On Wed, Jan 10, 2018 at 4:18 PM, Chengguang Xu <cgxu519@xxxxxxxxxx> wrote:
> 在 2018年1月10日，下午9:16，Amir Goldstein <amir73il@xxxxxxxxx> 写道：
>>
>> On Wed, Jan 10, 2018 at 2:32 PM, Chengguang Xu <cgxu519@xxxxxxxxxx> wrote:
>>> Currently syncfs(2) call on overlayfs just simply call sync_filesystem()
>>> on upper_sb to synchronize whole dirty inodes in upper filesystem
>>> regardless of the overlay ownership of the inode. It has obvious
>>> shortcomings as below.
>>>
>>> (1) Performance
>>> Synchronization is probably heavy in most cases, especially when upper
>>> filesystem is not dedicated to target overlayfs.
>>>
>>> (2) Interference
>>> Unplanned synchronization will probably impact IO performance of the
>>> processes which use other overlayfs on same upper filesystem or directly
>>> use upper filesystem.
>>>
>>> This patch iterates overlay inodes to only sync target dirty inodes in
>>> upper filesystem. By doing this, It is able to reduce cost of synchronization
>>> and will not seriously impact IO performance of irrelative processes.
>>>
>>> Signed-off-by: Chengguang Xu <cgxu519@xxxxxxxxxx>
>>> ---
>>> Changes since v1:
>>> - If upper filesystem is readonly mode then skip synchronization.
>>> - Introduce global wait list to replace temporary wait list for
>>> concurrent synchronization.
>>>
>>
>> Looks ok. A few more suggestions below.
>>
>>> fs/overlayfs/ovl_entry.h | 5 +++
>>> fs/overlayfs/super.c | 105 +++++++++++++++++++++++++++++++++++++++++++++--
>>> 2 files changed, 107 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/fs/overlayfs/ovl_entry.h b/fs/overlayfs/ovl_entry.h
>>> index 9d0bc03..ff935da 100644
>>> --- a/fs/overlayfs/ovl_entry.h
>>> +++ b/fs/overlayfs/ovl_entry.h
>>> @@ -52,6 +52,9 @@ struct ovl_fs {
>>> /* Did we take the inuse lock? */
>>> bool upperdir_locked;
>>> bool workdir_locked;
>>> + /* ovl inode sync list and lock */
>>> + spinlock_t ovl_sync_list_lock;
>>> + struct list_head ovl_sync_list;
>>> };
>>>
>>> /* private information held for every overlayfs dentry */
>>> @@ -80,6 +83,8 @@ struct ovl_inode {
>>>
>>> /* synchronize copy up and more */
>>> struct mutex lock;
>>> + /* ovl inode sync list */
>>> + struct list_head sync_list;
>>> };
>>>
>>> static inline struct ovl_inode *OVL_I(struct inode *inode)
>>> diff --git a/fs/overlayfs/super.c b/fs/overlayfs/super.c
>>> index 76440fe..c7b788b 100644
>>> --- a/fs/overlayfs/super.c
>>> +++ b/fs/overlayfs/super.c
>>> @@ -17,6 +17,8 @@
>>> #include <linux/statfs.h>
>>> #include <linux/seq_file.h>
>>> #include <linux/posix_acl_xattr.h>
>>> +#include <linux/writeback.h>
>>> +#include <linux/blkdev.h>
>>> #include "overlayfs.h"
>>>
>>> MODULE_AUTHOR("Miklos Szeredi <miklos@xxxxxxxxxx>");
>>> @@ -195,6 +197,7 @@ static struct inode *ovl_alloc_inode(struct super_block *sb)
>>> oi->__upperdentry = NULL;
>>> oi->lower = NULL;
>>> mutex_init(&oi->lock);
>>> + INIT_LIST_HEAD(&oi->sync_list);
>>>
>>> return &oi->vfs_inode;
>>> }
>>> @@ -252,6 +255,97 @@ static void ovl_put_super(struct super_block *sb)
>>> ovl_free_fs(ofs);
>>> }
>>>
>>> +/**
>>> + * ovl_sync_filesystem
>>> + * @sb: The overlayfs super block
>>> + *
>>> + * Sync underlying dirty inodes in upper filesystem and wait for completion.
>>> + */
>>> +static int ovl_sync_filesystem(struct super_block *sb)
>>> +{
>>> + struct ovl_fs *ofs = sb->s_fs_info;
>>> + struct super_block *upper_sb = ofs->upper_mnt->mnt_sb;
>>> + struct ovl_inode *oi, *oi_next;
>>> + struct inode *inode, *i_next;
>>> + struct inode *upper_inode;
>>> + struct blk_plug plug;
>>> +
>>> + struct writeback_control wbc = {
>>> + .sync_mode = WB_SYNC_ALL,
>>> + .for_sync = 1,
>>> + .range_start = 0,
>>> + .range_end = LLONG_MAX,
>>> + .nr_to_write = LONG_MAX,
>>> + };
>>> +
>>> + blk_start_plug(&plug);
>>> + spin_lock(&sb->s_inode_list_lock);
>>> + list_for_each_entry_safe(inode, i_next, &sb->s_inodes, i_sb_list) {
>>> + upper_inode = ovl_inode_upper(inode);
>>> + if (!upper_inode)
>>> + continue;
>>> +
>>> + spin_lock(&upper_inode->i_lock);
>>> + if (upper_inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE) ||
>>> + list_empty(&upper_inode->i_io_list)) {
>>> + spin_unlock(&upper_inode->i_lock);
>>> + continue;
>>> + }
>>> + spin_unlock(&upper_inode->i_lock);
>>> +
>>> + if (!igrab(inode))
>>> + continue;
>>> +
>>> + if (!igrab(upper_inode)) {
>>> + iput(inode);
>>> + continue;
>>> + }
>>> + spin_unlock(&sb->s_inode_list_lock);
>>> +
>>> + sync_inode(upper_inode, &wbc);
>>> + spin_lock(&ofs->ovl_sync_list_lock);
>>> + if (list_empty(&OVL_I(inode)->sync_list))
>>> + list_add(&OVL_I(inode)->sync_list, &ofs->ovl_sync_list);
>>> + else {
>>> + iput(upper_inode);
>>> + iput(inode);
>>> + }
>>> + spin_unlock(&ofs->ovl_sync_list_lock);
>>> +
>>> + if (need_resched()) {
>>> + blk_finish_plug(&plug);
>>> + cond_resched();
>>> + blk_start_plug(&plug);
>>> + }
>>> + spin_lock(&sb->s_inode_list_lock);
>>> + }
>>> + spin_unlock(&sb->s_inode_list_lock);
>>> + blk_finish_plug(&plug);
>>> +
>>
>> You may want to consider factoring out helpers ovl_sync_inodes()
>> and maybe also ovl_writeback_inodes() ovl_wait_inodes(),
>> so the code in nicer and resembles generic helper structure.
>
> hmmm，you mean imitating VFS structure? :-)
Yes. smaller functions is better coding and easier to review.
>
>>
>>> + mutex_lock(&upper_sb->s_sync_lock);
>>
>> So it makes some sense to use upper_sb sync lock
>> to synchronize with callers of sync(2)/syncfs(2) on the upper fs,
>> but that could result syncfs() on overlay to take much longer if
>> there is a syncfs() on upper fs in progress.
>
> I have two proposals for this.
>
> A.
> After changing wait list to global, I think waiting without
> upper_sb->s_sync_lock is still safe, so we can avoid coupling
> with syncfs of upper filesystem and I prefer this.
>
> B.
> If you are worried about solution A, then we can introduce a mutex_lock
> inside overlayfs for serializing concurrent sync waiting operations.
>
No need to introduce a new mutex you should just use sb->s_sync_lock
and BTW, option A is probably safe w.r.t races, because according to
comment above wait_sb_inodes(), s_sync_lock is meant to improve
performance, not to avoid races.
>>
>> If you think of ovl_sync_filesystem() as a selective fsync() iterator,
>> then it might make more sense to synchronize overlayfs syncfs()
>> callers with overlay sb sync lock.
>
> IIRC, fsync doing sync & wait one by one in good order, and maybe it also
> calls metadata sync after that. So I think it’s not so efficient way to
> sync fs and may cause write amplification, and finally it needs opening file.
>
I did not mean you should actually use fsync(), just that fsync() of upper
inodes does not take upper_sb->s_sync_lock, so ovl_sync_filesystem()
should not take it either.
Thanks,
Amir.
--
To unsubscribe from this list: send the line "unsubscribe linux-unionfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
