To avoid confusing the compiler about possible negative sizes, switch
"ssize" which can never be negative from int to u32. Seen with GCC 13:
../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
[-Warray-bounds=]
57 | #define __underlying_memcpy __builtin_memcpy
| ^
...
../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
950 | memcpy(ip->i_link, name, ssize);
| ^~~~~~
Cc: Dave Kleikamp <[email protected]>
Cc: Christian Brauner <[email protected]>
Cc: Dave Chinner <[email protected]>
Cc: [email protected]
Signed-off-by: Kees Cook <[email protected]>
---
fs/jfs/namei.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
index b29d68b5eec5..494b9f4043cf 100644
--- a/fs/jfs/namei.c
+++ b/fs/jfs/namei.c
@@ -876,7 +876,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
tid_t tid;
ino_t ino = 0;
struct component_name dname;
- int ssize; /* source pathname size */
+ u32 ssize; /* source pathname size */
struct btstack btstack;
struct inode *ip = d_inode(dentry);
s64 xlen = 0;
@@ -957,7 +957,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
if (ssize > sizeof (JFS_IP(ip)->i_inline))
JFS_IP(ip)->mode2 &= ~INLINEEA;
- jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
+ jfs_info("jfs_symlink: fast symlink added ssize:%u name:%s ",
ssize, name);
}
/*
@@ -987,7 +987,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
ip->i_size = ssize - 1;
while (ssize) {
/* This is kind of silly since PATH_MAX == 4K */
- int copy_size = min(ssize, PSIZE);
+ u32 copy_size = min_t(u32, ssize, PSIZE);
mp = get_metapage(ip, xaddr, PSIZE, 1);
--
2.34.1
* Kees Cook ([email protected]) wrote:
> To avoid confusing the compiler about possible negative sizes, switch
> "ssize" which can never be negative from int to u32. Seen with GCC 13:
>
> ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
> [-Warray-bounds=]
> 57 | #define __underlying_memcpy __builtin_memcpy
> | ^
> ...
> ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
> 950 | memcpy(ip->i_link, name, ssize);
> | ^~~~~~
>
> Cc: Dave Kleikamp <[email protected]>
> Cc: Christian Brauner <[email protected]>
> Cc: Dave Chinner <[email protected]>
> Cc: [email protected]
> Signed-off-by: Kees Cook <[email protected]>
> ---
> fs/jfs/namei.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> index b29d68b5eec5..494b9f4043cf 100644
> --- a/fs/jfs/namei.c
> +++ b/fs/jfs/namei.c
> @@ -876,7 +876,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> tid_t tid;
> ino_t ino = 0;
> struct component_name dname;
> - int ssize; /* source pathname size */
> + u32 ssize; /* source pathname size */
Had you considered using size_t - this is set from a strlen and used by a memcpy
that both talk size_t.
Dave
> struct btstack btstack;
> struct inode *ip = d_inode(dentry);
> s64 xlen = 0;
> @@ -957,7 +957,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> if (ssize > sizeof (JFS_IP(ip)->i_inline))
> JFS_IP(ip)->mode2 &= ~INLINEEA;
>
> - jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
> + jfs_info("jfs_symlink: fast symlink added ssize:%u name:%s ",
> ssize, name);
> }
> /*
> @@ -987,7 +987,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> ip->i_size = ssize - 1;
> while (ssize) {
> /* This is kind of silly since PATH_MAX == 4K */
> - int copy_size = min(ssize, PSIZE);
> + u32 copy_size = min_t(u32, ssize, PSIZE);
>
> mp = get_metapage(ip, xaddr, PSIZE, 1);
>
> --
> 2.34.1
>
--
-----Open up your eyes, open up your mind, open up your code -------
/ Dr. David Alan Gilbert | Running GNU/Linux | Happy \
\ dave @ treblig.org | | In Hex /
\ _________________________|_____ http://www.treblig.org |_______/
On Sat, Feb 04, 2023 at 08:25:45PM +0000, Dr. David Alan Gilbert wrote:
> * Kees Cook ([email protected]) wrote:
> > To avoid confusing the compiler about possible negative sizes, switch
> > "ssize" which can never be negative from int to u32. Seen with GCC 13:
> >
> > ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
> > [-Warray-bounds=]
> > 57 | #define __underlying_memcpy __builtin_memcpy
> > | ^
> > ...
> > ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
> > 950 | memcpy(ip->i_link, name, ssize);
> > | ^~~~~~
> >
> > Cc: Dave Kleikamp <[email protected]>
> > Cc: Christian Brauner <[email protected]>
> > Cc: Dave Chinner <[email protected]>
> > Cc: [email protected]
> > Signed-off-by: Kees Cook <[email protected]>
> > ---
> > fs/jfs/namei.c | 6 +++---
> > 1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> > index b29d68b5eec5..494b9f4043cf 100644
> > --- a/fs/jfs/namei.c
> > +++ b/fs/jfs/namei.c
> > @@ -876,7 +876,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> > tid_t tid;
> > ino_t ino = 0;
> > struct component_name dname;
> > - int ssize; /* source pathname size */
> > + u32 ssize; /* source pathname size */
>
> Had you considered using size_t - this is set from a strlen and used by a memcpy
> that both talk size_t.
I considered that, but I've had other maintainers upset about doubling
the variable size. I opted to keep the variable 32-bit here, so the
machine code would only change to lose signed-ness.
-Kees
--
Kees Cook
* Kees Cook ([email protected]) wrote:
> On Sat, Feb 04, 2023 at 08:25:45PM +0000, Dr. David Alan Gilbert wrote:
> > * Kees Cook ([email protected]) wrote:
> > > To avoid confusing the compiler about possible negative sizes, switch
> > > "ssize" which can never be negative from int to u32. Seen with GCC 13:
> > >
> > > ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
> > > [-Warray-bounds=]
> > > 57 | #define __underlying_memcpy __builtin_memcpy
> > > | ^
> > > ...
> > > ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
> > > 950 | memcpy(ip->i_link, name, ssize);
> > > | ^~~~~~
> > >
> > > Cc: Dave Kleikamp <[email protected]>
> > > Cc: Christian Brauner <[email protected]>
> > > Cc: Dave Chinner <[email protected]>
> > > Cc: [email protected]
> > > Signed-off-by: Kees Cook <[email protected]>
> > > ---
> > > fs/jfs/namei.c | 6 +++---
> > > 1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> > > index b29d68b5eec5..494b9f4043cf 100644
> > > --- a/fs/jfs/namei.c
> > > +++ b/fs/jfs/namei.c
> > > @@ -876,7 +876,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> > > tid_t tid;
> > > ino_t ino = 0;
> > > struct component_name dname;
> > > - int ssize; /* source pathname size */
> > > + u32 ssize; /* source pathname size */
> >
> > Had you considered using size_t - this is set from a strlen and used by a memcpy
> > that both talk size_t.
>
> I considered that, but I've had other maintainers upset about doubling
> the variable size.
I bet at least on some platforms it's cheaper as the 64 bit.
> I opted to keep the variable 32-bit here, so the
> machine code would only change to lose signed-ness.
Fair enough.
Dave
> -Kees
>
> --
> Kees Cook
--
-----Open up your eyes, open up your mind, open up your code -------
/ Dr. David Alan Gilbert | Running GNU/Linux | Happy \
\ dave @ treblig.org | | In Hex /
\ _________________________|_____ http://www.treblig.org |_______/
On Mon, Feb 06, 2023 at 07:50:42PM +0000, Dr. David Alan Gilbert wrote:
> * Kees Cook ([email protected]) wrote:
> > On Sat, Feb 04, 2023 at 08:25:45PM +0000, Dr. David Alan Gilbert wrote:
> > > * Kees Cook ([email protected]) wrote:
> > > > To avoid confusing the compiler about possible negative sizes, switch
> > > > "ssize" which can never be negative from int to u32. Seen with GCC 13:
> > > >
> > > > ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
> > > > [-Warray-bounds=]
> > > > 57 | #define __underlying_memcpy __builtin_memcpy
> > > > | ^
> > > > ...
> > > > ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
> > > > 950 | memcpy(ip->i_link, name, ssize);
> > > > | ^~~~~~
> > > >
> > > > Cc: Dave Kleikamp <[email protected]>
> > > > Cc: Christian Brauner <[email protected]>
> > > > Cc: Dave Chinner <[email protected]>
> > > > Cc: [email protected]
> > > > Signed-off-by: Kees Cook <[email protected]>
> > > > ---
> > > > fs/jfs/namei.c | 6 +++---
> > > > 1 file changed, 3 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> > > > index b29d68b5eec5..494b9f4043cf 100644
> > > > --- a/fs/jfs/namei.c
> > > > +++ b/fs/jfs/namei.c
> > > > @@ -876,7 +876,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> > > > tid_t tid;
> > > > ino_t ino = 0;
> > > > struct component_name dname;
> > > > - int ssize; /* source pathname size */
> > > > + u32 ssize; /* source pathname size */
> > >
> > > Had you considered using size_t - this is set from a strlen and used by a memcpy
> > > that both talk size_t.
> >
> > I considered that, but I've had other maintainers upset about doubling
> > the variable size.
>
> I bet at least on some platforms it's cheaper as the 64 bit.
>
> > I opted to keep the variable 32-bit here, so the
> > machine code would only change to lose signed-ness.
>
> Fair enough.
Thread ping. Can someone pick this up (or Ack it for my tree), please?
Thanks!
-Kees
--
Kees Cook
On Thu, Jun 1, 2023 at 9:55 AM Kees Cook <[email protected]> wrote:
>
> On Mon, Feb 06, 2023 at 07:50:42PM +0000, Dr. David Alan Gilbert wrote:
> > * Kees Cook ([email protected]) wrote:
> > > On Sat, Feb 04, 2023 at 08:25:45PM +0000, Dr. David Alan Gilbert wrote:
> > > > * Kees Cook ([email protected]) wrote:
> > > > > To avoid confusing the compiler about possible negative sizes, switch
> > > > > "ssize" which can never be negative from int to u32. Seen with GCC 13:
> > > > >
> > > > > ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
> > > > > [-Warray-bounds=]
> > > > > 57 | #define __underlying_memcpy __builtin_memcpy
> > > > > | ^
> > > > > ...
> > > > > ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
> > > > > 950 | memcpy(ip->i_link, name, ssize);
> > > > > | ^~~~~~
> > > > >
> > > > > Cc: Dave Kleikamp <[email protected]>
> > > > > Cc: Christian Brauner <[email protected]>
> > > > > Cc: Dave Chinner <[email protected]>
> > > > > Cc: [email protected]
> > > > > Signed-off-by: Kees Cook <[email protected]>
> > > > > ---
> > > > > fs/jfs/namei.c | 6 +++---
> > > > > 1 file changed, 3 insertions(+), 3 deletions(-)
> > > > >
> > > > > diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> > > > > index b29d68b5eec5..494b9f4043cf 100644
> > > > > --- a/fs/jfs/namei.c
> > > > > +++ b/fs/jfs/namei.c
> > > > > @@ -876,7 +876,7 @@ static int jfs_symlink(struct mnt_idmap *idmap, struct inode *dip,
> > > > > tid_t tid;
> > > > > ino_t ino = 0;
> > > > > struct component_name dname;
> > > > > - int ssize; /* source pathname size */
> > > > > + u32 ssize; /* source pathname size */
> > > >
> > > > Had you considered using size_t - this is set from a strlen and used by a memcpy
> > > > that both talk size_t.
> > >
> > > I considered that, but I've had other maintainers upset about doubling
> > > the variable size.
> >
> > I bet at least on some platforms it's cheaper as the 64 bit.
> >
> > > I opted to keep the variable 32-bit here, so the
> > > machine code would only change to lose signed-ness.
> >
> > Fair enough.
>
> Thread ping. Can someone pick this up (or Ack it for my tree), please?
>
Acked-by: Jeff Xu <[email protected]>
In case you ask someone to look at code, which I did.
Best
-Jeff
> Thanks!
>
> -Kees
>
> --
> Kees Cook
On Sat, 04 Feb 2023 10:33:56 -0800, Kees Cook wrote:
> To avoid confusing the compiler about possible negative sizes, switch
> "ssize" which can never be negative from int to u32. Seen with GCC 13:
>
> ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
> [-Warray-bounds=]
> 57 | #define __underlying_memcpy __builtin_memcpy
> | ^
> ...
> ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
> 950 | memcpy(ip->i_link, name, ssize);
> | ^~~~~~
>
> [...]
Applied to the vfs.misc branch of the vfs/vfs.git tree.
Patches in the vfs.misc branch should appear in linux-next soon.
Please report any outstanding bugs that were missed during review in a
new review to the original patch series allowing us to drop it.
It's encouraged to provide Acked-bys and Reviewed-bys even though the
patch has now been applied. If possible patch trailers will be updated.
tree: https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git
branch: vfs.misc
[1/1] jfs: Use unsigned variable for length calculations
https://git.kernel.org/vfs/vfs/c/2d6e1895d440
I apologize to all of you for letting this one go so long.
Acked-by: Dave Kleikamp <[email protected]>
On 6/2/23 3:32AM, Christian Brauner wrote:
> On Sat, 04 Feb 2023 10:33:56 -0800, Kees Cook wrote:
>> To avoid confusing the compiler about possible negative sizes, switch
>> "ssize" which can never be negative from int to u32. Seen with GCC 13:
>>
>> ../fs/jfs/namei.c: In function 'jfs_symlink': ../include/linux/fortify-string.h:57:33: warning: '__builtin_memcpy' pointer overflow between offset 0 and size [-2147483648, -1]
>> [-Warray-bounds=]
>> 57 | #define __underlying_memcpy __builtin_memcpy
>> | ^
>> ...
>> ../fs/jfs/namei.c:950:17: note: in expansion of macro 'memcpy'
>> 950 | memcpy(ip->i_link, name, ssize);
>> | ^~~~~~
>>
>> [...]
>
> Applied to the vfs.misc branch of the vfs/vfs.git tree.
> Patches in the vfs.misc branch should appear in linux-next soon.
>
> Please report any outstanding bugs that were missed during review in a
> new review to the original patch series allowing us to drop it.
>
> It's encouraged to provide Acked-bys and Reviewed-bys even though the
> patch has now been applied. If possible patch trailers will be updated.
>
> tree: https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git
> branch: vfs.misc
>
> [1/1] jfs: Use unsigned variable for length calculations
> https://git.kernel.org/vfs/vfs/c/2d6e1895d440
From: Kees Cook
> Sent: 01 June 2023 17:55
>
> On Mon, Feb 06, 2023 at 07:50:42PM +0000, Dr. David Alan Gilbert wrote:
> > * Kees Cook ([email protected]) wrote:
> > > On Sat, Feb 04, 2023 at 08:25:45PM +0000, Dr. David Alan Gilbert wrote:
...
> > > > > - int ssize; /* source pathname size */
> > > > > + u32 ssize; /* source pathname size */
> > > >
> > > > Had you considered using size_t - this is set from a strlen and used by a memcpy
> > > > that both talk size_t.
> > >
> > > I considered that, but I've had other maintainers upset about doubling
> > > the variable size.
> >
> > I bet at least on some platforms it's cheaper as the 64 bit.
> >
> > > I opted to keep the variable 32-bit here, so the
> > > machine code would only change to lose signed-ness.
On x86-64 'unsigned int' and 'signed long' are likely to
generate the best code.
If you use 'signed int' it will to have to be sign extended
if used in a 64-bit expression (eg as an array index).
(That is probably true of most 64bit arch.)
OTOH 'unsigned long' always requires a REX prefix - making the
code that bit larger.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)