Re: [git pull] vfs.git part 1

Re: [git pull] vfs.git part 1

[Michael Ellerman]

Al Viro <viro@ZenIV.linux.org.uk> writes:

> 	vfs.git topology is rather convoluted this cycle, so
> I'm afraid that it'll take more pull requests than usual ;-/
>
> The first pile is #work.misc-set_fs.  Assorted getting rid
> of cargo-culted access_ok(), cargo-culted set_fs() and
> field-by-field copyouts.  The same description applies to
> a lot of stuff in other branches - this is just the stuff that
> didn't fit into a more specific topical branch.
>
> The following changes since commit c86daad2c25bfd4a33d48b7691afaa96d9c5ab46:
>
>   Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input (2017-05-26 16:45:13 -0700)
>
> are available in the git repository at:
>
>   git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git work.misc-set_fs
>
> for you to fetch changes up to 8c6657cb50cb037ff58b3f6a547c6569568f3527:
>
>   Switch flock copyin/copyout primitives to copy_{from,to}_user() (2017-06-26 23:52:44 -0400)

This commit seems to have broken networking on a bunch of my PPC
machines (64-bit kernel, 32-bit userspace).

# first bad commit: [8c6657cb50cb037ff58b3f6a547c6569568f3527] Switch flock copyin/copyout primitives to copy_{from,to}_user()

The symptom is eth0 doesn't get address via dhcp.

Reverting it on top of master (9f45efb928) everything works OK again.


Trying to bring networking up manually gives:

# ifup eth0
ifup: failed to lock lockfile /run/network/ifstate.eth0: Invalid argument

strace shows:

5647  fcntl64(3, F_SETLK, {l_type=F_WRLCK, l_whence=SEEK_SET, l_start=0, l_len=0}) = -1 EINVAL (Invalid argument)
5647  write(2, "ifup: failed to lock lockfile /r"..., 74) = 74

vs the working case:

6005  fcntl64(3, F_SETLK, {l_type=F_WRLCK, l_whence=SEEK_SET, l_start=0, l_len=0}) = 0

Patch coming.

cheers

Re: [git pull] vfs.git part 1

[Linus Torvalds]

[-- Attachment #1: Type: text/plain, Size: 842 bytes --]

On Fri, Jul 7, 2017 at 5:46 AM, Michael Ellerman <mpe@ellerman.id.au> wrote:
> Al Viro <viro@ZenIV.linux.org.uk> writes:
>
>>
>>   Switch flock copyin/copyout primitives to copy_{from,to}_user() (2017-06-26 23:52:44 -0400)
>
> This commit seems to have broken networking on a bunch of my PPC
> machines (64-bit kernel, 32-bit userspace).

Bah. I think that commit is entirely broken, due to having the
arguments to the "copy_flock_fields()" in the wrong order.

The copy_flock_fields() macro has the arguments in order <from, to>,
but all the users seem to do it the other way around.

I think it would have been more obvious if the put_compat_flock*()
source argument had been "const".

> Patch coming.

I'm not seeing a patch, so I did my own. But it's _entirely_ untested.
Does the attached fix things for you?

                     Linus

[-- Attachment #2: patch.diff --]
[-- Type: text/plain, Size: 1885 bytes --]

 fs/fcntl.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/fs/fcntl.c b/fs/fcntl.c
index b6bd89628025..eeb19e22fd08 100644
--- a/fs/fcntl.c
+++ b/fs/fcntl.c
@@ -527,43 +527,43 @@ SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 	(to).l_len = (from).l_len;		\
 	(to).l_pid = (from).l_pid;
 
-static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
+static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
 {
 	struct compat_flock fl;
 
 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
 		return -EFAULT;
-	copy_flock_fields(*kfl, fl);
+	copy_flock_fields(fl, *kfl);
 	return 0;
 }
 
-static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
+static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
 {
 	struct compat_flock64 fl;
 
 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
 		return -EFAULT;
-	copy_flock_fields(*kfl, fl);
+	copy_flock_fields(fl, *kfl);
 	return 0;
 }
 
-static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
+static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
 {
 	struct compat_flock fl;
 
 	memset(&fl, 0, sizeof(struct compat_flock));
-	copy_flock_fields(fl, *kfl);
+	copy_flock_fields(*kfl, fl);
 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
 		return -EFAULT;
 	return 0;
 }
 
-static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
+static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
 {
 	struct compat_flock64 fl;
 
 	memset(&fl, 0, sizeof(struct compat_flock64));
-	copy_flock_fields(fl, *kfl);
+	copy_flock_fields(*kfl, fl);
 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
 		return -EFAULT;
 	return 0;

Re: [git pull] vfs.git part 1

[Linus Torvalds]

On Fri, Jul 7, 2017 at 8:59 AM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
>> Patch coming.
>
> I'm not seeing a patch, so I did my own. But it's _entirely_ untested.
> Does the attached fix things for you?

Oh, I see you sent a patch to the list but didn't cc me like in this thread.

Hmm. Al - I'd like to add the "const" parts at least. How the ordering
gets fixed (I changed it in the users of the macro, Michael changed
the macro itself) I don't much care about.

Can you get me a pull request soon since this presumably also breaks
every other compat case, and it just happened that power was the one
that noticed it first.. Or I can just commit my version, but I guess
Michael's is at least tested..

               Linus

Re: [git pull] vfs.git part 1

[Michael Ellerman]

Linus Torvalds <torvalds@linux-foundation.org> writes:

> On Fri, Jul 7, 2017 at 8:59 AM, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
>>
>>> Patch coming.
>>
>> I'm not seeing a patch, so I did my own. But it's _entirely_ untested.
>> Does the attached fix things for you?
>
> Oh, I see you sent a patch to the list but didn't cc me like in this thread.

Oops, I sent it To you, but I forgot to make it a reply to this thread
which was daft.

cheers

Re: [git pull] vfs.git part 1

[Linus Torvalds]

[-- Attachment #1: Type: text/plain, Size: 743 bytes --]

On Fri, Jul 7, 2017 at 8:59 AM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> The copy_flock_fields() macro has the arguments in order <from, to>,
> but all the users seem to do it the other way around.

Looking more at it, I think I'd also like copy_flock_fields() to take
pointer arguments, to match all the code around it (both
copy_to/from_user and the memset calls.

The actual order of arguments I suspect Michael's patch did better -
make the copy_flock_fields() just match the order of memcpy() and
copy_to/from_user(), both of which have <dest,src> order.

So I think my preferred patch would be something like this, even if it
is bigger than either.

Comments? Michael, does this work for your case?

              Linus

[-- Attachment #2: patch.diff --]
[-- Type: text/plain, Size: 2332 bytes --]

 fs/fcntl.c | 30 +++++++++++++++---------------
 1 file changed, 15 insertions(+), 15 deletions(-)

diff --git a/fs/fcntl.c b/fs/fcntl.c
index b6bd89628025..3b01b646e528 100644
--- a/fs/fcntl.c
+++ b/fs/fcntl.c
@@ -520,50 +520,50 @@ SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 
 #ifdef CONFIG_COMPAT
 /* careful - don't use anywhere else */
-#define copy_flock_fields(from, to)		\
-	(to).l_type = (from).l_type;		\
-	(to).l_whence = (from).l_whence;	\
-	(to).l_start = (from).l_start;		\
-	(to).l_len = (from).l_len;		\
-	(to).l_pid = (from).l_pid;
-
-static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
+#define copy_flock_fields(dst, src)		\
+	(dst)->l_type = (src)->l_type;		\
+	(dst)->l_whence = (src)->l_whence;	\
+	(dst)->l_start = (src)->l_start;	\
+	(dst)->l_len = (src)->l_len;		\
+	(dst)->l_pid = (src)->l_pid;
+
+static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
 {
 	struct compat_flock fl;
 
 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
 		return -EFAULT;
-	copy_flock_fields(*kfl, fl);
+	copy_flock_fields(kfl, &fl);
 	return 0;
 }
 
-static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
+static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
 {
 	struct compat_flock64 fl;
 
 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
 		return -EFAULT;
-	copy_flock_fields(*kfl, fl);
+	copy_flock_fields(kfl, &fl);
 	return 0;
 }
 
-static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
+static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
 {
 	struct compat_flock fl;
 
 	memset(&fl, 0, sizeof(struct compat_flock));
-	copy_flock_fields(fl, *kfl);
+	copy_flock_fields(&fl, kfl);
 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
 		return -EFAULT;
 	return 0;
 }
 
-static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
+static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
 {
 	struct compat_flock64 fl;
 
 	memset(&fl, 0, sizeof(struct compat_flock64));
-	copy_flock_fields(fl, *kfl);
+	copy_flock_fields(&fl, kfl);
 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
 		return -EFAULT;
 	return 0;

Re: [git pull] vfs.git part 1

[Al Viro]

On Fri, Jul 07, 2017 at 10:35:41AM -0700, Linus Torvalds wrote:

> Comments? Michael, does this work for your case?

Looks sane...

> +++ b/fs/fcntl.c
> @@ -520,50 +520,50 @@ SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
>  
>  #ifdef CONFIG_COMPAT
>  /* careful - don't use anywhere else */
> -#define copy_flock_fields(from, to)		\
> -	(to).l_type = (from).l_type;		\
> -	(to).l_whence = (from).l_whence;	\
> -	(to).l_start = (from).l_start;		\
> -	(to).l_len = (from).l_len;		\
> -	(to).l_pid = (from).l_pid;
> -
> -static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
> +#define copy_flock_fields(dst, src)		\
> +	(dst)->l_type = (src)->l_type;		\
> +	(dst)->l_whence = (src)->l_whence;	\
> +	(dst)->l_start = (src)->l_start;	\
> +	(dst)->l_len = (src)->l_len;		\
> +	(dst)->l_pid = (src)->l_pid;
> +
> +static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
>  {
>  	struct compat_flock fl;
>  
>  	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
>  		return -EFAULT;
> -	copy_flock_fields(*kfl, fl);
> +	copy_flock_fields(kfl, &fl);
>  	return 0;
>  }
>  
> -static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
> +static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
>  {
>  	struct compat_flock64 fl;
>  
>  	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
>  		return -EFAULT;
> -	copy_flock_fields(*kfl, fl);
> +	copy_flock_fields(kfl, &fl);
>  	return 0;
>  }
>  
> -static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
> +static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
>  {
>  	struct compat_flock fl;
>  
>  	memset(&fl, 0, sizeof(struct compat_flock));
> -	copy_flock_fields(fl, *kfl);
> +	copy_flock_fields(&fl, kfl);
>  	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
>  		return -EFAULT;
>  	return 0;
>  }
>  
> -static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
> +static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
>  {
>  	struct compat_flock64 fl;
>  
>  	memset(&fl, 0, sizeof(struct compat_flock64));
> -	copy_flock_fields(fl, *kfl);
> +	copy_flock_fields(&fl, kfl);
>  	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
>  		return -EFAULT;
>  	return 0;

Re: [git pull] vfs.git part 1

[Michael Ellerman]

Linus Torvalds <torvalds@linux-foundation.org> writes:

> On Fri, Jul 7, 2017 at 8:59 AM, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
>>
>> The copy_flock_fields() macro has the arguments in order <from, to>,
>> but all the users seem to do it the other way around.
>
> Looking more at it, I think I'd also like copy_flock_fields() to take
> pointer arguments, to match all the code around it (both
> copy_to/from_user and the memset calls.
>
> The actual order of arguments I suspect Michael's patch did better -
> make the copy_flock_fields() just match the order of memcpy() and
> copy_to/from_user(), both of which have <dest,src> order.
>
> So I think my preferred patch would be something like this, even if it
> is bigger than either.
>
> Comments? Michael, does this work for your case?

Yeah that works, as committed in your tree. Sorry for the slow reply,
our time zones don't line up all that well :)

cheers