netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Kuniyuki Iwashima <kuniyu@amazon.com>
To: "David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>
Cc: Ivan Babrou <ivan@cloudflare.com>,
	Kuniyuki Iwashima <kuniyu@amazon.com>,
	Kuniyuki Iwashima <kuni1840@gmail.com>, <netdev@vger.kernel.org>
Subject: [PATCH v4 net-next 3/4] af_unix: Run GC on only one CPU.
Date: Tue, 19 Dec 2023 12:01:01 +0900	[thread overview]
Message-ID: <20231219030102.27509-4-kuniyu@amazon.com> (raw)
In-Reply-To: <20231219030102.27509-1-kuniyu@amazon.com>

If more than 16000 inflight AF_UNIX sockets exist and the garbage
collector is not running, unix_(dgram|stream)_sendmsg() call unix_gc().
Also, they wait for unix_gc() to complete.

In unix_gc(), all inflight AF_UNIX sockets are traversed at least once,
and more if they are the GC candidate.  Thus, sendmsg() significantly
slows down with too many inflight AF_UNIX sockets.

There is a small window to invoke multiple unix_gc() instances, which
will then be blocked by the same spinlock except for one.

Let's convert unix_gc() to use struct work so that it will not consume
CPUs unnecessarily.

Note WRITE_ONCE(gc_in_progress, true) is moved to wait_for_unix_gc().
If we leave the WRITE_ONCE() as is and use the following test to call
flush_work(), a process might not call it.

    CPU 0                                     CPU 1
    ---                                       ---
                                              start work and call __unix_gc()
    if (work_pending(&unix_gc_work) ||        <-- false
        READ_ONCE(gc_in_progress))            <-- false
            flush_work();                     <-- missed!
	                                      WRITE_ONCE(gc_in_progress, true)

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 net/unix/garbage.c | 37 +++++++++++++++++++------------------
 1 file changed, 19 insertions(+), 18 deletions(-)

diff --git a/net/unix/garbage.c b/net/unix/garbage.c
index 4d634f5f6a55..0dba36b0bb95 100644
--- a/net/unix/garbage.c
+++ b/net/unix/garbage.c
@@ -86,7 +86,9 @@
 /* Internal data structures and random procedures: */
 
 static LIST_HEAD(gc_candidates);
-static DECLARE_WAIT_QUEUE_HEAD(unix_gc_wait);
+
+static void __unix_gc(struct work_struct *work);
+static DECLARE_WORK(unix_gc_work, __unix_gc);
 
 static void scan_inflight(struct sock *x, void (*func)(struct unix_sock *),
 			  struct sk_buff_head *hitlist)
@@ -188,17 +190,21 @@ void wait_for_unix_gc(void)
 {
 	/* If number of inflight sockets is insane,
 	 * force a garbage collect right now.
+	 *
 	 * Paired with the WRITE_ONCE() in unix_inflight(),
-	 * unix_notinflight() and gc_in_progress().
+	 * unix_notinflight(), and __unix_gc().
 	 */
 	if (READ_ONCE(unix_tot_inflight) > UNIX_INFLIGHT_TRIGGER_GC &&
-	    !READ_ONCE(gc_in_progress))
-		unix_gc();
-	wait_event(unix_gc_wait, gc_in_progress == false);
+	    !READ_ONCE(gc_in_progress)) {
+		WRITE_ONCE(gc_in_progress, true);
+		queue_work(system_unbound_wq, &unix_gc_work);
+	}
+
+	if (READ_ONCE(gc_in_progress))
+		flush_work(&unix_gc_work);
 }
 
-/* The external entry point: unix_gc() */
-void unix_gc(void)
+static void __unix_gc(struct work_struct *work)
 {
 	struct sk_buff *next_skb, *skb;
 	struct unix_sock *u;
@@ -209,13 +215,6 @@ void unix_gc(void)
 
 	spin_lock(&unix_gc_lock);
 
-	/* Avoid a recursive GC. */
-	if (gc_in_progress)
-		goto out;
-
-	/* Paired with READ_ONCE() in wait_for_unix_gc(). */
-	WRITE_ONCE(gc_in_progress, true);
-
 	/* First, select candidates for garbage collection.  Only
 	 * in-flight sockets are considered, and from those only ones
 	 * which don't have any external reference.
@@ -319,11 +318,13 @@ void unix_gc(void)
 	/* All candidates should have been detached by now. */
 	BUG_ON(!list_empty(&gc_candidates));
 
+	spin_unlock(&unix_gc_lock);
+
 	/* Paired with READ_ONCE() in wait_for_unix_gc(). */
 	WRITE_ONCE(gc_in_progress, false);
+}
 
-	wake_up(&unix_gc_wait);
-
- out:
-	spin_unlock(&unix_gc_lock);
+void unix_gc(void)
+{
+	queue_work(system_unbound_wq, &unix_gc_work);
 }
-- 
2.30.2


  parent reply	other threads:[~2023-12-19  3:02 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-12-19  3:00 [PATCH v4 net-next 0/4] af_unix: Random improvements for GC Kuniyuki Iwashima
2023-12-19  3:00 ` [PATCH v4 net-next 1/4] af_unix: Do not use atomic ops for unix_sk(sk)->inflight Kuniyuki Iwashima
2023-12-19  3:01 ` [PATCH v4 net-next 2/4] af_unix: Return struct unix_sock from unix_get_socket() Kuniyuki Iwashima
2023-12-19  3:01 ` Kuniyuki Iwashima [this message]
2023-12-19  3:01 ` [PATCH v4 net-next 4/4] af_unix: Try to run GC async Kuniyuki Iwashima
2024-01-02 23:29 ` [PATCH v4 net-next 0/4] af_unix: Random improvements for GC Jakub Kicinski
2024-01-09  2:01   ` Kuniyuki Iwashima

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=20231219030102.27509-4-kuniyu@amazon.com \
    --to=kuniyu@amazon.com \
    --cc=davem@davemloft.net \
    --cc=edumazet@google.com \
    --cc=ivan@cloudflare.com \
    --cc=kuba@kernel.org \
    --cc=kuni1840@gmail.com \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    /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).