| Message ID | 20161121214052.32428-1-andreas@gazonk.org (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Herbert Xu |
| Headers | show |
On Mon, Nov 21, 2016 at 10:40:52PM +0100, Andreas Bofjall wrote: > After dash had executed the exit trap handler, the trap was reset but > the pointer was never freed. This leak can be demonstrated by running > dash through valgrind and executing the following shell script: > foo() { > true > } > trap foo EXIT > Fix by properly freeing the trap pointer in exitshell(). > Signed-off-by: Andreas Bofjall <andreas@gazonk.org> > --- > src/trap.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/src/trap.c b/src/trap.c > index edb9938..5418b07 100644 > --- a/src/trap.c > +++ b/src/trap.c > @@ -389,6 +389,7 @@ exitshell(void) > trap[0] = NULL; > evalskip = 0; > evalstring(p, 0); > + ckfree(p); > } > out: > /* This patch will shut up valgrind in the common case, but does not handle the general case. The command string may contain an error or invoke the exit builtin and in either case the command string will be leaked (SIGINT might be expected to have a similar effect, but behaves strangely from an EXIT trap in dash). You can probably use the exception handling already present in the function to fix this. Note that ckfree() should only be used while INTOFF is in effect, both to avoid longjmp'ing out of free() and to ensure exactly one free in the presence of interruptions and errors.
On Tue, 22 Nov 2016, Jilles Tjoelker wrote: > On Mon, Nov 21, 2016 at 10:40:52PM +0100, Andreas Bofjall wrote: >> diff --git a/src/trap.c b/src/trap.c >> index edb9938..5418b07 100644 >> --- a/src/trap.c >> +++ b/src/trap.c >> @@ -389,6 +389,7 @@ exitshell(void) >> trap[0] = NULL; >> evalskip = 0; >> evalstring(p, 0); >> + ckfree(p); >> } >> out: >> /* > > This patch will shut up valgrind in the common case, but does not handle > the general case. The command string may contain an error or invoke the > exit builtin and in either case the command string will be leaked > (SIGINT might be expected to have a similar effect, but behaves > strangely from an EXIT trap in dash). > > You can probably use the exception handling already present in the > function to fix this. Note that ckfree() should only be used while > INTOFF is in effect, both to avoid longjmp'ing out of free() and to > ensure exactly one free in the presence of interruptions and errors. Thanks for the feedback! Would something simple like moving the call to ckfree() to after the out: label and wrapping it in INTOFF/INTON (thereby catching both the normal return path and the exception) be ok, or do you mean something more elaborate? /Andreas -- To unsubscribe from this list: send the line "unsubscribe dash" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/src/trap.c b/src/trap.c index edb9938..5418b07 100644 --- a/src/trap.c +++ b/src/trap.c @@ -389,6 +389,7 @@ exitshell(void) trap[0] = NULL; evalskip = 0; evalstring(p, 0); + ckfree(p); } out: /*
After dash had executed the exit trap handler, the trap was reset but the pointer was never freed. This leak can be demonstrated by running dash through valgrind and executing the following shell script: foo() { true } trap foo EXIT Fix by properly freeing the trap pointer in exitshell(). Signed-off-by: Andreas Bofjall <andreas@gazonk.org> --- src/trap.c | 1 + 1 file changed, 1 insertion(+)