for-6.8/io_uring-2024-01-18

-----BEGIN PGP SIGNATURE-----
 
 iQJEBAABCAAuFiEEwPw5LcreJtl1+l5K99NY+ylx4KYFAmWpoAYQHGF4Ym9lQGtl
 cm5lbC5kawAKCRD301j7KXHgphZ5D/47nZdRZq1dKWDxn8qQiVjjgXtu9F8z1cXo
 AKZ5pWR44i/bIvyk7RT8dtNS1OGR88hKdF15ZoKg701AnciAHvl03Inb2Mbnpocb
 3g54WPKSlRR1qhepjXicrr5onQiqC6jKw6/CMa6z+qtKtTu+LNAh2+k1TqfDZRA/
 oXCxBHH0P0LmfRzMiJaoDwwd8UnIzd7r09qdzsvzKL5bEs64vpTKQkS5D8l2ftIf
 4PMRfkvCIPzyEsZR4jz0PTgRsVbShKVxDJaaiAQfqYxVmSGQBc0S9yvhsSe7domE
 Xodx3qJxIXrqXAaT5sOQDoo5J985Z6JDaZzK3dp1ULk+F4gXLtG1YelPd5QoMFUW
 CKfqXEby9P3JljhyIokB+3eaUEjjKBSXw7vWW2GfrMEAfp1nHzDRAGKOUz66/L+8
 R/b3WsN1CDNgHviOPhcxOVaQ5ysdIT8vVFGpYiNCb5X4hXZQS2CBY1gVN1fpLhuV
 4bj5mV8/qUPlHVs+HegCqROLX2rThASGZea1XXrrYVakTBmKWnm4s+G2Ir+wup6s
 9Xtn+v32BodfbJmipsW0/jxVuTJsCq2mcCU2ltOiXO4xwcJhw+dx8Alb1cwSvP37
 E+Sr3XVEgtsywiCrl42ndXes/ztE6v9v+WZNlP82OwMMYMlDhPAjyi/r7YWwtWT9
 r6bXKfcOfQ==
 =rd3Y
 -----END PGP SIGNATURE-----

Merge tag 'for-6.8/io_uring-2024-01-18' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:
 "Nothing major in here, just a few fixes and cleanups that arrived
  after the initial merge window pull request got finalized, as well as
  a fix for a patch that got merged earlier"

* tag 'for-6.8/io_uring-2024-01-18' of git://git.kernel.dk/linux:
  io_uring: combine cq_wait_nr checks
  io_uring: clean *local_work_add var naming
  io_uring: clean up local tw add-wait sync
  io_uring: adjust defer tw counting
  io_uring/register: guard compat syscall with CONFIG_COMPAT
  io_uring/rsrc: improve code generation for fixed file assignment
  io_uring/rw: cleanup io_rw_done()

io_uring/io_uring.c

@@ -137,6 +137,14 @@ struct io_defer_entry {
 #define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL)
 #define IO_REQ_LINK_FLAGS (REQ_F_LINK | REQ_F_HARDLINK)
 
+/*
+ * No waiters. It's larger than any valid value of the tw counter
+ * so that tests against ->cq_wait_nr would fail and skip wake_up().
+ */
+#define IO_CQ_WAKE_INIT		(-1U)
+/* Forced wake up if there is a waiter regardless of ->cq_wait_nr */
+#define IO_CQ_WAKE_FORCE	(IO_CQ_WAKE_INIT >> 1)
+
 static bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
 					 struct task_struct *task,
 					 bool cancel_all);
@@ -303,6 +311,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 		goto err;
 
 	ctx->flags = p->flags;
+	atomic_set(&ctx->cq_wait_nr, IO_CQ_WAKE_INIT);
 	init_waitqueue_head(&ctx->sqo_sq_wait);
 	INIT_LIST_HEAD(&ctx->sqd_list);
 	INIT_LIST_HEAD(&ctx->cq_overflow_list);
@@ -1304,16 +1313,23 @@ static inline void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 	unsigned nr_wait, nr_tw, nr_tw_prev;
-	struct llist_node *first;
+	struct llist_node *head;
 
+	/* See comment above IO_CQ_WAKE_INIT */
+	BUILD_BUG_ON(IO_CQ_WAKE_FORCE <= IORING_MAX_CQ_ENTRIES);
+
+	/*
+	 * We don't know how many reuqests is there in the link and whether
+	 * they can even be queued lazily, fall back to non-lazy.
+	 */
 	if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK))
 		flags &= ~IOU_F_TWQ_LAZY_WAKE;
 
-	first = READ_ONCE(ctx->work_llist.first);
+	head = READ_ONCE(ctx->work_llist.first);
 	do {
 		nr_tw_prev = 0;
-		if (first) {
+		if (head) {
-			struct io_kiocb *first_req = container_of(first,
+			struct io_kiocb *first_req = container_of(head,
 							struct io_kiocb,
 							io_task_work.node);
 			/*
@@ -1322,17 +1338,29 @@ static inline void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
 			 */
 			nr_tw_prev = READ_ONCE(first_req->nr_tw);
 		}
+
+		/*
+		 * Theoretically, it can overflow, but that's fine as one of
+		 * previous adds should've tried to wake the task.
+		 */
 		nr_tw = nr_tw_prev + 1;
-		/* Large enough to fail the nr_wait comparison below */
 		if (!(flags & IOU_F_TWQ_LAZY_WAKE))
-			nr_tw = -1U;
+			nr_tw = IO_CQ_WAKE_FORCE;
 
 		req->nr_tw = nr_tw;
-		req->io_task_work.node.next = first;
+		req->io_task_work.node.next = head;
-	} while (!try_cmpxchg(&ctx->work_llist.first, &first,
+	} while (!try_cmpxchg(&ctx->work_llist.first, &head,
 			      &req->io_task_work.node));
 
-	if (!first) {
+	/*
+	 * cmpxchg implies a full barrier, which pairs with the barrier
+	 * in set_current_state() on the io_cqring_wait() side. It's used
+	 * to ensure that either we see updated ->cq_wait_nr, or waiters
+	 * going to sleep will observe the work added to the list, which
+	 * is similar to the wait/wawke task state sync.
+	 */
+
+	if (!head) {
 		if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
 			atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
 		if (ctx->has_evfd)
@@ -1340,14 +1368,12 @@ static inline void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
 	}
 
 	nr_wait = atomic_read(&ctx->cq_wait_nr);
-	/* no one is waiting */
+	/* not enough or no one is waiting */
-	if (!nr_wait)
+	if (nr_tw < nr_wait)
 		return;
-	/* either not enough or the previous add has already woken it up */
+	/* the previous add has already woken it up */
-	if (nr_wait > nr_tw || nr_tw_prev >= nr_wait)
+	if (nr_tw_prev >= nr_wait)
 		return;
-	/* pairs with set_current_state() in io_cqring_wait() */
-	smp_mb__after_atomic();
 	wake_up_state(ctx->submitter_task, TASK_INTERRUPTIBLE);
 }
 
@@ -2000,9 +2026,10 @@ inline struct file *io_file_get_fixed(struct io_kiocb *req, int fd,
 		goto out;
 	fd = array_index_nospec(fd, ctx->nr_user_files);
 	slot = io_fixed_file_slot(&ctx->file_table, fd);
-	file = io_slot_file(slot);
+	if (!req->rsrc_node)
+		__io_req_set_rsrc_node(req, ctx);
 	req->flags |= io_slot_flags(slot);
-	io_req_set_rsrc_node(req, ctx, 0);
+	file = io_slot_file(slot);
 out:
 	io_ring_submit_unlock(ctx, issue_flags);
 	return file;
@@ -2613,7 +2640,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 
 		ret = io_cqring_wait_schedule(ctx, &iowq);
 		__set_current_state(TASK_RUNNING);
-		atomic_set(&ctx->cq_wait_nr, 0);
+		atomic_set(&ctx->cq_wait_nr, IO_CQ_WAKE_INIT);
 
 		/*
 		 * Run task_work after scheduling and before io_should_wake().

io_uring/register.c

@@ -14,6 +14,7 @@
 #include <linux/slab.h>
 #include <linux/uaccess.h>
 #include <linux/nospec.h>
+#include <linux/compat.h>
 #include <linux/io_uring.h>
 #include <linux/io_uring_types.h>
 
@@ -278,13 +279,14 @@ static __cold int io_register_iowq_aff(struct io_ring_ctx *ctx,
 	if (len > cpumask_size())
 		len = cpumask_size();
 
-	if (in_compat_syscall()) {
+#ifdef CONFIG_COMPAT
+	if (in_compat_syscall())
 		ret = compat_get_bitmap(cpumask_bits(new_mask),
 					(const compat_ulong_t __user *)arg,
 					len * 8 /* CHAR_BIT */);
-	} else {
+	else
+#endif
 		ret = copy_from_user(new_mask, arg, len);
-	}
 
 	if (ret) {
 		free_cpumask_var(new_mask);

io_uring/rsrc.h

@@ -102,17 +102,21 @@ static inline void io_charge_rsrc_node(struct io_ring_ctx *ctx,
 	node->refs++;
 }
 
+static inline void __io_req_set_rsrc_node(struct io_kiocb *req,
+					  struct io_ring_ctx *ctx)
+{
+	lockdep_assert_held(&ctx->uring_lock);
+	req->rsrc_node = ctx->rsrc_node;
+	io_charge_rsrc_node(ctx, ctx->rsrc_node);
+}
+
 static inline void io_req_set_rsrc_node(struct io_kiocb *req,
 					struct io_ring_ctx *ctx,
 					unsigned int issue_flags)
 {
 	if (!req->rsrc_node) {
 		io_ring_submit_lock(ctx, issue_flags);
-
+		__io_req_set_rsrc_node(req, ctx);
-		lockdep_assert_held(&ctx->uring_lock);
-
-		req->rsrc_node = ctx->rsrc_node;
-		io_charge_rsrc_node(ctx, ctx->rsrc_node);
 		io_ring_submit_unlock(ctx, issue_flags);
 	}
 }

io_uring/rw.c

@@ -168,27 +168,6 @@ void io_readv_writev_cleanup(struct io_kiocb *req)
 	kfree(io->free_iovec);
 }
 
-static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
-{
-	switch (ret) {
-	case -EIOCBQUEUED:
-		break;
-	case -ERESTARTSYS:
-	case -ERESTARTNOINTR:
-	case -ERESTARTNOHAND:
-	case -ERESTART_RESTARTBLOCK:
-		/*
-		 * We can't just restart the syscall, since previously
-		 * submitted sqes may already be in progress. Just fail this
-		 * IO with EINTR.
-		 */
-		ret = -EINTR;
-		fallthrough;
-	default:
-		kiocb->ki_complete(kiocb, ret);
-	}
-}
-
 static inline loff_t *io_kiocb_update_pos(struct io_kiocb *req)
 {
 	struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw);
@@ -371,6 +350,33 @@ static void io_complete_rw_iopoll(struct kiocb *kiocb, long res)
 	smp_store_release(&req->iopoll_completed, 1);
 }
 
+static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
+{
+	/* IO was queued async, completion will happen later */
+	if (ret == -EIOCBQUEUED)
+		return;
+
+	/* transform internal restart error codes */
+	if (unlikely(ret < 0)) {
+		switch (ret) {
+		case -ERESTARTSYS:
+		case -ERESTARTNOINTR:
+		case -ERESTARTNOHAND:
+		case -ERESTART_RESTARTBLOCK:
+			/*
+			 * We can't just restart the syscall, since previously
+			 * submitted sqes may already be in progress. Just fail
+			 * this IO with EINTR.
+			 */
+			ret = -EINTR;
+			break;
+		}
+	}
+
+	INDIRECT_CALL_2(kiocb->ki_complete, io_complete_rw_iopoll,
+			io_complete_rw, kiocb, ret);
+}
+
 static int kiocb_done(struct io_kiocb *req, ssize_t ret,
 		       unsigned int issue_flags)
 {