Skip to contentSkip to conversation

of: reserved_mem: fix stack overflow and make dynamic reserved region configurable

PATCH In review Wandun Chen 3 participants 10 messages Jun 4 – Jun 5 Device tree

lore
Share & tools
lore
#0 lore raw

cover of: reserved_mem: fix stack overflow and make dynamic reserved region configurable

From: Wandun Chen <[email protected]>

Patch 1 fixes a stack buffer overflow in fdt_scan_reserved_mem() when the DTS contains more than 64 reserved-memory regions.

Patch 2 makes the limit configurable via Kconfig (OF_RESERVED_MEM_DYNAMIC_REGIONS) so platforms with many such regions can modify it without patching the source.

Wandun Chen (2):
of: reserved_mem: avoid stack buffer overflow in
fdt_scan_reserved_mem()
of: reserved_mem: add config to extend dynamic reserved memory regions

 drivers/of/Kconfig           | 11 +++++++++++
 drivers/of/of_private.h      |  2 +-
 drivers/of/of_reserved_mem.c |  6 ++++++
 3 files changed, 18 insertions(+), 1 deletion(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/2 of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

From: Wandun Chen <[email protected]>

Sashiko found a potential stack buffer overflow in fdt_scan_reserved_mem() due to missing bounds checking on dynamic_nodes_cnt [1].

Fix this by adding bounds check.

---
 drivers/of/of_reserved_mem.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/of/of_reserved_mem.c +6 -0
index ce1d5530ec0f..27dc98aa9bf9 100644
@@ -346,6 +346,12 @@ int __init fdt_scan_reserved_mem(void)
 		err = __reserved_mem_reserve_reg(child, uname);
 		if (!err)
 			count++;
+
+		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+				    MAX_RESERVED_REGIONS);
+			continue;
+		}
 		/*
 		 * Save the nodes for the dynamically-placed regions
 		 * into an array which will be used for allocation right
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH 1/2] of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] The newly added bounds check for `dynamic_nodes_cnt` is placed outside the block that evaluates…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added bounds check for dynamic_nodes_cnt is placed outside the block that evaluates whether a node is actually a dynamically-placed region, leading to spurious error logs. --

commit 6a79f526de5e1185b7c79fb469545f4094fb402f
Author: Wandun Chen <[email protected]>

of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

This commit adds bounds checking on dynamic_nodes_cnt to prevent a potential stack buffer overflow in fdt_scan_reserved_mem().

Quoted 3 lines
drivers/of/of_reserved_mem.c +0 -0

[ ... ]

drivers/of/of_reserved_mem.c int __init fdt_scan_reserved_mem(void) · line 349

Quoted 10 lines
drivers/of/of_reserved_mem.c +6 -0
@@ -341,6 +341,12 @@ int __init fdt_scan_reserved_mem(void)
 		err = __reserved_mem_reserve_reg(child, uname);
 		if (!err)
 			count++;
+
+		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+				    MAX_RESERVED_REGIONS);
+			continue;
+		}

[Severity: Medium]
Will this bounds check trigger spurious error logs for statically-placed regions?

Because the check is placed outside the block that evaluates whether a node is actually a dynamically-placed region, it evaluates for all subsequent nodes once the limit is reached.

If the number of dynamically-placed regions exactly reaches MAX_RESERVED_REGIONS, and a statically-placed region is processed afterwards, won't it incorrectly trigger the warning claiming the dynamic nodes limit was exceeded?

Quoted 11 lines
drivers/of/of_reserved_mem.c +0 -0
 		/*
 		 * Save the nodes for the dynamically-placed regions
 		 * into an array which will be used for allocation right
drivers/of/of_reserved_mem.c +0 -0
 		 * after all the statically-placed regions are reserved
 		 * or marked as no-map. This is done to avoid dynamically
 		 * allocating from one of the statically-placed regions.
 		 */
 		if (err == -ENOENT && of_get_flat_dt_prop(child, "size", NULL)) {
 			dynamic_nodes[dynamic_nodes_cnt] = child;
 			dynamic_nodes_cnt++;
 		}
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
author
#3 lore raw

Re: [PATCH 1/2] of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

Quoted 19 lines · On 6/4/26 15:45, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added bounds check for dynamic_nodes_cnt is placed outside the block that evaluates whether a node is actually a dynamically-placed region, leading to spurious error logs. --

commit 6a79f526de5e1185b7c79fb469545f4094fb402f
Author: Wandun Chen <[email protected]>

of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

This commit adds bounds checking on dynamic_nodes_cnt to prevent a potential stack buffer overflow in fdt_scan_reserved_mem().

Quoted 3 lines
drivers/of/of_reserved_mem.c +0 -0

[ ... ]

Quoted 13 lines

drivers/of/of_reserved_mem.c int __init fdt_scan_reserved_mem(void) · line 349

Quoted 10 lines
drivers/of/of_reserved_mem.c +6 -0
@@ -341,6 +341,12 @@ int __init fdt_scan_reserved_mem(void)
  		err = __reserved_mem_reserve_reg(child, uname);
  		if (!err)
  			count++;
+
+		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+				    MAX_RESERVED_REGIONS);
+			continue;
+		}

[Severity: Medium]
Will this bounds check trigger spurious error logs for statically-placed regions?

Yes, it indeed, will fix in v2.

Quoted 21 lines

Because the check is placed outside the block that evaluates whether a node is actually a dynamically-placed region, it evaluates for all subsequent nodes once the limit is reached.

If the number of dynamically-placed regions exactly reaches MAX_RESERVED_REGIONS, and a statically-placed region is processed afterwards, won't it incorrectly trigger the warning claiming the dynamic nodes limit was exceeded?

Quoted 11 lines
drivers/of/of_reserved_mem.c +0 -0
  		/*
  		 * Save the nodes for the dynamically-placed regions
  		 * into an array which will be used for allocation right

* after all the statically-placed regions are reserved
* or marked as no-map. This is done to avoid dynamically
* allocating from one of the statically-placed regions. */
if (err == -ENOENT && of_get_flat_dt_prop(child, "size", NULL)) {
dynamic_nodes[dynamic_nodes_cnt] = child; dynamic_nodes_cnt++;
}

maintainer
#4 lore raw

Re: [PATCH 1/2] of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

Quoted 35 lines · Wandun
Quoted 19 lines · On 6/4/26 15:45, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added bounds check for dynamic_nodes_cnt is placed outside the block that evaluates whether a node is actually a dynamically-placed region, leading to spurious error logs. --

commit 6a79f526de5e1185b7c79fb469545f4094fb402f
Author: Wandun Chen <[email protected]>

of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

This commit adds bounds checking on dynamic_nodes_cnt to prevent a potential stack buffer overflow in fdt_scan_reserved_mem().

Quoted 3 lines
drivers/of/of_reserved_mem.c +0 -0

[ ... ]

Quoted 13 lines

drivers/of/of_reserved_mem.c int __init fdt_scan_reserved_mem(void) · line 349

Quoted 10 lines
drivers/of/of_reserved_mem.c +6 -0
@@ -341,6 +341,12 @@ int __init fdt_scan_reserved_mem(void)
  		err = __reserved_mem_reserve_reg(child, uname);
  		if (!err)
  			count++;
+
+		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+				    MAX_RESERVED_REGIONS);
+			continue;
+		}

[Severity: Medium]
Will this bounds check trigger spurious error logs for statically-placed regions?

Quoted 1 line

Yes, it indeed, will fix in v2.

I now have the same change (and same problems) in another submission:

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

See my comment there.

Rob

author
#5 lore raw

Re: [PATCH 1/2] of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

Quoted 35 lines · Rob Herring
Quoted 34 lines · Wandun
Quoted 32 lines · On 6/4/26 15:45, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added bounds check for dynamic_nodes_cnt is placed outside the block that evaluates whether a node is actually a dynamically-placed region, leading to spurious error logs. --

commit 6a79f526de5e1185b7c79fb469545f4094fb402f
Author: Wandun Chen <[email protected]>

of: reserved_mem: avoid stack buffer overflow in fdt_scan_reserved_mem()

This commit adds bounds checking on dynamic_nodes_cnt to prevent a potential stack buffer overflow in fdt_scan_reserved_mem().

Quoted 3 lines
drivers/of/of_reserved_mem.c +0 -0

[ ... ]

drivers/of/of_reserved_mem.c int __init fdt_scan_reserved_mem(void) · line 349

Quoted 10 lines
drivers/of/of_reserved_mem.c +6 -0
@@ -341,6 +341,12 @@ int __init fdt_scan_reserved_mem(void)
   		err = __reserved_mem_reserve_reg(child, uname);
   		if (!err)
   			count++;
+
+		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+				    MAX_RESERVED_REGIONS);
+			continue;
+		}

[Severity: Medium]
Will this bounds check trigger spurious error logs for statically-placed regions?

Quoted 6 lines
Quoted 1 line

Yes, it indeed, will fix in v2.

I now have the same change (and same problems) in another submission:

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

See my comment there.

Thanks for pointing out the other submission. Since the same fix is already being worked on, I'll drop this patch. Best regards, Wandun

Quoted 2 lines

Rob

#6 lore raw

PATCH 2/2 of: reserved_mem: add config to extend dynamic reserved memory regions

In reply to Wandun Chen

From: Wandun Chen <[email protected]>

Nowadays, the dynamic reserved memory regions is 64 by default, If the count of reserved memory regions defined in DTS bigger than 64, only 64 reserved memory can be handled properly.

So add a config to configure the actual dynamic reserved memory regions count instead of modify the code.

---
 drivers/of/Kconfig           | 11 +++++++++++
 drivers/of/of_private.h      |  2 +-
 drivers/of/of_reserved_mem.c |  2 +-
 3 files changed, 13 insertions(+), 2 deletions(-)
drivers/of/Kconfig +11 -0
index 50697cc3b07e..d6496ec3765c 100644
@@ -99,6 +99,17 @@ config OF_IRQ
 config OF_RESERVED_MEM
 	def_bool OF_EARLY_FLATTREE
 
+config OF_RESERVED_MEM_DYNAMIC_REGIONS
+	int "Maximum count of the dynamic reserved memory regions"
+	depends on OF_RESERVED_MEM
+	default 64
+	range 1 256
+	help
+	  Allows to define proper dynamic reserved memory regions number
+	  according to DTS configuration.
+
+	  If unsure, leave the default value "64".
+
 config OF_RESOLVE
 	bool
 
drivers/of/of_private.h +1 -1
index 0ae16da066e2..6ad00798f39d 100644
@@ -9,7 +9,7 @@
  */
 
 #define FDT_ALIGN_SIZE 8
-#define MAX_RESERVED_REGIONS    64
+#define MAX_RESERVED_REGIONS	CONFIG_OF_RESERVED_MEM_DYNAMIC_REGIONS
 
 /**
  * struct alias_prop - Alias property in 'aliases' node
drivers/of/of_reserved_mem.c +1 -1
index 27dc98aa9bf9..d1680fc4fb38 100644
@@ -348,7 +348,7 @@ int __init fdt_scan_reserved_mem(void)
 			count++;
 
 		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
-			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d), try expanding CONFIG_OF_RESERVED_MEM_DYNAMIC_REGIONS.\n",
 				    MAX_RESERVED_REGIONS);
 			continue;
 		}
Signature · 1 line
2.43.0
#7 lore raw

Re: [PATCH 2/2] of: reserved_mem: add config to extend dynamic reserved memory regions

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Out-of-bounds write on the static reserved_mem_array if dynamic allocation of the reserved memory array fails. --

commit 716aa35cecc2a52f29b8fd8fd9a0aae87756ea2e
Author: Wandun Chen <[email protected]>

of: reserved_mem: add config to extend dynamic reserved memory regions

Nowadays, the dynamic reserved memory regions is 64 by default, If the count of reserved memory regions defined in DTS bigger than 64, only 64 reserved memory can be handled properly.

So add a config to configure the actual dynamic reserved memory regions count instead of modify the code.

drivers/of/of_reserved_mem.c int __init fdt_scan_reserved_mem(void) · line 349

Quoted 13 lines
drivers/of/of_reserved_mem.c +1 -1
index 056c4179f549c..73d7d7865518e 100644
@@ -343,7 +343,7 @@ int __init fdt_scan_reserved_mem(void)
 			count++;
 
 		if (dynamic_nodes_cnt >= MAX_RESERVED_REGIONS) {
-			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d)\n",
+			pr_err_once("Reserved memory: reached MAX_RESERVED_REGIONS(%d), try expanding CONFIG_OF_RESERVED_MEM_DYNAMIC_REGIONS.\n",
 				    MAX_RESERVED_REGIONS);
 			continue;
 		}

[Severity: High]
This is a pre-existing issue, but there appears to be an out-of-bounds write on the static reserved_mem_array if dynamic allocation fails.

In fdt_scan_reserved_mem(), the total count of reserved memory regions (and thus total_reserved_mem_cnt) can exceed MAX_RESERVED_REGIONS because static regions are not capped by the dynamic_nodes_cnt >= MAX_RESERVED_REGIONS check.

Later, alloc_reserved_mem_array() allocates a dynamically sized array of total_reserved_mem_cnt elements. If memblock_alloc() fails, it returns early without resetting total_reserved_mem_cnt to MAX_RESERVED_REGIONS:

drivers/of/of_reserved_mem.c:alloc_reserved_mem_array() {
	new_array = memblock_alloc(alloc_size, SMP_CACHE_BYTES);
	if (!new_array) {
		pr_err("Failed to allocate memory for reserved_mem array with err: %d", -ENOMEM);
		return;
	}

...
}

When fdt_scan_reserved_mem_late() subsequently initializes the static regions, it calls fdt_init_reserved_mem_node() for each. The bounds check relies on the uncapped total_reserved_mem_cnt:

drivers/of/of_reserved_mem.c:fdt_init_reserved_mem_node() {
	struct reserved_mem *rmem = &reserved_mem[reserved_mem_count];

	if (reserved_mem_count == total_reserved_mem_cnt) {
		pr_err("not enough space for all defined regions.\n");
		return;
	}

...
}

Since total_reserved_mem_cnt wasn't reset on failure, the check allows reserved_mem_count to exceed MAX_RESERVED_REGIONS. Does this result in out-of-bounds writes to the static reserved_mem_array?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
maintainer
#8 lore raw

Re: [PATCH 2/2] of: reserved_mem: add config to extend dynamic reserved memory regions

In reply to Wandun Chen
Quoted 21 lines · Wandun Chen

From: Wandun Chen <[email protected]>

Nowadays, the dynamic reserved memory regions is 64 by default, If the count of reserved memory regions defined in DTS bigger than 64, only 64 reserved memory can be handled properly.

So add a config to configure the actual dynamic reserved memory regions count instead of modify the code.

---
 drivers/of/Kconfig           | 11 +++++++++++
 drivers/of/of_private.h      |  2 +-
 drivers/of/of_reserved_mem.c |  2 +-
 3 files changed, 13 insertions(+), 2 deletions(-)
drivers/of/Kconfig +0 -0
index 50697cc3b07e..d6496ec3765c 100644

drivers/of/Kconfig config OF_IRQ · line 106

Quoted 8 lines
drivers/of/Kconfig +5 -0
@@ -99,6 +99,17 @@ config OF_IRQ
 config OF_RESERVED_MEM
 	def_bool OF_EARLY_FLATTREE
 
+config OF_RESERVED_MEM_DYNAMIC_REGIONS
+	int "Maximum count of the dynamic reserved memory regions"
+	depends on OF_RESERVED_MEM
+	default 64
+	range 1 256

My opinion on making this a config option is well documented. That's the primary reason we split dynamic regions.

How many do you need and why do you need so many regions? Seems like an abuse of reserved memory.

Rob

author
#9 lore raw

Re: [PATCH 2/2] of: reserved_mem: add config to extend dynamic reserved memory regions

Quoted 25 lines · Rob Herring
Quoted 24 lines · Wandun Chen

From: Wandun Chen <[email protected]>

Nowadays, the dynamic reserved memory regions is 64 by default, If the count of reserved memory regions defined in DTS bigger than 64, only 64 reserved memory can be handled properly.

So add a config to configure the actual dynamic reserved memory regions count instead of modify the code.

---
  drivers/of/Kconfig           | 11 +++++++++++
  drivers/of/of_private.h      |  2 +-
  drivers/of/of_reserved_mem.c |  2 +-
  3 files changed, 13 insertions(+), 2 deletions(-)
drivers/of/Kconfig +0 -0
index 50697cc3b07e..d6496ec3765c 100644
@@ -99,6 +99,17 @@ config OF_IRQ
  config OF_RESERVED_MEM
  	def_bool OF_EARLY_FLATTREE
 
Quoted 7 lines

drivers/of/Kconfig config OF_IRQ · line 106

Quoted 5 lines
drivers/of/Kconfig +5 -0
+config OF_RESERVED_MEM_DYNAMIC_REGIONS
+	int "Maximum count of the dynamic reserved memory regions"
+	depends on OF_RESERVED_MEM
+	default 64
+	range 1 256

My opinion on making this a config option is well documented. That's the primary reason we split dynamic regions.

Get it.

Quoted 3 lines

How many do you need and why do you need so many regions? Seems like an abuse of reserved memory.

I also agree that 64 dynamic regions should be sufficient for reasonable use cases.
To give some context on why I wrote this patch: the 64 dynamic region limit is implicit, there is no message to inform the user, but more than 64 regions also seems like an abuse of reserved memory. I will drop this patch as well, thanks for the review. Best regards, Wandun

Quoted 2 lines

Rob