Skip to contentSkip to conversation

Make dmi_name_in_vendors more focused

PATCH In review Jean Delvare 3 participants 6 messages Nov 4, 2011 – Nov 10, 2011 Dmi/smbios support

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Make dmi_name_in_vendors more focused

The current implementation of dmi_name_in_vendors() is an invitation to lazy coding and false positives [1]. Searching for a string in 8 different DMI fields is something nobody should ever need to do. You know what you're looking for, so you should know where to look. strstr isn't fast, especially when it fails, so we should avoid calling it when it just can't succeed.

Looking at the current users of the function, it seems clear to me that they are looking for a system or board vendor name, so let's limit dmi_name_in_vendors to these two DMI fields. This much better matches the function name, BTW.

[1] We currently have code looking for short names in DMI data, such as "IBM" or "ASUS". I let you guess what will happen the day other vendors ship products named, for example, "SCHREIBMEISTER" or "PEGASUS".

---
This patch was already sent on 2011-05-15. I thought Andrew had picked it up but apparently not, otherwise it should already be upstream by now.

 drivers/firmware/dmi_scan.c |    6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/firmware/dmi_scan.c +2 -4
@@ -585,14 +585,12 @@ int dmi_name_in_serial(const char *str)
 }
 
 /**
- *	dmi_name_in_vendors - Check if string is anywhere in the DMI vendor information.
+ *	dmi_name_in_vendors - Check if string is in the DMI system or board vendor name
  *	@str: 	Case sensitive Name
  */
 int dmi_name_in_vendors(const char *str)
 {
-	static int fields[] = { DMI_BIOS_VENDOR, DMI_BIOS_VERSION, DMI_SYS_VENDOR,
-				DMI_PRODUCT_NAME, DMI_PRODUCT_VERSION, DMI_BOARD_VENDOR,
-				DMI_BOARD_NAME, DMI_BOARD_VERSION, DMI_NONE };
+	static int fields[] = { DMI_SYS_VENDOR, DMI_BOARD_VENDOR, DMI_NONE };
 	int i;
 	for (i = 0; fields[i] != DMI_NONE; i++) {
 		int f = fields[i];
Signature · 1 line
Jean Delvare
#1 lore raw
Quoted 16 lines · On Fri, 4 Nov 2011 10:26:47 +0100 Jean Delvare <[email protected]> wrote:

The current implementation of dmi_name_in_vendors() is an invitation to lazy coding and false positives [1]. Searching for a string in 8 different DMI fields is something nobody should ever need to do. You know what you're looking for, so you should know where to look. strstr isn't fast, especially when it fails, so we should avoid calling it when it just can't succeed.

Looking at the current users of the function, it seems clear to me that they are looking for a system or board vendor name, so let's limit dmi_name_in_vendors to these two DMI fields. This much better matches the function name, BTW.

[1] We currently have code looking for short names in DMI data, such as "IBM" or "ASUS". I let you guess what will happen the day other vendors ship products named, for example, "SCHREIBMEISTER" or "PEGASUS".

Quoted 6 lines

---
This patch was already sent on 2011-05-15. I thought Andrew had picked it up but apparently not, otherwise it should already be upstream by now.

I did merge it in May and have carried it in -mm (and hence in linux-next) since then. I have sent it to Jesse at least twice, with no effect.

I'll send it to Linus for 3.3-rc1, OK?

#2 lore raw
Quoted 27 lines · On Wed, 9 Nov 2011 14:25:57 -0800 Andrew Morton <[email protected]>…
Quoted 17 lines · On Fri, 4 Nov 2011 10:26:47 +0100 Jean Delvare <[email protected]> wrote:

The current implementation of dmi_name_in_vendors() is an invitation to lazy coding and false positives [1]. Searching for a string in 8 different DMI fields is something nobody should ever need to do. You know what you're looking for, so you should know where to look. strstr isn't fast, especially when it fails, so we should avoid calling it when it just can't succeed.

Looking at the current users of the function, it seems clear to me that they are looking for a system or board vendor name, so let's limit dmi_name_in_vendors to these two DMI fields. This much better matches the function name, BTW.

[1] We currently have code looking for short names in DMI data, such as "IBM" or "ASUS". I let you guess what will happen the day other vendors ship products named, for example, "SCHREIBMEISTER" or "PEGASUS".

Quoted 6 lines

---
This patch was already sent on 2011-05-15. I thought Andrew had picked it up but apparently not, otherwise it should already be upstream by now.

Quoted 3 lines

I did merge it in May and have carried it in -mm (and hence in linux-next) since then. I have sent it to Jesse at least twice, with no effect.

I don't generally look after DMI stuff... I can queue it though if you want. My only worry is potential breakage for existing DMI matches, but it looks like Jean already did the audit for that.

Jesse

#3 lore raw
Quoted 9 lines · Jesse Barnes
Quoted 6 lines · On Wed, 9 Nov 2011 14:25:57 -0800 Andrew Morton <[email protected]>…
Quoted 3 lines · On Fri, 4 Nov 2011 10:26:47 +0100 Jean Delvare <[email protected]> wrote:

This patch was already sent on 2011-05-15. I thought Andrew had picked it up but apparently not, otherwise it should already be upstream by now.

Quoted 5 lines
Quoted 3 lines

I did merge it in May and have carried it in -mm (and hence in linux-next) since then. I have sent it to Jesse at least twice, with no effect.

I don't generally look after DMI stuff...

Indeed I'm not sure why Jesse would take this, it's not PCI-related in any way.

Quoted 3 lines

I can queue it though if you
want. My only worry is potential breakage for existing DMI matches, but it looks like Jean already did the audit for that.

Yes I did, both at the time of the original submission and again when resending.

Signature · 1 line
Jean Delvare
#4 lore raw
In reply to Andrew Morton
Quoted 26 lines · Andrew Morton
Quoted 16 lines · On Fri, 4 Nov 2011 10:26:47 +0100 Jean Delvare <[email protected]> wrote:

The current implementation of dmi_name_in_vendors() is an invitation to lazy coding and false positives [1]. Searching for a string in 8 different DMI fields is something nobody should ever need to do. You know what you're looking for, so you should know where to look. strstr isn't fast, especially when it fails, so we should avoid calling it when it just can't succeed.

Looking at the current users of the function, it seems clear to me that they are looking for a system or board vendor name, so let's limit dmi_name_in_vendors to these two DMI fields. This much better matches the function name, BTW.

[1] We currently have code looking for short names in DMI data, such as "IBM" or "ASUS". I let you guess what will happen the day other vendors ship products named, for example, "SCHREIBMEISTER" or "PEGASUS".

Quoted 6 lines

---
This patch was already sent on 2011-05-15. I thought Andrew had picked it up but apparently not, otherwise it should already be upstream by now.

Quoted 5 lines

I did merge it in May and have carried it in -mm (and hence in linux-next) since then. I have sent it to Jesse at least twice, with no effect.

I'll send it to Linus for 3.3-rc1, OK?

If you want, sure. Note however that it might be less risky to send it now, as I just audited the code. By 3.3-rc1, new callers might have appeared and nobody may remember to audit again. Committing this kind of change right after rc1 seems better than during the merge window, as it lets us ensure that no new use case sneaked it.

Signature · 1 line
Jean Delvare
#5 lore raw
Quoted 4 lines · On Thu, 10 Nov 2011 09:13:40 +0100 Jean Delvare <[email protected]> wrote:
Quoted 1 line

I'll send it to Linus for 3.3-rc1, OK?

If you want, sure. Note however that it might be less risky to send it now, as I just audited the code.

OK....