Skip to contentSkip to conversation

Adding support for carrying IMA measurement logs

PATCH In review Prakhar Srivastava 2 participants 9 messages Jun 18, 2020 – Jul 16, 2020

Subsystems (7)
lore
Share & tools
lore
#0 lore raw

v2 cover Adding support for carrying IMA measurement logs

Integrgity Measurement Architecture(IMA) during kexec(kexec file load) verifies the kernel signature and measures the signature of the kernel.

The signature in the measuremnt logs is used to verfiy the authenticity of the kernel in the subsequent kexec'd session, however in the current implementation IMA measurement logs are not carried over thus remote attesation cannot verify the signature of the running kernel.

Adding support to arm64 to carry over the IMA measurement logs over kexec.

Add a new chosen node entry linux,ima-kexec-buffer to hold the address and the size of the memory reserved to carry the IMA measurement log. Refactor existing powerpc code to be used by amr64 as well.

Changelog:

v2:
Break patches into separate patches.
- Powerpc related Refactoring
- Updating the docuemntation for chosen node
- Updating arm64 to support IMA buffer pass

v1:
Refactoring carrying over IMA measuremnet logs over Kexec. This patch
moves the non-architecture specific code out of powerpc and adds to security/ima.(Suggested by Thiago)
Add Documentation regarding the ima-kexec-buffer node in the chosen
node documentation

v0:
Add a layer of abstraction to use the memory reserved by device tree
for ima buffer pass.
Add support for ima buffer pass using reserved memory for arm64 kexec.
Update the arch sepcific code path in kexec file load to store the ima buffer in the reserved memory. The same reserved memory is read on kexec or cold boot.

Prakhar Srivastava (3):
Refactoring powerpc code for carrying over IMA measurement logs, to
move non architecture specific code to security/ima.
dt-bindings: chosen: Document ima-kexec-buffer carrying over IMA
measuremnt logs over kexec.
Add support for arm64 to carry over IMA measurement logs

 Documentation/devicetree/bindings/chosen.txt |  17 +++
 arch/arm64/Kconfig                           |   1 +
 arch/arm64/include/asm/ima.h                 |  17 +++
 arch/arm64/include/asm/kexec.h               |   3 +
 arch/arm64/kernel/machine_kexec_file.c       |  47 +++++--
 arch/powerpc/include/asm/ima.h               |  10 --
 arch/powerpc/kexec/ima.c                     | 126 ++-----------------
 security/integrity/ima/ima_kexec.c           | 116 +++++++++++++++++
 8 files changed, 201 insertions(+), 136 deletions(-)
 create mode 100644 arch/arm64/include/asm/ima.h
Signature · 1 line
2.25.1
#5 lore raw

PATCH v2 2/3 dt-bindings: chosen: Document ima-kexec-buffer

Integrity measurement architecture(IMA) validates if files have been accidentally or maliciously altered, both remotely and locally, appraise a file's measurement against a "good" value stored as an extended attribute, and enforce local file integrity.

IMA also measures singatures of kernel and initrd during kexec along with the command line used for kexec.
These measurements are critical to verify the seccurity posture of the OS.

Resering memory and adding the memory information to a device tree node acts as the mechanism to carry over IMA measurement logs.

Update devicetree documentation to reflect the addition of new property under the chosen node.

---
 Documentation/devicetree/bindings/chosen.txt | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
Documentation/devicetree/bindings/chosen.txt +17 -0
index 45e79172a646..a15f70c007ef 100644
@@ -135,3 +135,20 @@ e.g.
 		linux,initrd-end = <0x82800000>;
 	};
 };
+
+linux,ima-kexec-buffer
+----------------------
+
+This property(currently used by powerpc, arm64) holds the memory range,
+the address and the size, of the IMA measurement logs that are being carried
+over to the kexec session.
+
+/ {
+	chosen {
+		linux,ima-kexec-buffer = <0x9 0x82000000 0x0 0x00008000>;
+	};
+};
+
+This porperty does not represent real hardware, but the memory allocated for
+carrying the IMA measurement logs. The address and the suze are expressed in
+#address-cells and #size-cells, respectively of the root node.
Signature · 1 line
2.25.1
#1 lore raw

PATCH v2 1/3 Refactoring powerpc code for carrying over IMA measurement logs, to move non architecture specific code to security/ima.

In reply to Prakhar Srivastava

Powerpc has support to carry over the IMA measurement logs. Refatoring the non-architecture specific code out of arch/powerpc and into security/ima.

The code adds support for reserving and freeing up of memory for IMA measurement logs.

---
 arch/powerpc/include/asm/ima.h     |  10 ---
 arch/powerpc/kexec/ima.c           | 126 ++---------------------------
 security/integrity/ima/ima_kexec.c | 116 ++++++++++++++++++++++++++
 3 files changed, 124 insertions(+), 128 deletions(-)
arch/powerpc/include/asm/ima.h +0 -10
index ead488cf3981..c29ec86498f8 100644
@@ -4,15 +4,6 @@
 
 struct kimage;
 
-int ima_get_kexec_buffer(void **addr, size_t *size);
-int ima_free_kexec_buffer(void);
-
-#ifdef CONFIG_IMA
-void remove_ima_buffer(void *fdt, int chosen_node);
-#else
-static inline void remove_ima_buffer(void *fdt, int chosen_node) {}
-#endif
-
 #ifdef CONFIG_IMA_KEXEC
 int arch_ima_add_kexec_buffer(struct kimage *image, unsigned long load_addr,
 			      size_t size);
@@ -22,7 +13,6 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node);
 static inline int setup_ima_buffer(const struct kimage *image, void *fdt,
 				   int chosen_node)
 {
-	remove_ima_buffer(fdt, chosen_node);
 	return 0;
 }
 #endif /* CONFIG_IMA_KEXEC */
arch/powerpc/kexec/ima.c +8 -118

142 lines not on this page · raw message

security/integrity/ima/ima_kexec.c +116 -0

124 lines not on this page · raw message

Signature · 1 line
2.25.1
#8 lore raw

PATCH v2 3/3 Add support for arm64 to carry over IMA measurement logs

In reply to Prakhar Srivastava

Add support for arm64 to carry over IMA measurement logs. Update arm64 code to call into functions made available in patch 1/3.

---
 arch/arm64/Kconfig                     |  1 +
 arch/arm64/include/asm/ima.h           | 17 ++++++++++
 arch/arm64/include/asm/kexec.h         |  3 ++
 arch/arm64/kernel/machine_kexec_file.c | 47 +++++++++++++++++++++-----
 4 files changed, 60 insertions(+), 8 deletions(-)
 create mode 100644 arch/arm64/include/asm/ima.h
arch/arm64/Kconfig +1 -0
index 5d513f461957..3d544e2e25e6 100644
@@ -1070,6 +1070,7 @@ config KEXEC
 config KEXEC_FILE
 	bool "kexec file based system call"
 	select KEXEC_CORE
+	select HAVE_IMA_KEXEC
 	help
 	  This is new version of kexec system call. This system call is
 	  file based and takes file descriptors as system call argument
arch/arm64/include/asm/ima.h new file +17 -0
new file mode 100644
index 000000000000..70ac39b74607
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_ARCH_IMA_H
+#define _ASM_ARCH_IMA_H
+
+struct kimage;
+
+#ifdef CONFIG_IMA_KEXEC
+int arch_ima_add_kexec_buffer(struct kimage *image, unsigned long load_addr,
+			      size_t size);
+#else
+static inline int arch_ima_add_kexec_buffer(struct kimage *image,
+			unsigned long load_addr, size_t size)
+{
+	return 0;
+}
+#endif /* CONFIG_IMA_KEXEC */
+#endif /* _ASM_ARCH_IMA_H */
arch/arm64/include/asm/kexec.h +3 -0
index d24b527e8c00..7bd60c185ad3 100644
@@ -100,6 +100,9 @@ struct kimage_arch {
 	void *elf_headers;
 	unsigned long elf_headers_mem;
 	unsigned long elf_headers_sz;
+
+	phys_addr_t ima_buffer_addr;
+	size_t ima_buffer_size;
 };
 
 extern const struct kexec_file_ops kexec_image_ops;
arch/arm64/kernel/machine_kexec_file.c +39 -8

71 lines not on this page · raw message

Signature · 1 line
2.25.1
#2 lore raw

Re: [V2 PATCH 1/3] Refactoring powerpc code for carrying over IMA measurement logs, to move non architecture specific code to security/ima.

In reply to Prakhar Srivastava
Quoted 5 lines · Prakhar Srivastava

Powerpc has support to carry over the IMA measurement logs. Refatoring the non-architecture specific code out of arch/powerpc and into security/ima.

The code adds support for reserving and freeing up of memory for IMA measurement logs.

Last week, Mimi provided this feedback:

"From your patch description, this patch should be broken up. Moving the non-architecture specific code out of powerpc should be one patch.
Additional support should be in another patch. After each patch, the
code should work properly."

That's not what you do here. You move the code, but you also make other changes at the same time. This has two problems:

1. It makes the patch harder to review, because it's very easy to miss a change.

2. If in the future a git bisect later points to this patch, it's not clear whether the problem is because of the code movement, or because of the other changes.

When you move code, ideally the patch should only make the changes necessary to make the code work at its new location. The patch which does code movement should not cause any change in behavior.

Other changes should go in separate patches, either before or after the one moving the code.

More comments below.

Quoted 28 lines
---
 arch/powerpc/include/asm/ima.h     |  10 ---
 arch/powerpc/kexec/ima.c           | 126 ++---------------------------
 security/integrity/ima/ima_kexec.c | 116 ++++++++++++++++++++++++++
 3 files changed, 124 insertions(+), 128 deletions(-)
arch/powerpc/include/asm/ima.h +0 -9
index ead488cf3981..c29ec86498f8 100644
@@ -4,15 +4,6 @@
 
 struct kimage;
 
-int ima_get_kexec_buffer(void **addr, size_t *size);
-int ima_free_kexec_buffer(void);
-
-#ifdef CONFIG_IMA
-void remove_ima_buffer(void *fdt, int chosen_node);
-#else
-static inline void remove_ima_buffer(void *fdt, int chosen_node) {}
-#endif
-
 #ifdef CONFIG_IMA_KEXEC
 int arch_ima_add_kexec_buffer(struct kimage *image, unsigned long load_addr,
 			      size_t size);

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node); · line 17

Quoted 6 lines
arch/powerpc/include/asm/ima.h +0 -1
@@ -22,7 +13,6 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node);
 static inline int setup_ima_buffer(const struct kimage *image, void *fdt,
 				   int chosen_node)
 {
-	remove_ima_buffer(fdt, chosen_node);
 	return 0;
 }

This is wrong. Even if the currently running kernel doesn't have CONFIG_IMA_KEXEC, it should remove the IMA buffer property and memory reservation from the FDT that is being prepared for the next kernel.

This is because the IMA kexec buffer is useless for the next kernel, regardless of whether the current kernel supports CONFIG_IMA_KEXEC or not. Keeping it around would be a waste of memory.

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node) · line 182

Quoted 5 lines
arch/powerpc/include/asm/ima.h +0 -1
@@ -179,13 +64,18 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node)
 	int ret, addr_cells, size_cells, entry_size;
 	u8 value[16];
 
-	remove_ima_buffer(fdt, chosen_node);

This is wrong, for the same reason stated above.

Quoted 7 lines
arch/powerpc/include/asm/ima.h +2 -2
 	if (!image->arch.ima_buffer_size)
 		return 0;
 
-	ret = get_addr_size_cells(&addr_cells, &size_cells);
-	if (ret)
+	ret = fdt_address_cells(fdt, chosen_node);
+	if (ret < 0)

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node) · line 81

Quoted 10 lines
arch/powerpc/include/asm/ima.h +6 -0
+		return ret;
+	addr_cells = ret;
+
+	ret = fdt_size_cells(fdt, chosen_node);
+	if (ret < 0)
 		return ret;
+	size_cells = ret;
 
 	entry_size = 4 * (addr_cells + size_cells);
 

I liked this change. Thanks! I agree it's better to use fdt_address_cells() and fdt_size_cells() here.

But it should be in a separate patch. Either before or after the one moving the code.

Quoted 64 lines
security/integrity/ima/ima_kexec.c +54 -0
index 121de3e04af2..e1e6d6154015 100644
@@ -10,8 +10,124 @@
 #include <linux/seq_file.h>
 #include <linux/vmalloc.h>
 #include <linux/kexec.h>
+#include <linux/of.h>
+#include <linux/memblock.h>
+#include <linux/libfdt.h>
 #include "ima.h"
 
+static int get_addr_size_cells(int *addr_cells, int *size_cells)
+{
+	struct device_node *root;
+
+	root = of_find_node_by_path("/");
+	if (!root)
+		return -EINVAL;
+
+	*addr_cells = of_n_addr_cells(root);
+	*size_cells = of_n_size_cells(root);
+
+	of_node_put(root);
+
+	return 0;
+}
+
+static int do_get_kexec_buffer(const void *prop, int len, unsigned long *addr,
+			       size_t *size)
+{
+	int ret, addr_cells, size_cells;
+
+	ret = get_addr_size_cells(&addr_cells, &size_cells);
+	if (ret)
+		return ret;
+
+	if (len < 4 * (addr_cells + size_cells))
+		return -ENOENT;
+
+	*addr = of_read_number(prop, addr_cells);
+	*size = of_read_number(prop + 4 * addr_cells, size_cells);
+
+	return 0;
+}
+
+/**
+ * ima_get_kexec_buffer - get IMA buffer from the previous kernel
+ * @addr:	On successful return, set to point to the buffer contents.
+ * @size:	On successful return, set to the buffer size.
+ *
+ * Return: 0 on success, negative errno on error.
+ */
+int ima_get_kexec_buffer(void **addr, size_t *size)
+{
+	int ret, len;
+	unsigned long tmp_addr;
+	size_t tmp_size;
+	const void *prop;
+
+	prop = of_get_property(of_chosen, "linux,ima-kexec-buffer", &len);
+	if (!prop)
+		return -ENOENT;

security/integrity/ima/ima_kexec.c line 78

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+
+	ret = do_get_kexec_buffer(prop, len, &tmp_addr, &tmp_size);
+	if (ret)
+		return ret;
+
+	*addr = __va(tmp_addr);
+	*size = tmp_size;
+
+	return 0;
+}

The functions above were moved without being changed. Good.

Quoted 13 lines
security/integrity/ima/ima_kexec.c +13 -0
+/**
+ * ima_free_kexec_buffer - free memory used by the IMA buffer
+ */
+int ima_free_kexec_buffer(void)
+{
+	int ret;
+	unsigned long addr;
+	size_t size;
+	struct property *prop;
+
+	prop = of_find_property(of_chosen, "linux,ima-kexec-buffer", NULL);
+	if (!prop)
+		return -ENOENT;

security/integrity/ima/ima_kexec.c line 102

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+
+	ret = do_get_kexec_buffer(prop->value, prop->length, &addr, &size);
+	if (ret)
+		return ret;
+
+	ret = of_remove_property(of_chosen, prop);
+	if (ret)
+		return ret;
+
+	return memblock_free(__pa(addr), size);

Here you added a __pa() call. Do you store a virtual address in linux,ima-kexec-buffer property? Doesn't it make more sense to store a physical address?

Even if making this change is the correct thing to do, it should be a separate patch, unless it can't be avoided. And if that is the case, then it should be explained in the commit message.

Quoted 17 lines
security/integrity/ima/ima_kexec.c +17 -0
+
+}
+
+/**
+ * remove_ima_buffer - remove the IMA buffer property and reservation from @fdt
+ *
+ * The IMA measurement buffer is of no use to a subsequent kernel, so we always
+ * remove it from the device tree.
+ */
+void remove_ima_buffer(void *fdt, int chosen_node)
+{
+	int ret, len;
+	unsigned long addr;
+	size_t size;
+	const void *prop;
+
+	prop = fdt_getprop(fdt, chosen_node, "linux,ima-kexec-buffer", &len);

security/integrity/ima/ima_kexec.c line 129

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+	if (!prop)
+		return;
+
+	do_get_kexec_buffer(prop, len, &addr, &size);
+	ret = fdt_delprop(fdt, chosen_node, "linux,ima-kexec-buffer");
+	if (ret < 0)
+		return;
+
+	memblock_free(addr, size);
+}

Here is another function that changed when moved. This one I know to be wrong. You're confusing the purposes of remove_ima_buffer() and ima_free_kexec_buffer().

You did send me a question about them nearly three weeks ago which I only answered today, so I apologize. Also, their names could more clearly reflect their differences, so it's bad naming on my part.

With IMA kexec buffers, there are two kernels (and thus their two respective, separate device trees) to be concerned about:

1. the currently running kernel, which uses the live device tree (accessed with the of_* functions) and the memblock subsystem;

2. the kernel which is being loaded by kexec, which will use the FDT blob being passed around as argument to these functions, and the memory reservations in the memory reservation table of the FDT blob.

ima_free_kexec_buffer() is used by IMA in the currently running kernel. Therefore the device tree it is concerned about is the live one, and thus uses the of_* functions to access it. And uses memblock to change the memory reservation.

remove_ima_buffer() on the other hand is used by the kexec code to prepare an FDT blob for the kernel that is being loaded. It should not make any changes to live device tree, nor to memblock allocations. It should only make changes to the FDT blob.

--
Thiago Jung Bauermann
IBM Linux Technology Center

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#6 lore raw

Re: [V2 PATCH 2/3] dt-bindings: chosen: Document ima-kexec-buffer

In reply to Prakhar Srivastava
Quoted 9 lines · Prakhar Srivastava

Integrity measurement architecture(IMA) validates if files have been accidentally or maliciously altered, both remotely and locally, appraise a file's measurement against a "good" value stored as an extended attribute, and enforce local file integrity.

IMA also measures singatures of kernel and initrd during kexec along with the command line used for kexec.
These measurements are critical to verify the seccurity posture of the OS.

Quoted 5 lines

Resering memory and adding the memory information to a device tree node acts as the mechanism to carry over IMA measurement logs.

Update devicetree documentation to reflect the addition of new property under the chosen node.

Thank you for writing this documentation patch. It's something I should have done when I added the powerpc IMA kexec support.

You addressed Rob Herring's comments regarding the commit message, but not the ones regarding the patch contents.

When posting a new version of the patches, make sure to address all comments made so far. Addressing a comment doesn't necessarily mean implementing the requested change. If you don't then you should at least explain why you chose a different path.

I mention it because this has occurred before with this patch series, and it's hard to make forward progress if review comments get ignored.

Quoted 9 lines
---
 Documentation/devicetree/bindings/chosen.txt | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
Documentation/devicetree/bindings/chosen.txt +0 -0
index 45e79172a646..a15f70c007ef 100644

Documentation/devicetree/bindings/chosen.txt e.g. · line 142

Quoted 8 lines
Documentation/devicetree/bindings/chosen.txt +5 -0
@@ -135,3 +135,20 @@ e.g.
 		linux,initrd-end = <0x82800000>;
 	};
 };
+
+linux,ima-kexec-buffer
+----------------------
+
+This property(currently used by powerpc, arm64) holds the memory range,

space before the parenthesis.

Documentation/devicetree/bindings/chosen.txt e.g. · line 143

Quoted 1 line
Documentation/devicetree/bindings/chosen.txt +1 -0
+the address and the size, of the IMA measurement logs that are being carried

Maybe it's because English isn't my first language, but IMHO it's clearer if "the address and the size" is between parentheses rather than commas.

Documentation/devicetree/bindings/chosen.txt e.g. · line 144

Quoted 1 line
Documentation/devicetree/bindings/chosen.txt +1 -0
+over to the kexec session.

I don't think there's a "kexec session", but I'm not sure what a good term would be. "linux,booted-from-kexec" uses "new kernel" so perhaps that's a good option to use instead of "kexec session".

Documentation/devicetree/bindings/chosen.txt e.g. · line 154

Quoted 10 lines
Documentation/devicetree/bindings/chosen.txt +10 -0
+
+/ {
+	chosen {
+		linux,ima-kexec-buffer = <0x9 0x82000000 0x0 0x00008000>;
+	};
+};
+
+This porperty does not represent real hardware, but the memory allocated for
+carrying the IMA measurement logs. The address and the suze are expressed in
+#address-cells and #size-cells, respectively of the root node.

--
Thiago Jung Bauermann
IBM Linux Technology Center

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#3 lore raw

Re: [V2 PATCH 1/3] Refactoring powerpc code for carrying over IMA measurement logs, to move non architecture specific code to security/ima.

In reply to Thiago Jung Bauermann
Quoted 26 lines · Thiago Jung Bauermann
Quoted 5 lines · Prakhar Srivastava

Powerpc has support to carry over the IMA measurement logs. Refatoring the non-architecture specific code out of arch/powerpc and into security/ima.

The code adds support for reserving and freeing up of memory for IMA measurement logs.

Last week, Mimi provided this feedback:

"From your patch description, this patch should be broken up. Moving the non-architecture specific code out of powerpc should be one patch.
Additional support should be in another patch. After each patch, the
code should work properly."

That's not what you do here. You move the code, but you also make other changes at the same time. This has two problems:

1. It makes the patch harder to review, because it's very easy to miss a change.

2. If in the future a git bisect later points to this patch, it's not clear whether the problem is because of the code movement, or because of the other changes.

Quoted 9 lines

When you move code, ideally the patch should only make the changes necessary to make the code work at its new location. The patch which does code movement should not cause any change in behavior.

Other changes should go in separate patches, either before or after the one moving the code.

More comments below.

Hi Thiago,

Apologies for the delayed response i was away for a few days.
I am working on breaking up the changes so that its easier to review and update as well.

Thanks,
Prakhar Srivastava

Quoted 39 lines
Quoted 28 lines
---
  arch/powerpc/include/asm/ima.h     |  10 ---
  arch/powerpc/kexec/ima.c           | 126 ++---------------------------
  security/integrity/ima/ima_kexec.c | 116 ++++++++++++++++++++++++++
  3 files changed, 124 insertions(+), 128 deletions(-)
arch/powerpc/include/asm/ima.h +0 -9
index ead488cf3981..c29ec86498f8 100644
@@ -4,15 +4,6 @@
 
  struct kimage;
 
-int ima_get_kexec_buffer(void **addr, size_t *size);
-int ima_free_kexec_buffer(void);
-
-#ifdef CONFIG_IMA
-void remove_ima_buffer(void *fdt, int chosen_node);
-#else
-static inline void remove_ima_buffer(void *fdt, int chosen_node) {}
-#endif
-
  #ifdef CONFIG_IMA_KEXEC
  int arch_ima_add_kexec_buffer(struct kimage *image, unsigned long load_addr,
  			      size_t size);

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node); · line 17

Quoted 6 lines
arch/powerpc/include/asm/ima.h +0 -1
@@ -22,7 +13,6 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node);
  static inline int setup_ima_buffer(const struct kimage *image, void *fdt,
  				   int chosen_node)
  {
-	remove_ima_buffer(fdt, chosen_node);
  	return 0;
  }

This is wrong. Even if the currently running kernel doesn't have CONFIG_IMA_KEXEC, it should remove the IMA buffer property and memory reservation from the FDT that is being prepared for the next kernel.

Quoted 4 lines

This is because the IMA kexec buffer is useless for the next kernel, regardless of whether the current kernel supports CONFIG_IMA_KEXEC or not. Keeping it around would be a waste of memory.

I will keep it in my next revision.
My understanding was the reserved memory is freed and property removed when IMA loads the logs on init. During setup_fdt in kexec, a duplicate copy of the dt is used, but memory still needs to be allocated, thus the property itself indicats presence of reserved memory.

Quoted 128 lines

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node) · line 182

Quoted 5 lines
arch/powerpc/include/asm/ima.h +0 -1
@@ -179,13 +64,18 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node)
  	int ret, addr_cells, size_cells, entry_size;
  	u8 value[16];
 
-	remove_ima_buffer(fdt, chosen_node);

This is wrong, for the same reason stated above.

Quoted 7 lines
arch/powerpc/include/asm/ima.h +2 -2
  	if (!image->arch.ima_buffer_size)
  		return 0;
 
-	ret = get_addr_size_cells(&addr_cells, &size_cells);
-	if (ret)
+	ret = fdt_address_cells(fdt, chosen_node);
+	if (ret < 0)

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node) · line 81

Quoted 10 lines
arch/powerpc/include/asm/ima.h +6 -0
+		return ret;
+	addr_cells = ret;
+
+	ret = fdt_size_cells(fdt, chosen_node);
+	if (ret < 0)
  		return ret;
+	size_cells = ret;
 
  	entry_size = 4 * (addr_cells + size_cells);
 

I liked this change. Thanks! I agree it's better to use fdt_address_cells() and fdt_size_cells() here.

But it should be in a separate patch. Either before or after the one moving the code.

Quoted 64 lines
security/integrity/ima/ima_kexec.c +54 -0
index 121de3e04af2..e1e6d6154015 100644
@@ -10,8 +10,124 @@
  #include <linux/seq_file.h>
  #include <linux/vmalloc.h>
  #include <linux/kexec.h>
+#include <linux/of.h>
+#include <linux/memblock.h>
+#include <linux/libfdt.h>
  #include "ima.h"
 
+static int get_addr_size_cells(int *addr_cells, int *size_cells)
+{
+	struct device_node *root;
+
+	root = of_find_node_by_path("/");
+	if (!root)
+		return -EINVAL;
+
+	*addr_cells = of_n_addr_cells(root);
+	*size_cells = of_n_size_cells(root);
+
+	of_node_put(root);
+
+	return 0;
+}
+
+static int do_get_kexec_buffer(const void *prop, int len, unsigned long *addr,
+			       size_t *size)
+{
+	int ret, addr_cells, size_cells;
+
+	ret = get_addr_size_cells(&addr_cells, &size_cells);
+	if (ret)
+		return ret;
+
+	if (len < 4 * (addr_cells + size_cells))
+		return -ENOENT;
+
+	*addr = of_read_number(prop, addr_cells);
+	*size = of_read_number(prop + 4 * addr_cells, size_cells);
+
+	return 0;
+}
+
+/**
+ * ima_get_kexec_buffer - get IMA buffer from the previous kernel
+ * @addr:	On successful return, set to point to the buffer contents.
+ * @size:	On successful return, set to the buffer size.
+ *
+ * Return: 0 on success, negative errno on error.
+ */
+int ima_get_kexec_buffer(void **addr, size_t *size)
+{
+	int ret, len;
+	unsigned long tmp_addr;
+	size_t tmp_size;
+	const void *prop;
+
+	prop = of_get_property(of_chosen, "linux,ima-kexec-buffer", &len);
+	if (!prop)
+		return -ENOENT;

security/integrity/ima/ima_kexec.c line 78

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+
+	ret = do_get_kexec_buffer(prop, len, &tmp_addr, &tmp_size);
+	if (ret)
+		return ret;
+
+	*addr = __va(tmp_addr);
+	*size = tmp_size;
+
+	return 0;
+}

The functions above were moved without being changed. Good.

Quoted 19 lines
security/integrity/ima/ima_kexec.c +19 -0
+/**
+ * ima_free_kexec_buffer - free memory used by the IMA buffer
+ */
+int ima_free_kexec_buffer(void)
+{
+	int ret;
+	unsigned long addr;
+	size_t size;
+	struct property *prop;
+
+	prop = of_find_property(of_chosen, "linux,ima-kexec-buffer", NULL);
+	if (!prop)
+		return -ENOENT;
+
+	ret = do_get_kexec_buffer(prop->value, prop->length, &addr, &size);
+	if (ret)
+		return ret;
+
+	ret = of_remove_property(of_chosen, prop);
Quoted 9 lines

security/integrity/ima/ima_kexec.c line 102

Quoted 4 lines
security/integrity/ima/ima_kexec.c +4 -0
+	if (ret)
+		return ret;
+
+	return memblock_free(__pa(addr), size);

Here you added a __pa() call. Do you store a virtual address in linux,ima-kexec-buffer property? Doesn't it make more sense to store a physical address?

trying to minimize the changes here as do_get_kexec_buffer return the va. I will refactor this to remove the double translation.

Quoted 55 lines

Even if making this change is the correct thing to do, it should be a separate patch, unless it can't be avoided. And if that is the case, then it should be explained in the commit message.

Quoted 17 lines
security/integrity/ima/ima_kexec.c +17 -0
+
+}
+
+/**
+ * remove_ima_buffer - remove the IMA buffer property and reservation from @fdt
+ *
+ * The IMA measurement buffer is of no use to a subsequent kernel, so we always
+ * remove it from the device tree.
+ */
+void remove_ima_buffer(void *fdt, int chosen_node)
+{
+	int ret, len;
+	unsigned long addr;
+	size_t size;
+	const void *prop;
+
+	prop = fdt_getprop(fdt, chosen_node, "linux,ima-kexec-buffer", &len);

security/integrity/ima/ima_kexec.c line 129

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+	if (!prop)
+		return;
+
+	do_get_kexec_buffer(prop, len, &addr, &size);
+	ret = fdt_delprop(fdt, chosen_node, "linux,ima-kexec-buffer");
+	if (ret < 0)
+		return;
+
+	memblock_free(addr, size);
+}

Here is another function that changed when moved. This one I know to be wrong. You're confusing the purposes of remove_ima_buffer() and ima_free_kexec_buffer().

You did send me a question about them nearly three weeks ago which I only answered today, so I apologize. Also, their names could more clearly reflect their differences, so it's bad naming on my part.

With IMA kexec buffers, there are two kernels (and thus their two respective, separate device trees) to be concerned about:

1. the currently running kernel, which uses the live device tree (accessed with the of_* functions) and the memblock subsystem;

2. the kernel which is being loaded by kexec, which will use the FDT blob being passed around as argument to these functions, and the memory reservations in the memory reservation table of the FDT blob.

ima_free_kexec_buffer() is used by IMA in the currently running kernel. Therefore the device tree it is concerned about is the live one, and thus uses the of_* functions to access it. And uses memblock to change the memory reservation.

Quoted 5 lines

remove_ima_buffer() on the other hand is used by the kexec code to prepare an FDT blob for the kernel that is being loaded. It should not make any changes to live device tree, nor to memblock allocations. It should only make changes to the FDT blob.

Thank you for this, greatly appreciate clearing my misunderstandings.

Quoted 4 lines

--
Thiago Jung Bauermann
IBM Linux Technology Center

#7 lore raw

Re: [V2 PATCH 2/3] dt-bindings: chosen: Document ima-kexec-buffer

In reply to Thiago Jung Bauermann
Quoted 64 lines · Thiago Jung Bauermann
Quoted 9 lines · Prakhar Srivastava

Integrity measurement architecture(IMA) validates if files have been accidentally or maliciously altered, both remotely and locally, appraise a file's measurement against a "good" value stored as an extended attribute, and enforce local file integrity.

IMA also measures singatures of kernel and initrd during kexec along with the command line used for kexec.
These measurements are critical to verify the seccurity posture of the OS.

Quoted 5 lines

Resering memory and adding the memory information to a device tree node acts as the mechanism to carry over IMA measurement logs.

Update devicetree documentation to reflect the addition of new property under the chosen node.

Thank you for writing this documentation patch. It's something I should have done when I added the powerpc IMA kexec support.

You addressed Rob Herring's comments regarding the commit message, but not the ones regarding the patch contents.

When posting a new version of the patches, make sure to address all comments made so far. Addressing a comment doesn't necessarily mean implementing the requested change. If you don't then you should at least explain why you chose a different path.

I mention it because this has occurred before with this patch series, and it's hard to make forward progress if review comments get ignored.

Quoted 9 lines
---
  Documentation/devicetree/bindings/chosen.txt | 17 +++++++++++++++++
  1 file changed, 17 insertions(+)
Documentation/devicetree/bindings/chosen.txt +0 -0
index 45e79172a646..a15f70c007ef 100644

Documentation/devicetree/bindings/chosen.txt e.g. · line 142

Quoted 8 lines
Documentation/devicetree/bindings/chosen.txt +5 -0
@@ -135,3 +135,20 @@ e.g.
  		linux,initrd-end = <0x82800000>;
  	};
  };
+
+linux,ima-kexec-buffer
+----------------------
+
+This property(currently used by powerpc, arm64) holds the memory range,

space before the parenthesis.

Documentation/devicetree/bindings/chosen.txt e.g. · line 143

Quoted 1 line
Documentation/devicetree/bindings/chosen.txt +1 -0
+the address and the size, of the IMA measurement logs that are being carried

Maybe it's because English isn't my first language, but IMHO it's clearer if "the address and the size" is between parentheses rather than commas.

Documentation/devicetree/bindings/chosen.txt e.g. · line 144

Quoted 1 line
Documentation/devicetree/bindings/chosen.txt +1 -0
+over to the kexec session.

I don't think there's a "kexec session", but I'm not sure what a good term would be. "linux,booted-from-kexec" uses "new kernel" so perhaps that's a good option to use instead of "kexec session".

Quoted 12 lines

Documentation/devicetree/bindings/chosen.txt e.g. · line 154

Quoted 10 lines
Documentation/devicetree/bindings/chosen.txt +10 -0
+
+/ {
+	chosen {
+		linux,ima-kexec-buffer = <0x9 0x82000000 0x0 0x00008000>;
+	};
+};
+
+This porperty does not represent real hardware, but the memory allocated for
+carrying the IMA measurement logs. The address and the suze are expressed in
+#address-cells and #size-cells, respectively of the root node.

I will update the descriptions and ack the comments/changes in the patches as well.

Thankyou,
Prakhar Srivastava

Quoted 4 lines

--
Thiago Jung Bauermann
IBM Linux Technology Center

#4 lore raw

Re: [V2 PATCH 1/3] Refactoring powerpc code for carrying over IMA measurement logs, to move non architecture specific code to security/ima.

In reply to Prakhar Srivastava

Hello Prakhar,

Quoted 34 lines · Prakhar Srivastava
Quoted 33 lines · Thiago Jung Bauermann
Quoted 5 lines · Prakhar Srivastava

Powerpc has support to carry over the IMA measurement logs. Refatoring the non-architecture specific code out of arch/powerpc and into security/ima.

The code adds support for reserving and freeing up of memory for IMA measurement logs.

Last week, Mimi provided this feedback:

"From your patch description, this patch should be broken up. Moving the non-architecture specific code out of powerpc should be one patch.
Additional support should be in another patch. After each patch, the
code should work properly."

That's not what you do here. You move the code, but you also make other changes at the same time. This has two problems:

1. It makes the patch harder to review, because it's very easy to miss a change.

2. If in the future a git bisect later points to this patch, it's not clear whether the problem is because of the code movement, or because of the other changes.

When you move code, ideally the patch should only make the changes necessary to make the code work at its new location. The patch which does code movement should not cause any change in behavior.

Other changes should go in separate patches, either before or after the one moving the code.

Quoted 7 lines
Quoted 2 lines

More comments below.

Hi Thiago,

Apologies for the delayed response i was away for a few days.
I am working on breaking up the changes so that its easier to review and update as well.

No problem.

Quoted 43 lines

Thanks,
Prakhar Srivastava

Quoted 39 lines
Quoted 28 lines
---
  arch/powerpc/include/asm/ima.h     |  10 ---
  arch/powerpc/kexec/ima.c           | 126 ++---------------------------
  security/integrity/ima/ima_kexec.c | 116 ++++++++++++++++++++++++++
  3 files changed, 124 insertions(+), 128 deletions(-)
arch/powerpc/include/asm/ima.h +0 -9
index ead488cf3981..c29ec86498f8 100644
@@ -4,15 +4,6 @@
 
  struct kimage;
 
-int ima_get_kexec_buffer(void **addr, size_t *size);
-int ima_free_kexec_buffer(void);
-
-#ifdef CONFIG_IMA
-void remove_ima_buffer(void *fdt, int chosen_node);
-#else
-static inline void remove_ima_buffer(void *fdt, int chosen_node) {}
-#endif
-
  #ifdef CONFIG_IMA_KEXEC
  int arch_ima_add_kexec_buffer(struct kimage *image, unsigned long load_addr,
  			      size_t size);

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node); · line 17

Quoted 6 lines
arch/powerpc/include/asm/ima.h +0 -1
@@ -22,7 +13,6 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node);
  static inline int setup_ima_buffer(const struct kimage *image, void *fdt,
  				   int chosen_node)
  {
-	remove_ima_buffer(fdt, chosen_node);
  	return 0;
  }

This is wrong. Even if the currently running kernel doesn't have CONFIG_IMA_KEXEC, it should remove the IMA buffer property and memory reservation from the FDT that is being prepared for the next kernel.

Quoted 7 lines
Quoted 4 lines

This is because the IMA kexec buffer is useless for the next kernel, regardless of whether the current kernel supports CONFIG_IMA_KEXEC or not. Keeping it around would be a waste of memory.

I will keep it in my next revision.
My understanding was the reserved memory is freed and property removed when IMA loads the logs on init.

If CONFIG_IMA_KEXEC is set, then yes. If it isn't then that needs to happen in the function above.

Quoted 141 lines

During setup_fdt in kexec, a duplicate copy of the dt is
used, but memory still needs to be allocated, thus the property itself indicats presence of reserved memory.

Quoted 128 lines

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node) · line 182

Quoted 5 lines
arch/powerpc/include/asm/ima.h +0 -1
@@ -179,13 +64,18 @@ int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node)
  	int ret, addr_cells, size_cells, entry_size;
  	u8 value[16];
 
-	remove_ima_buffer(fdt, chosen_node);

This is wrong, for the same reason stated above.

Quoted 7 lines
arch/powerpc/include/asm/ima.h +2 -2
  	if (!image->arch.ima_buffer_size)
  		return 0;
 
-	ret = get_addr_size_cells(&addr_cells, &size_cells);
-	if (ret)
+	ret = fdt_address_cells(fdt, chosen_node);
+	if (ret < 0)

arch/powerpc/include/asm/ima.h int setup_ima_buffer(const struct kimage *image, void *fdt, int chosen_node) · line 81

Quoted 10 lines
arch/powerpc/include/asm/ima.h +6 -0
+		return ret;
+	addr_cells = ret;
+
+	ret = fdt_size_cells(fdt, chosen_node);
+	if (ret < 0)
  		return ret;
+	size_cells = ret;
 
  	entry_size = 4 * (addr_cells + size_cells);
 

I liked this change. Thanks! I agree it's better to use fdt_address_cells() and fdt_size_cells() here.

But it should be in a separate patch. Either before or after the one moving the code.

Quoted 64 lines
security/integrity/ima/ima_kexec.c +54 -0
index 121de3e04af2..e1e6d6154015 100644
@@ -10,8 +10,124 @@
  #include <linux/seq_file.h>
  #include <linux/vmalloc.h>
  #include <linux/kexec.h>
+#include <linux/of.h>
+#include <linux/memblock.h>
+#include <linux/libfdt.h>
  #include "ima.h"
 
+static int get_addr_size_cells(int *addr_cells, int *size_cells)
+{
+	struct device_node *root;
+
+	root = of_find_node_by_path("/");
+	if (!root)
+		return -EINVAL;
+
+	*addr_cells = of_n_addr_cells(root);
+	*size_cells = of_n_size_cells(root);
+
+	of_node_put(root);
+
+	return 0;
+}
+
+static int do_get_kexec_buffer(const void *prop, int len, unsigned long *addr,
+			       size_t *size)
+{
+	int ret, addr_cells, size_cells;
+
+	ret = get_addr_size_cells(&addr_cells, &size_cells);
+	if (ret)
+		return ret;
+
+	if (len < 4 * (addr_cells + size_cells))
+		return -ENOENT;
+
+	*addr = of_read_number(prop, addr_cells);
+	*size = of_read_number(prop + 4 * addr_cells, size_cells);
+
+	return 0;
+}
+
+/**
+ * ima_get_kexec_buffer - get IMA buffer from the previous kernel
+ * @addr:	On successful return, set to point to the buffer contents.
+ * @size:	On successful return, set to the buffer size.
+ *
+ * Return: 0 on success, negative errno on error.
+ */
+int ima_get_kexec_buffer(void **addr, size_t *size)
+{
+	int ret, len;
+	unsigned long tmp_addr;
+	size_t tmp_size;
+	const void *prop;
+
+	prop = of_get_property(of_chosen, "linux,ima-kexec-buffer", &len);
+	if (!prop)
+		return -ENOENT;

security/integrity/ima/ima_kexec.c line 78

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+
+	ret = do_get_kexec_buffer(prop, len, &tmp_addr, &tmp_size);
+	if (ret)
+		return ret;
+
+	*addr = __va(tmp_addr);
+	*size = tmp_size;
+
+	return 0;
+}

The functions above were moved without being changed. Good.

Quoted 19 lines
security/integrity/ima/ima_kexec.c +19 -0
+/**
+ * ima_free_kexec_buffer - free memory used by the IMA buffer
+ */
+int ima_free_kexec_buffer(void)
+{
+	int ret;
+	unsigned long addr;
+	size_t size;
+	struct property *prop;
+
+	prop = of_find_property(of_chosen, "linux,ima-kexec-buffer", NULL);
+	if (!prop)
+		return -ENOENT;
+
+	ret = do_get_kexec_buffer(prop->value, prop->length, &addr, &size);
+	if (ret)
+		return ret;
+
+	ret = of_remove_property(of_chosen, prop);
Quoted 9 lines

security/integrity/ima/ima_kexec.c line 102

Quoted 4 lines
security/integrity/ima/ima_kexec.c +4 -0
+	if (ret)
+		return ret;
+
+	return memblock_free(__pa(addr), size);

Here you added a __pa() call. Do you store a virtual address in linux,ima-kexec-buffer property? Doesn't it make more sense to store a physical address?

Quoted 2 lines

trying to minimize the changes here as do_get_kexec_buffer return the va. I will refactor this to remove the double translation.

In the powerpc version, do_get_kexec_buffer() returns the pa, and one of its callers does the va translation. I think that worked well.

Quoted 55 lines
Quoted 55 lines

Even if making this change is the correct thing to do, it should be a separate patch, unless it can't be avoided. And if that is the case, then it should be explained in the commit message.

Quoted 17 lines
security/integrity/ima/ima_kexec.c +17 -0
+
+}
+
+/**
+ * remove_ima_buffer - remove the IMA buffer property and reservation from @fdt
+ *
+ * The IMA measurement buffer is of no use to a subsequent kernel, so we always
+ * remove it from the device tree.
+ */
+void remove_ima_buffer(void *fdt, int chosen_node)
+{
+	int ret, len;
+	unsigned long addr;
+	size_t size;
+	const void *prop;
+
+	prop = fdt_getprop(fdt, chosen_node, "linux,ima-kexec-buffer", &len);

security/integrity/ima/ima_kexec.c line 129

Quoted 10 lines
security/integrity/ima/ima_kexec.c +10 -0
+	if (!prop)
+		return;
+
+	do_get_kexec_buffer(prop, len, &addr, &size);
+	ret = fdt_delprop(fdt, chosen_node, "linux,ima-kexec-buffer");
+	if (ret < 0)
+		return;
+
+	memblock_free(addr, size);
+}

Here is another function that changed when moved. This one I know to be wrong. You're confusing the purposes of remove_ima_buffer() and ima_free_kexec_buffer().

You did send me a question about them nearly three weeks ago which I only answered today, so I apologize. Also, their names could more clearly reflect their differences, so it's bad naming on my part.

With IMA kexec buffers, there are two kernels (and thus their two respective, separate device trees) to be concerned about:

1. the currently running kernel, which uses the live device tree (accessed with the of_* functions) and the memblock subsystem;

2. the kernel which is being loaded by kexec, which will use the FDT blob being passed around as argument to these functions, and the memory reservations in the memory reservation table of the FDT blob.

ima_free_kexec_buffer() is used by IMA in the currently running kernel. Therefore the device tree it is concerned about is the live one, and thus uses the of_* functions to access it. And uses memblock to change the memory reservation.

Quoted 6 lines
Quoted 5 lines

remove_ima_buffer() on the other hand is used by the kexec code to prepare an FDT blob for the kernel that is being loaded. It should not make any changes to live device tree, nor to memblock allocations. It should only make changes to the FDT blob.

Thank you for this, greatly appreciate clearing my misunderstandings.

You're welcome. Sorry again for not answering your question before you sent this patch series.

Signature · 2 lines
Thiago Jung Bauermann
IBM Linux Technology Center