Skip to contentSkip to conversation

rdmacg: Replace strncmp with str_has_prefix

PATCH In review Chuhong Yuan 5 participants 9 messages Jul 29, 2019 – Sep 2, 2019 cgroup

lore
Share & tools
lore
#0 lore raw

PATCH 1/12 rdmacg: Replace strncmp with str_has_prefix

strncmp(str, const, len) is error-prone.
We had better use newly introduced
str_has_prefix() instead of it.

---
 kernel/cgroup/rdma.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/cgroup/rdma.c +1 -1
index ae042c347c64..fd12a227f8e4 100644
@@ -379,7 +379,7 @@ static int parse_resource(char *c, int *intval)
 			return -EINVAL;
 		return i;
 	}
-	if (strncmp(value, RDMACG_MAX_STR, len) == 0) {
+	if (str_has_prefix(value, RDMACG_MAX_STR)) {
 		*intval = S32_MAX;
 		return i;
 	}
Signature · 1 line
2.20.1
#1 lore raw
Quoted 3 lines · Chuhong Yuan

strncmp(str, const, len) is error-prone.
We had better use newly introduced
str_has_prefix() instead of it.

Wait, stop. :) After Laura called my attention to your conversion series, mpe pointed out that str_has_prefix() is almost redundant to strstarts() (from 2009), and the latter has many more users. Let's fix strstarts() match str_has_prefix()'s return behavior (all the existing callers are doing boolean tests, so the change in return value won't matter), and then we can continue with this replacement. (And add some documentation to Documenation/process/deprecated.rst along with a checkpatch.pl test maybe too?)

Actually I'd focus first on the actually broken cases first (sizeof() without the "-1", etc):

$ git grep strncmp.*sizeof | grep -v -- '-' | wc -l
17

I expect the "copy/paste" changes could just be a Coccinelle script that Linus could run to fix all the cases (and should be added to the kernel source's list of Coccinelle scripts). Especially since the bulk of the usage pattern are doing literals like this:

arch/alpha/kernel/setup.c: if (strncmp(p, "mem=", 4) == 0) {

$ git grep -E 'strncmp.*(sizeof|, *[0-9]*)' | wc -l
2565

And some cases are weirdly backwards:

tools/perf/util/callchain.c: if (!strncmp(tok, "none", strlen(tok))) {

-Kees

Quoted 21 lines
---
 kernel/cgroup/rdma.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/cgroup/rdma.c +1 -1
index ae042c347c64..fd12a227f8e4 100644
@@ -379,7 +379,7 @@ static int parse_resource(char *c, int *intval)
 			return -EINVAL;
 		return i;
 	}
-	if (strncmp(value, RDMACG_MAX_STR, len) == 0) {
+	if (str_has_prefix(value, RDMACG_MAX_STR)) {
 		*intval = S32_MAX;
 		return i;
 	}
Signature · 1 line
2.20.1
Signature · 1 line
Kees Cook
#2 lore raw
Quoted 6 lines · Kees Cook <[email protected]> 于2019年7月30日周二 下午12:26写道:
Quoted 3 lines · Chuhong Yuan

strncmp(str, const, len) is error-prone.
We had better use newly introduced
str_has_prefix() instead of it.

Quoted 9 lines

Wait, stop. :) After Laura called my attention to your conversion series, mpe pointed out that str_has_prefix() is almost redundant to strstarts() (from 2009), and the latter has many more users. Let's fix strstarts() match str_has_prefix()'s return behavior (all the existing callers are doing boolean tests, so the change in return value won't matter), and then we can continue with this replacement. (And add some documentation to Documenation/process/deprecated.rst along with a checkpatch.pl test maybe too?)

Thanks for your advice!
Does that mean replacing strstarts()'s implementation with str_has_prefix()'s and then use strstarts() to substitute strncmp?

I am not very clear about how to add the test into checkpatch.pl. Should I write a check for this pattern or directly add strncmp into deprecated_apis?

Quoted 11 lines

Actually I'd focus first on the actually broken cases first (sizeof() without the "-1", etc):

$ git grep strncmp.*sizeof | grep -v -- '-' | wc -l
17

I expect the "copy/paste" changes could just be a Coccinelle script that Linus could run to fix all the cases (and should be added to the kernel source's list of Coccinelle scripts). Especially since the bulk of the usage pattern are doing literals like this:

Actually I am using a Coccinelle script to detect the cases and have found 800+ places of strncmp(str, const, len).
But the script still needs some improvement since it has false negatives and only focuses on detecting, not replacement. I can upload it after improvement.
In which form should I upload it? In a patch's description or put it in coccinelle scripts?

Quoted 11 lines

arch/alpha/kernel/setup.c: if (strncmp(p, "mem=", 4) == 0) {

$ git grep -E 'strncmp.*(sizeof|, *[0-9]*)' | wc -l
2565

And some cases are weirdly backwards:

tools/perf/util/callchain.c: if (!strncmp(tok, "none", strlen(tok))) {

-Kees

I think with the help of Coccinelle script, all strncmp(str, const, len) can be replaced and these problems will be eliminated. :)

Regards,
Chuhong

Quoted 24 lines
Quoted 19 lines
---
 kernel/cgroup/rdma.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/cgroup/rdma.c +1 -2
index ae042c347c64..fd12a227f8e4 100644
@@ -379,7 +379,7 @@ static int parse_resource(char *c, int *intval)
                      return -EINVAL;
              return i;
      }
-     if (strncmp(value, RDMACG_MAX_STR, len) == 0) {
+     if (str_has_prefix(value, RDMACG_MAX_STR)) {
              *intval = S32_MAX;
              return i;
      }
--
Quoted 2 lines

2.20.1

--
Kees Cook

#3 lore raw
Quoted 16 lines · Chuhong Yuan
Quoted 5 lines · Kees Cook <[email protected]> 于2019年7月30日周二 下午12:26写道:
Quoted 3 lines · Chuhong Yuan

strncmp(str, const, len) is error-prone.
We had better use newly introduced
str_has_prefix() instead of it.

Quoted 9 lines

Wait, stop. :) After Laura called my attention to your conversion series, mpe pointed out that str_has_prefix() is almost redundant to strstarts() (from 2009), and the latter has many more users. Let's fix strstarts() match str_has_prefix()'s return behavior (all the existing callers are doing boolean tests, so the change in return value won't matter), and then we can continue with this replacement. (And add some documentation to Documenation/process/deprecated.rst along with a checkpatch.pl test maybe too?)

Quoted 8 lines

Thanks for your advice!
Does that mean replacing strstarts()'s implementation with str_has_prefix()'s and then use strstarts() to substitute strncmp?

I am not very clear about how to add the test into checkpatch.pl. Should I write a check for this pattern or directly add strncmp into deprecated_apis?

After Nitin's patch gets applied: (which btw wasn't cc'd to lkml)

(sorry about the bad link, something about it hits some
spam filter)

open wall . com/lists/kernel-hardening/2019/07/28/1

Add it to deprecated_string_apis

And we've had this sort of discussion before:

https://lore.kernel.org/patchwork/patch/1026598/

I believe I'm still in favor of global conversion of strstarts to str_has_prefix.

#4 lore raw
Quoted 3 lines · On Tue, 30 Jul 2019 00:03:28 -0700 Joe Perches <[email protected]> wrote:

I believe I'm still in favor of global conversion of strstarts to str_has_prefix.

So am I.

-- Steve

#5 lore raw
In reply to Chuhong Yuan
Quoted 39 lines · Chuhong Yuan <[email protected]> 于2019年7月30日周二 下午2:39写道:
Quoted 6 lines · Kees Cook <[email protected]> 于2019年7月30日周二 下午12:26写道:
Quoted 3 lines · Chuhong Yuan

strncmp(str, const, len) is error-prone.
We had better use newly introduced
str_has_prefix() instead of it.

Quoted 9 lines

Wait, stop. :) After Laura called my attention to your conversion series, mpe pointed out that str_has_prefix() is almost redundant to strstarts() (from 2009), and the latter has many more users. Let's fix strstarts() match str_has_prefix()'s return behavior (all the existing callers are doing boolean tests, so the change in return value won't matter), and then we can continue with this replacement. (And add some documentation to Documenation/process/deprecated.rst along with a checkpatch.pl test maybe too?)

Thanks for your advice!
Does that mean replacing strstarts()'s implementation with str_has_prefix()'s and then use strstarts() to substitute strncmp?

I am not very clear about how to add the test into checkpatch.pl. Should I write a check for this pattern or directly add strncmp into deprecated_apis?

Quoted 11 lines

Actually I'd focus first on the actually broken cases first (sizeof() without the "-1", etc):

$ git grep strncmp.*sizeof | grep -v -- '-' | wc -l
17

I expect the "copy/paste" changes could just be a Coccinelle script that Linus could run to fix all the cases (and should be added to the kernel source's list of Coccinelle scripts). Especially since the bulk of the usage pattern are doing literals like this:

Quoted 17 lines

Actually I am using a Coccinelle script to detect the cases and have found 800+ places of strncmp(str, const, len).
But the script still needs some improvement since it has false negatives and only focuses on detecting, not replacement. I can upload it after improvement.
In which form should I upload it? In a patch's description or put it in coccinelle scripts?

Quoted 9 lines

arch/alpha/kernel/setup.c: if (strncmp(p, "mem=", 4) == 0) {

$ git grep -E 'strncmp.*(sizeof|, *[0-9]*)' | wc -l
2565

And some cases are weirdly backwards:

tools/perf/util/callchain.c: if (!strncmp(tok, "none", strlen(tok))) {

I find there are cases of this pattern are not wrong.
One example is kernel/irq/debugfs.c: if (!strncmp(buf, "trigger", size)) {

Thus I do not know whether I should include these cases in my script.

Quoted 33 lines
Quoted 2 lines

-Kees

I think with the help of Coccinelle script, all strncmp(str, const, len) can be replaced and these problems will be eliminated. :)

Regards,
Chuhong

Quoted 24 lines
Quoted 19 lines
---
 kernel/cgroup/rdma.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/cgroup/rdma.c +1 -2
index ae042c347c64..fd12a227f8e4 100644
@@ -379,7 +379,7 @@ static int parse_resource(char *c, int *intval)
                      return -EINVAL;
              return i;
      }
-     if (strncmp(value, RDMACG_MAX_STR, len) == 0) {
+     if (str_has_prefix(value, RDMACG_MAX_STR)) {
              *intval = S32_MAX;
              return i;
      }
--
Quoted 2 lines

2.20.1

--
Kees Cook

#6 lore raw
Quoted 51 lines · Chuhong Yuan
Quoted 50 lines · Chuhong Yuan <[email protected]> 于2019年7月30日周二 下午2:39写道:
Quoted 5 lines · Kees Cook <[email protected]> 于2019年7月30日周二 下午12:26写道:
Quoted 3 lines · Chuhong Yuan

strncmp(str, const, len) is error-prone.
We had better use newly introduced
str_has_prefix() instead of it.

Quoted 9 lines

Wait, stop. :) After Laura called my attention to your conversion series, mpe pointed out that str_has_prefix() is almost redundant to strstarts() (from 2009), and the latter has many more users. Let's fix strstarts() match str_has_prefix()'s return behavior (all the existing callers are doing boolean tests, so the change in return value won't matter), and then we can continue with this replacement. (And add some documentation to Documenation/process/deprecated.rst along with a checkpatch.pl test maybe too?)

Thanks for your advice!
Does that mean replacing strstarts()'s implementation with str_has_prefix()'s and then use strstarts() to substitute strncmp?

I am not very clear about how to add the test into checkpatch.pl. Should I write a check for this pattern or directly add strncmp into deprecated_apis?

Quoted 11 lines

Actually I'd focus first on the actually broken cases first (sizeof() without the "-1", etc):

$ git grep strncmp.*sizeof | grep -v -- '-' | wc -l
17

I expect the "copy/paste" changes could just be a Coccinelle script that Linus could run to fix all the cases (and should be added to the kernel source's list of Coccinelle scripts). Especially since the bulk of the usage pattern are doing literals like this:

Actually I am using a Coccinelle script to detect the cases and have found 800+ places of strncmp(str, const, len).
But the script still needs some improvement since it has false negatives and only focuses on detecting, not replacement. I can upload it after improvement.
In which form should I upload it? In a patch's description or put it in coccinelle scripts?

Quoted 5 lines

arch/alpha/kernel/setup.c: if (strncmp(p, "mem=", 4) == 0) {

$ git grep -E 'strncmp.*(sizeof|, *[0-9]*)' | wc -l
2565
Quoted 8 lines
Quoted 3 lines
Quoted 3 lines

And some cases are weirdly backwards:

tools/perf/util/callchain.c: if (!strncmp(tok, "none", strlen(tok))) {

I find there are cases of this pattern are not wrong.
One example is kernel/irq/debugfs.c: if (!strncmp(buf, "trigger", size)) {

Thus I do not know whether I should include these cases in my script.

That case isn't looking for a prefix AFAICS, so you should skip it.

I think Kees regexp was just slightly wrong, it should be:

git grep -E 'strncmp.*(sizeof|, *[0-9]+)'

ie. either literal "sizeof" or *at least one* digit.

cheers

#7 lore raw
In reply to Chuhong Yuan
Quoted 2 lines · Chuhong Yuan

I think with the help of Coccinelle script, all strncmp(str, const, len) can be replaced and these problems will be eliminated. :)

Hi! Just pinging this thread again. Any progress on this conversion?

Thanks!

Signature · 1 line
Kees Cook
#8 lore raw
Quoted 7 lines · Kees Cook
Quoted 2 lines · Chuhong Yuan

I think with the help of Coccinelle script, all strncmp(str, const, len) can be replaced and these problems will be eliminated. :)

Hi! Just pinging this thread again. Any progress on this conversion?

I didn't work further on this conversion since it seems that developers are not very interested in this problem (only half of my sent patches have been responded till now) and I am working on other projects recently.

If the Coccinelle script is needed, I can try to implement it next week.

Regards,
Chuhong

Quoted 4 lines

Thanks!

--
Kees Cook