[PATCH 1/1] alloc_tag: fix empty codetag module section handling

[PATCH 1/1] alloc_tag: fix empty codetag module section handling

[Suren Baghdasaryan]

When a module does not have any allocations, it's allocation tag section
is empty and codetag_alloc_module_section() returns NULL. However this
condition should never happen because codetag_needs_module_section() will
detect an empty section and avoid calling codetag_alloc_module_section().
Change codetag_alloc_module_section() to never return NULL, which should
prevent static checker warnings. Add a WARN_ON() and a proper error
reporting in case codetag_alloc_module_section() returns NULL, to prevent
future codetag type implementations from returning NULL from their
cttype->desc.alloc_section_mem() operation.

Fixes: 61c9e58f3a10 ("alloc_tag: load module tags into separate contiguous memory")
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/all/50f12fa1-17c1-4940-a6bf-beaf61f6b17a@stanley.mountain/
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
---
 kernel/module/main.c | 4 ++++
 lib/alloc_tag.c      | 2 +-
 lib/codetag.c        | 2 +-
 3 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/kernel/module/main.c b/kernel/module/main.c
index 129c98e6380d..00c16f5c5568 100644
--- a/kernel/module/main.c
+++ b/kernel/module/main.c
@@ -2316,6 +2316,10 @@ static int move_module(struct module *mod, struct load_info *info)
 		if (codetag_needs_module_section(mod, sname, shdr->sh_size)) {
 			dest = codetag_alloc_module_section(mod, sname, shdr->sh_size,
 					arch_mod_section_prepend(mod, i), shdr->sh_addralign);
+			if (WARN_ON(!dest)) {
+				ret = -EINVAL;
+				goto out_err;
+			}
 			if (IS_ERR(dest)) {
 				ret = PTR_ERR(dest);
 				goto out_err;
diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
index 1c74942e6dfd..00ab18ea452a 100644
--- a/lib/alloc_tag.c
+++ b/lib/alloc_tag.c
@@ -437,7 +437,7 @@ static void *reserve_module_tags(struct module *mod, unsigned long size,
 
 	/* If no tags return NULL */
 	if (size < sizeof(struct alloc_tag))
-		return NULL;
+		return ERR_PTR(-EINVAL);
 
 	/*
 	 * align is always power of 2, so we can use IS_ALIGNED and ALIGN.
diff --git a/lib/codetag.c b/lib/codetag.c
index 4949511b4933..42aadd6c1454 100644
--- a/lib/codetag.c
+++ b/lib/codetag.c
@@ -244,7 +244,7 @@ void *codetag_alloc_module_section(struct module *mod, const char *name,
 {
 	const char *type_name = name + strlen(CODETAG_SECTION_PREFIX);
 	struct codetag_type *cttype;
-	void *ret = NULL;
+	void *ret = ERR_PTR(-EINVAL);
 
 	mutex_lock(&codetag_lock);
 	list_for_each_entry(cttype, &codetag_types, link) {

base-commit: 758db9ca0107bc6c00f0ed4808974d66c8dc2fea
-- 
2.47.0.163.g1226f6d8fa-goog

Re: [PATCH 1/1] alloc_tag: fix empty codetag module section handling

[Suren Baghdasaryan]

On Thu, Oct 31, 2024 at 5:00 PM Suren Baghdasaryan <surenb@google.com> wrote:
>
> When a module does not have any allocations, it's allocation tag section
> is empty and codetag_alloc_module_section() returns NULL. However this
> condition should never happen because codetag_needs_module_section() will
> detect an empty section and avoid calling codetag_alloc_module_section().
> Change codetag_alloc_module_section() to never return NULL, which should
> prevent static checker warnings. Add a WARN_ON() and a proper error
> reporting in case codetag_alloc_module_section() returns NULL, to prevent
> future codetag type implementations from returning NULL from their
> cttype->desc.alloc_section_mem() operation.
>
> Fixes: 61c9e58f3a10 ("alloc_tag: load module tags into separate contiguous memory")
> Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
> Closes: https://lore.kernel.org/all/50f12fa1-17c1-4940-a6bf-beaf61f6b17a@stanley.mountain/
> Signed-off-by: Suren Baghdasaryan <surenb@google.com>

Andrew, I was going to respin v5 of my patchset and include all these
small fixes in it but it's a bit tricky because I would have to revert
another unrelated patch [1] from mm-unstable which refactors relevant
code. So far the fixes are rather small, so I think you should not
have much trouble folding them into the original patchset when the
time comes, but if that becomes a problem I can prepare a new version.

[1] d0d77a256a75 ("mm/codetag: uninline and move pgalloc_tag_copy and
pgalloc_tag_split")

> ---
>  kernel/module/main.c | 4 ++++
>  lib/alloc_tag.c      | 2 +-
>  lib/codetag.c        | 2 +-
>  3 files changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/module/main.c b/kernel/module/main.c
> index 129c98e6380d..00c16f5c5568 100644
> --- a/kernel/module/main.c
> +++ b/kernel/module/main.c
> @@ -2316,6 +2316,10 @@ static int move_module(struct module *mod, struct load_info *info)
>                 if (codetag_needs_module_section(mod, sname, shdr->sh_size)) {
>                         dest = codetag_alloc_module_section(mod, sname, shdr->sh_size,
>                                         arch_mod_section_prepend(mod, i), shdr->sh_addralign);
> +                       if (WARN_ON(!dest)) {
> +                               ret = -EINVAL;
> +                               goto out_err;
> +                       }
>                         if (IS_ERR(dest)) {
>                                 ret = PTR_ERR(dest);
>                                 goto out_err;
> diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> index 1c74942e6dfd..00ab18ea452a 100644
> --- a/lib/alloc_tag.c
> +++ b/lib/alloc_tag.c
> @@ -437,7 +437,7 @@ static void *reserve_module_tags(struct module *mod, unsigned long size,
>
>         /* If no tags return NULL */
>         if (size < sizeof(struct alloc_tag))
> -               return NULL;
> +               return ERR_PTR(-EINVAL);
>
>         /*
>          * align is always power of 2, so we can use IS_ALIGNED and ALIGN.
> diff --git a/lib/codetag.c b/lib/codetag.c
> index 4949511b4933..42aadd6c1454 100644
> --- a/lib/codetag.c
> +++ b/lib/codetag.c
> @@ -244,7 +244,7 @@ void *codetag_alloc_module_section(struct module *mod, const char *name,
>  {
>         const char *type_name = name + strlen(CODETAG_SECTION_PREFIX);
>         struct codetag_type *cttype;
> -       void *ret = NULL;
> +       void *ret = ERR_PTR(-EINVAL);
>
>         mutex_lock(&codetag_lock);
>         list_for_each_entry(cttype, &codetag_types, link) {
>
> base-commit: 758db9ca0107bc6c00f0ed4808974d66c8dc2fea
> --
> 2.47.0.163.g1226f6d8fa-goog
>

Re: [PATCH 1/1] alloc_tag: fix empty codetag module section handling

[Andrew Morton]

On Thu, 31 Oct 2024 17:13:58 -0700 Suren Baghdasaryan <surenb@google.com> wrote:

> On Thu, Oct 31, 2024 at 5:00 PM Suren Baghdasaryan <surenb@google.com> wrote:
> >
> > When a module does not have any allocations, it's allocation tag section
> > is empty and codetag_alloc_module_section() returns NULL. However this
> > condition should never happen because codetag_needs_module_section() will
> > detect an empty section and avoid calling codetag_alloc_module_section().
> > Change codetag_alloc_module_section() to never return NULL, which should
> > prevent static checker warnings. Add a WARN_ON() and a proper error
> > reporting in case codetag_alloc_module_section() returns NULL, to prevent
> > future codetag type implementations from returning NULL from their
> > cttype->desc.alloc_section_mem() operation.
> >
> > Fixes: 61c9e58f3a10 ("alloc_tag: load module tags into separate contiguous memory")
> > Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
> > Closes: https://lore.kernel.org/all/50f12fa1-17c1-4940-a6bf-beaf61f6b17a@stanley.mountain/
> > Signed-off-by: Suren Baghdasaryan <surenb@google.com>
> 
> Andrew, I was going to respin v5 of my patchset and include all these
> small fixes in it but it's a bit tricky because I would have to revert
> another unrelated patch [1] from mm-unstable which refactors relevant
> code. So far the fixes are rather small, so I think you should not
> have much trouble folding them into the original patchset when the
> time comes, but if that becomes a problem I can prepare a new version.
> 

No probs, thanks, I figured it out.  The descriptions of
where-it-fits-and-why are helpful.

Of course we could just pile everything onto mm-unstable HEAD and live
with a messier commit history, but I think things are manageable at
present.

Re: [PATCH 1/1] alloc_tag: fix empty codetag module section handling

[Dan Carpenter]

On Thu, Oct 31, 2024 at 05:00:17PM -0700, Suren Baghdasaryan wrote:
> diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> index 1c74942e6dfd..00ab18ea452a 100644
> --- a/lib/alloc_tag.c
> +++ b/lib/alloc_tag.c
> @@ -437,7 +437,7 @@ static void *reserve_module_tags(struct module *mod, unsigned long size,
>  
>  	/* If no tags return NULL */
                      ^^^^^^^^^^^

>  	if (size < sizeof(struct alloc_tag))
> -		return NULL;
> +		return ERR_PTR(-EINVAL);
>  

Thanks for this.  We'd want to update the comment as well.

regards,
dan carpenter