Skip to contentSkip to conversation

hfsplus: add a check for hfs_bnode_find

PATCH In review Chuhong Yuan 3 participants 5 messages Oct 16, 2019 – Oct 18, 2019

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

PATCH 2/2 hfsplus: add a check for hfs_bnode_find

hfs_brec_update_parent misses a check for hfs_bnode_find and may miss the failure.
Add a check for it like what is done in again.

---
 fs/hfsplus/brec.c | 2 ++
 1 file changed, 2 insertions(+)
fs/hfsplus/brec.c +2 -0
index 1918544a7871..22bada8288c4 100644
@@ -434,6 +434,8 @@ static int hfs_brec_update_parent(struct hfs_find_data *fd)
 			new_node->parent = tree->root;
 		}
 		fd->bnode = hfs_bnode_find(tree, new_node->parent);
+		if (IS_ERR(fd->bnode))
+			return PTR_ERR(fd->bnode);
 		/* create index key and entry */
 		hfs_bnode_read_key(new_node, fd->search_key, 14);
 		cnid = cpu_to_be32(new_node->this);
Signature · 1 line
2.20.1
#1 lore raw

Hi,

Quoted 14 lines · Chuhong Yuan

hfs_brec_update_parent misses a check for hfs_bnode_find and may miss the failure.
Add a check for it like what is done in again.

---
 fs/hfsplus/brec.c | 2 ++
 1 file changed, 2 insertions(+)
fs/hfsplus/brec.c +0 -0
index 1918544a7871..22bada8288c4 100644

fs/hfsplus/brec.c static int hfs_brec_update_parent(struct hfs_find_data *fd) · line 438

Quoted 5 lines
fs/hfsplus/brec.c +2 -0
@@ -434,6 +434,8 @@ static int hfs_brec_update_parent(struct hfs_find_data *fd)
 			new_node->parent = tree->root;
 		}
 		fd->bnode = hfs_bnode_find(tree, new_node->parent);
+		if (IS_ERR(fd->bnode))
+			return PTR_ERR(fd->bnode);

You shouldn't just return here, you still hold a reference to new_node. The call to hfs_bnode_find() after the again label seems to be making a similar mistake.

I don't think either one can actually fail though, because the parent nodes have all been read and hashed before, haven't they?

Quoted 6 lines
fs/hfsplus/brec.c +0 -0
 		/* create index key and entry */
 		hfs_bnode_read_key(new_node, fd->search_key, 14);
 		cnid = cpu_to_be32(new_node->this);
Signature · 1 line
2.20.1
#2 lore raw
Quoted 28 lines · Ernesto A. Fernández

Hi,

Quoted 14 lines · Chuhong Yuan

hfs_brec_update_parent misses a check for hfs_bnode_find and may miss the failure.
Add a check for it like what is done in again.

---
 fs/hfsplus/brec.c | 2 ++
 1 file changed, 2 insertions(+)
fs/hfsplus/brec.c +0 -0
index 1918544a7871..22bada8288c4 100644

fs/hfsplus/brec.c static int hfs_brec_update_parent(struct hfs_find_data *fd) · line 438

Quoted 5 lines
fs/hfsplus/brec.c +2 -0
@@ -434,6 +434,8 @@ static int hfs_brec_update_parent(struct hfs_find_data *fd)
                      new_node->parent = tree->root;
              }
              fd->bnode = hfs_bnode_find(tree, new_node->parent);
+             if (IS_ERR(fd->bnode))
+                     return PTR_ERR(fd->bnode);

You shouldn't just return here, you still hold a reference to new_node. The call to hfs_bnode_find() after the again label seems to be making a similar mistake.

Quoted 3 lines

I don't think either one can actually fail though, because the parent nodes have all been read and hashed before, haven't they?

I find that after hfs_bnode_findhash in hfs_bnode_find, there is a test for HFS_BNODE_ERROR and may return an error. I'm not sure whether it can happen here.

Quoted 6 lines
Quoted 6 lines
fs/hfsplus/brec.c +0 -1
              /* create index key and entry */
              hfs_bnode_read_key(new_node, fd->search_key, 14);
              cnid = cpu_to_be32(new_node->this);
--

2.20.1

#3 lore raw
Quoted 30 lines · Chuhong Yuan
Quoted 28 lines · Ernesto A. Fernández

Hi,

Quoted 14 lines · Chuhong Yuan

hfs_brec_update_parent misses a check for hfs_bnode_find and may miss the failure.
Add a check for it like what is done in again.

---
 fs/hfsplus/brec.c | 2 ++
 1 file changed, 2 insertions(+)
fs/hfsplus/brec.c +0 -0
index 1918544a7871..22bada8288c4 100644

fs/hfsplus/brec.c static int hfs_brec_update_parent(struct hfs_find_data *fd) · line 438

Quoted 5 lines
fs/hfsplus/brec.c +2 -0
@@ -434,6 +434,8 @@ static int hfs_brec_update_parent(struct hfs_find_data *fd)
                      new_node->parent = tree->root;
              }
              fd->bnode = hfs_bnode_find(tree, new_node->parent);
+             if (IS_ERR(fd->bnode))
+                     return PTR_ERR(fd->bnode);

You shouldn't just return here, you still hold a reference to new_node. The call to hfs_bnode_find() after the again label seems to be making a similar mistake.

Quoted 7 lines
Quoted 3 lines

I don't think either one can actually fail though, because the parent nodes have all been read and hashed before, haven't they?

I find that after hfs_bnode_findhash in hfs_bnode_find, there is a test for HFS_BNODE_ERROR and may return an error. I'm not sure whether it can happen here.

That would require a race between hfs_bnode_find() and hfs_bnode_create(), but the node has already been created.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
fs/hfsplus/brec.c +0 -1
              /* create index key and entry */
              hfs_bnode_read_key(new_node, fd->search_key, 14);
              cnid = cpu_to_be32(new_node->this);
--

2.20.1

#4 lore raw

Sorry, I had some glitch during message sending. I am repeating the message sending.

Quoted 37 lines · Ernesto A. Fernández
Quoted 34 lines · Chuhong Yuan
Quoted 31 lines · Ernesto A. Fernández

Hi,

Quoted 14 lines · Chuhong Yuan

hfs_brec_update_parent misses a check for hfs_bnode_find and may miss the failure.
Add a check for it like what is done in again.

---
fs/hfsplus/brec.c | 2 ++

1 file changed, 2 insertions(+)
fs/hfsplus/brec.c +0 -0
index 1918544a7871..22bada8288c4 100644

fs/hfsplus/brec.c static int hfs_brec_update_parent(struct hfs_find_data *fd) · line 438

Quoted 5 lines
fs/hfsplus/brec.c +2 -0
@@ -434,6 +434,8 @@ static int hfs_brec_update_parent(struct hfs_find_data *fd)
                     new_node->parent = tree->root;
             }
             fd->bnode = hfs_bnode_find(tree, new_node->parent);
+             if (IS_ERR(fd->bnode))
+                     return PTR_ERR(fd->bnode);

You shouldn't just return here, you still hold a reference to new_node. The call to hfs_bnode_find() after the again label seems to be making a similar mistake.

I don't think either one can actually fail though, because the parent nodes have all been read and hashed before, haven't they?

Quoted 7 lines
Quoted 3 lines

I find that after hfs_bnode_findhash in hfs_bnode_find, there is a test for HFS_BNODE_ERROR and may return an error. I'm not sure whether it can happen here.

That would require a race between hfs_bnode_find() and hfs_bnode_create(), but the node has already been created.

The whole function hfs_brec_update_parent() looks like the cycle. And there are several places where PTR_ERR(node) is returned with error ([1] - [2]). So, it sounds that it needs to follow this pattern or to rework these cases too. And, by the way, what if the node pointer will be NULL?

Thanks,
Viacheslav Dubeyko.

[1] https://elixir.bootlin.com/linux/latest/source/fs/hfsplus/brec.c#L371
[2] https://elixir.bootlin.com/linux/latest/source/fs/hfsplus/brec.c#L402

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
fs/hfsplus/brec.c +0 -1
             /* create index key and entry */
             hfs_bnode_read_key(new_node, fd->search_key, 14);
             cnid = cpu_to_be32(new_node->this);
--

2.20.1