Skip to contentSkip to conversation

Checking signal handler implementations with SmPL

Cocci SF Markus Elfring 2 participants 6 messages Jul 15, 2015 – Jul 17, 2015

lore
Share & tools
lore
#0 lore raw

Hello,

I am looking for a specific programming mistake with the help of the semantic patch language once more.

1. A SmPL approach like the following seems to work to some degree.

@find_unsafe@
identifier receiver =~ "^(?x)
(?:
(?:alarm|parent|sig(?:chld|hup|term))_handler
|
handle_sigterm
|
(?:CancelJ|cancel_j)ob
)$";
@@
void receiver(...)
{
...
(
- exit(...);
|
- fprintf(...);
)
...
}

elfring at Sonne:~/Projekte/CUPS> spatch.opt -sp-file async-unsafe1.cocci -dir lokal > async-unsafe1.diff && cat async-unsafe1.diff
init_defs_builtins: /usr/local/lib/coccinelle/standard.h
?
Skipping:lokal/vcnet/regex/regerror.c

backend/ipp.c +0 -2
diff -u -p a/backend/ipp.c b/backend/ipp.c
@@ -3271,8 +3271,6 @@ sigterm_handler(int sig)          /* I - Signal
 
   if (tmpfilename[0])
     unlink(tmpfilename);
-
-  exit(1);
 
 
 

2. I have tried another SmPL script variant out.

@show_unsafe@
identifier receiver =~ "^(?x)
(?:
(?:alarm|parent|sig(?:chld|hup|term))_handler
|
handle_sigterm
|
(?:CancelJ|cancel_j)ob
)$";
@@
void receiver(...)
{
...
(
* exit
|
* fprintf
)
(...);
...
}

elfring at Sonne:~/Projekte/CUPS> spatch.opt -sp-file async-unsafe2.cocci -dir lokal > async-unsafe2.diff && cat async-unsafe2.diff
init_defs_builtins: /usr/local/lib/coccinelle/standard.h
?

tmp/nothing/backend/dnssd.c renamed from backend/dnssd.c +0 -1
diff -u -p lokal/backend/dnssd.c /tmp/nothing/backend/dnssd.c
@@ -1289,7 +1289,6 @@ sigterm_handler(int sig)          /* I - Signal
   (void)sig;
 
   if (job_canceled)
-    exit(CUPS_BACKEND_OK);
   else
     job_canceled = 1;
 }
tmp/nothing/backend/usb-darwin.c renamed from backend/usb-darwin.c +0 -2
diff -u -p lokal/backend/usb-darwin.c /tmp/nothing/backend/usb-darwin.c
@@ -2262,9 +2262,7 @@ sigterm_handler(int sig)          /* I - Signal
     while (waitpid(child_pid, &status, 0) < 0 && errno == EINTR);
 
     if (WIFEXITED(status))
-      exit(WEXITSTATUS(status));
     else if (status == SIGTERM || status == SIGKILL)
-      exit(0);
     else
     {
       fprintf(stderr, "DEBUG: Child crashed on signal %d\n", status);
tmp/nothing/backend/ipp.c renamed from backend/ipp.c +0 -1
diff -u -p lokal/backend/ipp.c /tmp/nothing/backend/ipp.c
@@ -3272,7 +3272,6 @@ sigterm_handler(int sig)          /* I - Signal
   if (tmpfilename[0])
     unlink(tmpfilename);
 
-  exit(1);
 }
 
 

Why did the first approach find less?

Should the call of the function "fprintf" in the source file "usb-darwin.c" also be marked for further considerations here?
Are there any more update candidates to consider?

3. If I add the parameter "-jobs 4" to the shown commands for a parallel source code analysis, I wonder about a message like the following.

[Pid 5399]: Error creating async-unsafe2 : File exists; proceeding without stdout/stderr redirection

Should it be avoided usually?

Regards,
Markus

#1 lore raw
Quoted 12 lines · SF Markus Elfring

Hello,

I am looking for a specific programming mistake with the help of the semantic patch language once more.

1. A SmPL approach like the following seems to work to some degree.

@find_unsafe@
identifier receiver =~ "^(?x)
(?:

   (?:alarm|parent|sig(?:chld|hup|term))_handler
Quoted 16 lines

|
handle_sigterm
|
(?:CancelJ|cancel_j)ob
)$";
@@
void receiver(...)
{
...
(
- exit(...);
|
- fprintf(...);
)
...
}

When you make a transformation, you get a forall semantics - every control flow path in the function must respect your pattern. When you search for something with * you get an exists semantics - there must exist a path that has the patern you specify. If you don't like these defaults, you can change them, by putting exists or forall in the initial @@, or by putting when exists or when forall on the ...

Furthermore, ... matches the shortest path between two points. That is, in your rule, you will only match the case where a control-flow patch contains exactly one call to either exit or fprintf. If you want to allow more occurrences, you can use <... ...> with searches for 0 or more or something, or <+... ...+> which searches for one or more.

I believe the parallelism error has been fixed. I'm not sure if it was in the release, though. If your semantic patch is called foo.cocci, check that you don't have a file/directory called foo in the current directory.

julia

Quoted 102 lines

elfring at Sonne:~/Projekte/CUPS> spatch.opt -sp-file async-unsafe1.cocci -dir lokal > async-unsafe1.diff && cat async-unsafe1.diff
init_defs_builtins: /usr/local/lib/coccinelle/standard.h
?
Skipping:lokal/vcnet/regex/regerror.c
diff -u -p a/backend/ipp.c b/backend/ipp.c

backend/ipp.c +0 -2
@@ -3271,8 +3271,6 @@ sigterm_handler(int sig)          /* I - Signal
 
   if (tmpfilename[0])
     unlink(tmpfilename);
-
-  exit(1);
 
 
 

2. I have tried another SmPL script variant out.

@show_unsafe@
identifier receiver =~ "^(?x)
(?:

   (?:alarm|parent|sig(?:chld|hup|term))_handler

|
handle_sigterm
|
(?:CancelJ|cancel_j)ob
)$";
@@
void receiver(...)
{
...
(
* exit
|f
* fprintf
)
(...);
...
}

elfring at Sonne:~/Projekte/CUPS> spatch.opt -sp-file async-unsafe2.cocci -dir lokal > async-unsafe2.diff && cat async-unsafe2.diff
init_defs_builtins: /usr/local/lib/coccinelle/standard.h
?
diff -u -p lokal/backend/dnssd.c /tmp/nothing/backend/dnssd.c

tmp/nothing/backend/dnssd.c renamed from backend/dnssd.c +0 -1
@@ -1289,7 +1289,6 @@ sigterm_handler(int sig)          /* I - Signal
   (void)sig;
 
   if (job_canceled)
-    exit(CUPS_BACKEND_OK);
   else
     job_canceled = 1;
 }

diff -u -p lokal/backend/usb-darwin.c /tmp/nothing/backend/usb-darwin.c

tmp/nothing/backend/usb-darwin.c renamed from backend/usb-darwin.c +0 -2
@@ -2262,9 +2262,7 @@ sigterm_handler(int sig)          /* I - Signal
     while (waitpid(child_pid, &status, 0) < 0 && errno == EINTR);
 
     if (WIFEXITED(status))
-      exit(WEXITSTATUS(status));
     else if (status == SIGTERM || status == SIGKILL)
-      exit(0);
     else
     {
       fprintf(stderr, "DEBUG: Child crashed on signal %d\n", status);

diff -u -p lokal/backend/ipp.c /tmp/nothing/backend/ipp.c

tmp/nothing/backend/ipp.c renamed from backend/ipp.c +0 -1
@@ -3272,7 +3272,6 @@ sigterm_handler(int sig)          /* I - Signal
   if (tmpfilename[0])
     unlink(tmpfilename);
 
-  exit(1);
 }
 
 

Why did the first approach find less?

Should the call of the function "fprintf" in the source file "usb-darwin.c" also be marked for further considerations here?
Are there any more update candidates to consider?

3. If I add the parameter "-jobs 4" to the shown commands for a parallel source code analysis, I wonder about a message like the following.

[Pid 5399]: Error creating async-unsafe2 : File exists; proceeding without stdout/stderr redirection

Should it be avoided usually?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci at systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

#2 lore raw
Quoted 1 line

Furthermore, ... matches the shortest path between two points.

It seems that I am still not so used to all combinations of the SmPL ellipsis.

Quoted 4 lines

That is, in your rule, you will only match the case where a control-flow patch contains exactly one call to either exit or fprintf. If you want to allow more occurrences, you can use <... ...> with searches for 0 or more or something, or <+... ...+> which searches for one or more.

I guess that the last option is more appropriate for my source code search pattern.

Thanks for your explanation.

Regards,
Markus

#3 lore raw

[Cocci] Deletion of blank lines?

In reply to SF Markus Elfring
Quoted 15 lines

void receiver(...)
{
...
(
- exit(...);
|
- fprintf(...);
)
...
}
?
diff -u -p a/backend/ipp.c b/backend/ipp.c

backend/ipp.c sigterm_handler(int sig) /* I - Signal · line 3275

Quoted 5 lines
backend/ipp.c +0 -2
@@ -3271,8 +3271,6 @@ sigterm_handler(int sig)          /* I - Signal
 
   if (tmpfilename[0])
     unlink(tmpfilename);
-
-  exit(1);

I find another detail interesting in this generated patch. The Coccinelle software suggests here to delete also a blank line at a specific source code place. I would usually interpret this SmPL script example in the way that I did not express the removal of blank lines.

How often do you care if additional source code reformatting should happen around a specific change (or not)?

Regards,
Markus

#4 lore raw

[Cocci] Deletion of blank lines?

Quoted 16 lines · SF Markus Elfring
Quoted 16 lines

void receiver(...)
{
...
(
- exit(...);
|
- fprintf(...);
)
...
}
?
diff -u -p a/backend/ipp.c b/backend/ipp.c

backend/ipp.c +0 -0
@@ -3271,8 +3271,6 @@ sigterm_handler(int sig)          /* I - Signal
 
Quoted 10 lines

backend/ipp.c sigterm_handler(int sig) /* I - Signal · line 3275

Quoted 4 lines
backend/ipp.c +0 -2
   if (tmpfilename[0])
     unlink(tmpfilename);
-
-  exit(1);

I find another detail interesting in this generated patch. The Coccinelle software suggests here to delete also a blank line at a specific source code place. I would usually interpret this SmPL script example in the way that I did not express the removal of blank lines.

Is there a blank line after the call to exit(1); also? I think that it tries toavoid creeating two blank lines.

julia

Quoted 11 lines

How often do you care if additional source code reformatting should happen around a specific change (or not)?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci at systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

#5 lore raw

[Cocci] Deletion of blank lines?

Quoted 1 line

Is there a blank line after the call to exit(1); also?

No.

The text line which follows the shown exit() call should only contain the closing curly bracket for the sigterm_handler() function according to the commit "Load cups into easysw/current." (ef416fc25c4af449e930416117bedb12fc9924ba) from 2006-01-13.

Quoted 1 line

I think that it tries toavoid creeating two blank lines.

Should the blank line between the shown calls be usually preserved by default?

When would the suggested functionality (like a bit of extra source code clean-up) be really needed?

Regards,
Markus