RE: [PATCH v3] hfs: remove BUG() from hfs_release_folio()/hfs_test_inode()/hfs_write_inode()

[Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>]

On Tue, 2025-07-22 at 15:22 +0100, Matthew Wilcox wrote:
> On Tue, Jul 22, 2025 at 11:04:30PM +0900, Tetsuo Handa wrote:
> > On 2025/07/22 22:30, Matthew Wilcox wrote:
> > > On Tue, Jul 22, 2025 at 07:42:35PM +0900, Tetsuo Handa wrote:
> > > > I can update patch description if you have one, but I don't plan to try something like below.
> > > 
> > > Why not?  Papering over the underlying problem is what I rejected in v1,
> > > and here we are months later with you trying a v4.
> > 
> > Because I don't know how HFS/HFS+ filesystems work.

The patch definitely should be rework. The phrase "I don't know how it works"
cannot be accepted as excuse. :)

> > I just want to close these nearly 1000 days old bugs.
> > 

We are not in a hurry. We must fix the reason of the bug but
not try to hide the real reason of the issue. 

> > You can write your patches.
> 
> I don't understand this attitude at all.  Are you in QA and being paid
> by "number of bugs closed per week"?

OK. Let's return to hfs_read_inode() again [1]. We have such logic here:

switch (rec->type) {
	case HFS_CDR_FIL:
<skipped>
		inode->i_ino = be32_to_cpu(rec->file.FlNum);
<skipped>
		break;
	case HFS_CDR_DIR:
		inode->i_ino = be32_to_cpu(rec->dir.DirID);
<skipped>
		break;
	default:
		make_bad_inode(inode);
}

So, if rec->type is OK (HFS_CDR_FIL, HFS_CDR_DIR) then we process
a particular type of record, otherwise, we create the bad inode. So, we simply
need to extend this logic. If rec->file.FlNum or rec->dir.DirID is equal or
bigger than HFS_FIRSTUSER_CNID, then we can create normal inode. Otherwise,
we need to create the bad inode. We simply need to add the checking logic
here. Tetsuo, does it make sense to you? :) Because, if we have corrupted value
of rec->file.FlNum or rec->dir.DirID, then it doesn't make sense to create
the normal inode with invalid i_ino. Simply, take a look here [2]:

/* Some special File ID numbers */
#define HFS_POR_CNID		1	/* Parent Of the Root */
#define HFS_ROOT_CNID		2	/* ROOT directory */
#define HFS_EXT_CNID		3	/* EXTents B-tree */
#define HFS_CAT_CNID		4	/* CATalog B-tree */
#define HFS_BAD_CNID		5	/* BAD blocks file */
#define HFS_ALLOC_CNID		6	/* ALLOCation file (HFS+) */
#define HFS_START_CNID		7	/* STARTup file (HFS+) */
#define HFS_ATTR_CNID		8	/* ATTRibutes file (HFS+) */
#define HFS_EXCH_CNID		15	/* ExchangeFiles temp id */
#define HFS_FIRSTUSER_CNID	16

Zero inode ID is completely invalid. And values from 1 - 15 are reserved
for HFS metadata structures.

Thanks,
Slava.

[1] https://elixir.bootlin.com/linux/v6.16-rc6/source/fs/hfs/inode.c#L350
[2] https://elixir.bootlin.com/linux/v6.16-rc6/source/fs/hfs/hfs.h#L40