From: Imran Khan <imran.f.khan@oracle.com>
To: Tejun Heo <tj@kernel.org>
Cc: viro@zeniv.linux.org.uk, gregkh@linuxfoundation.org,
ebiederm@xmission.com, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v8 02/10] kernfs: make ->attr.open RCU protected.
Date: Tue, 26 Apr 2022 11:54:36 +1000 [thread overview]
Message-ID: <4a778d83-df05-14e6-038a-9e0fb86f068e@oracle.com> (raw)
In-Reply-To: <YmLVrfI4UN8ajoUM@slm.duckdns.org>
Hello Tejun,
On 23/4/22 2:19 am, Tejun Heo wrote:
> On Sun, Apr 10, 2022 at 12:37:11PM +1000, Imran Khan wrote:
>> static int kernfs_seq_show(struct seq_file *sf, void *v)
>> {
>> struct kernfs_open_file *of = sf->private;
>> + struct kernfs_open_node *on = rcu_dereference_raw(of->kn->attr.open);
>
> I suppose raw deref is safe because @on can't go away while @of is alive,
> right?
Yes.
If that's the case, please factor out of -> on dereferencing into a
> helper and put a comment there explaining why the raw deref is safe.
>
Sure, will put dereferncing in a separate helper in next version.
>> @@ -201,7 +203,8 @@ static ssize_t kernfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
>> goto out_free;
>> }
>>
>> - of->event = atomic_read(&of->kn->attr.open->event);
>> + on = rcu_dereference_raw(of->kn->attr.open);
>
> cuz we don't wanna sprinkle raw derefs in multiple places without
> explanation like this.
>
Agree.
> ...
>> /**
>> @@ -566,24 +567,30 @@ static int kernfs_get_open_node(struct kernfs_node *kn,
>> static void kernfs_put_open_node(struct kernfs_node *kn,
>> struct kernfs_open_file *of)
>> {
>> - struct kernfs_open_node *on = kn->attr.open;
>> - unsigned long flags;
>> + struct kernfs_open_node *on;
>> +
>> + /* ->attr.open NULL means there are no more open files */
>> + if (rcu_dereference_raw(kn->attr.open) == NULL)
>> + return;
>
> For pointer value check, what you want is rcu_access_pointer(). That said,
> tho, why is this being called if no one is linked on it? Before removing the
> refcnt, that'd be the same as trying to put a 0 ref. How does that happen?
Yeah this check surely should not be needed. I will remove it in next
version.
> Also, can you please rename it to unlink or something of the sort? It's
> confusing to call it put when there's no refcnt.
>
sure I will rename _put_ to _unlink_.
>>
>> mutex_lock(&kernfs_open_file_mutex);
>> - spin_lock_irqsave(&kernfs_open_node_lock, flags);
>> +
>> + on = rcu_dereference_protected(kn->attr.open,
>> + lockdep_is_held(&kernfs_open_file_mutex));
>
> Again, a better way to do it would be defining a kn -> on accessor which
> encodes the safe way to deref and use it. The deref rule is tied to the
> deref itself not the callsite.
>
Okay I will factor this out in a separate helper.
[...]
>> static int kernfs_fop_open(struct inode *inode, struct file *file)
>> @@ -765,12 +772,13 @@ void kernfs_drain_open_files(struct kernfs_node *kn)
>> if (!(kn->flags & (KERNFS_HAS_MMAP | KERNFS_HAS_RELEASE)))
>> return;
>>
>> - on = kn->attr.open;
>> - if (!on)
>> + if (rcu_dereference_raw(kn->attr.open) == NULL)
>> return;
>
> rcu_access_pointer again and you gotta explain why the lockless check is
> safe.
>
The lockless check is safe because no one will be adding to ->attr.open
at this point of time. This allows early bail out if ->attr.open in
already NULL. And if kernfs_put_open_node makes ->attr.open NULL, it
does this under open_file_mutex so subsequent check under
open_file_mutex will make sure to bail out if kernfs_put_open_node won
the race.
I will put an explanatory comment in the code, explaining the same.
[...]
>> @@ -912,14 +920,13 @@ void kernfs_notify(struct kernfs_node *kn)
>> return;
>>
>> /* kick poll immediately */
>> - spin_lock_irqsave(&kernfs_open_node_lock, flags);
>> - on = kn->attr.open;
>> + rcu_read_lock();
>> + on = rcu_dereference(kn->attr.open);
>> if (on) {
>> atomic_inc(&on->event);
>> wake_up_interruptible(&on->poll);
>> }
>> - spin_unlock_irqrestore(&kernfs_open_node_lock, flags);
>> -
>> + rcu_read_unlock();
>
> An explanation of why this is safe in terms of event ordering would be great
> here.
>
This is safe because here we don't need to refcnt ->on in this case. If
writer (kernfs_put_open_node) has already made ->attr.open NULL we will
bail out. If kernfs_notify got an old ->attr.open we can still safely
process the event, even if kernfs_put_open_node updates ->attr.open to
NULL in parallel.
In both the cases the behaviour/order will be same as earlier code that
used kernfs_open_node_lock.
Please let me know if this answers your query or if something is still
missing.
Thanks
-- Imran
next prev parent reply other threads:[~2022-04-26 1:54 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-04-10 2:37 [PATCH v8 00/10] kernfs: Remove reference counting for kernfs_open_node Imran Khan
2022-04-10 2:37 ` [PATCH v8 01/10] " Imran Khan
2022-04-22 16:03 ` Tejun Heo
2022-04-26 1:43 ` Imran Khan
2022-04-26 18:29 ` Tejun Heo
2022-04-26 20:13 ` Al Viro
2022-04-26 20:16 ` Tejun Heo
2022-04-10 2:37 ` [PATCH v8 02/10] kernfs: make ->attr.open RCU protected Imran Khan
2022-04-22 16:19 ` Tejun Heo
2022-04-26 1:54 ` Imran Khan [this message]
2022-04-26 18:37 ` Tejun Heo
2022-04-10 2:37 ` [PATCH v8 03/10] kernfs: Change kernfs_notify_list to llist Imran Khan
2022-04-22 16:41 ` Tejun Heo
2022-04-10 2:37 ` [PATCH v8 04/10] kernfs: Introduce interface to access global kernfs_open_file_mutex Imran Khan
2022-04-10 2:37 ` [PATCH v8 05/10] kernfs: Replace global kernfs_open_file_mutex with hashed mutexes Imran Khan
2022-04-22 17:00 ` Tejun Heo
2022-04-10 2:37 ` [PATCH v8 06/10] kernfs: Use a per-fs rwsem to protect per-fs list of kernfs_super_info Imran Khan
2022-04-10 2:37 ` [PATCH v8 07/10] kernfs: Change kernfs_rename_lock into a read-write lock Imran Khan
2022-04-10 2:37 ` [PATCH v8 08/10] kernfs: Introduce interface to access per-fs rwsem Imran Khan
2022-04-10 2:37 ` [PATCH v8 09/10] kernfs: Replace per-fs rwsem with hashed rwsems Imran Khan
2022-04-10 2:37 ` [PATCH v8 10/10] kernfs: Add a document to describe hashed locks used in kernfs Imran Khan
2022-04-22 17:03 ` [PATCH v8 00/10] kernfs: Remove reference counting for kernfs_open_node Tejun Heo
2022-04-23 8:49 ` Imran Khan
2022-04-25 17:21 ` Tejun Heo
2022-04-28 17:28 ` Imran Khan
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4a778d83-df05-14e6-038a-9e0fb86f068e@oracle.com \
--to=imran.f.khan@oracle.com \
--cc=ebiederm@xmission.com \
--cc=gregkh@linuxfoundation.org \
--cc=linux-kernel@vger.kernel.org \
--cc=tj@kernel.org \
--cc=viro@zeniv.linux.org.uk \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).