Skip to contentSkip to conversation

hfsplus: fix null-ptr-deref in hfsplus_cat_write_inode

PATCH In review Deepanshu Kartikey 2 participants 4 messages Apr 15 – Apr 22

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 hfsplus: fix null-ptr-deref in hfsplus_cat_write_inode

When a corrupt HFS+ image is mounted where the hidden directory is absent, hfsplus_fill_super() handles -ENOENT from hfsplus_get_hidden_dir_entry() silently and continues with sbi->hidden_dir left as NULL.

hfsplus_link() and hfsplus_unlink() call hfsplus_cat_write_inode(sbi->hidden_dir) unconditionally without checking whether hidden_dir is NULL, triggering a general protection fault caught by KASAN as a null-ptr-deref at offset 0x28.

Other call sites in dir.c already guard against this by checking hidden_dir for NULL before use. Apply the same guard to the two unprotected call sites in hfsplus_link() and hfsplus_unlink().

Changes in v3:
- Correct fix location: guard sbi->hidden_dir before calling hfsplus_cat_write_inode() in hfsplus_link() and hfsplus_unlink() in dir.c, as suggested by Vyacheslav Dubeyko.

Changes in v2:
- Fixed commit message: hfsplus_delete_cat() is called from multiple callers, not just hfsplus_unlink() as incorrectly stated in v1.

---
 fs/hfsplus/dir.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
fs/hfsplus/dir.c +8 -7
index d559bf8625f8..a7feef53d8cb 100644
@@ -362,9 +362,11 @@ static int hfsplus_link(struct dentry *src_dentry, struct inode *dst_dir,
 	if (res)
 		goto out;
 
-	res = hfsplus_cat_write_inode(sbi->hidden_dir);
-	if (res)
-		goto out;
+	if (sbi->hidden_dir) {
+		res = hfsplus_cat_write_inode(sbi->hidden_dir);
+		if (res)
+			goto out;
+	}
 
 	res = hfsplus_cat_write_inode(inode);
 
@@ -431,11 +433,10 @@ static int hfsplus_unlink(struct inode *dir, struct dentry *dentry)
 out:
 	if (!res) {
 		res = hfsplus_cat_write_inode(dir);
-		if (!res) {
+		if (!res && sbi->hidden_dir)
 			res = hfsplus_cat_write_inode(sbi->hidden_dir);
-			if (!res)
-				res = hfsplus_cat_write_inode(inode);
-		}
+		if (!res)
+			res = hfsplus_cat_write_inode(inode);
 	}
 
 	mutex_unlock(&sbi->vh_mutex);
Signature · 1 line
2.43.0
#1 lore raw
Quoted 56 lines · Deepanshu Kartikey

When a corrupt HFS+ image is mounted where the hidden directory is absent, hfsplus_fill_super() handles -ENOENT from hfsplus_get_hidden_dir_entry() silently and continues with sbi->hidden_dir left as NULL.

hfsplus_link() and hfsplus_unlink() call hfsplus_cat_write_inode(sbi->hidden_dir) unconditionally without checking whether hidden_dir is NULL, triggering a general protection fault caught by KASAN as a null-ptr-deref at offset 0x28.

Other call sites in dir.c already guard against this by checking hidden_dir for NULL before use. Apply the same guard to the two unprotected call sites in hfsplus_link() and hfsplus_unlink().

Changes in v3:
- Correct fix location: guard sbi->hidden_dir before calling hfsplus_cat_write_inode() in hfsplus_link() and hfsplus_unlink() in dir.c, as suggested by Vyacheslav Dubeyko.

Changes in v2:
- Fixed commit message: hfsplus_delete_cat() is called from multiple callers, not just hfsplus_unlink() as incorrectly stated in v1.

---
 fs/hfsplus/dir.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
fs/hfsplus/dir.c +5 -4
index d559bf8625f8..a7feef53d8cb 100644
@@ -362,9 +362,11 @@ static int hfsplus_link(struct dentry *src_dentry, struct inode *dst_dir,
 	if (res)
 		goto out;
 
-	res = hfsplus_cat_write_inode(sbi->hidden_dir);
-	if (res)
-		goto out;
+	if (sbi->hidden_dir) {
+		res = hfsplus_cat_write_inode(sbi->hidden_dir);
+		if (res)
+			goto out;
+	}
 
 	res = hfsplus_cat_write_inode(inode);
 
@@ -431,11 +433,10 @@ static int hfsplus_unlink(struct inode *dir, struct dentry *dentry)
 out:
 	if (!res) {
 		res = hfsplus_cat_write_inode(dir);
-		if (!res) {

fs/hfsplus/dir.c static int hfsplus_unlink(struct inode *dir, struct dentry *dentry) · line 442

Quoted 10 lines
fs/hfsplus/dir.c +3 -3
+		if (!res && sbi->hidden_dir)
 			res = hfsplus_cat_write_inode(sbi->hidden_dir);
-			if (!res)
-				res = hfsplus_cat_write_inode(inode);
-		}
+		if (!res)
+			res = hfsplus_cat_write_inode(inode);
 	}
 
 	mutex_unlock(&sbi->vh_mutex);

As far as I can see, we have such logic in hfsplus_fill_super() [1]:

if (!sb_rdonly(sb)) {
<skipped>

    if (!sbi->hidden_dir) {
        <create hidden dir>
    }

<skipped>
}

So, I have two questions:
(1) If we mounted volume in READ-ONLY mode initially and we re-mount it into READ-WRITE mode, then we skip hidden directory creation. Should we process this in hfsplus_reconfigure()?
(2) Should we create hidden directory if corrupted volume hasn't hidden directory?

If we don't create the hidden dir, then we could have troubles in multiple
places: hfsplus_rename_cat() [2,3], hfsplus_delete_cat() [4,5]. It looks like that we need to create the hidden dir. Otherwise, we could trigger the kernel crash in multiple places.

Thanks,
Slava.

[1] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/super.c#L589
[2] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L326
[3] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L397
[4] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L418
[5] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/inode.c#L257

#2 lore raw
Quoted 23 lines · Viacheslav Dubeyko

As far as I can see, we have such logic in hfsplus_fill_super() [1]:

if (!sb_rdonly(sb)) {
<skipped>

    if (!sbi->hidden_dir) {
        <create hidden dir>
    }

<skipped>
}

So, I have two questions:
(1) If we mounted volume in READ-ONLY mode initially and we re-mount it into READ-WRITE mode, then we skip hidden directory creation. Should we process this in hfsplus_reconfigure()?
(2) Should we create hidden directory if corrupted volume hasn't hidden directory?

If we don't create the hidden dir, then we could have troubles in multiple
places: hfsplus_rename_cat() [2,3], hfsplus_delete_cat() [4,5]. It looks like that we need to create the hidden dir. Otherwise, we could trigger the kernel crash in multiple places.

Quoted 9 lines

Thanks,
Slava.

[1] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/super.c#L589
[2] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L326
[3] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L397
[4] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L418
[5] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/inode.c#L257

Hi Vyacheslav,

Thank you for the detailed feedback and for pointing out the deeper design issue.

You are absolutely right on both counts.

(1) hfsplus_reconfigure() does not create the hidden directory when remounting from read-only to read-write. This means sbi->hidden_dir can be NULL on a writable mount even if the volume is otherwise healthy. This needs to be fixed in hfsplus_reconfigure().

(2) The correct approach is not to add NULL checks at individual call sites — that is a band-aid. The real fix is to ensure sbi->hidden_dir is always valid on any read-write mount, enforcing this invariant at mount and remount time.

My plan for v4:
- Extract the hidden directory creation logic from hfsplus_fill_super() into a helper function hfsplus_create_hidden_dir().
- Call it from hfsplus_fill_super() as before.
- Also call it from hfsplus_reconfigure() when switching from read-only to read-write and hidden_dir is NULL.

This way hidden_dir is guaranteed valid for all write operations and the individual call sites in dir.c and inode.c are safe without needing NULL guards.

I will send v4 shortly.

Regards,
Deeanshu Kartikey

#3 lore raw
Quoted 58 lines · Deepanshu Kartikey
Quoted 23 lines · Viacheslav Dubeyko

As far as I can see, we have such logic in hfsplus_fill_super() [1]:

if (!sb_rdonly(sb)) {
<skipped>

    if (!sbi->hidden_dir) {
        <create hidden dir>
    }

<skipped>
}

So, I have two questions:
(1) If we mounted volume in READ-ONLY mode initially and we re-mount it into READ-WRITE mode, then we skip hidden directory creation. Should we process this in hfsplus_reconfigure()?
(2) Should we create hidden directory if corrupted volume hasn't hidden directory?

If we don't create the hidden dir, then we could have troubles in multiple
places: hfsplus_rename_cat() [2,3], hfsplus_delete_cat() [4,5]. It looks like that we need to create the hidden dir. Otherwise, we could trigger the kernel crash in multiple places.

Quoted 9 lines

Thanks,
Slava.

[1] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/super.c#L589
[2] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L326
[3] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L397
[4] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/dir.c#L418
[5] https://elixir.bootlin.com/linux/v7.0/source/fs/hfsplus/inode.c#L257

Hi Vyacheslav,

Thank you for the detailed feedback and for pointing out the deeper design issue.

You are absolutely right on both counts.

(1) hfsplus_reconfigure() does not create the hidden directory when remounting from read-only to read-write. This means sbi->hidden_dir can be NULL on a writable mount even if the volume is otherwise healthy. This needs to be fixed in hfsplus_reconfigure().

(2) The correct approach is not to add NULL checks at individual call sites — that is a band-aid. The real fix is to ensure sbi->hidden_dir is always valid on any read-write mount, enforcing this invariant at mount and remount time.

My plan for v4:
- Extract the hidden directory creation logic from hfsplus_fill_super() into a helper function hfsplus_create_hidden_dir().
- Call it from hfsplus_fill_super() as before.
- Also call it from hfsplus_reconfigure() when switching from read-only to read-write and hidden_dir is NULL.

Quoted 4 lines

This way hidden_dir is guaranteed valid for all write operations and the individual call sites in dir.c and inode.c are safe without needing NULL guards.

Makes sense.

Quoted 3 lines

I will send v4 shortly.

Sounds good.

Thanks,
Slava.