Skip to contentSkip to conversation

cxl/mbox: Do not allow immediate mode in SET_PARTITION_INFO

PATCH In review [email protected] 3 participants 6 messages Jan 3, 2022 – Jan 5, 2022 CXL

lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/1 cxl/mbox: Do not allow immediate mode in SET_PARTITION_INFO

From: Alison Schofield <[email protected]>

User space may send the SET_PARTITION_INFO mailbox command using the IOCTL interface. Inspect the input payload and fail if the immediate flag is set.

This is the first instance of the driver inspecting an input payload from user space. Assume there will be more such cases and implement with an extensible helper.

Note: At this time immediate partitioning is not allowed because the kernel will need to react immediately to this configuration change and that support is not yet implemented.

---
 drivers/cxl/core/mbox.c | 43 +++++++++++++++++++++++++++++++++++++++++
 drivers/cxl/cxlmem.h    |  7 +++++++
 2 files changed, 50 insertions(+)
drivers/cxl/core/mbox.c +43 -0
index be61a0d8016b..2cf5ccdea7df 100644
@@ -352,6 +352,41 @@ int cxl_query_cmd(struct cxl_memdev *cxlmd,
 	return 0;
 }
 
+/**
+ * cxl_payload_from_user_allowed() - Check contents of in_payload.
+ * @opcode: The mailbox command opcode.
+ * @payload_in: Pointer to the input payload passed in from user space.
+ *
+ * Return:
+ *  * true	- payload_in passes check for @opcode.
+ *  * false	- payload_in contains invalid or unsupported values.
+ *
+ * The driver may inspect payload contents before sending a mailbox
+ * command from user space to the device. The intent is to reject
+ * commands with input payloads that are known to be unsafe. This
+ * check is not intended to replace the users careful selection of
+ * mailbox command parameters and makes no guarantee that the user
+ * command will succeed, nor that it is appropriate.
+ *
+ * The specific checks are determined by the opcode.
+ */
+static bool cxl_payload_from_user_allowed(u16 opcode, void *payload_in)
+{
+	switch (opcode) {
+	case CXL_MBOX_OP_SET_PARTITION_INFO: {
+		struct cxl_mbox_set_partition_info *pi;
+
+		pi = (struct cxl_mbox_set_partition_info *)payload_in;
+		if (pi->flags && CXL_SET_PARTITION_IMMEDIATE_FLAG)
+			return false;
+		break;
+	}
+	default:
+		break;
+	}
+	return true;
+}
+
 /**
  * handle_mailbox_cmd_from_user() - Dispatch a mailbox command for userspace.
  * @cxlds: The device data for the operation
@@ -405,6 +440,14 @@ static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds,
 		}
 	}
 
+	if (!cxl_payload_from_user_allowed(mbox_cmd.opcode,
+					   mbox_cmd.payload_in)) {
+		dev_dbg(dev, "%s: input payload not allowed\n",
+			cxl_command_names[cmd->info.id].name);
+		rc = -EINVAL;
+		goto out;
+	}
+
 	dev_dbg(dev,
 		"Submitting %s command for user\n"
 		"\topcode: %x\n"
drivers/cxl/cxlmem.h +7 -0
index 8d96d009ad90..e10b86f06c75 100644
@@ -231,6 +231,13 @@ struct cxl_mbox_set_lsa {
 	u8 data[];
 } __packed;
 
+struct cxl_mbox_set_partition_info {
+	u64 volatile_capacity;
+	u8 flags;
+} __packed;
+
+#define  CXL_SET_PARTITION_IMMEDIATE_FLAG	BIT(0)
+
 /**
  * struct cxl_mem_command - Driver representation of a memory device command
  * @info: Command information as it exists for the UAPI

base-commit: 53989fad1286e652ea3655ae3367ba698da8d2ff

Signature · 1 line
2.31.1
#1 lore raw
Quoted 7 lines · On Mon, Jan 3, 2022 at 12:15 PM <[email protected]> wrote:

From: Alison Schofield <[email protected]>

User space may send the SET_PARTITION_INFO mailbox command using the IOCTL interface. Inspect the input payload and fail if the immediate flag is set.

Quoted 7 lines

This is the first instance of the driver inspecting an input payload from user space. Assume there will be more such cases and implement with an extensible helper.

Note: At this time immediate partitioning is not allowed because the kernel will need to react immediately to this configuration change and that support is not yet implemented.

I would say:

In order for the kernel to react to an immediate partition change it needs to assert that the change will not affect any active decode. At a minimum this requires validating that the device is using HDM decoders instead of the CXL DVSEC for decode, and that none of the active HDM decoders are affected by the partition change. For now, just fail until that support arrives.

Quoted 30 lines
---
 drivers/cxl/core/mbox.c | 43 +++++++++++++++++++++++++++++++++++++++++
 drivers/cxl/cxlmem.h    |  7 +++++++
 2 files changed, 50 insertions(+)
drivers/cxl/core/mbox.c +15 -0
index be61a0d8016b..2cf5ccdea7df 100644
@@ -352,6 +352,41 @@ int cxl_query_cmd(struct cxl_memdev *cxlmd,
        return 0;
 }
 
+/**
+ * cxl_payload_from_user_allowed() - Check contents of in_payload.
+ * @opcode: The mailbox command opcode.
+ * @payload_in: Pointer to the input payload passed in from user space.
+ *
+ * Return:
+ *  * true     - payload_in passes check for @opcode.
+ *  * false    - payload_in contains invalid or unsupported values.
+ *
+ * The driver may inspect payload contents before sending a mailbox
+ * command from user space to the device. The intent is to reject
+ * commands with input payloads that are known to be unsafe. This
+ * check is not intended to replace the users careful selection of
+ * mailbox command parameters and makes no guarantee that the user
+ * command will succeed, nor that it is appropriate.

drivers/cxl/core/mbox.c int cxl_query_cmd(struct cxl_memdev *cxlmd, · line 379

Quoted 10 lines
drivers/cxl/core/mbox.c +10 -0
+ *
+ * The specific checks are determined by the opcode.
+ */
+static bool cxl_payload_from_user_allowed(u16 opcode, void *payload_in)
+{
+       switch (opcode) {
+       case CXL_MBOX_OP_SET_PARTITION_INFO: {
+               struct cxl_mbox_set_partition_info *pi;
+
+               pi = (struct cxl_mbox_set_partition_info *)payload_in;

No need for a cast since @payload_in is a 'void *'

Quoted 33 lines
drivers/cxl/core/mbox.c +18 -0
+               if (pi->flags && CXL_SET_PARTITION_IMMEDIATE_FLAG)
+                       return false;
+               break;
+       }
+       default:
+               break;
+       }
+       return true;
+}
+
 /**
  * handle_mailbox_cmd_from_user() - Dispatch a mailbox command for userspace.
  * @cxlds: The device data for the operation
@@ -405,6 +440,14 @@ static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds,
                }
        }
 
+       if (!cxl_payload_from_user_allowed(mbox_cmd.opcode,
+                                          mbox_cmd.payload_in)) {
+               dev_dbg(dev, "%s: input payload not allowed\n",
+                       cxl_command_names[cmd->info.id].name);
+               rc = -EINVAL;
+               goto out;
+       }
+
        dev_dbg(dev,
                "Submitting %s command for user\n"
                "\topcode: %x\n"
drivers/cxl/cxlmem.h +0 -0
index 8d96d009ad90..e10b86f06c75 100644

drivers/cxl/cxlmem.h struct cxl_mbox_set_lsa { · line 235

Quoted 5 lines
drivers/cxl/cxlmem.h +2 -0
@@ -231,6 +231,13 @@ struct cxl_mbox_set_lsa {
        u8 data[];
 } __packed;
 
+struct cxl_mbox_set_partition_info {
+       u64 volatile_capacity;

This should be __le64, and it looks like 'struct cxl_mbox_get_lsa' and 'struct cxl_mbox_set_lsa' need similar fixups.

Quoted 13 lines
drivers/cxl/cxlmem.h +5 -0
+       u8 flags;
+} __packed;
+
+#define  CXL_SET_PARTITION_IMMEDIATE_FLAG      BIT(0)
+
 /**
  * struct cxl_mem_command - Driver representation of a memory device command
  * @info: Command information as it exists for the UAPI

base-commit: 53989fad1286e652ea3655ae3367ba698da8d2ff --
2.31.1

author maintainer
#2 lore raw

Thanks for the review...

Quoted 16 lines · Dan Williams
Quoted 7 lines · On Mon, Jan 3, 2022 at 12:15 PM <[email protected]> wrote:

From: Alison Schofield <[email protected]>

User space may send the SET_PARTITION_INFO mailbox command using the IOCTL interface. Inspect the input payload and fail if the immediate flag is set.

Quoted 7 lines

This is the first instance of the driver inspecting an input payload from user space. Assume there will be more such cases and implement with an extensible helper.

Note: At this time immediate partitioning is not allowed because the kernel will need to react immediately to this configuration change and that support is not yet implemented.

Quoted 9 lines

I would say:

In order for the kernel to react to an immediate partition change it needs to assert that the change will not affect any active decode. At a minimum this requires validating that the device is using HDM decoders instead of the CXL DVSEC for decode, and that none of the active HDM decoders are affected by the partition change. For now, just fail until that support arrives.

Got it!

Quoted 1 line
Quoted 1 line

snip

Quoted 10 lines

drivers/cxl/core/mbox.c int cxl_query_cmd(struct cxl_memdev *cxlmd, · line 379

Quoted 7 lines
drivers/cxl/core/mbox.c +7 -0
+static bool cxl_payload_from_user_allowed(u16 opcode, void *payload_in)
+{
+       switch (opcode) {
+       case CXL_MBOX_OP_SET_PARTITION_INFO: {
+               struct cxl_mbox_set_partition_info *pi;
+
+               pi = (struct cxl_mbox_set_partition_info *)payload_in;

No need for a cast since @payload_in is a 'void *'

Got it!

Quoted 10 lines
Quoted 10 lines
drivers/cxl/core/mbox.c +10 -0
+               if (pi->flags && CXL_SET_PARTITION_IMMEDIATE_FLAG)
+                       return false;
+               break;
+       }
+       default:
+               break;
+       }
+       return true;
+}
+

snip

Quoted 8 lines
Quoted 8 lines
drivers/cxl/cxlmem.h +0 -0
index 8d96d009ad90..e10b86f06c75 100644
@@ -231,6 +231,13 @@ struct cxl_mbox_set_lsa {
        u8 data[];
 } __packed;
 
Quoted 6 lines

drivers/cxl/cxlmem.h struct cxl_mbox_set_lsa { · line 235

Quoted 2 lines
drivers/cxl/cxlmem.h +2 -0
+struct cxl_mbox_set_partition_info {
+       u64 volatile_capacity;

This should be __le64, and it looks like 'struct cxl_mbox_get_lsa' and 'struct cxl_mbox_set_lsa' need similar fixups.

Got it and will do!

Quoted 13 lines
Quoted 13 lines
drivers/cxl/cxlmem.h +5 -0
+       u8 flags;
+} __packed;
+
+#define  CXL_SET_PARTITION_IMMEDIATE_FLAG      BIT(0)
+
 /**
  * struct cxl_mem_command - Driver representation of a memory device command
  * @info: Command information as it exists for the UAPI

base-commit: 53989fad1286e652ea3655ae3367ba698da8d2ff --
2.31.1

#3 lore raw
In reply to [email protected]
Quoted 9 lines · On 22-01-03 12:21:00, [email protected] wrote:

From: Alison Schofield <[email protected]>

User space may send the SET_PARTITION_INFO mailbox command using the IOCTL interface. Inspect the input payload and fail if the immediate flag is set.

This is the first instance of the driver inspecting an input payload from user space. Assume there will be more such cases and implement with an extensible helper.

Not sure if it's useful, but this was implemented at some point:
https://lore.kernel.org/linux-cxl/[email protected]/

Quoted 59 lines

Note: At this time immediate partitioning is not allowed because the kernel will need to react immediately to this configuration change and that support is not yet implemented.

---
 drivers/cxl/core/mbox.c | 43 +++++++++++++++++++++++++++++++++++++++++
 drivers/cxl/cxlmem.h    |  7 +++++++
 2 files changed, 50 insertions(+)
drivers/cxl/core/mbox.c +35 -0
index be61a0d8016b..2cf5ccdea7df 100644
@@ -352,6 +352,41 @@ int cxl_query_cmd(struct cxl_memdev *cxlmd,
 	return 0;
 }
 
+/**
+ * cxl_payload_from_user_allowed() - Check contents of in_payload.
+ * @opcode: The mailbox command opcode.
+ * @payload_in: Pointer to the input payload passed in from user space.
+ *
+ * Return:
+ *  * true	- payload_in passes check for @opcode.
+ *  * false	- payload_in contains invalid or unsupported values.
+ *
+ * The driver may inspect payload contents before sending a mailbox
+ * command from user space to the device. The intent is to reject
+ * commands with input payloads that are known to be unsafe. This
+ * check is not intended to replace the users careful selection of
+ * mailbox command parameters and makes no guarantee that the user
+ * command will succeed, nor that it is appropriate.
+ *
+ * The specific checks are determined by the opcode.
+ */
+static bool cxl_payload_from_user_allowed(u16 opcode, void *payload_in)
+{
+	switch (opcode) {
+	case CXL_MBOX_OP_SET_PARTITION_INFO: {
+		struct cxl_mbox_set_partition_info *pi;
+
+		pi = (struct cxl_mbox_set_partition_info *)payload_in;
+		if (pi->flags && CXL_SET_PARTITION_IMMEDIATE_FLAG)
+			return false;
+		break;
+	}
+	default:
+		break;
+	}
+	return true;
+}
+
 /**
  * handle_mailbox_cmd_from_user() - Dispatch a mailbox command for userspace.
  * @cxlds: The device data for the operation
@@ -405,6 +440,14 @@ static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds,
 		}

drivers/cxl/core/mbox.c static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds, · line 450

Quoted 10 lines
drivers/cxl/core/mbox.c +8 -0
 	}
 
+	if (!cxl_payload_from_user_allowed(mbox_cmd.opcode,
+					   mbox_cmd.payload_in)) {
+		dev_dbg(dev, "%s: input payload not allowed\n",
+			cxl_command_names[cmd->info.id].name);
+		rc = -EINVAL;
+		goto out;
+	}
+

Perhaps foolishly, the kdocs for handle_mailbox_cmd_from_user() documents the error conditions. Would you mind adding EINVAL?

Also, cxl_validate_cmd_from_user() was supposed to handle this kind of stuff. All validation from user commands should spawn from that. Is there some reason this one is different?

Quoted 8 lines
drivers/cxl/core/mbox.c +0 -0
 	dev_dbg(dev,
 		"Submitting %s command for user\n"
 		"\topcode: %x\n"
drivers/cxl/cxlmem.h +0 -0
index 8d96d009ad90..e10b86f06c75 100644

drivers/cxl/cxlmem.h struct cxl_mbox_set_lsa { · line 240

Quoted 10 lines
drivers/cxl/cxlmem.h +7 -0
@@ -231,6 +231,13 @@ struct cxl_mbox_set_lsa {
 	u8 data[];
 } __packed;
 
+struct cxl_mbox_set_partition_info {
+	u64 volatile_capacity;
+	u8 flags;
+} __packed;
+
+#define  CXL_SET_PARTITION_IMMEDIATE_FLAG	BIT(0)
+

I think these defines belong in cxl.h

Quoted 8 lines
drivers/cxl/cxlmem.h +0 -0
 /**
  * struct cxl_mem_command - Driver representation of a memory device command
  * @info: Command information as it exists for the UAPI

base-commit: 53989fad1286e652ea3655ae3367ba698da8d2ff

Signature · 1 line
2.31.1
author maintainer
#4 lore raw

Thanks for the review Ben...

Quoted 7 lines · Ben Widawsky
Quoted 6 lines · On 22-01-03 12:21:00, [email protected] wrote:

From: Alison Schofield <[email protected]>

User space may send the SET_PARTITION_INFO mailbox command using the IOCTL interface. Inspect the input payload and fail if the immediate flag is set.

Quoted 6 lines
Quoted 3 lines

This is the first instance of the driver inspecting an input payload from user space. Assume there will be more such cases and implement with an extensible helper.

Not sure if it's useful, but this was implemented at some point:
https://lore.kernel.org/linux-cxl/[email protected]/

Thanks. Looking.

Quoted 1 line

snip

Quoted 4 lines
Quoted 4 lines
(quoted diff) +0 -0
@@ -405,6 +440,14 @@ static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds,
 		}
 	}
 
Quoted 11 lines

drivers/cxl/core/mbox.c static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds, · line 450

Quoted 8 lines
drivers/cxl/core/mbox.c +8 -0
+	if (!cxl_payload_from_user_allowed(mbox_cmd.opcode,
+					   mbox_cmd.payload_in)) {
+		dev_dbg(dev, "%s: input payload not allowed\n",
+			cxl_command_names[cmd->info.id].name);
+		rc = -EINVAL;
+		goto out;
+	}
+

Perhaps foolishly, the kdocs for handle_mailbox_cmd_from_user() documents the error conditions. Would you mind adding EINVAL?

I'm going to try to mv the allowable check, so this will go away.

Quoted 4 lines

Also, cxl_validate_cmd_from_user() was supposed to handle this kind of stuff. All validation from user commands should spawn from that. Is there some reason this one is different?

The existing cxl_validate_cmd_from_user() happens before we copy the payload from user - so this payload check didn't fit in there. Let me look at reorganizing a bit so validating the cmd and it's payload can both spawn from cxl_validate_cmd_from_user().

Quoted 1 line

snip

Quoted 6 lines

drivers/cxl/cxlmem.h struct cxl_mbox_set_lsa { · line 240

Quoted 3 lines
drivers/cxl/cxlmem.h +2 -0
+#define  CXL_SET_PARTITION_IMMEDIATE_FLAG	BIT(0)
+

I think these defines belong in cxl.h

It seems this is OK in cxl.h where other CXL spec defines now live.

Quoted 8 lines
Quoted 8 lines

/**
* struct cxl_mem_command - Driver representation of a memory device command
* @info: Command information as it exists for the UAPI

base-commit: 53989fad1286e652ea3655ae3367ba698da8d2ff

Signature · 1 line
2.31.1
#5 lore raw
Quoted 39 lines · Alison Schofield

Thanks for the review Ben...

Quoted 7 lines · Ben Widawsky
Quoted 6 lines · On 22-01-03 12:21:00, [email protected] wrote:

From: Alison Schofield <[email protected]>

User space may send the SET_PARTITION_INFO mailbox command using the IOCTL interface. Inspect the input payload and fail if the immediate flag is set.

Quoted 6 lines
Quoted 3 lines

This is the first instance of the driver inspecting an input payload from user space. Assume there will be more such cases and implement with an extensible helper.

Not sure if it's useful, but this was implemented at some point:
https://lore.kernel.org/linux-cxl/[email protected]/

Thanks. Looking.

Quoted 1 line

snip

Quoted 4 lines
Quoted 4 lines
(quoted diff) +0 -0
@@ -405,6 +440,14 @@ static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds,
            }
    }
 
Quoted 11 lines

drivers/cxl/core/mbox.c static int handle_mailbox_cmd_from_user(struct cxl_dev_state *cxlds, · line 450

Quoted 8 lines
drivers/cxl/core/mbox.c +8 -0
+   if (!cxl_payload_from_user_allowed(mbox_cmd.opcode,
+                                      mbox_cmd.payload_in)) {
+           dev_dbg(dev, "%s: input payload not allowed\n",
+                   cxl_command_names[cmd->info.id].name);
+           rc = -EINVAL;
+           goto out;
+   }
+

Perhaps foolishly, the kdocs for handle_mailbox_cmd_from_user() documents the error conditions. Would you mind adding EINVAL?

Quoted 1 line

I'm going to try to mv the allowable check, so this will go away.

Wouldn't the error code be EBUSY just like the other commands the kernel blocks based on the state of the device?

It's moot if this is moving.

Quoted 10 lines
Quoted 5 lines

Also, cxl_validate_cmd_from_user() was supposed to handle this kind of stuff. All validation from user commands should spawn from that. Is there some reason this one is different?

The existing cxl_validate_cmd_from_user() happens before we copy the payload from user - so this payload check didn't fit in there. Let me look at reorganizing a bit so validating the cmd and it's payload can both spawn from cxl_validate_cmd_from_user().

Having cxl_validate_cmd_from_user() setup the 'struct cxl_mbox_cmd' and then forward that along to handle_mailbox_cmd_from_user() doesn't sound too bad. It does mean we need to remember that if the kernel grows internal paths for calling commands that might be disallowed by the state of the device we might need to move this validation out of the "from user" context.