Skip to contentSkip to conversation

cxl/region: Avoid null pointer dereference in region lookup

PATCH Superseded [email protected] 3 participants 4 messages Jun 4, 2024 – Jun 5, 2024 CXL

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/1 cxl/region: Avoid null pointer dereference in region lookup

From: Alison Schofield <[email protected]>

cxl_dpa_to_region() looks up a region based on a memdev and DPA. It wrongly assumes an endpoint found mapping the DPA is also of a fully assembled region. When not true it leads to a null pointer dereference looking up the region name.

This appears during testing of region lookup after a failure to assemble a BIOS defined region or if the lookup raced with the assembly of the BIOS defined region.

Failure to clean up BIOS defined regions that fail assembly is an issue in itself and a fix to that problem will alleviate some of the impact. It will not alleviate the race condition so let's harden this path.

The behavior change is that the kernel oops due to a null pointer dereference is replaced with a dev_dbg() message noting that an endpoint was mapped.

Additional comments are added so that future users of this function can more clearly understand what it provides.

Fixes 0a105ab28a4d ("cxl/memdev: Warn of poison inject or clear to a mapped region")
---
 drivers/cxl/core/region.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)
drivers/cxl/core/region.c +15 -4
index 3c2b6144be23..88051bb673bf 100644
@@ -2688,22 +2688,33 @@ static int __cxl_dpa_to_region(struct device *dev, void *arg)
 {
 	struct cxl_dpa_to_region_context *ctx = arg;
 	struct cxl_endpoint_decoder *cxled;
+	struct cxl_region *cxlr;
 	u64 dpa = ctx->dpa;
 
 	if (!is_endpoint_decoder(dev))
 		return 0;
 
 	cxled = to_cxl_endpoint_decoder(dev);
-	if (!cxled->dpa_res || !resource_size(cxled->dpa_res))
+	if (!cxled || !cxled->dpa_res || !resource_size(cxled->dpa_res))
 		return 0;
 
 	if (dpa > cxled->dpa_res->end || dpa < cxled->dpa_res->start)
 		return 0;
 
-	dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
-		dev_name(&cxled->cxld.region->dev));
+	/*
+	 * Stop the region search (return 1) when an endpoint mapping is
+	 * found. The region may not be fully constructed so offering
+	 * the cxlr in the context structure is not guaranteed.
+	 */
+	cxlr = cxled->cxld.region;
+	if (cxlr)
+		dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
+			dev_name(&cxlr->dev));
+	else
+		dev_dbg(dev, "dpa:0x%llx mapped in endpoint:%s\n", dpa,
+			dev_name(dev));
 
-	ctx->cxlr = cxled->cxld.region;
+	ctx->cxlr = cxlr;
 
 	return 1;
 }

base-commit: 49ba7b515c4c0719b866d16f068e62d16a8a3dd1

Signature · 1 line
2.37.3
Reviewed-by on v1 1/1
#1 lore raw
Quoted 20 lines · On Mon, 3 Jun 2024 17:36:09 -0700 [email protected] wrote:

From: Alison Schofield <[email protected]>

cxl_dpa_to_region() looks up a region based on a memdev and DPA. It wrongly assumes an endpoint found mapping the DPA is also of a fully assembled region. When not true it leads to a null pointer dereference looking up the region name.

This appears during testing of region lookup after a failure to assemble a BIOS defined region or if the lookup raced with the assembly of the BIOS defined region.

Failure to clean up BIOS defined regions that fail assembly is an issue in itself and a fix to that problem will alleviate some of the impact. It will not alleviate the race condition so let's harden this path.

The behavior change is that the kernel oops due to a null pointer dereference is replaced with a dev_dbg() message noting that an endpoint was mapped.

Quoted 5 lines

Additional comments are added so that future users of this function can more clearly understand what it provides.

Fixes 0a105ab28a4d ("cxl/memdev: Warn of poison inject or clear to a mapped region")

Nasty corner case. Fix looks correct to me.

Quoted 48 lines
---
 drivers/cxl/core/region.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)
drivers/cxl/core/region.c +15 -4
index 3c2b6144be23..88051bb673bf 100644
@@ -2688,22 +2688,33 @@ static int __cxl_dpa_to_region(struct device *dev, void *arg)
 {
 	struct cxl_dpa_to_region_context *ctx = arg;
 	struct cxl_endpoint_decoder *cxled;
+	struct cxl_region *cxlr;
 	u64 dpa = ctx->dpa;
 
 	if (!is_endpoint_decoder(dev))
 		return 0;
 
 	cxled = to_cxl_endpoint_decoder(dev);
-	if (!cxled->dpa_res || !resource_size(cxled->dpa_res))
+	if (!cxled || !cxled->dpa_res || !resource_size(cxled->dpa_res))
 		return 0;
 
 	if (dpa > cxled->dpa_res->end || dpa < cxled->dpa_res->start)
 		return 0;
 
-	dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
-		dev_name(&cxled->cxld.region->dev));
+	/*
+	 * Stop the region search (return 1) when an endpoint mapping is
+	 * found. The region may not be fully constructed so offering
+	 * the cxlr in the context structure is not guaranteed.
+	 */
+	cxlr = cxled->cxld.region;
+	if (cxlr)
+		dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
+			dev_name(&cxlr->dev));
+	else
+		dev_dbg(dev, "dpa:0x%llx mapped in endpoint:%s\n", dpa,
+			dev_name(dev));
 
-	ctx->cxlr = cxled->cxld.region;
+	ctx->cxlr = cxlr;
 
 	return 1;
 }

base-commit: 49ba7b515c4c0719b866d16f068e62d16a8a3dd1

#2 lore raw
In reply to [email protected]

Perhaps change the subject with: s/Avoid/Fix/?

Quoted 6 lines · alison.schofield@ wrote:

From: Alison Schofield <[email protected]>

cxl_dpa_to_region() looks up a region based on a memdev and DPA. It wrongly assumes an endpoint found mapping the DPA is also of a fully assembled region. When not true it leads to a null pointer dereference looking up the region name.

Can you place the a snippet of the Call Trace here, that helps support that the writeup came to the right conclusion, and it helps folks hitting this error to find this patch in a web search.

Quoted 29 lines

This appears during testing of region lookup after a failure to assemble a BIOS defined region or if the lookup raced with the assembly of the BIOS defined region.

Failure to clean up BIOS defined regions that fail assembly is an issue in itself and a fix to that problem will alleviate some of the impact. It will not alleviate the race condition so let's harden this path.

The behavior change is that the kernel oops due to a null pointer dereference is replaced with a dev_dbg() message noting that an endpoint was mapped.

Additional comments are added so that future users of this function can more clearly understand what it provides.

Fixes 0a105ab28a4d ("cxl/memdev: Warn of poison inject or clear to a mapped region")
---
 drivers/cxl/core/region.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)
drivers/cxl/core/region.c +0 -0
index 3c2b6144be23..88051bb673bf 100644
@@ -2688,22 +2688,33 @@ static int __cxl_dpa_to_region(struct device *dev, void *arg)
 {
 	struct cxl_dpa_to_region_context *ctx = arg;

drivers/cxl/core/region.c static int __cxl_dpa_to_region(struct device *dev, void *arg) · line 2698

Quoted 10 lines
drivers/cxl/core/region.c +2 -1
 	struct cxl_endpoint_decoder *cxled;
+	struct cxl_region *cxlr;
 	u64 dpa = ctx->dpa;
 
 	if (!is_endpoint_decoder(dev))
 		return 0;
 
 	cxled = to_cxl_endpoint_decoder(dev);
-	if (!cxled->dpa_res || !resource_size(cxled->dpa_res))
+	if (!cxled || !cxled->dpa_res || !resource_size(cxled->dpa_res))

Why add the !cxled check? to_cxl_endpoint_decoder() only returns NULL if someone messes up and sends a !is_endpoint_decoder(@dev) to this conversion routine and that check was already performed above.

drivers/cxl/core/region.c static int __cxl_dpa_to_region(struct device *dev, void *arg) · line 2708

Quoted 12 lines
drivers/cxl/core/region.c +5 -2
 		return 0;
 
 	if (dpa > cxled->dpa_res->end || dpa < cxled->dpa_res->start)
 		return 0;
 
-	dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
-		dev_name(&cxled->cxld.region->dev));
+	/*
+	 * Stop the region search (return 1) when an endpoint mapping is
+	 * found. The region may not be fully constructed so offering
+	 * the cxlr in the context structure is not guaranteed.
+	 */

Comment looks good...

drivers/cxl/core/region.c static int __cxl_dpa_to_region(struct device *dev, void *arg) · line 2715

Quoted 7 lines
drivers/cxl/core/region.c +7 -0
+	cxlr = cxled->cxld.region;
+	if (cxlr)
+		dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
+			dev_name(&cxlr->dev));
+	else
+		dev_dbg(dev, "dpa:0x%llx mapped in endpoint:%s\n", dpa,
+			dev_name(dev));

...but I am not sure gymanistics to improve a debug message are worth it, especially because all of the callers either emit the found region in a tracepoint or print a warning, so I would just be in favor of deleting the debug message altogether, or at most just trimming to the "mapped in endpoint" one.

author maintainer
#3 lore raw
Quoted 1 line · Dan Williams

Perhaps change the subject with: s/Avoid/Fix/?

OK

Quoted 4 lines
Quoted 2 lines · alison.schofield@ wrote:

From: Alison Schofield <[email protected]>

Quoted 8 lines
Quoted 4 lines

cxl_dpa_to_region() looks up a region based on a memdev and DPA. It wrongly assumes an endpoint found mapping the DPA is also of a fully assembled region. When not true it leads to a null pointer dereference looking up the region name.

Can you place the a snippet of the Call Trace here, that helps support that the writeup came to the right conclusion, and it helps folks hitting this error to find this patch in a web search.

Will add in v2

snip...

Quoted 15 lines
Quoted 15 lines
drivers/cxl/core/region.c +1 -0
index 3c2b6144be23..88051bb673bf 100644
@@ -2688,22 +2688,33 @@ static int __cxl_dpa_to_region(struct device *dev, void *arg)
 {
 	struct cxl_dpa_to_region_context *ctx = arg;
 	struct cxl_endpoint_decoder *cxled;
+	struct cxl_region *cxlr;
 	u64 dpa = ctx->dpa;
 
 	if (!is_endpoint_decoder(dev))
 		return 0;
 
Quoted 8 lines

drivers/cxl/core/region.c static int __cxl_dpa_to_region(struct device *dev, void *arg) · line 2698

Quoted 3 lines
drivers/cxl/core/region.c +1 -1
 	cxled = to_cxl_endpoint_decoder(dev);
-	if (!cxled->dpa_res || !resource_size(cxled->dpa_res))
+	if (!cxled || !cxled->dpa_res || !resource_size(cxled->dpa_res))

Why add the !cxled check? to_cxl_endpoint_decoder() only returns NULL if someone messes up and sends a !is_endpoint_decoder(@dev) to this conversion routine and that check was already performed above.

Abundance of caution I suppose. I'll remove.

Quoted 15 lines

drivers/cxl/core/region.c static int __cxl_dpa_to_region(struct device *dev, void *arg) · line 2708

Quoted 12 lines
drivers/cxl/core/region.c +5 -2
 		return 0;
 
 	if (dpa > cxled->dpa_res->end || dpa < cxled->dpa_res->start)
 		return 0;
 
-	dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
-		dev_name(&cxled->cxld.region->dev));
+	/*
+	 * Stop the region search (return 1) when an endpoint mapping is
+	 * found. The region may not be fully constructed so offering
+	 * the cxlr in the context structure is not guaranteed.
+	 */

Comment looks good...

Quoted 13 lines

drivers/cxl/core/region.c static int __cxl_dpa_to_region(struct device *dev, void *arg) · line 2715

Quoted 7 lines
drivers/cxl/core/region.c +7 -0
+	cxlr = cxled->cxld.region;
+	if (cxlr)
+		dev_dbg(dev, "dpa:0x%llx mapped in region:%s\n", dpa,
+			dev_name(&cxlr->dev));
+	else
+		dev_dbg(dev, "dpa:0x%llx mapped in endpoint:%s\n", dpa,
+			dev_name(dev));

...but I am not sure gymanistics to improve a debug message are worth it, especially because all of the callers either emit the found region in a tracepoint or print a warning, so I would just be in favor of deleting the debug message altogether, or at most just trimming to the "mapped in endpoint" one.

Yeah, probably went overboard here. Previously was offering the name of the region so wanted to keep same - but seeing as this is a debug only message, I think the endpoint alone, as you suggest is good.

Thanks for the review,
Alison