| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <2103722d0e10bbd71ad6f93550668cea717381bc.camel@ibm.com>
Date: Fri, 25 Jul 2025 17:42:55 +0000
From: Viacheslav Dubeyko <Slava.Dubeyko@....com>
To: "penguin-kernel@...ove.SAKURA.ne.jp" <penguin-kernel@...ove.SAKURA.ne.jp>,
"willy@...radead.org" <willy@...radead.org>
CC: "glaubitz@...sik.fu-berlin.de" <glaubitz@...sik.fu-berlin.de>,
"frank.li@...o.com" <frank.li@...o.com>,
"slava@...eyko.com"
<slava@...eyko.com>,
"linux-fsdevel@...r.kernel.org"
<linux-fsdevel@...r.kernel.org>,
"linux-kernel@...r.kernel.org"
<linux-kernel@...r.kernel.org>,
"akpm@...ux-foundation.org"
<akpm@...ux-foundation.org>
Subject: RE: [PATCH v3] hfs: remove BUG() from
hfs_release_folio()/hfs_test_inode()/hfs_write_inode()
On Fri, 2025-07-25 at 07:05 +0900, Tetsuo Handa wrote:
> On 2025/07/25 4:49, Viacheslav Dubeyko wrote:
> > On Thu, 2025-07-24 at 15:55 +0900, Tetsuo Handa wrote:
> > > Then, something like below change?
> > >
> > > --- a/fs/hfs/inode.c
> > > +++ b/fs/hfs/inode.c
> > > @@ -318,6 +318,9 @@ static int hfs_read_inode(struct inode *inode, void *data)
> > > struct hfs_iget_data *idata = data;
> > > struct hfs_sb_info *hsb = HFS_SB(inode->i_sb);
> > > hfs_cat_rec *rec;
> > > + /* https://developer.apple.com/library/archive/technotes/tn/tn1150.html#CNID */
> >
> > We already have all declarations in hfs.h:
> >
> > /* 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
>
> These declarations does not define 14, and some flags are never used despite
> being declared here.
>
> >
> > So, adding the link here doesn't make any sense.
> >
> > > + static const u16 bad_cnid_list = (1 << 0) | (1 << 6) | (1 << 7) | (1 << 8) |
> > > + (1 << 9) | (1 << 10) | (1 << 11) | (1 << 12) | (1 << 13);
>
> Some of values in this constant are not declared.
>
It means that we need to declare the missing values. But hardcoded bare numbers
are really bad practice.
> >
> > I don't see any sense to introduce flags here. First of all, please, don't use
> > hardcoded values but you should use declared constants from hfs.h (for example,
> > HFS_EXT_CNID instead of 3). Secondly, you can simply compare the i_ino with
> > constants, for example:
>
> This will save a lot of computational power compared to switch().
>
Even if you would like to use flags, then the logic must to be simple and
understandable. You still can use special inline function and do not create a
mess in hfs_read_inode(). Especially, you can declare the mask one time in
header, for example, but not to prepare the bad_cnid_list for every function
call. Currently, the code looks really messy.
> >
> > bool is_inode_id_invalid(u64 ino) {
> > switch (inode->i_ino) {
> > case HFS_EXT_CNID:
> > ...
> > return true;
> >
> > }
> >
> > return false;
> > }
> >
> > Thirdly, you can introduce an inline function that can do such check. And it
> > make sense to introduce constant for the case of zero value.
> >
> > Why have you missed HFS_EXT_CNID, HFS_CAT_CNID? These values cannot used in
> > hfs_read_inode().
>
> Is hfs_read_inode() never called for HFS_EXT_CNID and HFS_CAT_CNID ?
>
The location of Catalog File and Extents File are defined in superblock. As a
result, Catalog File cannot contain a record with CNID HFS_EXT_CNID or
HFS_CAT_CNID. And if hfs_read_inode() receives these values, then it is some
corruption of Catalog File.
> >
> > >
> > > HFS_I(inode)->flags = 0;
> > > HFS_I(inode)->rsrc_inode = NULL;
> > > @@ -358,6 +361,8 @@ static int hfs_read_inode(struct inode *inode, void *data)
> > > inode->i_op = &hfs_file_inode_operations;
> > > inode->i_fop = &hfs_file_operations;
> > > inode->i_mapping->a_ops = &hfs_aops;
> > > + if (inode->i_ino < HFS_FIRSTUSER_CNID && ((1U << inode->i_ino) & bad_cnid_list))
> > > + make_bad_inode(inode);
> >
> > It looks pretty complicated. You can simply use one above-mentioned function
> > with the check:
> >
> > if (is_inode_id_invalid(be32_to_cpu(rec->dir.DirID)))
> > <goto to make bad inode>
> >
> > We can simply check the the inode ID in the beginning of the whole action:
> >
> > <Make the check here>
> > inode->i_ino = be32_to_cpu(rec->file.FlNum);
> > inode->i_mode = S_IRUGO | S_IXUGO;
> > if (!(rec->file.Flags & HFS_FIL_LOCK))
> > inode->i_mode |= S_IWUGO;
> > inode->i_mode &= ~hsb->s_file_umask;
> > inode->i_mode |= S_IFREG;
> > inode_set_mtime_to_ts(inode,
> > inode_set_atime_to_ts(inode,
> > inode_set_ctime_to_ts(inode, hfs_m_to_utime(rec->file.MdDat))));
> > inode->i_op = &hfs_file_inode_operations;
> > inode->i_fop = &hfs_file_operations;
> > inode->i_mapping->a_ops = &hfs_aops;
> >
> > It doesn't make any sense to construct inode if we will make in bad inode,
> > finally. Don't waste computational power. :)
> >
> > > break;
> > > case HFS_CDR_DIR:
> > > inode->i_ino = be32_to_cpu(rec->dir.DirID);
> > > @@ -368,6 +373,8 @@ static int hfs_read_inode(struct inode *inode, void *data)
> > > inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, hfs_m_to_utime(rec->dir.MdDat))));
> > > inode->i_op = &hfs_dir_inode_operations;
> > > inode->i_fop = &hfs_dir_operations;
> > > + if (inode->i_ino < HFS_FIRSTUSER_CNID && ((1U << inode->i_ino) & bad_cnid_list))
> > > + make_bad_inode(inode);
> >
> > We already have make_bad_inode(inode) as default action. So, simply jump there.
> >
> > > break;
> > > default:
> > > make_bad_inode(inode);
> > >
> > >
> > >
> > > But I can't be convinced that above change is sufficient, for if I do
> > >
> > > + static u8 serial;
> > > + if (inode->i_ino < HFS_FIRSTUSER_CNID && ((1U << inode->i_ino) & bad_cnid_list))
> > > + inode->i_ino = (serial++) % 16;
> >
> > I don't see the point in flags introduction. It makes logic very complicated.
>
> The point of this change is to excecise inode->i_ino for all values between 0 and 15.
> Some of values between 0 and 15 must be valid as inode->i_ino , doesn't these? Then,
>
If you have mask of valid or/and invalid, then you can simply check that this
mask contain the flag. It will bed simple bit state checking. Currently, the
code looks weird, not clear, complicated, and inefficient.
> >
> > >
> > > instead of
> > >
> > > + if (inode->i_ino < HFS_FIRSTUSER_CNID && ((1U << inode->i_ino) & bad_cnid_list))
> > > + make_bad_inode(inode);
> > >
> > > , the reproducer still hits BUG() for 0, 1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 and 15
> > > because hfs_write_inode() handles only 2, 3 and 4.
> > >
> >
> > How can we go into hfs_write_inode() if we created the bad inode for invalid
> > inode ID? How is it possible?
>
> are all of 0, 1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 and 15 invalid value for hfs_read_inode() ?
>
> If all of 0, 1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 and 15 are invalid value for hfs_read_inode(),
> and 3 and 4 are also invalid value for hfs_read_inode(), hfs_read_inode() would accept only 2.
> Something is crazily wrong.
>
> Can we really filter some of values between 0 and 15 at hfs_read_inode() ?
0 value is invalid.
#define HFS_POR_CNID 1 /* Parent Of the Root */
#define HFS_ROOT_CNID 2 /* ROOT directory */
These values are legitimate values.
#define HFS_EXT_CNID 3 /* EXTents B-tree */
#define HFS_CAT_CNID 4 /* CATalog B-tree */
This metadata structures are defined in MDB. This is invalid values for
hfs_read_inode().
#define HFS_BAD_CNID 5 /* BAD blocks file */
This could be defined in Catalog File because MDB has nothing for this metadata
structure. However, it's ancient technology.
#define HFS_ALLOC_CNID 6 /* ALLOCation file (HFS+) */
#define HFS_START_CNID 7 /* STARTup file (HFS+) */
#define HFS_ATTR_CNID 8 /* ATTRibutes file (HFS+) */
These value are invalid for HFS.
9, 10, 11, 12, 13, 14 can be defined as constants and it is invalid values. Foe
example:
#define HFS_RESERVED_CNID_9 9
#define HFS_RESERVED_CNID_10 10
...
#define HFS_RESERVED_CNID_14 14
#define HFS_EXCH_CNID 15 /* ExchangeFiles temp id */
This could be defined in Catalog File (maybe not). I didn't find anything
related to this in HFS specification.
So, 1, 2, 5, 15, etc can be accepted by hfs_read_inode().
0, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14 is invalid values for hfs_read_inode().
Thanks,
Slava.
Powered by blists - more mailing lists