Re: [BUG] kernel BUG at fs/userfaultfd.c:385 after 04f5866e41fb

[Oleg Nesterov <oleg@redhat.com>]

On 08/14, Oleg Nesterov wrote:
>
> On 08/14, Kefeng Wang wrote:
> >
> > On 2019/8/14 21:53, Michal Hocko wrote:
> > > On Tue 13-08-19 17:08:05, Kefeng Wang wrote:
> > >>
> > >> Syzkaller reproducer:
> > >> # {Threaded:true Collide:true Repeat:false RepeatTimes:0 Procs:1 Sandbox:none Fault:false FaultCall:-1 FaultNth:0 EnableTun:true EnableNetDev:true EnableNetReset:false EnableCgroups:false EnableBinfmtMisc:true EnableCloseFds:true UseTmpDir:true HandleSegv:true Repro:false Trace:false}
> > >> r0 = userfaultfd(0x80800)
> > >> ioctl$UFFDIO_API(r0, 0xc018aa3f, &(0x7f0000000200))
> > >> ioctl$UFFDIO_REGISTER(r0, 0xc020aa00, &(0x7f0000000080)={{&(0x7f0000ff2000/0xe000)=nil, 0xe000}, 0x1})
> > >> ioctl$UFFDIO_COPY(r0, 0xc028aa03, 0x0)
> > >> ioctl$UFFDIO_COPY(r0, 0xc028aa03, &(0x7f0000000000)={&(0x7f0000ffc000/0x3000)=nil, &(0x7f0000ffd000/0x2000)=nil, 0x3000})
> > >> syz_execute_func(&(0x7f00000000c0)="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")
> > >> poll(&(0x7f00000000c0)=[{}], 0x1, 0x0)
> > >
> > > Is there any way to decypher the above?
> >
> > no, I also want to know the way :(
>
> perhaps you can run it under strace?
>
> I am wondering if "goto skip_mm" in userfaultfd_release() is correct...
> shouldn't it clear VM_UFFD_* and reset vm_userfaultfd_ctx.ctx even if
> !mmget_still_valid ?

Heh, I didn't notice you too mentioned userfaultfd_release() in your email.
can you try the patch below?

Oleg.


--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -880,6 +880,7 @@ static int userfaultfd_release(struct inode *inode, struct file *file)
 	/* len == 0 means wake all */
 	struct userfaultfd_wake_range range = { .len = 0, };
 	unsigned long new_flags;
+	bool xxx;
 
 	WRITE_ONCE(ctx->released, true);
 
@@ -895,8 +896,7 @@ static int userfaultfd_release(struct inode *inode, struct file *file)
 	 * taking the mmap_sem for writing.
 	 */
 	down_write(&mm->mmap_sem);
-	if (!mmget_still_valid(mm))
-		goto skip_mm;
+	xxx = mmget_still_valid(mm);
 	prev = NULL;
 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
 		cond_resched();
@@ -907,19 +907,20 @@ static int userfaultfd_release(struct inode *inode, struct file *file)
 			continue;
 		}
 		new_flags = vma->vm_flags & ~(VM_UFFD_MISSING | VM_UFFD_WP);
-		prev = vma_merge(mm, prev, vma->vm_start, vma->vm_end,
-				 new_flags, vma->anon_vma,
-				 vma->vm_file, vma->vm_pgoff,
-				 vma_policy(vma),
-				 NULL_VM_UFFD_CTX);
-		if (prev)
-			vma = prev;
-		else
-			prev = vma;
+		if (xxx) {
+			prev = vma_merge(mm, prev, vma->vm_start, vma->vm_end,
+					 new_flags, vma->anon_vma,
+					 vma->vm_file, vma->vm_pgoff,
+					 vma_policy(vma),
+					 NULL_VM_UFFD_CTX);
+			if (prev)
+				vma = prev;
+			else
+				prev = vma;
+		}
 		vma->vm_flags = new_flags;
 		vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 	}
-skip_mm:
 	up_write(&mm->mmap_sem);
 	mmput(mm);
 wakeup: