Skip to contentSkip to conversation

ovl: Add check for missing lookup operation on inode

PATCH In review Vasiliy Kovalev 4 participants 8 messages Nov 18, 2024 – Nov 23, 2024 Overlay filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ovl: Add check for missing lookup operation on inode

Ensure that the lookup operation is present for the inode in the overlay filesystem. If the operation is missing, log a warning and return an EIO error to prevent further issues in the lookup process.

---
 fs/overlayfs/namei.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/overlayfs/namei.c +7 -0
index 5764f91d283e7..a73f37e401cf0 100644
@@ -1115,6 +1115,13 @@ struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 	for (i = 0; !d.stop && i < ovl_numlower(poe); i++) {
 		struct ovl_path lower = ovl_lowerstack(poe)[i];
 
+		if (!lower.dentry->d_inode->i_op->lookup) {
+			err = -EIO;
+			pr_warn_ratelimited("missing lookup operation for inode %p\n",
+								lower.dentry->d_inode);
+			goto out_put;
+		}
+
 		if (!ovl_redirect_follow(ofs))
 			d.last = i == ovl_numlower(poe) - 1;
 		else if (d.is_dir || !ofs->numdatalayer)
Signature · 1 line
2.33.8
maintainer
#1 lore raw
Quoted 17 lines · Vasiliy Kovalev

Ensure that the lookup operation is present for the inode in the overlay filesystem. If the operation is missing, log a warning and return an EIO error to prevent further issues in the lookup process.

---
 fs/overlayfs/namei.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/overlayfs/namei.c +0 -0
index 5764f91d283e7..a73f37e401cf0 100644

fs/overlayfs/namei.c struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry, · line 1124

Quoted 10 lines
fs/overlayfs/namei.c +7 -0
@@ -1115,6 +1115,13 @@ struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
        for (i = 0; !d.stop && i < ovl_numlower(poe); i++) {
                struct ovl_path lower = ovl_lowerstack(poe)[i];
 
+               if (!lower.dentry->d_inode->i_op->lookup) {
+                       err = -EIO;
+                       pr_warn_ratelimited("missing lookup operation for inode %p\n",
+                                                               lower.dentry->d_inode);
+                       goto out_put;
+               }
+

This looks like it is papering over a bug.
The dentries in the poe lower stack are supposed to be d_can_lookup(), which means that they should have a ->lookup op.

See in ovl_lookup_single():

         if (!d_can_lookup(this)) {
                if (d->is_dir || !last_element) {
                        d->stop = true;
                        goto put_and_out;
                }

Can you analyse what went wrong with the reproducer? How did we get to a state where lowerstack of parent has a dentry which is !d_can_lookup?

Thanks,
Amir.

maintainer
#2 lore raw
Quoted 3 lines · Amir Goldstein

Can you analyse what went wrong with the reproducer? How did we get to a state where lowerstack of parent has a dentry which is !d_can_lookup?

Theoretically we could still get a an S_ISDIR inode, because ovl_get_inode() doesn't look at the is_dir value that lookup found. I.e. lookup thinks it found a non-dir, but iget will create a dir because of the backing inode's type.

AFAICS this can only happen if i_op->lookup is not set on S_ISDIR for the backing inode, which shouldn't happen on normal filesystems. Reproducer seems to use bfs, which *should* be normal, and bfs_iget certainly doesn't do anything weird in that case, so I still don't understand what is happening.

In any case something like the following should filter out such weirdness:

bool ovl_dentry_weird(struct dentry *dentry)
{
+ if (!d_can_lookup(dentry) && !d_is_file(dentry) &&
!d_is_symlink(dentry))
+ return true;
+

        return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |

Thanks,
Miklos

#3 lore raw

Hi,

Quoted 11 lines · Miklos Szeredi
Quoted 3 lines · Amir Goldstein

Can you analyse what went wrong with the reproducer? How did we get to a state where lowerstack of parent has a dentry which is !d_can_lookup?

Theoretically we could still get a an S_ISDIR inode, because ovl_get_inode() doesn't look at the is_dir value that lookup found. I.e. lookup thinks it found a non-dir, but iget will create a dir because of the backing inode's type.

Quoted 5 lines

AFAICS this can only happen if i_op->lookup is not set on S_ISDIR for the backing inode, which shouldn't happen on normal filesystems. Reproducer seems to use bfs, which *should* be normal, and bfs_iget certainly doesn't do anything weird in that case, so I still don't understand what is happening.

During testing, it was discovered that BFS can return a directory inode without a lookup operation. Adding the following check in bfs_iget:

struct inode *bfs_iget(struct super_block *sb, unsigned long ino) {

...

	brelse(bh);

+ if (S_ISDIR(inode->i_mode) && !inode->i_op->lookup) {
+ printf("Directory inode missing lookup %s:%08lx\n",

						inode->i_sb->s_id, ino);
+		goto error;
+	}
+
	unlock_new_inode(inode);
	return inode;

error:

	iget_failed(inode);
	return ERR_PTR(-EIO);
}

prevents the error but exposes an invalid inode:

loop0: detected capacity change from 0 to 64
BFS-fs: bfs_iget(): Directory inode missing lookup loop0:00000002
overlayfs: overlapping lowerdir path

Would this be considered a valid workaround, or does BFS require further fixes?

Review

Quoted 9 lines

In any case something like the following should filter out such weirdness:

bool ovl_dentry_weird(struct dentry *dentry)
{
+ if (!d_can_lookup(dentry) && !d_is_file(dentry) &&
!d_is_symlink(dentry))

(quoted diff) +2 -0
+               return true;
+
         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |

I tested this addition successfully.

Additionally, fixes for BFS seem to be discussed reluctantly.
For instance, this patch set [1] has remained unanswered.
Perhaps it would be worth considering discarding invalid inodes at the overlayfs level?

[1] https://lore.kernel.org/all/[email protected]/

Quoted 2 lines

Thanks,
Miklos

--
Thanks,
Vasiliy Kovalev

maintainer
#4 lore raw
Quoted 63 lines · Vasiliy Kovalev

Hi,

Quoted 11 lines · Miklos Szeredi
Quoted 3 lines · Amir Goldstein

Can you analyse what went wrong with the reproducer? How did we get to a state where lowerstack of parent has a dentry which is !d_can_lookup?

Theoretically we could still get a an S_ISDIR inode, because ovl_get_inode() doesn't look at the is_dir value that lookup found. I.e. lookup thinks it found a non-dir, but iget will create a dir because of the backing inode's type.

Quoted 5 lines

AFAICS this can only happen if i_op->lookup is not set on S_ISDIR for the backing inode, which shouldn't happen on normal filesystems. Reproducer seems to use bfs, which *should* be normal, and bfs_iget certainly doesn't do anything weird in that case, so I still don't understand what is happening.

During testing, it was discovered that BFS can return a directory inode without a lookup operation. Adding the following check in bfs_iget:

struct inode *bfs_iget(struct super_block *sb, unsigned long ino) {

...

        brelse(bh);
(quoted diff) +5 -0
+       if (S_ISDIR(inode->i_mode) && !inode->i_op->lookup) {
+               printf("Directory inode missing lookup %s:%08lx\n",
                                                inode->i_sb->s_id, ino);
+               goto error;
+       }
+
        unlock_new_inode(inode);
        return inode;

error:

        iget_failed(inode);
        return ERR_PTR(-EIO);
}

prevents the error but exposes an invalid inode:

loop0: detected capacity change from 0 to 64
BFS-fs: bfs_iget(): Directory inode missing lookup loop0:00000002
overlayfs: overlapping lowerdir path

Would this be considered a valid workaround, or does BFS require further fixes?

Review

Quoted 9 lines

In any case something like the following should filter out such weirdness:

bool ovl_dentry_weird(struct dentry *dentry)
{
+ if (!d_can_lookup(dentry) && !d_is_file(dentry) &&
!d_is_symlink(dentry))

(quoted diff) +2 -0
+               return true;
+
         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
Quoted 6 lines

I tested this addition successfully.

Additionally, fixes for BFS seem to be discussed reluctantly. For instance, this patch set [1] has remained unanswered. Perhaps it would be worth considering discarding invalid inodes at the overlayfs level?

Sure. please send proper patch following Miklos' suggestion

Thanks,
Amir.

#5 lore raw

PATCH v2 1/1 ovl: Filter invalid inodes with missing lookup function

Add a check to the ovl_dentry_weird() function to prevent the processing of directory inodes that lack the lookup function. This is important because such inodes can cause errors in overlayfs when passed to the lowerstack.

---
 fs/overlayfs/util.c | 3 +++
 1 file changed, 3 insertions(+)
fs/overlayfs/util.c +3 -0
index 3bb107471fb42..9aa7493b1e103 100644
@@ -202,6 +202,9 @@ void ovl_dentry_init_flags(struct dentry *dentry, struct dentry *upperdentry,
 
 bool ovl_dentry_weird(struct dentry *dentry)
 {
+	if (!d_can_lookup(dentry) && !d_is_file(dentry) && !d_is_symlink(dentry))
+		return true;
+
 	return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 				  DCACHE_MANAGE_TRANSIT |
 				  DCACHE_OP_HASH |
Signature · 1 line
2.33.8
maintainer Applied
#6 lore raw

Re: [PATCH v2] ovl: Filter invalid inodes with missing lookup function

Quoted 30 lines · Vasiliy Kovalev

Add a check to the ovl_dentry_weird() function to prevent the processing of directory inodes that lack the lookup function. This is important because such inodes can cause errors in overlayfs when passed to the lowerstack.

---
 fs/overlayfs/util.c | 3 +++
 1 file changed, 3 insertions(+)
fs/overlayfs/util.c +3 -1
index 3bb107471fb42..9aa7493b1e103 100644
@@ -202,6 +202,9 @@ void ovl_dentry_init_flags(struct dentry *dentry, struct dentry *upperdentry,
 
 bool ovl_dentry_weird(struct dentry *dentry)
 {
+       if (!d_can_lookup(dentry) && !d_is_file(dentry) && !d_is_symlink(dentry))
+               return true;
+
        return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
                                  DCACHE_MANAGE_TRANSIT |
                                  DCACHE_OP_HASH |
--
Quoted 2 lines

2.33.8

Applied to overlayfs-next. Will send along with 6.13 PR

Thanks,
Amir.

#7 lore raw
In reply to Vasiliy Kovalev
Quoted 22 lines · Vasiliy Kovalev

without a lookup operation. Adding the following check in bfs_iget:

struct inode *bfs_iget(struct super_block *sb, unsigned long ino) {

...

	brelse(bh);
(quoted diff) +2 -0
+	if (S_ISDIR(inode->i_mode) && !inode->i_op->lookup) {
+		printf("Directory inode missing lookup %s:%08lx\n",
						inode->i_sb->s_id, ino);
fs/overlayfs/namei.c +3 -0
+		goto error;
+	}
+
	unlock_new_inode(inode);
	return inode;

error:

	iget_failed(inode);
	return ERR_PTR(-EIO);
}
Quoted 8 lines

prevents the error but exposes an invalid inode:

loop0: detected capacity change from 0 to 64
BFS-fs: bfs_iget(): Directory inode missing lookup loop0:00000002
overlayfs: overlapping lowerdir path

Would this be considered a valid workaround, or does BFS require further fixes?

Yes, it does. Note that this

        inode->i_mode = 0x0000FFFF & le32_to_cpu(di->i_mode);

sets the bits 0..15, which includes not only the permissions (0..11), but the file type as well. And those |= are not going to be enough to prevent trouble - what we have there is

	0x1000 => FIFO
	0x2000 => CHR
	0x4000 => DIR
	0x6000 => BLK
	0x8000 => REG
	0xa000 => LNK
	0xe000 => SOCK

So depending upon ->i_vtype you get one of
* ->i_op and ->i_fop set for directory, type bits - 0x4000 | junk
* ->i_op and ->i_fop set for regular file, type bits - 0x8000 | junk
* ->i_op and ->i_fop left empty, type bits - junk.

Frankly, I would rather ignore bits 12..15 (i.e.

        inode->i_mode = 0x00000FFF & le32_to_cpu(di->i_mode);
instead of
        inode->i_mode = 0x0000FFFF & le32_to_cpu(di->i_mode);

) and complain (and fail) if ->i_vtype value is fucked up.