3 files changed, 56 insertions, 15 deletions

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 44c8512a0355..efd53316be6b 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -3452,9 +3452,9 @@ static void *io_uring_validate_mmap_request(struct file *file,
 		unsigned int bgid;
 
 		bgid = (offset & ~IORING_OFF_MMAP_MASK) >> IORING_OFF_PBUF_SHIFT;
-		mutex_lock(&ctx->uring_lock);
+		rcu_read_lock();
 		ptr = io_pbuf_get_address(ctx, bgid);
-		mutex_unlock(&ctx->uring_lock);
+		rcu_read_unlock();
 		if (!ptr)
 			return ERR_PTR(-EINVAL);
 		break;
diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c
index f6e5ae026e4b..cd8c688c1a97 100644
--- a/io_uring/kbuf.c
+++ b/io_uring/kbuf.c
@@ -31,19 +31,35 @@ struct io_provide_buf {
 	__u16				bid;
 };
 
+static struct io_buffer_list *__io_buffer_get_list(struct io_ring_ctx *ctx,
+						   struct io_buffer_list *bl,
+						   unsigned int bgid)
+{
+	if (bl && bgid < BGID_ARRAY)
+		return &bl[bgid];
+
+	return xa_load(&ctx->io_bl_xa, bgid);
+}
+
 static inline struct io_buffer_list *io_buffer_get_list(struct io_ring_ctx *ctx,
 							unsigned int bgid)
 {
-	if (ctx->io_bl && bgid < BGID_ARRAY)
-		return &ctx->io_bl[bgid];
+	lockdep_assert_held(&ctx->uring_lock);
 
-	return xa_load(&ctx->io_bl_xa, bgid);
+	return __io_buffer_get_list(ctx, ctx->io_bl, bgid);
 }
 
 static int io_buffer_add_list(struct io_ring_ctx *ctx,
 			      struct io_buffer_list *bl, unsigned int bgid)
 {
+	/*
+	 * Store buffer group ID and finally mark the list as visible.
+	 * The normal lookup doesn't care about the visibility as we're
+	 * always under the ->uring_lock, but the RCU lookup from mmap does.
+	 */
 	bl->bgid = bgid;
+	smp_store_release(&bl->is_ready, 1);
+
 	if (bgid < BGID_ARRAY)
 		return 0;
 
@@ -194,18 +210,19 @@ void __user *io_buffer_select(struct io_kiocb *req, size_t *len,
 
 static __cold int io_init_bl_list(struct io_ring_ctx *ctx)
 {
+	struct io_buffer_list *bl;
 	int i;
 
-	ctx->io_bl = kcalloc(BGID_ARRAY, sizeof(struct io_buffer_list),
-				GFP_KERNEL);
-	if (!ctx->io_bl)
+	bl = kcalloc(BGID_ARRAY, sizeof(struct io_buffer_list), GFP_KERNEL);
+	if (!bl)
 		return -ENOMEM;
 
 	for (i = 0; i < BGID_ARRAY; i++) {
-		INIT_LIST_HEAD(&ctx->io_bl[i].buf_list);
-		ctx->io_bl[i].bgid = i;
+		INIT_LIST_HEAD(&bl[i].buf_list);
+		bl[i].bgid = i;
 	}
 
+	smp_store_release(&ctx->io_bl, bl);
 	return 0;
 }
 
@@ -270,7 +287,7 @@ void io_destroy_buffers(struct io_ring_ctx *ctx)
 	xa_for_each(&ctx->io_bl_xa, index, bl) {
 		xa_erase(&ctx->io_bl_xa, bl->bgid);
 		__io_remove_buffers(ctx, bl, -1U);
-		kfree(bl);
+		kfree_rcu(bl, rcu);
 	}
 
 	while (!list_empty(&ctx->io_buffers_pages)) {
@@ -455,7 +472,16 @@ int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
 		INIT_LIST_HEAD(&bl->buf_list);
 		ret = io_buffer_add_list(ctx, bl, p->bgid);
 		if (ret) {
-			kfree(bl);
+			/*
+			 * Doesn't need rcu free as it was never visible, but
+			 * let's keep it consistent throughout. Also can't
+			 * be a lower indexed array group, as adding one
+			 * where lookup failed cannot happen.
+			 */
+			if (p->bgid >= BGID_ARRAY)
+				kfree_rcu(bl, rcu);
+			else
+				WARN_ON_ONCE(1);
 			goto err;
 		}
 	}
@@ -550,6 +576,8 @@ int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 	struct io_buffer_list *bl, *free_bl = NULL;
 	int ret;
 
+	lockdep_assert_held(&ctx->uring_lock);
+
 	if (copy_from_user(&reg, arg, sizeof(reg)))
 		return -EFAULT;
 
@@ -604,7 +632,7 @@ int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 		return 0;
 	}
 
-	kfree(free_bl);
+	kfree_rcu(free_bl, rcu);
 	return ret;
 }
 
@@ -613,6 +641,8 @@ int io_unregister_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 	struct io_uring_buf_reg reg;
 	struct io_buffer_list *bl;
 
+	lockdep_assert_held(&ctx->uring_lock);
+
 	if (copy_from_user(&reg, arg, sizeof(reg)))
 		return -EFAULT;
 	if (reg.resv[0] || reg.resv[1] || reg.resv[2])
@@ -629,7 +659,7 @@ int io_unregister_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 	__io_remove_buffers(ctx, bl, -1U);
 	if (bl->bgid >= BGID_ARRAY) {
 		xa_erase(&ctx->io_bl_xa, bl->bgid);
-		kfree(bl);
+		kfree_rcu(bl, rcu);
 	}
 	return 0;
 }
@@ -638,7 +668,15 @@ void *io_pbuf_get_address(struct io_ring_ctx *ctx, unsigned long bgid)
 {
 	struct io_buffer_list *bl;
 
-	bl = io_buffer_get_list(ctx, bgid);
+	bl = __io_buffer_get_list(ctx, smp_load_acquire(&ctx->io_bl), bgid);
+
+	/*
+	 * Ensure the list is fully setup. Only strictly needed for RCU lookup
+	 * via mmap, and in that case only for the array indexed groups. For
+	 * the xarray lookups, it's either visible and ready, or not at all.
+	 */
+	if (!smp_load_acquire(&bl->is_ready))
+		return NULL;
 	if (!bl || !bl->is_mmap)
 		return NULL;
 
diff --git a/io_uring/kbuf.h b/io_uring/kbuf.h
index d14345ef61fc..274f8b7fb8ab 100644
--- a/io_uring/kbuf.h
+++ b/io_uring/kbuf.h
@@ -15,6 +15,7 @@ struct io_buffer_list {
 			struct page **buf_pages;
 			struct io_uring_buf_ring *buf_ring;
 		};
+		struct rcu_head rcu;
 	};
 	__u16 bgid;
 
@@ -28,6 +29,8 @@ struct io_buffer_list {
 	__u8 is_mapped;
 	/* ring mapped provided buffers, but mmap'ed by application */
 	__u8 is_mmap;
+	/* bl is visible from an RCU point of view for lookup */
+	__u8 is_ready;
 };
 
 struct io_buffer {