Skip to contentSkip to conversation

dma-buf: Move dma_buf_release() from fops to dentry_ops

PATCH In review Sumit Semwal 3 participants 6 messages Jun 11, 2020 – Jun 22, 2020

Subsystems (3)
lore
Share & tools
lore

Sumit Semwal

author maintainer
#0 lore raw

PATCH v2 1/1 dma-buf: Move dma_buf_release() from fops to dentry_ops

Charan Teja reported a 'use-after-free' in dmabuffs_dname [1], which happens if the dma_buf_release() is called while the userspace is accessing the dma_buf pseudo fs's dmabuffs_dname() in another process, and dma_buf_release() releases the dmabuf object when the last reference to the struct file goes away.

I discussed with Arnd Bergmann, and he suggested that rather than tying the dma_buf_release() to the file_operations' release(), we can tie it to the dentry_operations' d_release(), which will be called when the last ref to the dentry is removed.

The path exercised by __fput() calls f_op->release() first, and then calls dput, which eventually calls d_op->d_release().

In the 'normal' case, when no userspace access is happening via dma_buf pseudo fs, there should be exactly one fd, file, dentry and inode, so closing the fd will kill of everything right away.

In the presented case, the dentry's d_release() will be called only when the dentry's last ref is released.

Therefore, lets move dma_buf_release() from fops->release() to d_ops->d_release()

Many thanks to Arnd for his FS insights :)

[1]: https://lore.kernel.org/patchwork/patch/1238278/

Fixes bb2bb9030425 ("dma-buf: add DMA_BUF_SET_NAME ioctls")
Reported-by [email protected]
Cc <[email protected]> [5.3+]
Cc Arnd Bergmann <[email protected]>

---
v2: per Arnd: Moved dma_buf_release() above to avoid forward declaration;
removed dentry_ops check.

---
 drivers/dma-buf/dma-buf.c | 54 ++++++++++++++++++---------------------
 1 file changed, 25 insertions(+), 29 deletions(-)
drivers/dma-buf/dma-buf.c +25 -29
index 01ce125f8e8d..412629601ad3 100644
@@ -54,37 +54,11 @@ static char *dmabuffs_dname(struct dentry *dentry, char *buffer, int buflen)
 			     dentry->d_name.name, ret > 0 ? name : "");
 }
 
-static const struct dentry_operations dma_buf_dentry_ops = {
-	.d_dname = dmabuffs_dname,
-};
-
-static struct vfsmount *dma_buf_mnt;
-
-static int dma_buf_fs_init_context(struct fs_context *fc)
-{
-	struct pseudo_fs_context *ctx;
-
-	ctx = init_pseudo(fc, DMA_BUF_MAGIC);
-	if (!ctx)
-		return -ENOMEM;
-	ctx->dops = &dma_buf_dentry_ops;
-	return 0;
-}
-
-static struct file_system_type dma_buf_fs_type = {
-	.name = "dmabuf",
-	.init_fs_context = dma_buf_fs_init_context,
-	.kill_sb = kill_anon_super,
-};
-
-static int dma_buf_release(struct inode *inode, struct file *file)
+static void dma_buf_release(struct dentry *dentry)
 {
 	struct dma_buf *dmabuf;
 
-	if (!is_dma_buf_file(file))
-		return -EINVAL;
-
-	dmabuf = file->private_data;
+	dmabuf = dentry->d_fsdata;
 
 	BUG_ON(dmabuf->vmapping_counter);
 
@@ -110,9 +84,32 @@ static int dma_buf_release(struct inode *inode, struct file *file)
 	module_put(dmabuf->owner);
 	kfree(dmabuf->name);
 	kfree(dmabuf);
+}
+
+static const struct dentry_operations dma_buf_dentry_ops = {
+	.d_dname = dmabuffs_dname,
+	.d_release = dma_buf_release,
+};
+
+static struct vfsmount *dma_buf_mnt;
+
+static int dma_buf_fs_init_context(struct fs_context *fc)
+{
+	struct pseudo_fs_context *ctx;
+
+	ctx = init_pseudo(fc, DMA_BUF_MAGIC);
+	if (!ctx)
+		return -ENOMEM;
+	ctx->dops = &dma_buf_dentry_ops;
 	return 0;
 }
 
+static struct file_system_type dma_buf_fs_type = {
+	.name = "dmabuf",
+	.init_fs_context = dma_buf_fs_init_context,
+	.kill_sb = kill_anon_super,
+};
+
 static int dma_buf_mmap_internal(struct file *file, struct vm_area_struct *vma)
 {
 	struct dma_buf *dmabuf;
@@ -412,7 +409,6 @@ static void dma_buf_show_fdinfo(struct seq_file *m, struct file *file)
 }
 
 static const struct file_operations dma_buf_fops = {
-	.release	= dma_buf_release,
 	.mmap		= dma_buf_mmap_internal,
 	.llseek		= dma_buf_llseek,
 	.poll		= dma_buf_poll,
Signature · 1 line
2.27.0
author maintainer
#1 lore raw

Hello,

If there are no objections to this, I will plan to merge it soon.

Daniel, Chris, Chenbo?

Quoted 130 lines · Sumit Semwal

Charan Teja reported a 'use-after-free' in dmabuffs_dname [1], which happens if the dma_buf_release() is called while the userspace is accessing the dma_buf pseudo fs's dmabuffs_dname() in another process, and dma_buf_release() releases the dmabuf object when the last reference to the struct file goes away.

I discussed with Arnd Bergmann, and he suggested that rather than tying the dma_buf_release() to the file_operations' release(), we can tie it to the dentry_operations' d_release(), which will be called when the last ref to the dentry is removed.

The path exercised by __fput() calls f_op->release() first, and then calls dput, which eventually calls d_op->d_release().

In the 'normal' case, when no userspace access is happening via dma_buf pseudo fs, there should be exactly one fd, file, dentry and inode, so closing the fd will kill of everything right away.

In the presented case, the dentry's d_release() will be called only when the dentry's last ref is released.

Therefore, lets move dma_buf_release() from fops->release() to d_ops->d_release()

Many thanks to Arnd for his FS insights :)

[1]: https://lore.kernel.org/patchwork/patch/1238278/

Fixes bb2bb9030425 ("dma-buf: add DMA_BUF_SET_NAME ioctls")
Reported-by [email protected]
Cc <[email protected]> [5.3+]
Cc Arnd Bergmann <[email protected]>

---
v2: per Arnd: Moved dma_buf_release() above to avoid forward declaration;
removed dentry_ops check.

---
 drivers/dma-buf/dma-buf.c | 54 ++++++++++++++++++---------------------
 1 file changed, 25 insertions(+), 29 deletions(-)
drivers/dma-buf/dma-buf.c +25 -30
index 01ce125f8e8d..412629601ad3 100644
@@ -54,37 +54,11 @@ static char *dmabuffs_dname(struct dentry *dentry, char *buffer, int buflen)
                             dentry->d_name.name, ret > 0 ? name : "");
 }
 
-static const struct dentry_operations dma_buf_dentry_ops = {
-       .d_dname = dmabuffs_dname,
-};
-
-static struct vfsmount *dma_buf_mnt;
-
-static int dma_buf_fs_init_context(struct fs_context *fc)
-{
-       struct pseudo_fs_context *ctx;
-
-       ctx = init_pseudo(fc, DMA_BUF_MAGIC);
-       if (!ctx)
-               return -ENOMEM;
-       ctx->dops = &dma_buf_dentry_ops;
-       return 0;
-}
-
-static struct file_system_type dma_buf_fs_type = {
-       .name = "dmabuf",
-       .init_fs_context = dma_buf_fs_init_context,
-       .kill_sb = kill_anon_super,
-};
-
-static int dma_buf_release(struct inode *inode, struct file *file)
+static void dma_buf_release(struct dentry *dentry)
 {
        struct dma_buf *dmabuf;
 
-       if (!is_dma_buf_file(file))
-               return -EINVAL;
-
-       dmabuf = file->private_data;
+       dmabuf = dentry->d_fsdata;
 
        BUG_ON(dmabuf->vmapping_counter);
 
@@ -110,9 +84,32 @@ static int dma_buf_release(struct inode *inode, struct file *file)
        module_put(dmabuf->owner);
        kfree(dmabuf->name);
        kfree(dmabuf);
+}
+
+static const struct dentry_operations dma_buf_dentry_ops = {
+       .d_dname = dmabuffs_dname,
+       .d_release = dma_buf_release,
+};
+
+static struct vfsmount *dma_buf_mnt;
+
+static int dma_buf_fs_init_context(struct fs_context *fc)
+{
+       struct pseudo_fs_context *ctx;
+
+       ctx = init_pseudo(fc, DMA_BUF_MAGIC);
+       if (!ctx)
+               return -ENOMEM;
+       ctx->dops = &dma_buf_dentry_ops;
        return 0;
 }
 
+static struct file_system_type dma_buf_fs_type = {
+       .name = "dmabuf",
+       .init_fs_context = dma_buf_fs_init_context,
+       .kill_sb = kill_anon_super,
+};
+
 static int dma_buf_mmap_internal(struct file *file, struct vm_area_struct *vma)
 {
        struct dma_buf *dmabuf;
@@ -412,7 +409,6 @@ static void dma_buf_show_fdinfo(struct seq_file *m, struct file *file)
 }
 
 static const struct file_operations dma_buf_fops = {
-       .release        = dma_buf_release,
        .mmap           = dma_buf_mmap_internal,
        .llseek         = dma_buf_llseek,
        .poll           = dma_buf_poll,
--
Quoted 2 lines

2.27.0

Best,
Sumit.
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#2 lore raw

Quoting Sumit Semwal (2020-06-16 13:42:13)

Quoted 3 lines

Hello,

If there are no objections to this, I will plan to merge it soon.

I was going to suggest running it against our CI, but that's unavailable at the moment.

There's a particularly nasty BUG_ON() in dma_buf_release that we hit irregularly, that this might help with.
-Chris
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#3 lore raw

Hi Chris,

Quoted 11 lines · Chris Wilson

Quoting Sumit Semwal (2020-06-16 13:42:13)

Quoted 3 lines

Hello,

If there are no objections to this, I will plan to merge it soon.

I was going to suggest running it against our CI, but that's unavailable at the moment.

There's a particularly nasty BUG_ON() in dma_buf_release that we hit irregularly, that this might help with.

Thanks for your reply; if the CI is going to be available in a couple of days, we could wait - it'd be definitely good to see a bug being splattered out!

Quoted 1 line

-Chris

Best,
Sumit.
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Tested-by on v2 1/1
#4 lore raw
In reply to Sumit Semwal

Thanks Sumit for the fix.

Quoted 28 lines · Sumit Semwal

Charan Teja reported a 'use-after-free' in dmabuffs_dname [1], which happens if the dma_buf_release() is called while the userspace is accessing the dma_buf pseudo fs's dmabuffs_dname() in another process, and dma_buf_release() releases the dmabuf object when the last reference to the struct file goes away.

I discussed with Arnd Bergmann, and he suggested that rather than tying the dma_buf_release() to the file_operations' release(), we can tie it to the dentry_operations' d_release(), which will be called when the last ref to the dentry is removed.

The path exercised by __fput() calls f_op->release() first, and then calls dput, which eventually calls d_op->d_release().

In the 'normal' case, when no userspace access is happening via dma_buf pseudo fs, there should be exactly one fd, file, dentry and inode, so closing the fd will kill of everything right away.

In the presented case, the dentry's d_release() will be called only when the dentry's last ref is released.

Therefore, lets move dma_buf_release() from fops->release() to d_ops->d_release()

Many thanks to Arnd for his FS insights :)

[1]: https://lore.kernel.org/patchwork/patch/1238278/

Quoted 8 lines
Fixes bb2bb9030425 ("dma-buf: add DMA_BUF_SET_NAME ioctls")
Reported-by [email protected]
Cc <[email protected]> [5.3+]
Cc Arnd Bergmann <[email protected]>

Tested this patch for Android running on Snapdragon hardware and see no issues.

Quoted 93 lines

---
v2: per Arnd: Moved dma_buf_release() above to avoid forward declaration;
removed dentry_ops check.

---
 drivers/dma-buf/dma-buf.c | 54 ++++++++++++++++++---------------------
 1 file changed, 25 insertions(+), 29 deletions(-)
drivers/dma-buf/dma-buf.c +25 -29
index 01ce125f8e8d..412629601ad3 100644
@@ -54,37 +54,11 @@ static char *dmabuffs_dname(struct dentry *dentry, char *buffer, int buflen)
 			     dentry->d_name.name, ret > 0 ? name : "");
 }
 
-static const struct dentry_operations dma_buf_dentry_ops = {
-	.d_dname = dmabuffs_dname,
-};
-
-static struct vfsmount *dma_buf_mnt;
-
-static int dma_buf_fs_init_context(struct fs_context *fc)
-{
-	struct pseudo_fs_context *ctx;
-
-	ctx = init_pseudo(fc, DMA_BUF_MAGIC);
-	if (!ctx)
-		return -ENOMEM;
-	ctx->dops = &dma_buf_dentry_ops;
-	return 0;
-}
-
-static struct file_system_type dma_buf_fs_type = {
-	.name = "dmabuf",
-	.init_fs_context = dma_buf_fs_init_context,
-	.kill_sb = kill_anon_super,
-};
-
-static int dma_buf_release(struct inode *inode, struct file *file)
+static void dma_buf_release(struct dentry *dentry)
 {
 	struct dma_buf *dmabuf;
 
-	if (!is_dma_buf_file(file))
-		return -EINVAL;
-
-	dmabuf = file->private_data;
+	dmabuf = dentry->d_fsdata;
 
 	BUG_ON(dmabuf->vmapping_counter);
 
@@ -110,9 +84,32 @@ static int dma_buf_release(struct inode *inode, struct file *file)
 	module_put(dmabuf->owner);
 	kfree(dmabuf->name);
 	kfree(dmabuf);
+}
+
+static const struct dentry_operations dma_buf_dentry_ops = {
+	.d_dname = dmabuffs_dname,
+	.d_release = dma_buf_release,
+};
+
+static struct vfsmount *dma_buf_mnt;
+
+static int dma_buf_fs_init_context(struct fs_context *fc)
+{
+	struct pseudo_fs_context *ctx;
+
+	ctx = init_pseudo(fc, DMA_BUF_MAGIC);
+	if (!ctx)
+		return -ENOMEM;
+	ctx->dops = &dma_buf_dentry_ops;
 	return 0;
 }
 
+static struct file_system_type dma_buf_fs_type = {
+	.name = "dmabuf",
+	.init_fs_context = dma_buf_fs_init_context,
+	.kill_sb = kill_anon_super,
+};
+
 static int dma_buf_mmap_internal(struct file *file, struct vm_area_struct *vma)
 {
 	struct dma_buf *dmabuf;
@@ -412,7 +409,6 @@ static void dma_buf_show_fdinfo(struct seq_file *m, struct file *file)
 }
 
 static const struct file_operations dma_buf_fops = {
-	.release	= dma_buf_release,
 	.mmap		= dma_buf_mmap_internal,
 	.llseek		= dma_buf_llseek,
 	.poll		= dma_buf_poll,
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
Forum, a Linux Foundation Collaborative Project
author maintainer
#5 lore raw

Hello Charan,

Quoted 41 lines · Charan Teja Kalla

Thanks Sumit for the fix.

Quoted 28 lines · Sumit Semwal

Charan Teja reported a 'use-after-free' in dmabuffs_dname [1], which happens if the dma_buf_release() is called while the userspace is accessing the dma_buf pseudo fs's dmabuffs_dname() in another process, and dma_buf_release() releases the dmabuf object when the last reference to the struct file goes away.

I discussed with Arnd Bergmann, and he suggested that rather than tying the dma_buf_release() to the file_operations' release(), we can tie it to the dentry_operations' d_release(), which will be called when the last ref to the dentry is removed.

The path exercised by __fput() calls f_op->release() first, and then calls dput, which eventually calls d_op->d_release().

In the 'normal' case, when no userspace access is happening via dma_buf pseudo fs, there should be exactly one fd, file, dentry and inode, so closing the fd will kill of everything right away.

In the presented case, the dentry's d_release() will be called only when the dentry's last ref is released.

Therefore, lets move dma_buf_release() from fops->release() to d_ops->d_release()

Many thanks to Arnd for his FS insights :)

[1]: https://lore.kernel.org/patchwork/patch/1238278/

Quoted 8 lines
Fixes bb2bb9030425 ("dma-buf: add DMA_BUF_SET_NAME ioctls")
Reported-by [email protected]
Cc <[email protected]> [5.3+]
Cc Arnd Bergmann <[email protected]>
Quoted 3 lines

Tested this patch for Android running on Snapdragon hardware and see no issues.

Thanks for your tested-by, appreciate it!

Chris,
any luck with your CI to test if this also helps your dma_buf_release() bug that you guys have seen?

If you've not been able to test, and there are no more objections by EOD today, I will merge this to the drm-misc-fixes branch.

Quoted 5 lines
Quoted 5 lines

---
v2: per Arnd: Moved dma_buf_release() above to avoid forward declaration;
removed dentry_ops check.
---

<snip>
Best,
Sumit.