Merge tag 'io_uring-6.11-20240726' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:

 - Fix a syzbot issue for the msg ring cache added in this release. No
   ill effects from this one, but it did make KMSAN unhappy (me)

 - Sanitize the NAPI timeout handling, by unifying the value handling
   into all ktime_t rather than converting back and forth (Pavel)

 - Fail NAPI registration for IOPOLL rings, it's not supported (Pavel)

 - Fix a theoretical issue with ring polling and cancelations (Pavel)

 - Various little cleanups and fixes (Pavel)

* tag 'io_uring-6.11-20240726' of git://git.kernel.dk/linux:
  io_uring/napi: pass ktime to io_napi_adjust_timeout
  io_uring/napi: use ktime in busy polling
  io_uring/msg_ring: fix uninitialized use of target_req->flags
  io_uring: align iowq and task request error handling
  io_uring: kill REQ_F_CANCEL_SEQ
  io_uring: simplify io_uring_cmd return
  io_uring: fix io_match_task must_hold
  io_uring: don't allow netpolling with SETUP_IOPOLL
  io_uring: tighten task exit cancellations

include/linux/io_uring_types.h

@@ -404,7 +404,7 @@ struct io_ring_ctx {
 	spinlock_t		napi_lock;	/* napi_list lock */
 
 	/* napi busy poll default timeout */
-	unsigned int		napi_busy_poll_to;
+	ktime_t			napi_busy_poll_dt;
 	bool			napi_prefer_busy_poll;
 	bool			napi_enabled;
 
@@ -461,7 +461,6 @@ enum {
 	REQ_F_SUPPORT_NOWAIT_BIT,
 	REQ_F_ISREG_BIT,
 	REQ_F_POLL_NO_LAZY_BIT,
-	REQ_F_CANCEL_SEQ_BIT,
 	REQ_F_CAN_POLL_BIT,
 	REQ_F_BL_EMPTY_BIT,
 	REQ_F_BL_NO_RECYCLE_BIT,
@@ -536,8 +535,6 @@ enum {
 	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
 	/* don't use lazy poll wake for this request */
 	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
-	/* cancel sequence is set and valid */
-	REQ_F_CANCEL_SEQ	= IO_REQ_FLAG(REQ_F_CANCEL_SEQ_BIT),
 	/* file is pollable */
 	REQ_F_CAN_POLL		= IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
 	/* buffer list was empty after selection of buffer */

io_uring/io_uring.c

@@ -1849,7 +1849,7 @@ void io_wq_submit_work(struct io_wq_work *work)
 	} while (1);
 
 	/* avoid locking problems by failing it from a clean context */
-	if (ret < 0)
+	if (ret)
 		io_req_task_queue_fail(req, ret);
 }
 
@@ -2416,12 +2416,14 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 
 	if (uts) {
 		struct timespec64 ts;
+		ktime_t dt;
 
 		if (get_timespec64(&ts, uts))
 			return -EFAULT;
 
-		iowq.timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns());
-		io_napi_adjust_timeout(ctx, &iowq, &ts);
+		dt = timespec64_to_ktime(ts);
+		iowq.timeout = ktime_add(dt, ktime_get());
+		io_napi_adjust_timeout(ctx, &iowq, dt);
 	}
 
 	if (sig) {
@@ -3031,8 +3033,11 @@ __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd)
 		bool loop = false;
 
 		io_uring_drop_tctx_refs(current);
+		if (!tctx_inflight(tctx, !cancel_all))
+			break;
+
 		/* read completions before cancelations */
-		inflight = tctx_inflight(tctx, !cancel_all);
+		inflight = tctx_inflight(tctx, false);
 		if (!inflight)
 			break;
 

io_uring/io_uring.h

@@ -43,7 +43,7 @@ struct io_wait_queue {
 	ktime_t timeout;
 
 #ifdef CONFIG_NET_RX_BUSY_POLL
-	unsigned int napi_busy_poll_to;
+	ktime_t napi_busy_poll_dt;
 	bool napi_prefer_busy_poll;
 #endif
 };

io_uring/msg_ring.c

@@ -110,10 +110,10 @@ static struct io_kiocb *io_msg_get_kiocb(struct io_ring_ctx *ctx)
 	if (spin_trylock(&ctx->msg_lock)) {
 		req = io_alloc_cache_get(&ctx->msg_cache);
 		spin_unlock(&ctx->msg_lock);
-	}
 		if (req)
 			return req;
-	return kmem_cache_alloc(req_cachep, GFP_KERNEL | __GFP_NOWARN);
+	}
+	return kmem_cache_alloc(req_cachep, GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
 }
 
 static int io_msg_data_remote(struct io_kiocb *req)

io_uring/napi.c

@@ -33,6 +33,12 @@ static struct io_napi_entry *io_napi_hash_find(struct hlist_head *hash_list,
 	return NULL;
 }
 
+static inline ktime_t net_to_ktime(unsigned long t)
+{
+	/* napi approximating usecs, reverse busy_loop_current_time */
+	return ns_to_ktime(t << 10);
+}
+
 void __io_napi_add(struct io_ring_ctx *ctx, struct socket *sock)
 {
 	struct hlist_head *hash_list;
@@ -102,14 +108,14 @@ static inline void io_napi_remove_stale(struct io_ring_ctx *ctx, bool is_stale)
 		__io_napi_remove_stale(ctx);
 }
 
-static inline bool io_napi_busy_loop_timeout(unsigned long start_time,
-					     unsigned long bp_usec)
+static inline bool io_napi_busy_loop_timeout(ktime_t start_time,
+					     ktime_t bp)
 {
-	if (bp_usec) {
-		unsigned long end_time = start_time + bp_usec;
-		unsigned long now = busy_loop_current_time();
+	if (bp) {
+		ktime_t end_time = ktime_add(start_time, bp);
+		ktime_t now = net_to_ktime(busy_loop_current_time());
 
-		return time_after(now, end_time);
+		return ktime_after(now, end_time);
 	}
 
 	return true;
@@ -124,7 +130,8 @@ static bool io_napi_busy_loop_should_end(void *data,
 		return true;
 	if (io_should_wake(iowq) || io_has_work(iowq->ctx))
 		return true;
-	if (io_napi_busy_loop_timeout(start_time, iowq->napi_busy_poll_to))
+	if (io_napi_busy_loop_timeout(net_to_ktime(start_time),
+				      iowq->napi_busy_poll_dt))
 		return true;
 
 	return false;
@@ -181,10 +188,12 @@ static void io_napi_blocking_busy_loop(struct io_ring_ctx *ctx,
  */
 void io_napi_init(struct io_ring_ctx *ctx)
 {
+	u64 sys_dt = READ_ONCE(sysctl_net_busy_poll) * NSEC_PER_USEC;
+
 	INIT_LIST_HEAD(&ctx->napi_list);
 	spin_lock_init(&ctx->napi_lock);
 	ctx->napi_prefer_busy_poll = false;
-	ctx->napi_busy_poll_to = READ_ONCE(sysctl_net_busy_poll);
+	ctx->napi_busy_poll_dt = ns_to_ktime(sys_dt);
 }
 
 /*
@@ -217,11 +226,13 @@ void io_napi_free(struct io_ring_ctx *ctx)
 int io_register_napi(struct io_ring_ctx *ctx, void __user *arg)
 {
 	const struct io_uring_napi curr = {
-		.busy_poll_to 	  = ctx->napi_busy_poll_to,
+		.busy_poll_to 	  = ktime_to_us(ctx->napi_busy_poll_dt),
 		.prefer_busy_poll = ctx->napi_prefer_busy_poll
 	};
 	struct io_uring_napi napi;
 
+	if (ctx->flags & IORING_SETUP_IOPOLL)
+		return -EINVAL;
 	if (copy_from_user(&napi, arg, sizeof(napi)))
 		return -EFAULT;
 	if (napi.pad[0] || napi.pad[1] || napi.pad[2] || napi.resv)
@@ -230,7 +241,7 @@ int io_register_napi(struct io_ring_ctx *ctx, void __user *arg)
 	if (copy_to_user(arg, &curr, sizeof(curr)))
 		return -EFAULT;
 
-	WRITE_ONCE(ctx->napi_busy_poll_to, napi.busy_poll_to);
+	WRITE_ONCE(ctx->napi_busy_poll_dt, napi.busy_poll_to * NSEC_PER_USEC);
 	WRITE_ONCE(ctx->napi_prefer_busy_poll, !!napi.prefer_busy_poll);
 	WRITE_ONCE(ctx->napi_enabled, true);
 	return 0;
@@ -247,14 +258,14 @@ int io_register_napi(struct io_ring_ctx *ctx, void __user *arg)
 int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg)
 {
 	const struct io_uring_napi curr = {
-		.busy_poll_to 	  = ctx->napi_busy_poll_to,
+		.busy_poll_to 	  = ktime_to_us(ctx->napi_busy_poll_dt),
 		.prefer_busy_poll = ctx->napi_prefer_busy_poll
 	};
 
 	if (arg && copy_to_user(arg, &curr, sizeof(curr)))
 		return -EFAULT;
 
-	WRITE_ONCE(ctx->napi_busy_poll_to, 0);
+	WRITE_ONCE(ctx->napi_busy_poll_dt, 0);
 	WRITE_ONCE(ctx->napi_prefer_busy_poll, false);
 	WRITE_ONCE(ctx->napi_enabled, false);
 	return 0;
@@ -271,25 +282,14 @@ int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg)
  * the NAPI timeout accordingly.
  */
 void __io_napi_adjust_timeout(struct io_ring_ctx *ctx, struct io_wait_queue *iowq,
-			      struct timespec64 *ts)
+			      ktime_t to_wait)
 {
-	unsigned int poll_to = READ_ONCE(ctx->napi_busy_poll_to);
-
-	if (ts) {
-		struct timespec64 poll_to_ts;
-
-		poll_to_ts = ns_to_timespec64(1000 * (s64)poll_to);
-		if (timespec64_compare(ts, &poll_to_ts) < 0) {
-			s64 poll_to_ns = timespec64_to_ns(ts);
-			if (poll_to_ns > 0) {
-				u64 val = poll_to_ns + 999;
-				do_div(val, 1000);
-				poll_to = val;
-			}
-		}
-	}
+	ktime_t poll_dt = READ_ONCE(ctx->napi_busy_poll_dt);
+
+	if (to_wait)
+		poll_dt = min(poll_dt, to_wait);
 
-	iowq->napi_busy_poll_to = poll_to;
+	iowq->napi_busy_poll_dt = poll_dt;
 }
 
 /*
@@ -318,7 +318,7 @@ int io_napi_sqpoll_busy_poll(struct io_ring_ctx *ctx)
 	LIST_HEAD(napi_list);
 	bool is_stale = false;
 
-	if (!READ_ONCE(ctx->napi_busy_poll_to))
+	if (!READ_ONCE(ctx->napi_busy_poll_dt))
 		return 0;
 	if (list_empty_careful(&ctx->napi_list))
 		return 0;

io_uring/napi.h

@@ -18,7 +18,7 @@ int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg);
 void __io_napi_add(struct io_ring_ctx *ctx, struct socket *sock);
 
 void __io_napi_adjust_timeout(struct io_ring_ctx *ctx,
-		struct io_wait_queue *iowq, struct timespec64 *ts);
+		struct io_wait_queue *iowq, ktime_t to_wait);
 void __io_napi_busy_loop(struct io_ring_ctx *ctx, struct io_wait_queue *iowq);
 int io_napi_sqpoll_busy_poll(struct io_ring_ctx *ctx);
 
@@ -29,11 +29,11 @@ static inline bool io_napi(struct io_ring_ctx *ctx)
 
 static inline void io_napi_adjust_timeout(struct io_ring_ctx *ctx,
 					  struct io_wait_queue *iowq,
-					  struct timespec64 *ts)
+					  ktime_t to_wait)
 {
 	if (!io_napi(ctx))
 		return;
-	__io_napi_adjust_timeout(ctx, iowq, ts);
+	__io_napi_adjust_timeout(ctx, iowq, to_wait);
 }
 
 static inline void io_napi_busy_loop(struct io_ring_ctx *ctx,
@@ -55,7 +55,7 @@ static inline void io_napi_add(struct io_kiocb *req)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct socket *sock;
 
-	if (!READ_ONCE(ctx->napi_busy_poll_to))
+	if (!READ_ONCE(ctx->napi_busy_poll_dt))
 		return;
 
 	sock = sock_from_file(req->file);
@@ -88,7 +88,7 @@ static inline void io_napi_add(struct io_kiocb *req)
 }
 static inline void io_napi_adjust_timeout(struct io_ring_ctx *ctx,
 					  struct io_wait_queue *iowq,
-					  struct timespec64 *ts)
+					  ktime_t to_wait)
 {
 }
 static inline void io_napi_busy_loop(struct io_ring_ctx *ctx,

io_uring/timeout.c

@@ -639,7 +639,7 @@ void io_queue_linked_timeout(struct io_kiocb *req)
 
 static bool io_match_task(struct io_kiocb *head, struct task_struct *task,
 			  bool cancel_all)
-	__must_hold(&req->ctx->timeout_lock)
+	__must_hold(&head->ctx->timeout_lock)
 {
 	struct io_kiocb *req;
 

io_uring/uring_cmd.c

@@ -265,7 +265,7 @@ int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
 		req_set_fail(req);
 	io_req_uring_cleanup(req, issue_flags);
 	io_req_set_res(req, ret, 0);
-	return ret < 0 ? ret : IOU_OK;
+	return IOU_OK;
 }
 
 int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,