Content-Type: text/plain;
        charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 11.1 \(3445.4.7\))
Subject: Re: [PATCH] mm: save current->journal_info before calling
 fault/page_mkwrite
From: "Yan, Zheng" <zyan@redhat.com>
In-Reply-To: <20171213165923.0ea4eb3e996b7d8bf1fff72f@linux-foundation.org>
Date: Thu, 14 Dec 2017 10:09:58 +0800
Cc: LKML <linux-kernel@vger.kernel.org>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        ceph-devel <ceph-devel@vger.kernel.org>, linux-ext4@vger.kernel.org,
        viro@zeniv.linux.org.uk, jlayton@redhat.com, linux-mm@kvack.org
Content-Transfer-Encoding: 8BIT
Message-Id: <12AE4806-72D3-4AA2-A483-693375DA7D36@redhat.com>
References: <20171213035836.916-1-zyan@redhat.com>
 <20171213165923.0ea4eb3e996b7d8bf1fff72f@linux-foundation.org>
To: Andrew Morton <akpm@linux-foundation.org>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 4616
Lines: 119


> On 14 Dec 2017, at 08:59, Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> On Wed, 13 Dec 2017 11:58:36 +0800 "Yan, Zheng" <zyan@redhat.com> wrote:
> 
>> We recently got an Oops report:
>> 
>> BUG: unable to handle kernel NULL pointer dereference at (null)
>> IP: jbd2__journal_start+0x38/0x1a2
>> [...]
>> Call Trace:
>>  ext4_page_mkwrite+0x307/0x52b
>>  _ext4_get_block+0xd8/0xd8
>>  do_page_mkwrite+0x6e/0xd8
>>  handle_mm_fault+0x686/0xf9b
>>  mntput_no_expire+0x1f/0x21e
>>  __do_page_fault+0x21d/0x465
>>  dput+0x4a/0x2f7
>>  page_fault+0x22/0x30
>>  copy_user_generic_string+0x2c/0x40
>>  copy_page_to_iter+0x8c/0x2b8
>>  generic_file_read_iter+0x26e/0x845
>>  timerqueue_del+0x31/0x90
>>  ceph_read_iter+0x697/0xa33 [ceph]
>>  hrtimer_cancel+0x23/0x41
>>  futex_wait+0x1c8/0x24d
>>  get_futex_key+0x32c/0x39a
>>  __vfs_read+0xe0/0x130
>>  vfs_read.part.1+0x6c/0x123
>>  handle_mm_fault+0x831/0xf9b
>>  __fget+0x7e/0xbf
>>  SyS_read+0x4d/0xb5
>> 
>> The reason is that page fault can happen when one filesystem copies
>> data from/to userspace, the filesystem may set current->journal_info.
>> If the userspace memory is mapped to a file on another filesystem,
>> the later filesystem may also want to use current->journal_info.
>> 
> 
> whoops.
> 
> A cc:stable will be needed here...
> 
> A filesystem doesn't "copy data from/to userspace".  I assume here
> we're referring to a read() where the source is a pagecache page for
> filesystem A and the destination is a MAP_SHARED page in filesystem B?
> 
> But in that case I don't see why filesystem A would have a live
> ->journal_info?  It's just doing a read.


Background: when there are multiple cephfs clients read/write a file at time same time, read/write should go directly to object store daemon, using page cache is disabled.

ceph_read_iter() uses current->journal_info to pass context information to ceph_readpages().  ceph_readpages() needs to know if its caller has already gotten capability of using page cache (distinguish read from readahead/fadvise). If not, it tries getting the capability by itself. I checked other filesystem, btrfs probably suffers similar problem for its readpages. (verify_parent_transid() uses current->journal_info and it can be called by by btrfs_get_extent())

Regards
Yan, Zheng

> 
> So can we please have more detailed info on the exact scenario here?
> 
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -2347,12 +2347,22 @@ static int do_page_mkwrite(struct vm_fault *vmf)
>> {
>> 	int ret;
>> 	struct page *page = vmf->page;
>> +	void *old_journal_info = current->journal_info;
>> 	unsigned int old_flags = vmf->flags;
>> 
>> +	/*
>> +	 * If the fault happens during read_iter() copies data to
>> +	 * userspace, filesystem may have set current->journal_info.
>> +	 * If the userspace memory is mapped to a file on another
>> +	 * filesystem, page_mkwrite() of the later filesystem may
>> +	 * want to access/modify current->journal_info.
>> +	 */
>> +	current->journal_info = NULL;
>> 	vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
>> 
>> 	ret = vmf->vma->vm_ops->page_mkwrite(vmf);
>> -	/* Restore original flags so that caller is not surprised */
>> +	/* Restore original journal_info and flags */
>> +	current->journal_info = old_journal_info;
>> 	vmf->flags = old_flags;
>> 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
>> 		return ret;
>> @@ -3191,9 +3201,20 @@ static int do_anonymous_page(struct vm_fault *vmf)
>> static int __do_fault(struct vm_fault *vmf)
>> {
>> 	struct vm_area_struct *vma = vmf->vma;
>> +	void *old_journal_info = current->journal_info;
>> 	int ret;
>> 
>> +	/*
>> +	 * If the fault happens during write_iter() copies data from
>> +	 * userspace, filesystem may have set current->journal_info.
>> +	 * If the userspace memory is mapped to a file on another
>> +	 * filesystem, fault handler of the later filesystem may want
>> +	 * to access/modify current->journal_info.
>> +	 */
>> +	current->journal_info = NULL;
>> 	ret = vma->vm_ops->fault(vmf);
>> +	/* Restore original journal_info */
>> +	current->journal_info = old_journal_info;
>> 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY |
>> 			    VM_FAULT_DONE_COW)))
>> 		return ret;
> 
> Can you explain why you chose these two sites?  Rather than, for
> example, way up in handle_mm_fault()?
> 
> It's hard to believe that a fault handler will alter ->journal_info if
> it is handling a read fault, so perhaps we only need to do this for a
> write fault?  Although such an optimization probably isn't worthwhile. 
> The whole thing is only about three instructions.
> 
>