automatic use of checkpatch.pl for security?
Hi,
In an effort to continue the constification work, it'd be nice to not accidentally introduce regressions or add additional work. Since checkpatch.pl already knows to warn about a lot of things including const structures, it would be great to have all commits going through linux-next (or something) have to pass at least a subset of checkpatch.pl's checks.
For example, Lionel Debroux pointed out to me that looking at the last 1000 commits, there are a lot of warnings, including things like:
WARNING: struct dma_map_ops should normally be const #499: FILE: arch/mips/mm/dma-default.c:301:
+static struct dma_map_ops mips_default_dma_map_ops = {
Can we add some kind of automatic checking to actually give checkpatch.pl some real teeth for at least some of its checks?
-Kees
Signature · 2 lines
Kees Cook Ubuntu Security Team
Quoted 11 lines · Kees Cook
Hi,
In an effort to continue the constification work, it'd be nice to not accidentally introduce regressions or add additional work. Since checkpatch.pl already knows to warn about a lot of things including const structures, it would be great to have all commits going through linux-next (or something) have to pass at least a subset of checkpatch.pl's checks.
For example, Lionel Debroux pointed out to me that looking at the last 1000 commits, there are a lot of warnings, including things like:
Quoted 7 lines
WARNING: struct
dma_map_opsshould normally be const #499: FILE:arch/mips/mm/dma-default.c:301:
+static structdma_map_opsmips_default_dma_map_ops= {Can we add some kind of automatic checking to actually give checkpatch.pl some real teeth for at least some of its checks?
Ok, did you actually try to make it const as suggested? If you had, you would have found that there are declarations throughout the code base that conflict with checkpatch.pl's suggestion.
There are several things we could do:
1) Force people to clean up the entire kernel tree before making trivial changes that checkpatch.pl might complain about.
2) Change checkpatch.pl so that it doesn't complain about this.
3) Make reasonable changes and ignore the checkpatch.pl warning.
In that specific case you cite, #3 was chosen.
If you gate admission to linux-next with some sort of automated check like this, I fear the wrath of the disgruntled masses may fall upon you.
David Daney
Hi David,
Quoted 27 lines · David Daney
Quoted 9 lines · Kees Cook
In an effort to continue the constification work, it'd be nice to not accidentally introduce regressions or add additional work. Since checkpatch.pl already knows to warn about a lot of things including const structures, it would be great to have all commits going through linux-next (or something) have to pass at least a subset of checkpatch.pl's checks.
For example, Lionel Debroux pointed out to me that looking at the last 1000 commits, there are a lot of warnings, including things like:
Quoted 7 lines
WARNING: struct
dma_map_opsshould normally be const #499: FILE:arch/mips/mm/dma-default.c:301:
+static structdma_map_opsmips_default_dma_map_ops= {Can we add some kind of automatic checking to actually give checkpatch.pl some real teeth for at least some of its checks?
Ok, did you actually try to make it const as suggested? If you had, you would have found that there are declarations throughout the code base that conflict with checkpatch.pl's suggestion.
There are several things we could do:
1) Force people to clean up the entire kernel tree before making trivial changes that checkpatch.pl might complain about.
Quoted 6 lines
2) Change checkpatch.pl so that it doesn't complain about this.
3) Make reasonable changes and ignore the checkpatch.pl warning.
In that specific case you cite, #3 was chosen.
Right, I don't want to suggest unreasonable changes; I want to try and start a discussion about what might make a good addition to help avoid obvious problems. (The chosen example was, perhaps, not a good one.)
Quoted 3 lines
If you gate admission to linux-next with some sort of automated check like this, I fear the wrath of the disgruntled masses may fall upon you.
But it seems like it might be nice to do at least something there?
-Kees
Signature · 2 lines
Kees Cook Ubuntu Security Team
Hi,
Quoted 36 lines · Kees Cook
Hi David,
Quoted 34 lines · David Daney
Quoted 11 lines · Kees Cook
In an effort to continue the constification work, it'd be nice to not accidentally introduce regressions or add additional work. Since checkpatch.pl already knows to warn about a lot of things including const structures, it would be great to have all commits going through linux-next (or something) have to pass at least a subset of checkpatch.pl's checks.
For example, Lionel Debroux pointed out to me that looking at the last 1000 commits, there are a lot of warnings, including things like:
Quoted 7 lines
WARNING: struct
dma_map_opsshould normally be const #499: FILE:arch/mips/mm/dma-default.c:301:
+static structdma_map_opsmips_default_dma_map_ops= {Can we add some kind of automatic checking to actually give checkpatch.pl some real teeth for at least some of its checks?
Ok, did you actually try to make it const as suggested? If you had, you would have found that there are declarations throughout the code base that conflict with checkpatch.pl's suggestion.
There are several things we could do:
1) Force people to clean up the entire kernel tree before making trivial changes that checkpatch.pl might complain about.
2) Change checkpatch.pl so that it doesn't complain about this.
3) Make reasonable changes and ignore the checkpatch.pl warning.
Quoted 6 lines
Quoted 1 line
In that specific case you cite, #3 was chosen.
Right, I don't want to suggest unreasonable changes; I want to try and start a discussion about what might make a good addition to help avoid obvious problems. (The chosen example was, perhaps, not a good one.)
My bad, sorry.backlight_ops and platform_suspend_ops, for which I sent patches to linux-janitors, may be better examples: several new static mutable instances of those structs have been added after 79404849e90a41ea2109bd0e2f7c7164b0c4ce73, which adds backlight_ops, platform_suspend_ops and others to the list of "should be const" structures. backlight_device_register() has been taking a "const struct backlight_ops *ops" argument since 9905a43b2d563e6f89e4c63c4278ada03f2ebb14, nearly 11 months ago.
Quoted 4 lines
Quoted 3 lines
If you gate admission to linux-next with some sort of automated check like this, I fear the wrath of the disgruntled masses may fall upon you.
But it seems like it might be nice to do at least something there?
I think so, in order to help janitorial work.
Lionel.
Quoted 55 lines · On Tue, 09 Nov 2010 21:49:33 +0100 Lionel Debroux wrote:
Hi,
Quoted 36 lines · Kees Cook
Hi David,
Quoted 34 lines · David Daney
Quoted 11 lines · Kees Cook
In an effort to continue the constification work, it'd be nice to not accidentally introduce regressions or add additional work. Since checkpatch.pl already knows to warn about a lot of things including const structures, it would be great to have all commits going through linux-next (or something) have to pass at least a subset of checkpatch.pl's checks.
For example, Lionel Debroux pointed out to me that looking at the last 1000 commits, there are a lot of warnings, including things like:
Quoted 7 lines
WARNING: struct
dma_map_opsshould normally be const #499: FILE:arch/mips/mm/dma-default.c:301:
+static structdma_map_opsmips_default_dma_map_ops= {Can we add some kind of automatic checking to actually give checkpatch.pl some real teeth for at least some of its checks?
Ok, did you actually try to make it const as suggested? If you had, you would have found that there are declarations throughout the code base that conflict with checkpatch.pl's suggestion.
There are several things we could do:
1) Force people to clean up the entire kernel tree before making trivial changes that checkpatch.pl might complain about.
2) Change checkpatch.pl so that it doesn't complain about this.
3) Make reasonable changes and ignore the checkpatch.pl warning.
Quoted 6 lines
Quoted 1 line
In that specific case you cite, #3 was chosen.
Right, I don't want to suggest unreasonable changes; I want to try and start a discussion about what might make a good addition to help avoid obvious problems. (The chosen example was, perhaps, not a good one.)
My bad, sorry.
backlight_opsandplatform_suspend_ops, for which I sent patches to linux-janitors, may be better examples: several new static mutable instances of those structs have been added after79404849e90a41ea2109bd0e2f7c7164b0c4ce73, which addsbacklight_ops,platform_suspend_opsand others to the list of "should be const" structures.backlight_device_register()has been taking a "const structbacklight_ops*ops" argument since9905a43b2d563e6f89e4c63c4278ada03f2ebb14, nearly 11 months ago.
Quoted 5 lines
Quoted 4 lines
Quoted 3 lines
If you gate admission to linux-next with some sort of automated check like this, I fear the wrath of the disgruntled masses may fall upon you.
But it seems like it might be nice to do at least something there?
I think so, in order to help janitorial work.
linux-next of 2010.1109 has these const warnings from checkpatch:
drivers_tty_pty.c.chk:WARNING: struct file_operations should normally be const drivers_tty_pty.c.chk-#706: FILE: drivers/tty/pty.c:703:
drivers_tty_pty.c.chk-+static struct file_operations ptmx_fops;
--
drivers_tty_tty_io.c.chk:WARNING: struct file_operations should normally be const drivers_tty_tty_io.c.chk-#3187: FILE: drivers/tty/tty_io.c:3184:
drivers_tty_tty_io.c.chk-+void tty_default_fops(struct file_operations *fops)
---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***