Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Pavel Begunkov <asml.silence@gmail.com>
To:     io-uring@vger.kernel.org
Cc:     Jens Axboe <axboe@kernel.dk>, asml.silence@gmail.com,
        linux-kernel@vger.kernel.org
Subject: [PATCH 05/11] io_uring/rsrc: protect node refs with uring_lock
Date:   Thu, 30 Mar 2023 15:53:23 +0100
Message-Id: <af5eecb33fd78144d97c1228c54fe1a5dcc9ee9a.1680187408.git.asml.silence@gmail.com>
In-Reply-To: <cover.1680187408.git.asml.silence@gmail.com>
References: <cover.1680187408.git.asml.silence@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Currently, for nodes we have an atomic counter and some cached
(non-atomic) refs protected by uring_lock. Let's put all ref
manipulations under uring_lock and get rid of the atomic part.
It's free as in all cases we care about we already hold the lock.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/io_uring.c | 18 ++++++++++++------
 io_uring/rsrc.c     | 30 ++++--------------------------
 io_uring/rsrc.h     | 29 +++++------------------------
 3 files changed, 21 insertions(+), 56 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 52a88da65f57..e55c48d91209 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -967,7 +967,7 @@ bool io_aux_cqe(struct io_ring_ctx *ctx, bool defer, u64 user_data, s32 res, u32
 	return true;
 }
 
-static void __io_req_complete_post(struct io_kiocb *req)
+static void __io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 	struct io_rsrc_node *rsrc_node = NULL;
@@ -1003,7 +1003,11 @@ static void __io_req_complete_post(struct io_kiocb *req)
 	}
 	io_cq_unlock_post(ctx);
 
-	io_put_rsrc_node(rsrc_node);
+	if (rsrc_node) {
+		io_ring_submit_lock(ctx, issue_flags);
+		io_put_rsrc_node(rsrc_node);
+		io_ring_submit_unlock(ctx, issue_flags);
+	}
 }
 
 void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)
@@ -1013,12 +1017,12 @@ void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)
 		io_req_task_work_add(req);
 	} else if (!(issue_flags & IO_URING_F_UNLOCKED) ||
 		   !(req->ctx->flags & IORING_SETUP_IOPOLL)) {
-		__io_req_complete_post(req);
+		__io_req_complete_post(req, issue_flags);
 	} else {
 		struct io_ring_ctx *ctx = req->ctx;
 
 		mutex_lock(&ctx->uring_lock);
-		__io_req_complete_post(req);
+		__io_req_complete_post(req, issue_flags & ~IO_URING_F_UNLOCKED);
 		mutex_unlock(&ctx->uring_lock);
 	}
 }
@@ -1120,7 +1124,10 @@ __cold void io_free_req_tw(struct io_kiocb *req, struct io_tw_state *ts)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 
-	io_put_rsrc_node(req->rsrc_node);
+	if (req->rsrc_node) {
+		io_tw_lock(ctx, ts);
+		io_put_rsrc_node(req->rsrc_node);
+	}
 	io_dismantle_req(req);
 	io_put_task_remote(req->task, 1);
 
@@ -2791,7 +2798,6 @@ static void io_req_caches_free(struct io_ring_ctx *ctx)
 static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
 {
 	io_sq_thread_finish(ctx);
-	io_rsrc_refs_drop(ctx);
 	/* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */
 	io_wait_rsrc_data(ctx->buf_data);
 	io_wait_rsrc_data(ctx->file_data);
diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
index 1e7c960737fd..1237fc77c250 100644
--- a/io_uring/rsrc.c
+++ b/io_uring/rsrc.c
@@ -27,23 +27,10 @@ static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
 				  struct io_mapped_ubuf **pimu,
 				  struct page **last_hpage);
 
-#define IO_RSRC_REF_BATCH	100
-
 /* only define max */
 #define IORING_MAX_FIXED_FILES	(1U << 20)
 #define IORING_MAX_REG_BUFFERS	(1U << 14)
 
-void io_rsrc_refs_drop(struct io_ring_ctx *ctx)
-	__must_hold(&ctx->uring_lock)
-{
-	struct io_rsrc_node *node = ctx->rsrc_node;
-
-	if (node && node->cached_refs) {
-		io_rsrc_put_node(node, node->cached_refs);
-		node->cached_refs = 0;
-	}
-}
-
 int __io_account_mem(struct user_struct *user, unsigned long nr_pages)
 {
 	unsigned long page_limit, cur_pages, new_pages;
@@ -153,13 +140,6 @@ static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slo
 	*slot = NULL;
 }
 
-void io_rsrc_refs_refill(struct io_ring_ctx *ctx, struct io_rsrc_node *node)
-	__must_hold(&ctx->uring_lock)
-{
-	node->cached_refs += IO_RSRC_REF_BATCH;
-	refcount_add(IO_RSRC_REF_BATCH, &node->refs);
-}
-
 static void __io_rsrc_put_work(struct io_rsrc_node *ref_node)
 {
 	struct io_rsrc_data *rsrc_data = ref_node->rsrc_data;
@@ -225,7 +205,8 @@ void io_rsrc_node_destroy(struct io_rsrc_node *ref_node)
 	kfree(ref_node);
 }
 
-__cold void io_rsrc_node_ref_zero(struct io_rsrc_node *node)
+void io_rsrc_node_ref_zero(struct io_rsrc_node *node)
+	__must_hold(&node->rsrc_data->ctx->uring_lock)
 {
 	struct io_ring_ctx *ctx = node->rsrc_data->ctx;
 	unsigned long flags;
@@ -269,7 +250,7 @@ static struct io_rsrc_node *io_rsrc_node_alloc(void)
 	if (!ref_node)
 		return NULL;
 
-	refcount_set(&ref_node->refs, 1);
+	ref_node->refs = 1;
 	INIT_LIST_HEAD(&ref_node->node);
 	INIT_LIST_HEAD(&ref_node->rsrc_list);
 	ref_node->done = false;
@@ -283,8 +264,6 @@ void io_rsrc_node_switch(struct io_ring_ctx *ctx,
 	WARN_ON_ONCE(!ctx->rsrc_backup_node);
 	WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node);
 
-	io_rsrc_refs_drop(ctx);
-
 	if (data_to_kill) {
 		struct io_rsrc_node *rsrc_node = ctx->rsrc_node;
 
@@ -295,14 +274,13 @@ void io_rsrc_node_switch(struct io_ring_ctx *ctx,
 
 		atomic_inc(&data_to_kill->refs);
 		/* put master ref */
-		io_rsrc_put_node(rsrc_node, 1);
+		io_put_rsrc_node(rsrc_node);
 		ctx->rsrc_node = NULL;
 	}
 
 	if (!ctx->rsrc_node) {
 		ctx->rsrc_node = ctx->rsrc_backup_node;
 		ctx->rsrc_backup_node = NULL;
-		ctx->rsrc_node->cached_refs = 0;
 	}
 }
 
diff --git a/io_uring/rsrc.h b/io_uring/rsrc.h
index 8164777279ba..fd3bdd30a993 100644
--- a/io_uring/rsrc.h
+++ b/io_uring/rsrc.h
@@ -37,13 +37,12 @@ struct io_rsrc_data {
 };
 
 struct io_rsrc_node {
-	refcount_t			refs;
+	int refs;
 	struct list_head		node;
 	struct list_head		rsrc_list;
 	struct io_rsrc_data		*rsrc_data;
 	struct llist_node		llist;
 	bool				done;
-	int				cached_refs;
 };
 
 struct io_mapped_ubuf {
@@ -57,10 +56,8 @@ struct io_mapped_ubuf {
 void io_rsrc_put_tw(struct callback_head *cb);
 void io_rsrc_node_ref_zero(struct io_rsrc_node *node);
 void io_rsrc_put_work(struct work_struct *work);
-void io_rsrc_refs_refill(struct io_ring_ctx *ctx, struct io_rsrc_node *node);
 void io_wait_rsrc_data(struct io_rsrc_data *data);
 void io_rsrc_node_destroy(struct io_rsrc_node *ref_node);
-void io_rsrc_refs_drop(struct io_ring_ctx *ctx);
 int io_rsrc_node_switch_start(struct io_ring_ctx *ctx);
 int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
 			  struct io_rsrc_node *node, void *rsrc);
@@ -109,38 +106,22 @@ int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg,
 int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg,
 			unsigned int size, unsigned int type);
 
-static inline void io_rsrc_put_node(struct io_rsrc_node *node, int nr)
-{
-	if (refcount_sub_and_test(nr, &node->refs))
-		io_rsrc_node_ref_zero(node);
-}
-
 static inline void io_put_rsrc_node(struct io_rsrc_node *node)
 {
-	if (node)
-		io_rsrc_put_node(node, 1);
+	if (node && !--node->refs)
+		io_rsrc_node_ref_zero(node);
 }
 
 static inline void io_req_put_rsrc_locked(struct io_kiocb *req,
 					  struct io_ring_ctx *ctx)
-	__must_hold(&ctx->uring_lock)
 {
-	struct io_rsrc_node *node = req->rsrc_node;
-
-	if (node) {
-		if (node == ctx->rsrc_node)
-			node->cached_refs++;
-		else
-			io_rsrc_put_node(node, 1);
-	}
+	io_put_rsrc_node(req->rsrc_node);
 }
 
 static inline void io_charge_rsrc_node(struct io_ring_ctx *ctx,
 				       struct io_rsrc_node *node)
 {
-	node->cached_refs--;
-	if (unlikely(node->cached_refs < 0))
-		io_rsrc_refs_refill(ctx, node);
+	node->refs++;
 }
 
 static inline void io_req_set_rsrc_node(struct io_kiocb *req,
-- 
2.39.1