[PATCH 01/17] fork: fold legacy_clone_args_valid() into _do_fork()

[PATCH 01/17] fork: fold legacy_clone_args_valid() into _do_fork()

[Christian Brauner]

This separate helper only existed to guarantee the mutual exclusivity of
CLONE_PIDFD and CLONE_PARENT_SETTID for legacy clone since CLONE_PIDFD
abuses the parent_tid field to return the pidfd. But we can actually handle
this uniformely thus removing the helper. For legacy clone we can detect
that CLONE_PIDFD is specified in conjunction with CLONE_PARENT_SETTID
because they will share the same memory which is invalid and for clone3()
setting the separate pidfd and parent_tid fields to the same memory is
bogus as well. So fold that helper directly into _do_fork() by detecting
this case.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org>
Cc: linux-m68k@lists.linux-m68k.org
Cc: x86@kernel.org
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
---
 arch/m68k/kernel/process.c |  3 ---
 arch/x86/kernel/sys_ia32.c |  3 ---
 include/linux/sched/task.h |  1 -
 kernel/fork.c              | 30 ++++++++++++++----------------
 4 files changed, 14 insertions(+), 23 deletions(-)

diff --git a/arch/m68k/kernel/process.c b/arch/m68k/kernel/process.c
index 90ae376b7ab1..0608439ba452 100644
--- a/arch/m68k/kernel/process.c
+++ b/arch/m68k/kernel/process.c
@@ -125,9 +125,6 @@ asmlinkage int m68k_clone(struct pt_regs *regs)
 		.tls		= regs->d5,
 	};
 
-	if (!legacy_clone_args_valid(&args))
-		return -EINVAL;
-
 	return _do_fork(&args);
 }
 
diff --git a/arch/x86/kernel/sys_ia32.c b/arch/x86/kernel/sys_ia32.c
index f8d65c99feb8..720cde885042 100644
--- a/arch/x86/kernel/sys_ia32.c
+++ b/arch/x86/kernel/sys_ia32.c
@@ -251,9 +251,6 @@ COMPAT_SYSCALL_DEFINE5(ia32_clone, unsigned long, clone_flags,
 		.tls		= tls_val,
 	};
 
-	if (!legacy_clone_args_valid(&args))
-		return -EINVAL;
-
 	return _do_fork(&args);
 }
 #endif /* CONFIG_IA32_EMULATION */
diff --git a/include/linux/sched/task.h b/include/linux/sched/task.h
index 38359071236a..ddce0ea515d1 100644
--- a/include/linux/sched/task.h
+++ b/include/linux/sched/task.h
@@ -96,7 +96,6 @@ extern void exit_files(struct task_struct *);
 extern void exit_itimers(struct signal_struct *);
 
 extern long _do_fork(struct kernel_clone_args *kargs);
-extern bool legacy_clone_args_valid(const struct kernel_clone_args *kargs);
 extern long do_fork(unsigned long, unsigned long, unsigned long, int __user *, int __user *);
 struct task_struct *fork_idle(int);
 struct mm_struct *copy_init_mm(void);
diff --git a/kernel/fork.c b/kernel/fork.c
index 142b23645d82..9875aeb2ba41 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -2422,6 +2422,20 @@ long _do_fork(struct kernel_clone_args *args)
 	int trace = 0;
 	long nr;
 
+	/*
+	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
+	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
+	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
+	 * field in struct clone_args and it still doesn't make sense to have
+	 * them both point at the same memory location. Performing this check
+	 * here has the advantage that we don't need to have a separate helper
+	 * to check for legacy clone().
+	 */
+	if ((args->flags & CLONE_PIDFD) &&
+	    (args->flags & CLONE_PARENT_SETTID) &&
+	    (args->pidfd == args->parent_tid))
+		return -EINVAL;
+
 	/*
 	 * Determine whether and which event to report to ptracer.  When
 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
@@ -2479,16 +2493,6 @@ long _do_fork(struct kernel_clone_args *args)
 	return nr;
 }
 
-bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
-{
-	/* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
-	if ((kargs->flags & CLONE_PIDFD) &&
-	    (kargs->flags & CLONE_PARENT_SETTID))
-		return false;
-
-	return true;
-}
-
 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
 /* For compatibility with architectures that call do_fork directly rather than
  * using the syscall entry points below. */
@@ -2508,9 +2512,6 @@ long do_fork(unsigned long clone_flags,
 		.stack_size	= stack_size,
 	};
 
-	if (!legacy_clone_args_valid(&args))
-		return -EINVAL;
-
 	return _do_fork(&args);
 }
 #endif
@@ -2593,9 +2594,6 @@ SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
 		.tls		= tls,
 	};
 
-	if (!legacy_clone_args_valid(&args))
-		return -EINVAL;
-
 	return _do_fork(&args);
 }
 #endif
-- 
2.27.0

Re: [fork] 11689456e6: ltp.clone302.fail

[Christian Brauner]

On Sat, Jun 27, 2020 at 04:27:48PM +0800, kernel test robot wrote:
> Greeting,
> 
> FYI, we noticed the following commit (built with gcc-9):
> 
> commit: 11689456e6df828b7917a558a36212e68fa9aa69 ("[PATCH 01/17] fork: fold legacy_clone_args_valid() into _do_fork()")
> url: https://github.com/0day-ci/linux/commits/Christian-Brauner/arch-remove-do_fork-and-HAVE_COPY_THREAD_TLS/20200623-080105
> base: https://git.kernel.org/cgit/linux/kernel/git/davem/sparc.git master
> 
> in testcase: ltp
> with following parameters:
> 
> 	disk: 1HDD
> 	fs: ext4
> 	test: syscalls_part1
> 
> test-description: The LTP testsuite contains a collection of tools for testing the Linux kernel and related features.
> test-url: http://linux-test-project.github.io/
> 
> 
> on test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G
> 
> caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
> 
> 
> 
> 
> If you fix the issue, kindly add following tag
> Reported-by: kernel test robot <rong.a.chen@intel.com>
> 
> 
> <<<test_start>>>
> tag=clone302 stime=1593153327
> cmdline="clone302"
> contacts=""
> analysis=exit
> <<<test_output>>>
> tst_buffers.c:55: INFO: Test is using guarded buffers
> tst_test.c:1247: INFO: Timeout per run is 0h 05m 00s
> clone302.c:92: PASS: invalid args: clone3() failed as expected: EFAULT (14)
> clone302.c:92: PASS: zero size: clone3() failed as expected: EINVAL (22)
> clone302.c:92: PASS: short size: clone3() failed as expected: EINVAL (22)
> clone302.c:92: PASS: extra size: clone3() failed as expected: EFAULT (14)
> clone302.c:92: PASS: sighand-no-VM: clone3() failed as expected: EINVAL (22)
> clone302.c:92: PASS: thread-no-sighand: clone3() failed as expected: EINVAL (22)
> clone302.c:92: PASS: fs-newns: clone3() failed as expected: EINVAL (22)
> clone302.c:92: PASS: invalid pidfd: clone3() failed as expected: EFAULT (14)
> clone302.c:92: PASS: invalid childtid: clone3() failed as expected: EFAULT (14)
> clone302.c:88: FAIL: invalid parenttid: clone3() should fail with EFAULT: EINVAL (22)

In short, this is a change in failure behavior for clone3() I did expect
and am willing to risk. Here's why in the short form:
- clone3() is extremely new
- this failed before
- setting both CLONE_PIDFD and CLONE_PARENT_SETTID is extremely rare
  (haven't seen it in any codebases I know that use clone3())
- setting both CLONE_PIDFD and CLONE_PARENT_SETTID __and__ pointing them
  to the same adress doesn't work
  (haven't seen it in any codebases I know that use clone3() but see
  some more notes on that below)
- the change makes a special case go away and simplifies multiple
  call-sites

So a few notes about the test. I did stare at it for a while and was
confused why you expect EFAULT to be returned when CLONE_PARENT_SETTID
is set to an invalid memory address. Because that doesn't make sense.
When the parent tid is written to the memory location for
CLONE_PARENT_SETTID we're past the point of no return of process
creation, i.e. the return value from put_user() isn't checked and can't
be checked anymore so you'd never receive EFAULT for a bogus parent_tid
memory address. This is not something new. This has been the case since
the introduction of pid namespaces and specifically since commit
30e49c263e36 ("pid namespaces: allow cloning of new namespace").

But then it dawned on me. You're setting CLONE_PIDFD |
CLONE_PARENT_SETTID and you're pointing:
- args->parent_tid	= <invalid-address>
- args->pidfd		= NULL
so the EFAULT you've seen so far in your test-suite has never been for
CLONE_PARENT_SETTID but for CLONE_PIDFD since that value is written
before the point of no return and consequently put_user() is checked and
the EFAULT is surfaced. So independent of that issue here you might want
to adapt that test so it really tests what you want. :) (And maybe it's
worth documenting on the manpage for clone{3}() that failures for
CLONE_PARENT_SETTID and CLONE_CHILD_SETTID are not seen.)

(Also, note that for some reason, args->pidfd and pargs->parent_tid
must've ended up pointing to the same address in your test-suite. So my
guess is that args->pidfd pointed to garbage which got turned into a
useable address after tst_get_bad_addr() returned &invalid_address.
Maybe I'm missing something subtle though.)

Christian