[PATCH v2 1/2] erofs-utils: fuse: add missing return on getattr error

[PATCH v2 1/2] erofs-utils: fuse: add missing return on getattr error

[Ajay Rajera]

erofsfuse_getattr() calls fuse_reply_err() when erofs_read_inode_from_disk()
fails, but does not return afterwards. This causes the function to fall through
to erofsfuse_fill_stat() with uninitialized inode data and then call
fuse_reply_attr(), sending a second reply to the same FUSE request.

Sending two replies to a single FUSE request is undefined behavior in libfuse
and typically triggers an assertion failure or crash. The uninitialized inode
data may also expose garbage values to userspace.

Fix by adding the missing return after fuse_reply_err().

Signed-off-by: Ajay Rajera <newajay.11r@gmail.com>
---
 fuse/main.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fuse/main.c b/fuse/main.c
index 82aca8c..b634782 100644
--- a/fuse/main.c
+++ b/fuse/main.c
@@ -265,8 +265,10 @@ static void erofsfuse_getattr(fuse_req_t req, fuse_ino_t ino,
 	struct erofs_inode vi = { .sbi = &g_sbi, .nid = erofsfuse_to_nid(ino) };
 
 	ret = erofs_read_inode_from_disk(&vi);
-	if (ret < 0)
+	if (ret < 0) {
 		fuse_reply_err(req, -ret);
+		return;
+	}
 
 	erofsfuse_fill_stat(&vi, &stbuf);
 	stbuf.st_ino = ino;
-- 
2.51.0.windows.1

[PATCH v2 2/2] erofs-utils: lib: fix memory leak in erofs_gzran_builder_init error path

[Ajay Rajera]

When inflateInit2() fails, erofs_gzran_builder_init() returns an ERR_PTR(-EFAULT)
but forgets to free the previously allocated erofs_gzran_builder struct (gb),
resulting in a memory leak.

Fix by calling free(gb) before returning the error.

Signed-off-by: Ajay Rajera <newajay.11r@gmail.com>
---
 lib/gzran.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/lib/gzran.c b/lib/gzran.c
index dffb20a..8a01825 100644
--- a/lib/gzran.c
+++ b/lib/gzran.c
@@ -50,8 +50,10 @@ struct erofs_gzran_builder *erofs_gzran_builder_init(struct erofs_vfile *vf,
 	strm->avail_in = 0;
 	strm->next_in = Z_NULL;
 	ret = inflateInit2(strm, 47);	/* automatic zlib or gzip decoding */
-	if (ret != Z_OK)
+	if (ret != Z_OK) {
+		free(gb);
 		return ERR_PTR(-EFAULT);
+	}
 	gb->vf = vf;
 	gb->span_size = span_size;
 	gb->totout = gb->totin = 0;
-- 
2.51.0.windows.1

Re: [PATCH v2 1/2] erofs-utils: fuse: add missing return on getattr error

[Gao Xiang]

On 2026/3/21 14:26, Ajay Rajera wrote:
> erofsfuse_getattr() calls fuse_reply_err() when erofs_read_inode_from_disk()
> fails, but does not return afterwards. This causes the function to fall through
> to erofsfuse_fill_stat() with uninitialized inode data and then call
> fuse_reply_attr(), sending a second reply to the same FUSE request.
> 
> Sending two replies to a single FUSE request is undefined behavior in libfuse
> and typically triggers an assertion failure or crash. The uninitialized inode
> data may also expose garbage values to userspace.
> 
> Fix by adding the missing return after fuse_reply_err().

Each line of the commit message should not exceed 72 chars.

> 
> Signed-off-by: Ajay Rajera <newajay.11r@gmail.com>
> ---
>   fuse/main.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/fuse/main.c b/fuse/main.c
> index 82aca8c..b634782 100644
> --- a/fuse/main.c
> +++ b/fuse/main.c
> @@ -265,8 +265,10 @@ static void erofsfuse_getattr(fuse_req_t req, fuse_ino_t ino,
>   	struct erofs_inode vi = { .sbi = &g_sbi, .nid = erofsfuse_to_nid(ino) };
>   
>   	ret = erofs_read_inode_from_disk(&vi);
> -	if (ret < 0)
> +	if (ret < 0) {
>   		fuse_reply_err(req, -ret);
> +		return;
> +	}
>   
>   	erofsfuse_fill_stat(&vi, &stbuf);
>   	stbuf.st_ino = ino;

Re: [PATCH v2 1/2] erofs-utils: fuse: add missing return on getattr error

[Ajay Rajera]

Yeah, I apologize for the mistake.
I just sent patch v3
Thanks, Ajay.


On Sat, 21 Mar 2026 at 12:13, Gao Xiang <hsiangkao@linux.alibaba.com> wrote:
>
>
>
> On 2026/3/21 14:26, Ajay Rajera wrote:
> > erofsfuse_getattr() calls fuse_reply_err() when erofs_read_inode_from_disk()
> > fails, but does not return afterwards. This causes the function to fall through
> > to erofsfuse_fill_stat() with uninitialized inode data and then call
> > fuse_reply_attr(), sending a second reply to the same FUSE request.
> >
> > Sending two replies to a single FUSE request is undefined behavior in libfuse
> > and typically triggers an assertion failure or crash. The uninitialized inode
> > data may also expose garbage values to userspace.
> >
> > Fix by adding the missing return after fuse_reply_err().
>
> Each line of the commit message should not exceed 72 chars.
>
> >
> > Signed-off-by: Ajay Rajera <newajay.11r@gmail.com>
> > ---
> >   fuse/main.c | 4 +++-
> >   1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/fuse/main.c b/fuse/main.c
> > index 82aca8c..b634782 100644
> > --- a/fuse/main.c
> > +++ b/fuse/main.c
> > @@ -265,8 +265,10 @@ static void erofsfuse_getattr(fuse_req_t req, fuse_ino_t ino,
> >       struct erofs_inode vi = { .sbi = &g_sbi, .nid = erofsfuse_to_nid(ino) };
> >
> >       ret = erofs_read_inode_from_disk(&vi);
> > -     if (ret < 0)
> > +     if (ret < 0) {
> >               fuse_reply_err(req, -ret);
> > +             return;
> > +     }
> >
> >       erofsfuse_fill_stat(&vi, &stbuf);
> >       stbuf.st_ino = ino;
>

Re: [PATCH v2 2/2] erofs-utils: lib: fix memory leak in erofs_gzran_builder_init error path

[Gao Xiang]

On 2026/3/21 14:26, Ajay Rajera wrote:
> When inflateInit2() fails, erofs_gzran_builder_init() returns an ERR_PTR(-EFAULT)
> but forgets to free the previously allocated erofs_gzran_builder struct (gb),
> resulting in a memory leak.
> 
> Fix by calling free(gb) before returning the error.
> 
> Signed-off-by: Ajay Rajera <newajay.11r@gmail.com>

The same issue.

Re: [PATCH v2 2/2] erofs-utils: lib: fix memory leak in erofs_gzran_builder_init error path

[Ajay Rajera]

yeah, I just sent patch v3.
Thanks, Ajay.

On Sat, 21 Mar 2026 at 12:51, Gao Xiang <hsiangkao@linux.alibaba.com> wrote:
>
>
>
> On 2026/3/21 14:26, Ajay Rajera wrote:
> > When inflateInit2() fails, erofs_gzran_builder_init() returns an ERR_PTR(-EFAULT)
> > but forgets to free the previously allocated erofs_gzran_builder struct (gb),
> > resulting in a memory leak.
> >
> > Fix by calling free(gb) before returning the error.
> >
> > Signed-off-by: Ajay Rajera <newajay.11r@gmail.com>
>
> The same issue.