gemini-3.6-flash

---

Please rebase on top of ring-buffer/for-next, as I added Vincent's patches to that.

Quoted 16 lines
 kernel/trace/ring_buffer.c |   35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)
kernel/trace/ring_buffer.c +0 -1
index 2667992f0aa2..6d180689ad59 100644
@@ -6957,11 +6957,11 @@ 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);
 
Quoted 6 lines
Quoted 3 lines

kernel/trace/ring_buffer.c ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) · line 6964

Quoted 1 line
kernel/trace/ring_buffer.c +1 -0
+	bpage->order = buffer->subbuf_order;

This is still needed.

I am not sure, the lock is per-cpu_buffer but subbuf_order. is global to trace_buffer?

Yeah, we don't need this.
I got new ring_buffer_read_page() ensures the data_page->order == cpu_buffer->reader_page->order. This is a bit tricky but it allows us to asynchronously allocate reader_page in different order outside of locking. (and if free_page exists, it is copied with its order, so it is safe.)

Quoted 39 lines
Quoted 39 lines
Quoted 8 lines
kernel/trace/ring_buffer.c +0 -0
 	if (cpu_buffer->free_page) {
 		bpage->data = cpu_buffer->free_page;
 		cpu_buffer->free_page = NULL;
@@ -7010,13 +7010,13 @@ 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

kernel/trace/ring_buffer.c void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu, · line 7019

Quoted 10 lines
kernel/trace/ring_buffer.c +2 -2
 	 */
-	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 (data_page->order == buffer->subbuf_order && !cpu_buffer->free_page) {

Swap the order please. It has to check both to continue and if one fails it will not continue. Checking for cpu_buffer->free_page to be NULL first is the quicker check. And also the more likely one to fail.

Quoted 16 lines
kernel/trace/ring_buffer.c +0 -3
 		cpu_buffer->free_page = dpage;
 		dpage = NULL;
 	}
@@ -7094,15 +7094,15 @@ 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);
 
Quoted 7 lines
Quoted 3 lines

kernel/trace/ring_buffer.c int ring_buffer_read_page(struct trace_buffer *buffer, · line 7105

Quoted 3 lines
kernel/trace/ring_buffer.c +3 -0
+	if (data_page->order != buffer->subbuf_order)
+		return -1;
+

I have modified this as part of

tracing: Fix subbuf resize races with trace_pipe_raw readers

Got it. Yours changes the data structure. Mine is for older kernels as a minimal change.

Quoted 23 lines
Quoted 23 lines
Quoted 23 lines
kernel/trace/ring_buffer.c +15 -0
 	reader = rb_get_reader_page(cpu_buffer);
 	if (!reader)
 		return -1;
@@ -7350,6 +7350,27 @@ int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
 	/* Make sure all commits have finished */
 	synchronize_rcu();
 
+	/* Flush any cached free_page allocated with old_order */
+	for_each_buffer_cpu(buffer, cpu) {
+		struct buffer_data_page *old_free;
+		unsigned long flags;
+
+		if (!cpumask_test_cpu(cpu, buffer->cpumask))
+			continue;
+
+		cpu_buffer = buffer->buffers[cpu];
+
+		local_irq_save(flags);
+		arch_spin_lock(&cpu_buffer->lock);
+		old_free = cpu_buffer->free_page;
+		cpu_buffer->free_page = NULL;
+		arch_spin_unlock(&cpu_buffer->lock);
Quoted 8 lines
Quoted 8 lines

kernel/trace/ring_buffer.c int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) · line 7372

Quoted 5 lines
kernel/trace/ring_buffer.c +5 -0
+		local_irq_restore(flags);
+
+		if (old_free)
+			free_pages((unsigned long)old_free, old_order);
+	}

Honestly, this should be a separate patch. The first part of this patch is data races with adding and freeing, but this is about changes to the size.

Since Vincent's patch[1] handles data page pointer with its order, we don't this part anymore.

[1] ring-buffer: Make cpu_buffer::free_page a buffer_data_read_page

So I confirmed Vincent's patches with Fixed tag are fixing the racing problem in the different way.

Thank you,

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>