[PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Taylor Blau]

While trying to fix a move based on an uninitialized value (along with a
declaration after the first statement), be0fd57228
(maintenance --unregister: fix uninit'd data use &
-Wdeclaration-after-statement, 2022-11-15) unintentionally introduced a
use-after-free.

The problem arises when `maintenance_unregister()` sees a non-NULL
`config_file` string and thus tries to call
git_configset_get_value_multi() to lookup the corresponding values.

We store the result off, and then call git_configset_clear(), which
frees the pointer that we just stored. We then try to read that
now-freed pointer a few lines below, and there we have our
use-after-free:

    $ ./t7900-maintenance.sh -vxi --run=23 --valgrind
    [...]
    + git maintenance unregister --config-file ./other
    ==3048727== Invalid read of size 8
    ==3048727==    at 0x1869CA: maintenance_unregister (gc.c:1590)
    ==3048727==    by 0x188F42: cmd_maintenance (gc.c:2651)
    ==3048727==    by 0x128C62: run_builtin (git.c:466)
    ==3048727==    by 0x12907E: handle_builtin (git.c:721)
    ==3048727==    by 0x1292EC: run_argv (git.c:788)
    ==3048727==    by 0x12988E: cmd_main (git.c:926)
    ==3048727==    by 0x21ED39: main (common-main.c:57)
    ==3048727==  Address 0x4b38bc8 is 24 bytes inside a block of size 64 free'd
    ==3048727==    at 0x484617B: free (vg_replace_malloc.c:872)
    ==3048727==    by 0x2D207E: free_individual_entries (hashmap.c:188)
    ==3048727==    by 0x2D2153: hashmap_clear_ (hashmap.c:207)
    ==3048727==    by 0x270B5C: git_configset_clear (config.c:2375)
    ==3048727==    by 0x1869AC: maintenance_unregister (gc.c:1585)
    ==3048727==    by 0x188F42: cmd_maintenance (gc.c:2651)
    ==3048727==    by 0x128C62: run_builtin (git.c:466)
    ==3048727==    by 0x12907E: handle_builtin (git.c:721)
    ==3048727==    by 0x1292EC: run_argv (git.c:788)
    ==3048727==    by 0x12988E: cmd_main (git.c:926)
    ==3048727==    by 0x21ED39: main (common-main.c:57)
    [...]

Resolve this via a partial-revert of be0fd57228. The config_set struct
now gets a zero initialization, which makes free()-ing it a noop even
without calling git_configset_init(). When we do initialize it to a
non-zero value, it is only free()'d after our last read of `list`.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
I am disappointed in myself for finding this only after I pushed out a
hotfix to 'next' and rebuild the downstream branches.

This should be a minimal fix on top of Ævar's patch to get 'next'
building again.

 builtin/gc.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/builtin/gc.c b/builtin/gc.c
index d87cf84041..38882a1e35 100644
--- a/builtin/gc.c
+++ b/builtin/gc.c
@@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 	int found = 0;
 	struct string_list_item *item;
 	const struct string_list *list;
+	struct config_set cs = { { 0 } };

 	argc = parse_options(argc, argv, prefix, options,
 			     builtin_maintenance_unregister_usage, 0);
@@ -1551,12 +1552,9 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 				   options);

 	if (config_file) {
-		struct config_set cs;
-
 		git_configset_init(&cs);
 		git_configset_add_file(&cs, config_file);
 		list = git_configset_get_value_multi(&cs, key);
-		git_configset_clear(&cs);
 	} else {
 		list = git_config_get_value_multi(key);
 	}
@@ -1592,6 +1590,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
 		die(_("repository '%s' is not registered"), maintpath);
 	}

+	git_configset_clear(&cs);
 	free(maintpath);
 	return 0;
 }
--
2.38.0.16.g393fd4c6db

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Derrick Stolee]

On 11/15/2022 1:53 PM, Taylor Blau wrote:
> While trying to fix a move based on an uninitialized value (along with a
> declaration after the first statement), be0fd57228
> (maintenance --unregister: fix uninit'd data use &
> -Wdeclaration-after-statement, 2022-11-15) unintentionally introduced a
> use-after-free.
> 
> The problem arises when `maintenance_unregister()` sees a non-NULL
> `config_file` string and thus tries to call
> git_configset_get_value_multi() to lookup the corresponding values.
> 
> We store the result off, and then call git_configset_clear(), which
> frees the pointer that we just stored. We then try to read that
> now-freed pointer a few lines below, and there we have our
> use-after-free:

Makes sense why this needs to be pulled out to a larger scope, but
also why it's so easy to make this mistake.

> +	struct config_set cs = { { 0 } };
> 
>  	argc = parse_options(argc, argv, prefix, options,
>  			     builtin_maintenance_unregister_usage, 0);
> @@ -1551,12 +1552,9 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>  				   options);
> 
>  	if (config_file) {
> -		struct config_set cs;
> -
>  		git_configset_init(&cs);
>  		git_configset_add_file(&cs, config_file);
>  		list = git_configset_get_value_multi(&cs, key);
> -		git_configset_clear(&cs);

That the list depends on the configset and not exist as an
independent entity is non-obvious, but I'm sure is rooted
in some kind of memory-saving optimization.

>  	} else {
>  		list = git_config_get_value_multi(key);
>  	}
> @@ -1592,6 +1590,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>  		die(_("repository '%s' is not registered"), maintpath);
>  	}
> 
> +	git_configset_clear(&cs);
>  	free(maintpath);
>  	return 0;
>  }

Thanks for drilling down on this. LGTM.

-Stolee

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Ævar Arnfjörð Bjarmason]

On Tue, Nov 15 2022, Taylor Blau wrote:

> While trying to fix a move based on an uninitialized value (along with a
> declaration after the first statement), be0fd57228
> (maintenance --unregister: fix uninit'd data use &
> -Wdeclaration-after-statement, 2022-11-15) unintentionally introduced a
> use-after-free.
>
> The problem arises when `maintenance_unregister()` sees a non-NULL
> `config_file` string and thus tries to call
> git_configset_get_value_multi() to lookup the corresponding values.
>
> We store the result off, and then call git_configset_clear(), which
> frees the pointer that we just stored. We then try to read that
> now-freed pointer a few lines below, and there we have our
> use-after-free:
>
>     $ ./t7900-maintenance.sh -vxi --run=23 --valgrind
>     [...]
>     + git maintenance unregister --config-file ./other
>     ==3048727== Invalid read of size 8
>     ==3048727==    at 0x1869CA: maintenance_unregister (gc.c:1590)
>     ==3048727==    by 0x188F42: cmd_maintenance (gc.c:2651)
>     ==3048727==    by 0x128C62: run_builtin (git.c:466)
>     ==3048727==    by 0x12907E: handle_builtin (git.c:721)
>     ==3048727==    by 0x1292EC: run_argv (git.c:788)
>     ==3048727==    by 0x12988E: cmd_main (git.c:926)
>     ==3048727==    by 0x21ED39: main (common-main.c:57)
>     ==3048727==  Address 0x4b38bc8 is 24 bytes inside a block of size 64 free'd
>     ==3048727==    at 0x484617B: free (vg_replace_malloc.c:872)
>     ==3048727==    by 0x2D207E: free_individual_entries (hashmap.c:188)
>     ==3048727==    by 0x2D2153: hashmap_clear_ (hashmap.c:207)
>     ==3048727==    by 0x270B5C: git_configset_clear (config.c:2375)
>     ==3048727==    by 0x1869AC: maintenance_unregister (gc.c:1585)
>     ==3048727==    by 0x188F42: cmd_maintenance (gc.c:2651)
>     ==3048727==    by 0x128C62: run_builtin (git.c:466)
>     ==3048727==    by 0x12907E: handle_builtin (git.c:721)
>     ==3048727==    by 0x1292EC: run_argv (git.c:788)
>     ==3048727==    by 0x12988E: cmd_main (git.c:926)
>     ==3048727==    by 0x21ED39: main (common-main.c:57)
>     [...]
>
> Resolve this via a partial-revert of be0fd57228. The config_set struct
> now gets a zero initialization, which makes free()-ing it a noop even
> without calling git_configset_init(). When we do initialize it to a
> non-zero value, it is only free()'d after our last read of `list`.
>
> Signed-off-by: Taylor Blau <me@ttaylorr.com>
> ---
> I am disappointed in myself for finding this only after I pushed out a
> hotfix to 'next' and rebuild the downstream branches.
>
> This should be a minimal fix on top of Ævar's patch to get 'next'
> building again.

I'm also disappointed in myself, sorry. I *did* test it locally with
valgrind, but obviously fat-fingered it somehow and tested the wrong
version. Sorry!

>  builtin/gc.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/builtin/gc.c b/builtin/gc.c
> index d87cf84041..38882a1e35 100644
> --- a/builtin/gc.c
> +++ b/builtin/gc.c
> @@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>  	int found = 0;
>  	struct string_list_item *item;
>  	const struct string_list *list;
> +	struct config_set cs = { { 0 } };

Just "{ 0 }" here instead? I see it may have been copied from some older
pre-image though, and they'll do the same in either case, so it's not
important...

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Ævar Arnfjörð Bjarmason]

On Tue, Nov 15 2022, Derrick Stolee wrote:

> On 11/15/2022 1:53 PM, Taylor Blau wrote:
>> While trying to fix a move based on an uninitialized value (along with a
>> declaration after the first statement), be0fd57228
>> (maintenance --unregister: fix uninit'd data use &
>> -Wdeclaration-after-statement, 2022-11-15) unintentionally introduced a
>> use-after-free.
>> 
>> The problem arises when `maintenance_unregister()` sees a non-NULL
>> `config_file` string and thus tries to call
>> git_configset_get_value_multi() to lookup the corresponding values.
>> 
>> We store the result off, and then call git_configset_clear(), which
>> frees the pointer that we just stored. We then try to read that
>> now-freed pointer a few lines below, and there we have our
>> use-after-free:
>
> Makes sense why this needs to be pulled out to a larger scope, but
> also why it's so easy to make this mistake.

Yeah, the config API's full of foot-guns, although here we return a
"const struct string_list *", not a "struct string_list *", so in
retrospect this should be rather obvious...

But still, we should probably as #leftoverbits make it behave
consistently wrt naming. I.e. in this case it's
git_configset_get_value_multi() really behaves like a
git_configset_get_string_tmp(), and there's no equivalent of a
git_configset_get_string() (i.e. xstrdup()'d) for *_multi().

>> +	struct config_set cs = { { 0 } };
>> 
>>  	argc = parse_options(argc, argv, prefix, options,
>>  			     builtin_maintenance_unregister_usage, 0);
>> @@ -1551,12 +1552,9 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>>  				   options);
>> 
>>  	if (config_file) {
>> -		struct config_set cs;
>> -
>>  		git_configset_init(&cs);
>>  		git_configset_add_file(&cs, config_file);
>>  		list = git_configset_get_value_multi(&cs, key);
>> -		git_configset_clear(&cs);
>
> That the list depends on the configset and not exist as an
> independent entity is non-obvious, but I'm sure is rooted
> in some kind of memory-saving optimization.

Yes, and it's probably worth keeping that, but I haven't benchmarked
etc. This is only a problem in practice if you're constructing your own
configset, e.g. here because we have a custom config file. So for most
users this API is safe in general, i.e. we free() it, but it's the
config that's in "the_repository" normally, so it outlives any "normal"
code.

>>  	} else {
>>  		list = git_config_get_value_multi(key);
>>  	}
>> @@ -1592,6 +1590,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>>  		die(_("repository '%s' is not registered"), maintpath);
>>  	}
>> 
>> +	git_configset_clear(&cs);
>>  	free(maintpath);
>>  	return 0;
>>  }
>
> Thanks for drilling down on this. LGTM.

On the related subject of config API foot-guns, it would be great if you
could look over the in-flight series I have to make related parts of the
config API safe by default [1].

8/9 there fixes 6 segfaults, 3 of which are git blame'd to you :), and
9/9 a foot-gun-y interaction with the strvec API, which you'll also
probably find interesting...

1. https://lore.kernel.org/git/cover-v2-0.9-00000000000-20221101T225822Z-avarab@gmail.com/

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Taylor Blau]

On Tue, Nov 15, 2022 at 08:41:44PM +0100, Ævar Arnfjörð Bjarmason wrote:
> > I am disappointed in myself for finding this only after I pushed out a
> > hotfix to 'next' and rebuild the downstream branches.
> >
> > This should be a minimal fix on top of Ævar's patch to get 'next'
> > building again.
>
> I'm also disappointed in myself, sorry. I *did* test it locally with
> valgrind, but obviously fat-fingered it somehow and tested the wrong
> version. Sorry!

It's OK. Let's not beat ourselves up too much, and instead focusing on
making sure the quality for the next pushout is higher (which is more on
me than it is on you).

> >  builtin/gc.c | 5 ++---
> >  1 file changed, 2 insertions(+), 3 deletions(-)
> >
> > diff --git a/builtin/gc.c b/builtin/gc.c
> > index d87cf84041..38882a1e35 100644
> > --- a/builtin/gc.c
> > +++ b/builtin/gc.c
> > @@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
> >  	int found = 0;
> >  	struct string_list_item *item;
> >  	const struct string_list *list;
> > +	struct config_set cs = { { 0 } };
>
> Just "{ 0 }" here instead? I see it may have been copied from some older
> pre-image though, and they'll do the same in either case, so it's not
> important...

Copying from other zero-initializations of `struct config_set`:

    $ git grep -oh 'struct config_set.*= {.*' | sort | uniq -c
          3 struct config_set cs = { { 0 } };

Thanks,
Taylor

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Derrick Stolee]

On 11/15/22 2:54 PM, Taylor Blau wrote:
> On Tue, Nov 15, 2022 at 08:41:44PM +0100, Ævar Arnfjörð Bjarmason wrote:
>>> @@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>>>  	int found = 0;
>>>  	struct string_list_item *item;
>>>  	const struct string_list *list;
>>> +	struct config_set cs = { { 0 } };
>>
>> Just "{ 0 }" here instead? I see it may have been copied from some older
>> pre-image though, and they'll do the same in either case, so it's not
>> important...
> 
> Copying from other zero-initializations of `struct config_set`:
> 
>     $ git grep -oh 'struct config_set.*= {.*' | sort | uniq -c
>           3 struct config_set cs = { { 0 } };

Yes, without the double braces the compiler will complain on
macOS, I believe.

Thanks,
-Stolee

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Ævar Arnfjörð Bjarmason]

On Wed, Nov 16 2022, Derrick Stolee wrote:

> On 11/15/22 2:54 PM, Taylor Blau wrote:
>> On Tue, Nov 15, 2022 at 08:41:44PM +0100, Ævar Arnfjörð Bjarmason wrote:
>>>> @@ -1543,6 +1543,7 @@ static int maintenance_unregister(int argc, const char **argv, const char *prefi
>>>>  	int found = 0;
>>>>  	struct string_list_item *item;
>>>>  	const struct string_list *list;
>>>> +	struct config_set cs = { { 0 } };
>>>
>>> Just "{ 0 }" here instead? I see it may have been copied from some older
>>> pre-image though, and they'll do the same in either case, so it's not
>>> important...
>> 
>> Copying from other zero-initializations of `struct config_set`:
>> 
>>     $ git grep -oh 'struct config_set.*= {.*' | sort | uniq -c
>>           3 struct config_set cs = { { 0 } };
>
> Yes, without the double braces the compiler will complain on
> macOS, I believe.

Ah, that was sorted in 54795d37d9e (config.mak.dev: disable suggest
braces error on old clang versions, 2022-10-10).

It's fine here, we can follow-up for the #leftoverbits of changing those
some other time.

Re: [PATCH] builtin/gc.c: fix use-after-free in maintenance_unregister()

[Taylor Blau]

On Wed, Nov 16, 2022 at 04:14:30PM +0100, Ævar Arnfjörð Bjarmason wrote:
> >> Copying from other zero-initializations of `struct config_set`:
> >>
> >>     $ git grep -oh 'struct config_set.*= {.*' | sort | uniq -c
> >>           3 struct config_set cs = { { 0 } };
> >
> > Yes, without the double braces the compiler will complain on
> > macOS, I believe.
>
> Ah, that was sorted in 54795d37d9e (config.mak.dev: disable suggest
> braces error on old clang versions, 2022-10-10).
>
> It's fine here, we can follow-up for the #leftoverbits of changing those
> some other time.

Thanks for the reference to 54795d37d9e, Ævar.

There are a small handful of these sitting around in our codebase:

    $ git grep '{ { '
    builtin/gc.c:   struct config_set cs = { { 0 } };
    builtin/worktree.c:             struct config_set cs = { { 0 } };
    oidmap.h:#define OIDMAP_INIT { { NULL } }
    oidset.h:#define OIDSET_INIT { { 0 } }
    reftable/stack_test.c:  struct reftable_ref_record refs[2] = { { NULL } };
    reftable/stack_test.c:  struct reftable_log_record logs[2] = { { NULL } };
    reftable/stack_test.c:  struct reftable_ref_record refs[2] = { { NULL } };
    reftable/stack_test.c:  struct reftable_log_record logs[2] = { { NULL } };
    reftable/stack_test.c:  struct reftable_log_record logs[20] = { { NULL } };
    reset.c:        struct tree_desc desc[2] = { { NULL }, { NULL } };
    t/helper/test-ref-store.c:static struct flag_definition empty_flags[] = { { NULL, 0 } };
    worktree.c:     struct config_set cs = { { 0 } };

But I tend to agree that I'm not all that eager to go and change these
for the sake of changing them. I think we can slowly let them age out of
the codebase when we're touching nearby areas in the future.

Thanks,
Taylor