Re: [PATCH 4.19] memfd: Fix locking when tagging pins

[Sasha Levin <sashal@kernel.org>]

On Sat, Oct 26, 2019 at 10:03:49AM +0800, zhong jiang wrote:
>On 2019/10/26 0:58, Matthew Wilcox wrote:
>> From: "Matthew Wilcox (Oracle)" <willy@infradead.org>
>>
>> The RCU lock is insufficient to protect the radix tree iteration as
>> a deletion from the tree can occur before we take the spinlock to
>> tag the entry.  In 4.19, this has manifested as a bug with the following
>> trace:
>>
>> kernel BUG at lib/radix-tree.c:1429!
>> invalid opcode: 0000 [#1] SMP KASAN PTI
>> CPU: 7 PID: 6935 Comm: syz-executor.2 Not tainted 4.19.36 #25
>> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
>> RIP: 0010:radix_tree_tag_set+0x200/0x2f0 lib/radix-tree.c:1429
>> Code: 00 00 5b 5d 41 5c 41 5d 41 5e 41 5f c3 48 89 44 24 10 e8 a3 29 7e fe 48 8b 44 24 10 48 0f ab 03 e9 d2 fe ff ff e8 90 29 7e fe <0f> 0b 48 c7 c7 e0 5a 87 84 e8 f0 e7 08 ff 4c 89 ef e8 4a ff ac fe
>> RSP: 0018:ffff88837b13fb60 EFLAGS: 00010016
>> RAX: 0000000000040000 RBX: ffff8883c5515d58 RCX: ffffffff82cb2ef0
>> RDX: 0000000000000b72 RSI: ffffc90004cf2000 RDI: ffff8883c5515d98
>> RBP: ffff88837b13fb98 R08: ffffed106f627f7e R09: ffffed106f627f7e
>> R10: 0000000000000001 R11: ffffed106f627f7d R12: 0000000000000004
>> R13: ffffea000d7fea80 R14: 1ffff1106f627f6f R15: 0000000000000002
>> FS:  00007fa1b8df2700(0000) GS:ffff8883e2fc0000(0000) knlGS:0000000000000000
>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> CR2: 00007fa1b8df1db8 CR3: 000000037d4d2001 CR4: 0000000000160ee0
>> Call Trace:
>>  memfd_tag_pins mm/memfd.c:51 [inline]
>>  memfd_wait_for_pins+0x2c5/0x12d0 mm/memfd.c:81
>>  memfd_add_seals mm/memfd.c:215 [inline]
>>  memfd_fcntl+0x33d/0x4a0 mm/memfd.c:247
>>  do_fcntl+0x589/0xeb0 fs/fcntl.c:421
>>  __do_sys_fcntl fs/fcntl.c:463 [inline]
>>  __se_sys_fcntl fs/fcntl.c:448 [inline]
>>  __x64_sys_fcntl+0x12d/0x180 fs/fcntl.c:448
>>  do_syscall_64+0xc8/0x580 arch/x86/entry/common.c:293
>>
>> The problem does not occur in mainline due to the XArray rewrite which
>> changed the locking to exclude modification of the tree during iteration.
>> At the time, nobody realised this was a bugfix.  Backport the locking
>> changes to stable.
>>
>> Cc: stable@vger.kernel.org
>> Reported-by: zhong jiang <zhongjiang@huawei.com>
>> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
>> ---
>>  mm/memfd.c | 18 ++++++++++--------
>>  1 file changed, 10 insertions(+), 8 deletions(-)
>>
>> diff --git a/mm/memfd.c b/mm/memfd.c
>> index 2bb5e257080e..5859705dafe1 100644
>> --- a/mm/memfd.c
>> +++ b/mm/memfd.c
>> @@ -34,11 +34,12 @@ static void memfd_tag_pins(struct address_space *mapping)
>>  	void __rcu **slot;
>>  	pgoff_t start;
>>  	struct page *page;
>> +	unsigned int tagged = 0;
>>
>>  	lru_add_drain();
>>  	start = 0;
>> -	rcu_read_lock();
>>
>> +	xa_lock_irq(&mapping->i_pages);
>>  	radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
>>  		page = radix_tree_deref_slot(slot);
>>  		if (!page || radix_tree_exception(page)) {
>> @@ -47,18 +48,19 @@ static void memfd_tag_pins(struct address_space *mapping)
>>  				continue;
>>  			}
>>  		} else if (page_count(page) - page_mapcount(page) > 1) {
>> -			xa_lock_irq(&mapping->i_pages);
>>  			radix_tree_tag_set(&mapping->i_pages, iter.index,
>>  					   MEMFD_TAG_PINNED);
>> -			xa_unlock_irq(&mapping->i_pages);
>>  		}
>>
>> -		if (need_resched()) {
>> -			slot = radix_tree_iter_resume(slot, &iter);
>> -			cond_resched_rcu();
>> -		}
>> +		if (++tagged % 1024)
>> +			continue;
>> +
>> +		slot = radix_tree_iter_resume(slot, &iter);
>> +		xa_unlock_irq(&mapping->i_pages);
>> +		cond_resched();
>> +		xa_lock_irq(&mapping->i_pages);
>>  	}
>> -	rcu_read_unlock();
>> +	xa_unlock_irq(&mapping->i_pages);
>>  }
>>
>>  /*
>The patch looks good to me.   thanks for your review and efforts.
>
>Sasha,  The patch was correct,  It should go into stable instead of my patch.

I've queued up this series for all respective branches (fixing up 4.19),
thanks!

-- 
Thanks,
Sasha