[v2] block: add blktrace support for zoned block device commands

[PATCH v2 09/15] blktrace: add definitions for struct blk_io_trace2

Posted by Johannes Thumshirn 4 months, 2 weeks ago

Add definitions for the extended version of the blktrace protocol using a
wider action type to be able to record new actions in the kernel.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
 include/uapi/linux/blktrace_api.h | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/include/uapi/linux/blktrace_api.h b/include/uapi/linux/blktrace_api.h
index a3b1f35ac026..d58ef484de49 100644
--- a/include/uapi/linux/blktrace_api.h
+++ b/include/uapi/linux/blktrace_api.h
@@ -94,6 +94,7 @@ enum blktrace_notify {
 
 #define BLK_IO_TRACE_MAGIC	0x65617400
 #define BLK_IO_TRACE_VERSION	0x07
+#define BLK_IO_TRACE2_VERSION	0x08
 
 /*
  * The trace itself
@@ -113,6 +114,21 @@ struct blk_io_trace {
 	/* cgroup id will be stored here if exists */
 };
 
+struct blk_io_trace2 {
+	__u32 magic;		/* MAGIC << 8 | BLK_IO_TRACE2_VERSION */
+	__u32 sequence;		/* event number */
+	__u64 time;		/* in nanoseconds */
+	__u64 sector;		/* disk offset */
+	__u32 bytes;		/* transfer length */
+	__u32 pid;		/* who did it */
+	__u64 action;		/* what happened */
+	__u32 device;		/* device number */
+	__u32 cpu;		/* on what cpu did it happen */
+	__u16 error;		/* completion error */
+	__u16 pdu_len;		/* length of data after this trace */
+	__u8 pad[8];
+	/* cgroup id will be stored here if exists */
+};
 /*
  * The remap event
  */
-- 
2.51.0

Re: [PATCH v2 09/15] blktrace: add definitions for struct blk_io_trace2

Posted by Christoph Hellwig 4 months, 1 week ago

On Thu, Sep 25, 2025 at 05:02:25PM +0200, Johannes Thumshirn wrote:
> Add definitions for the extended version of the blktrace protocol using a
> wider action type to be able to record new actions in the kernel.
> 
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
>  include/uapi/linux/blktrace_api.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/include/uapi/linux/blktrace_api.h b/include/uapi/linux/blktrace_api.h
> index a3b1f35ac026..d58ef484de49 100644
> --- a/include/uapi/linux/blktrace_api.h
> +++ b/include/uapi/linux/blktrace_api.h
> @@ -94,6 +94,7 @@ enum blktrace_notify {
>  
>  #define BLK_IO_TRACE_MAGIC	0x65617400
>  #define BLK_IO_TRACE_VERSION	0x07
> +#define BLK_IO_TRACE2_VERSION	0x08
>  
>  /*
>   * The trace itself
> @@ -113,6 +114,21 @@ struct blk_io_trace {
>  	/* cgroup id will be stored here if exists */
>  };
>  
> +struct blk_io_trace2 {
> +	__u32 magic;		/* MAGIC << 8 | BLK_IO_TRACE2_VERSION */
> +	__u32 sequence;		/* event number */
> +	__u64 time;		/* in nanoseconds */
> +	__u64 sector;		/* disk offset */
> +	__u32 bytes;		/* transfer length */
> +	__u32 pid;		/* who did it */
> +	__u64 action;		/* what happened */
> +	__u32 device;		/* device number */
> +	__u32 cpu;		/* on what cpu did it happen */
> +	__u16 error;		/* completion error */
> +	__u16 pdu_len;		/* length of data after this trace */
> +	__u8 pad[8];

This will cause mismatching sizes between x86_32 and other
architectures because the size is not 8-byte aligned.  You'll need
to add or remove 4 bytes of padding to ensure that.

I really wish we could have good helpers to check for that.

Re: [PATCH v2 09/15] blktrace: add definitions for struct blk_io_trace2

Posted by Damien Le Moal 4 months, 1 week ago

On 9/26/25 00:02, Johannes Thumshirn wrote:
> Add definitions for the extended version of the blktrace protocol using a
> wider action type to be able to record new actions in the kernel.
> 
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
>  include/uapi/linux/blktrace_api.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/include/uapi/linux/blktrace_api.h b/include/uapi/linux/blktrace_api.h
> index a3b1f35ac026..d58ef484de49 100644
> --- a/include/uapi/linux/blktrace_api.h
> +++ b/include/uapi/linux/blktrace_api.h
> @@ -94,6 +94,7 @@ enum blktrace_notify {
>  
>  #define BLK_IO_TRACE_MAGIC	0x65617400
>  #define BLK_IO_TRACE_VERSION	0x07
> +#define BLK_IO_TRACE2_VERSION	0x08
>  
>  /*
>   * The trace itself
> @@ -113,6 +114,21 @@ struct blk_io_trace {
>  	/* cgroup id will be stored here if exists */
>  };
>  
> +struct blk_io_trace2 {
> +	__u32 magic;		/* MAGIC << 8 | BLK_IO_TRACE2_VERSION */
> +	__u32 sequence;		/* event number */
> +	__u64 time;		/* in nanoseconds */
> +	__u64 sector;		/* disk offset */
> +	__u32 bytes;		/* transfer length */
> +	__u32 pid;		/* who did it */
> +	__u64 action;		/* what happened */
> +	__u32 device;		/* device number */
> +	__u32 cpu;		/* on what cpu did it happen */
> +	__u16 error;		/* completion error */
> +	__u16 pdu_len;		/* length of data after this trace */
> +	__u8 pad[8];

Why 8 ? that makes the structure 4B 60 B. Padding to 12 would make it nicely
aligned to 64B...

> +	/* cgroup id will be stored here if exists */
> +};
>  /*
>   * The remap event
>   */


-- 
Damien Le Moal
Western Digital Research