[PATCHSET v6] Add io_uring futex/futexv support

[PATCHSET v6] Add io_uring futex/futexv support

[Jens Axboe]

Hi,

This patchset adds support for first futex wake and wait, and then
futexv.

For both wait/wake/waitv, we support the bitset variant, as the
"normal" variants can be easily implemented on top of that.

PI and requeue are not supported through io_uring, just the above
mentioned parts. This may change in the future, but in the spirit
of keeping this small (and based on what people have been asking for),
this is what we currently have.

When I did these patches, I forgot that Pavel had previously posted a
futex variant for io_uring. The major thing that had been holding me
back from people asking about futexes and io_uring, is that I wanted
to do this what I consider the right way - no usage of io-wq or thread
offload, an actually async implementation that is efficient to use
and don't rely on a blocking thread for futex wait/waitv. This is what
this patchset attempts to do, while being minimally invasive on the
futex side. I believe the diffstat reflects that.

As far as I can recall, the first request for futex support with
io_uring came from Andres Freund, working on postgres. His aio rework
of postgres was one of the early adopters of io_uring, and futex
support was a natural extension for that. This is relevant from both
a usability point of view, as well as for effiency and performance.
In Andres's words, for the former:

"Futex wait support in io_uring makes it a lot easier to avoid deadlocks
in concurrent programs that have their own buffer pool: Obviously pages in
the application buffer pool have to be locked during IO. If the initiator
of IO A needs to wait for a held lock B, the holder of lock B might wait
for the IO A to complete.  The ability to wait for a lock and IO
completions at the same time provides an efficient way to avoid such
deadlocks."

and in terms of effiency, even without unlocking the full potential yet,
Andres says:

"Futex wake support in io_uring is useful because it allows for more
efficient directed wakeups.  For some "locks" postgres has queues
implemented in userspace, with wakeup logic that cannot easily be
implemented with FUTEX_WAKE_BITSET on a single "futex word" (imagine
waiting for journal flushes to have completed up to a certain point). Thus
a "lock release" sometimes need to wake up many processes in a row.  A
quick-and-dirty conversion to doing these wakeups via io_uring lead to a
3% throughput increase, with 12% fewer context switches, albeit in a
fairly extreme workload."

Some basic io_uring futex support and test cases are available in the
liburing 'futex' branch:

https://git.kernel.dk/cgit/liburing/log/?h=futex

testing all of the variants. I originally wrote this code about a
month ago and Andres has been using it with postgres, and I'm not
aware of any bugs in it. That's not to say it's perfect, obviously,
and I welcome some feedback so we can move this forward and hash out
any potential issues.

In terms of testing, there's a functionality and beat-up test case
in liburing, and I've run all the ltp futex test cases as well to
ensure we didn't inadvertently break anything. It's also been in
linux-next for a long time and haven't heard any complaints.

 include/linux/io_uring_types.h |   5 +
 include/uapi/linux/io_uring.h  |   4 +
 io_uring/Makefile              |   1 +
 io_uring/cancel.c              |   5 +
 io_uring/cancel.h              |   4 +
 io_uring/futex.c               | 386 +++++++++++++++++++++++++++++++++
 io_uring/futex.h               |  36 +++
 io_uring/io_uring.c            |   7 +
 io_uring/opdef.c               |  34 +++
 kernel/futex/futex.h           |  20 ++
 kernel/futex/requeue.c         |   3 +-
 kernel/futex/syscalls.c        |  18 +-
 kernel/futex/waitwake.c        |  49 +++--
 13 files changed, 545 insertions(+), 27 deletions(-)

You can also find the code here:

https://git.kernel.dk/cgit/linux/log/?h=io_uring-futex

V6:
- Expand the two main commit messages describing command layout
- Fix double conversion of FUTEX2_* flags (Peter)
- Use FLAGS_STRICT for IORING_OP_FUTEX_WAKE, so we return 0 wakes
  when the caller asked for 0 (Peter)
- Fix issue with IORING_OP_FUTEX_WAITV and futex_wait_multiple_setup()
  doing futex_unqueue_multiple() if we had a wakeup while setting up.
- Cleanup IORING_OP_FUTEX_WAITV issue path
- Don't use sqe->futex_flags for the FUTEX2_* flags, reserve it for
  future internal use.
- Add more liburing test cases.
- Rebase on current tree (for-6.7/io_uring + tip locking/core)

-- 
Jens Axboe

[PATCH 1/8] futex: move FUTEX2_VALID_MASK to futex.h

[Jens Axboe]

We need this for validating the futex2 flags outside of the normal
futex syscalls.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    | 2 ++
 kernel/futex/syscalls.c | 2 --
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index a06030a1a27b..a173a9d501e1 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -52,6 +52,8 @@ static inline unsigned int futex_to_flags(unsigned int op)
 	return flags;
 }
 
+#define FUTEX2_VALID_MASK (FUTEX2_SIZE_MASK | FUTEX2_PRIVATE)
+
 /* FUTEX2_ to FLAGS_ */
 static inline unsigned int futex2_to_flags(unsigned int flags2)
 {
diff --git a/kernel/futex/syscalls.c b/kernel/futex/syscalls.c
index 8200d86d30e1..2b5cafdfdc50 100644
--- a/kernel/futex/syscalls.c
+++ b/kernel/futex/syscalls.c
@@ -179,8 +179,6 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
 	return do_futex(uaddr, op, val, tp, uaddr2, (unsigned long)utime, val3);
 }
 
-#define FUTEX2_VALID_MASK (FUTEX2_SIZE_MASK | FUTEX2_PRIVATE)
-
 /**
  * futex_parse_waitv - Parse a waitv array from userspace
  * @futexv:	Kernel side list of waiters to be filled
-- 
2.40.1

[PATCH 2/8] futex: factor out the futex wake handling

[Jens Axboe]

In preparation for having another waker that isn't futex_wake_mark(),
add a wake handler in futex_q. No extra data is associated with the
handler outside of struct futex_q itself. futex_wake_mark() is defined as
the standard wakeup helper, now set through futex_q_init like other
defaults.

Normal sync futex waiting relies on wake_q holding tasks that should
be woken up. This is what futex_wake_mark() does, it'll unqueue the
futex and add the associated task to the wake queue. For async usage of
futex waiting, rather than having tasks sleeping on the futex, we'll
need to deal with a futex wake differently. For the planned io_uring
case, that means posting a completion event for the task in question.
Having a definable wake handler can help support that use case.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    | 5 +++++
 kernel/futex/requeue.c  | 3 ++-
 kernel/futex/waitwake.c | 6 +++---
 3 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index a173a9d501e1..547f509b2c87 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -139,11 +139,15 @@ struct futex_pi_state {
 	union futex_key key;
 } __randomize_layout;
 
+struct futex_q;
+typedef void (futex_wake_fn)(struct wake_q_head *wake_q, struct futex_q *q);
+
 /**
  * struct futex_q - The hashed futex queue entry, one per waiting task
  * @list:		priority-sorted list of tasks waiting on this futex
  * @task:		the task waiting on the futex
  * @lock_ptr:		the hash bucket lock
+ * @wake:		the wake handler for this queue
  * @key:		the key the futex is hashed on
  * @pi_state:		optional priority inheritance state
  * @rt_waiter:		rt_waiter storage for use with requeue_pi
@@ -168,6 +172,7 @@ struct futex_q {
 
 	struct task_struct *task;
 	spinlock_t *lock_ptr;
+	futex_wake_fn *wake;
 	union futex_key key;
 	struct futex_pi_state *pi_state;
 	struct rt_mutex_waiter *rt_waiter;
diff --git a/kernel/futex/requeue.c b/kernel/futex/requeue.c
index a0a79954f506..9dc789399a1a 100644
--- a/kernel/futex/requeue.c
+++ b/kernel/futex/requeue.c
@@ -58,6 +58,7 @@ enum {
 
 const struct futex_q futex_q_init = {
 	/* list gets initialized in futex_queue()*/
+	.wake		= futex_wake_mark,
 	.key		= FUTEX_KEY_INIT,
 	.bitset		= FUTEX_BITSET_MATCH_ANY,
 	.requeue_state	= ATOMIC_INIT(Q_REQUEUE_PI_NONE),
@@ -593,7 +594,7 @@ int futex_requeue(u32 __user *uaddr1, unsigned int flags1,
 		/* Plain futexes just wake or requeue and are done */
 		if (!requeue_pi) {
 			if (++task_count <= nr_wake)
-				futex_wake_mark(&wake_q, this);
+				this->wake(&wake_q, this);
 			else
 				requeue_futex(this, hb1, hb2, &key2);
 			continue;
diff --git a/kernel/futex/waitwake.c b/kernel/futex/waitwake.c
index 37860f794bf7..35c6a637a4bb 100644
--- a/kernel/futex/waitwake.c
+++ b/kernel/futex/waitwake.c
@@ -177,7 +177,7 @@ int futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
 			if (!(this->bitset & bitset))
 				continue;
 
-			futex_wake_mark(&wake_q, this);
+			this->wake(&wake_q, this);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -292,7 +292,7 @@ int futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 				ret = -EINVAL;
 				goto out_unlock;
 			}
-			futex_wake_mark(&wake_q, this);
+			this->wake(&wake_q, this);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -306,7 +306,7 @@ int futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 					ret = -EINVAL;
 					goto out_unlock;
 				}
-				futex_wake_mark(&wake_q, this);
+				this->wake(&wake_q, this);
 				if (++op_ret >= nr_wake2)
 					break;
 			}
-- 
2.40.1

[PATCH 3/8] futex: abstract out a __futex_wake_mark() helper

[Jens Axboe]

Move the unqueue and lock_ptr clear into a helper that futex_wake_mark()
calls. Add it to the public functions as well, in preparation for using
it outside the core futex code.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    |  1 +
 kernel/futex/waitwake.c | 33 ++++++++++++++++++++++-----------
 2 files changed, 23 insertions(+), 11 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index 547f509b2c87..33835b81e0c3 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -219,6 +219,7 @@ extern int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
 			    struct futex_q *q, struct futex_hash_bucket **hb);
 extern void futex_wait_queue(struct futex_hash_bucket *hb, struct futex_q *q,
 				   struct hrtimer_sleeper *timeout);
+extern bool __futex_wake_mark(struct futex_q *q);
 extern void futex_wake_mark(struct wake_q_head *wake_q, struct futex_q *q);
 
 extern int fault_in_user_writeable(u32 __user *uaddr);
diff --git a/kernel/futex/waitwake.c b/kernel/futex/waitwake.c
index 35c6a637a4bb..6fcf5f723719 100644
--- a/kernel/futex/waitwake.c
+++ b/kernel/futex/waitwake.c
@@ -106,20 +106,11 @@
  * double_lock_hb() and double_unlock_hb(), respectively.
  */
 
-/*
- * The hash bucket lock must be held when this is called.
- * Afterwards, the futex_q must not be accessed. Callers
- * must ensure to later call wake_up_q() for the actual
- * wakeups to occur.
- */
-void futex_wake_mark(struct wake_q_head *wake_q, struct futex_q *q)
+bool __futex_wake_mark(struct futex_q *q)
 {
-	struct task_struct *p = q->task;
-
 	if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
-		return;
+		return false;
 
-	get_task_struct(p);
 	__futex_unqueue(q);
 	/*
 	 * The waiting task can free the futex_q as soon as q->lock_ptr = NULL
@@ -130,6 +121,26 @@ void futex_wake_mark(struct wake_q_head *wake_q, struct futex_q *q)
 	 */
 	smp_store_release(&q->lock_ptr, NULL);
 
+	return true;
+}
+
+/*
+ * The hash bucket lock must be held when this is called.
+ * Afterwards, the futex_q must not be accessed. Callers
+ * must ensure to later call wake_up_q() for the actual
+ * wakeups to occur.
+ */
+void futex_wake_mark(struct wake_q_head *wake_q, struct futex_q *q)
+{
+	struct task_struct *p = q->task;
+
+	get_task_struct(p);
+
+	if (!__futex_wake_mark(q)) {
+		put_task_struct(p);
+		return;
+	}
+
 	/*
 	 * Queue the task for later wakeup for after we've released
 	 * the hb->lock.
-- 
2.40.1

[PATCH 4/8] io_uring: add support for futex wake and wait

[Jens Axboe]

Add support for FUTEX_WAKE/WAIT primitives.

IORING_OP_FUTEX_WAKE is mix of FUTEX_WAKE and FUTEX_WAKE_BITSET, as
it does support passing in a bitset.

Similary, IORING_OP_FUTEX_WAIT is a mix of FUTEX_WAIT and
FUTEX_WAIT_BITSET.

For both of them, they are using the futex2 interface.

FUTEX_WAKE is straight forward, as those can always be done directly from
the io_uring submission without needing async handling. For FUTEX_WAIT,
things are a bit more complicated. If the futex isn't ready, then we
rely on a callback via futex_queue->wake() when someone wakes up the
futex. From that calback, we queue up task_work with the original task,
which will post a CQE and wake it, if necessary.

Cancelations are supported, both from the application point-of-view,
but also to be able to cancel pending waits if the ring exits before
all events have occurred. The return value of futex_unqueue() is used
to gate who wins the potential race between cancelation and futex
wakeups. Whomever gets a 'ret == 1' return from that claims ownership
of the io_uring futex request.

This is just the barebones wait/wake support. PI or REQUEUE support is
not added at this point, unclear if we might look into that later.

Likewise, explicit timeouts are not supported either. It is expected
that users that need timeouts would do so via the usual io_uring
mechanism to do that using linked timeouts.

The SQE format is as follows:

`addr`		Address of futex
`fd`		futex2(2) FUTEX2_* flags
`futex_flags`	io_uring specific command flags. None valid now.
`addr2`		Value of futex
`addr3`		Mask to wake/wait

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/io_uring_types.h |   5 +
 include/uapi/linux/io_uring.h  |   3 +
 io_uring/Makefile              |   1 +
 io_uring/cancel.c              |   5 +
 io_uring/cancel.h              |   4 +
 io_uring/futex.c               | 235 +++++++++++++++++++++++++++++++++
 io_uring/futex.h               |  34 +++++
 io_uring/io_uring.c            |   7 +
 io_uring/opdef.c               |  23 ++++
 9 files changed, 317 insertions(+)
 create mode 100644 io_uring/futex.c
 create mode 100644 io_uring/futex.h

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index e178461fa513..990984614fca 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -321,6 +321,11 @@ struct io_ring_ctx {
 
 	struct hlist_head	waitid_list;
 
+#ifdef CONFIG_FUTEX
+	struct hlist_head	futex_list;
+	struct io_alloc_cache	futex_cache;
+#endif
+
 	const struct cred	*sq_creds;	/* cred used for __io_sq_thread() */
 	struct io_sq_data	*sq_data;	/* if using sq thread polling */
 
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 425f64eee44e..04f9fba38d4b 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -66,6 +66,7 @@ struct io_uring_sqe {
 		__u32		msg_ring_flags;
 		__u32		uring_cmd_flags;
 		__u32		waitid_flags;
+		__u32		futex_flags;
 	};
 	__u64	user_data;	/* data to be passed back at completion time */
 	/* pack this to avoid bogus arm OABI complaints */
@@ -243,6 +244,8 @@ enum io_uring_op {
 	IORING_OP_SENDMSG_ZC,
 	IORING_OP_READ_MULTISHOT,
 	IORING_OP_WAITID,
+	IORING_OP_FUTEX_WAIT,
+	IORING_OP_FUTEX_WAKE,
 
 	/* this goes last, obviously */
 	IORING_OP_LAST,
diff --git a/io_uring/Makefile b/io_uring/Makefile
index 7bd64e442567..e5be47e4fc3b 100644
--- a/io_uring/Makefile
+++ b/io_uring/Makefile
@@ -10,3 +10,4 @@ obj-$(CONFIG_IO_URING)		+= io_uring.o xattr.o nop.o fs.o splice.o \
 					cancel.o kbuf.o rsrc.o rw.o opdef.o \
 					notif.o waitid.o
 obj-$(CONFIG_IO_WQ)		+= io-wq.o
+obj-$(CONFIG_FUTEX)		+= futex.o
diff --git a/io_uring/cancel.c b/io_uring/cancel.c
index eb77a51c5a79..3c19cccb1aec 100644
--- a/io_uring/cancel.c
+++ b/io_uring/cancel.c
@@ -16,6 +16,7 @@
 #include "poll.h"
 #include "timeout.h"
 #include "waitid.h"
+#include "futex.h"
 #include "cancel.h"
 
 struct io_cancel {
@@ -124,6 +125,10 @@ int io_try_cancel(struct io_uring_task *tctx, struct io_cancel_data *cd,
 	if (ret != -ENOENT)
 		return ret;
 
+	ret = io_futex_cancel(ctx, cd, issue_flags);
+	if (ret != -ENOENT)
+		return ret;
+
 	spin_lock(&ctx->completion_lock);
 	if (!(cd->flags & IORING_ASYNC_CANCEL_FD))
 		ret = io_timeout_cancel(ctx, cd);
diff --git a/io_uring/cancel.h b/io_uring/cancel.h
index fc98622e6166..c0a8e7c520b6 100644
--- a/io_uring/cancel.h
+++ b/io_uring/cancel.h
@@ -1,4 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
+#ifndef IORING_CANCEL_H
+#define IORING_CANCEL_H
 
 #include <linux/io_uring_types.h>
 
@@ -22,3 +24,5 @@ void init_hash_table(struct io_hash_table *table, unsigned size);
 
 int io_sync_cancel(struct io_ring_ctx *ctx, void __user *arg);
 bool io_cancel_req_match(struct io_kiocb *req, struct io_cancel_data *cd);
+
+#endif
diff --git a/io_uring/futex.c b/io_uring/futex.c
new file mode 100644
index 000000000000..eb4406ac46fb
--- /dev/null
+++ b/io_uring/futex.c
@@ -0,0 +1,235 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/fs.h>
+#include <linux/file.h>
+#include <linux/io_uring.h>
+
+#include <uapi/linux/io_uring.h>
+
+#include "../kernel/futex/futex.h"
+#include "io_uring.h"
+#include "rsrc.h"
+#include "futex.h"
+
+struct io_futex {
+	struct file	*file;
+	u32 __user	*uaddr;
+	unsigned long	futex_val;
+	unsigned long	futex_mask;
+	u32		futex_flags;
+};
+
+struct io_futex_data {
+	union {
+		struct futex_q		q;
+		struct io_cache_entry	cache;
+	};
+	struct io_kiocb	*req;
+};
+
+void io_futex_cache_init(struct io_ring_ctx *ctx)
+{
+	io_alloc_cache_init(&ctx->futex_cache, IO_NODE_ALLOC_CACHE_MAX,
+				sizeof(struct io_futex_data));
+}
+
+static void io_futex_cache_entry_free(struct io_cache_entry *entry)
+{
+	kfree(container_of(entry, struct io_futex_data, cache));
+}
+
+void io_futex_cache_free(struct io_ring_ctx *ctx)
+{
+	io_alloc_cache_free(&ctx->futex_cache, io_futex_cache_entry_free);
+}
+
+static void io_futex_complete(struct io_kiocb *req, struct io_tw_state *ts)
+{
+	struct io_futex_data *ifd = req->async_data;
+	struct io_ring_ctx *ctx = req->ctx;
+
+	io_tw_lock(ctx, ts);
+	if (!io_alloc_cache_put(&ctx->futex_cache, &ifd->cache))
+		kfree(ifd);
+	req->async_data = NULL;
+	hlist_del_init(&req->hash_node);
+	io_req_task_complete(req, ts);
+}
+
+static bool __io_futex_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req)
+{
+	struct io_futex_data *ifd = req->async_data;
+
+	/* futex wake already done or in progress */
+	if (!futex_unqueue(&ifd->q))
+		return false;
+
+	hlist_del_init(&req->hash_node);
+	io_req_set_res(req, -ECANCELED, 0);
+	req->io_task_work.func = io_futex_complete;
+	io_req_task_work_add(req);
+	return true;
+}
+
+int io_futex_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd,
+		    unsigned int issue_flags)
+{
+	struct hlist_node *tmp;
+	struct io_kiocb *req;
+	int nr = 0;
+
+	if (cd->flags & (IORING_ASYNC_CANCEL_FD|IORING_ASYNC_CANCEL_FD_FIXED))
+		return -ENOENT;
+
+	io_ring_submit_lock(ctx, issue_flags);
+	hlist_for_each_entry_safe(req, tmp, &ctx->futex_list, hash_node) {
+		if (req->cqe.user_data != cd->data &&
+		    !(cd->flags & IORING_ASYNC_CANCEL_ANY))
+			continue;
+		if (__io_futex_cancel(ctx, req))
+			nr++;
+		if (!(cd->flags & IORING_ASYNC_CANCEL_ALL))
+			break;
+	}
+	io_ring_submit_unlock(ctx, issue_flags);
+
+	if (nr)
+		return nr;
+
+	return -ENOENT;
+}
+
+bool io_futex_remove_all(struct io_ring_ctx *ctx, struct task_struct *task,
+			 bool cancel_all)
+{
+	struct hlist_node *tmp;
+	struct io_kiocb *req;
+	bool found = false;
+
+	lockdep_assert_held(&ctx->uring_lock);
+
+	hlist_for_each_entry_safe(req, tmp, &ctx->futex_list, hash_node) {
+		if (!io_match_task_safe(req, task, cancel_all))
+			continue;
+		__io_futex_cancel(ctx, req);
+		found = true;
+	}
+
+	return found;
+}
+
+int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+	u32 flags;
+
+	if (unlikely(sqe->len || sqe->futex_flags || sqe->buf_index ||
+		     sqe->file_index))
+		return -EINVAL;
+
+	iof->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	iof->futex_val = READ_ONCE(sqe->addr2);
+	iof->futex_mask = READ_ONCE(sqe->addr3);
+	flags = READ_ONCE(sqe->fd);
+
+	if (flags & ~FUTEX2_VALID_MASK)
+		return -EINVAL;
+
+	iof->futex_flags = futex2_to_flags(flags);
+	if (!futex_flags_valid(iof->futex_flags))
+		return -EINVAL;
+
+	if (!futex_validate_input(iof->futex_flags, iof->futex_val) ||
+	    !futex_validate_input(iof->futex_flags, iof->futex_mask))
+		return -EINVAL;
+
+	return 0;
+}
+
+static void io_futex_wake_fn(struct wake_q_head *wake_q, struct futex_q *q)
+{
+	struct io_futex_data *ifd = container_of(q, struct io_futex_data, q);
+	struct io_kiocb *req = ifd->req;
+
+	if (unlikely(!__futex_wake_mark(q)))
+		return;
+
+	io_req_set_res(req, 0, 0);
+	req->io_task_work.func = io_futex_complete;
+	io_req_task_work_add(req);
+}
+
+static struct io_futex_data *io_alloc_ifd(struct io_ring_ctx *ctx)
+{
+	struct io_cache_entry *entry;
+
+	entry = io_alloc_cache_get(&ctx->futex_cache);
+	if (entry)
+		return container_of(entry, struct io_futex_data, cache);
+
+	return kmalloc(sizeof(struct io_futex_data), GFP_NOWAIT);
+}
+
+int io_futex_wait(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+	struct io_ring_ctx *ctx = req->ctx;
+	struct io_futex_data *ifd = NULL;
+	struct futex_hash_bucket *hb;
+	int ret;
+
+	if (!iof->futex_mask) {
+		ret = -EINVAL;
+		goto done;
+	}
+
+	io_ring_submit_lock(ctx, issue_flags);
+	ifd = io_alloc_ifd(ctx);
+	if (!ifd) {
+		ret = -ENOMEM;
+		goto done_unlock;
+	}
+
+	req->async_data = ifd;
+	ifd->q = futex_q_init;
+	ifd->q.bitset = iof->futex_mask;
+	ifd->q.wake = io_futex_wake_fn;
+	ifd->req = req;
+
+	ret = futex_wait_setup(iof->uaddr, iof->futex_val, iof->futex_flags,
+			       &ifd->q, &hb);
+	if (!ret) {
+		hlist_add_head(&req->hash_node, &ctx->futex_list);
+		io_ring_submit_unlock(ctx, issue_flags);
+
+		futex_queue(&ifd->q, hb);
+		return IOU_ISSUE_SKIP_COMPLETE;
+	}
+
+done_unlock:
+	io_ring_submit_unlock(ctx, issue_flags);
+done:
+	if (ret < 0)
+		req_set_fail(req);
+	io_req_set_res(req, ret, 0);
+	kfree(ifd);
+	return IOU_OK;
+}
+
+int io_futex_wake(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+	int ret;
+
+	/*
+	 * Strict flags - ensure that waking 0 futexes yields a 0 result.
+	 * See commit 43adf8449510 ("futex: FLAGS_STRICT") for details.
+	 */
+	ret = futex_wake(iof->uaddr, FLAGS_STRICT | iof->futex_flags,
+			 iof->futex_val, iof->futex_mask);
+	if (ret < 0)
+		req_set_fail(req);
+	io_req_set_res(req, ret, 0);
+	return IOU_OK;
+}
diff --git a/io_uring/futex.h b/io_uring/futex.h
new file mode 100644
index 000000000000..ddc9e0d73c52
--- /dev/null
+++ b/io_uring/futex.h
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "cancel.h"
+
+int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_futex_wait(struct io_kiocb *req, unsigned int issue_flags);
+int io_futex_wake(struct io_kiocb *req, unsigned int issue_flags);
+
+#if defined(CONFIG_FUTEX)
+int io_futex_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd,
+		    unsigned int issue_flags);
+bool io_futex_remove_all(struct io_ring_ctx *ctx, struct task_struct *task,
+			 bool cancel_all);
+void io_futex_cache_init(struct io_ring_ctx *ctx);
+void io_futex_cache_free(struct io_ring_ctx *ctx);
+#else
+static inline int io_futex_cancel(struct io_ring_ctx *ctx,
+				  struct io_cancel_data *cd,
+				  unsigned int issue_flags)
+{
+	return 0;
+}
+static inline bool io_futex_remove_all(struct io_ring_ctx *ctx,
+				       struct task_struct *task, bool cancel_all)
+{
+	return false;
+}
+static inline void io_futex_cache_init(struct io_ring_ctx *ctx)
+{
+}
+static inline void io_futex_cache_free(struct io_ring_ctx *ctx)
+{
+}
+#endif
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 08c9ea46bb95..3c1c111d02cb 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -93,6 +93,7 @@
 #include "net.h"
 #include "notif.h"
 #include "waitid.h"
+#include "futex.h"
 
 #include "timeout.h"
 #include "poll.h"
@@ -330,6 +331,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 			    sizeof(struct async_poll));
 	io_alloc_cache_init(&ctx->netmsg_cache, IO_ALLOC_CACHE_MAX,
 			    sizeof(struct io_async_msghdr));
+	io_futex_cache_init(ctx);
 	init_completion(&ctx->ref_comp);
 	xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
 	mutex_init(&ctx->uring_lock);
@@ -350,6 +352,9 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 	ctx->submit_state.free_list.next = NULL;
 	INIT_WQ_LIST(&ctx->locked_free_list);
 	INIT_HLIST_HEAD(&ctx->waitid_list);
+#ifdef CONFIG_FUTEX
+	INIT_HLIST_HEAD(&ctx->futex_list);
+#endif
 	INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func);
 	INIT_WQ_LIST(&ctx->submit_state.compl_reqs);
 	INIT_HLIST_HEAD(&ctx->cancelable_uring_cmd);
@@ -2895,6 +2900,7 @@ static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	io_eventfd_unregister(ctx);
 	io_alloc_cache_free(&ctx->apoll_cache, io_apoll_cache_free);
 	io_alloc_cache_free(&ctx->netmsg_cache, io_netmsg_cache_free);
+	io_futex_cache_free(ctx);
 	io_destroy_buffers(ctx);
 	mutex_unlock(&ctx->uring_lock);
 	if (ctx->sq_creds)
@@ -3338,6 +3344,7 @@ static __cold bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
 	mutex_lock(&ctx->uring_lock);
 	ret |= io_poll_remove_all(ctx, task, cancel_all);
 	ret |= io_waitid_remove_all(ctx, task, cancel_all);
+	ret |= io_futex_remove_all(ctx, task, cancel_all);
 	ret |= io_uring_try_cancel_uring_cmd(ctx, task, cancel_all);
 	mutex_unlock(&ctx->uring_lock);
 	ret |= io_kill_timeouts(ctx, task, cancel_all);
diff --git a/io_uring/opdef.c b/io_uring/opdef.c
index aadcbf7136b0..31a3a421e94d 100644
--- a/io_uring/opdef.c
+++ b/io_uring/opdef.c
@@ -34,6 +34,7 @@
 #include "cancel.h"
 #include "rw.h"
 #include "waitid.h"
+#include "futex.h"
 
 static int io_no_issue(struct io_kiocb *req, unsigned int issue_flags)
 {
@@ -444,6 +445,22 @@ const struct io_issue_def io_issue_defs[] = {
 		.prep			= io_waitid_prep,
 		.issue			= io_waitid,
 	},
+	[IORING_OP_FUTEX_WAIT] = {
+#if defined(CONFIG_FUTEX)
+		.prep			= io_futex_prep,
+		.issue			= io_futex_wait,
+#else
+		.prep			= io_eopnotsupp_prep,
+#endif
+	},
+	[IORING_OP_FUTEX_WAKE] = {
+#if defined(CONFIG_FUTEX)
+		.prep			= io_futex_prep,
+		.issue			= io_futex_wake,
+#else
+		.prep			= io_eopnotsupp_prep,
+#endif
+	},
 };
 
 const struct io_cold_def io_cold_defs[] = {
@@ -670,6 +687,12 @@ const struct io_cold_def io_cold_defs[] = {
 		.name			= "WAITID",
 		.async_size		= sizeof(struct io_waitid_async),
 	},
+	[IORING_OP_FUTEX_WAIT] = {
+		.name			= "FUTEX_WAIT",
+	},
+	[IORING_OP_FUTEX_WAKE] = {
+		.name			= "FUTEX_WAKE",
+	},
 };
 
 const char *io_uring_get_opcode(u8 opcode)
-- 
2.40.1

[PATCH 5/8] futex: add wake_data to struct futex_q

[Jens Axboe]

With handling multiple futex_q for waitv, we cannot easily go from the
futex_q to data related to that request or queue. Add a wake_data
argument that belongs to the wake handler assigned.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index 33835b81e0c3..76f6c2e0f539 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -148,6 +148,7 @@ typedef void (futex_wake_fn)(struct wake_q_head *wake_q, struct futex_q *q);
  * @task:		the task waiting on the futex
  * @lock_ptr:		the hash bucket lock
  * @wake:		the wake handler for this queue
+ * @wake_data:		data associated with the wake handler
  * @key:		the key the futex is hashed on
  * @pi_state:		optional priority inheritance state
  * @rt_waiter:		rt_waiter storage for use with requeue_pi
@@ -173,6 +174,7 @@ struct futex_q {
 	struct task_struct *task;
 	spinlock_t *lock_ptr;
 	futex_wake_fn *wake;
+	void *wake_data;
 	union futex_key key;
 	struct futex_pi_state *pi_state;
 	struct rt_mutex_waiter *rt_waiter;
-- 
2.40.1

[PATCH 6/8] futex: make futex_parse_waitv() available as a helper

[Jens Axboe]

To make it more generically useful, augment it with allowing the caller
to pass in the wake handler and wake data. Convert the futex_waitv()
syscall, passing in the default handlers.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    |  5 +++++
 kernel/futex/syscalls.c | 16 +++++++++++-----
 2 files changed, 16 insertions(+), 5 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index 76f6c2e0f539..6b6a6b3da103 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -361,6 +361,11 @@ struct futex_vector {
 	struct futex_q q;
 };
 
+extern int futex_parse_waitv(struct futex_vector *futexv,
+			     struct futex_waitv __user *uwaitv,
+			     unsigned int nr_futexes, futex_wake_fn *wake,
+			     void *wake_data);
+
 extern int futex_wait_multiple(struct futex_vector *vs, unsigned int count,
 			       struct hrtimer_sleeper *to);
 
diff --git a/kernel/futex/syscalls.c b/kernel/futex/syscalls.c
index 2b5cafdfdc50..4b6da9116aa6 100644
--- a/kernel/futex/syscalls.c
+++ b/kernel/futex/syscalls.c
@@ -184,12 +184,15 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
  * @futexv:	Kernel side list of waiters to be filled
  * @uwaitv:     Userspace list to be parsed
  * @nr_futexes: Length of futexv
+ * @wake:	Wake to call when futex is woken
+ * @wake_data:	Data for the wake handler
  *
  * Return: Error code on failure, 0 on success
  */
-static int futex_parse_waitv(struct futex_vector *futexv,
-			     struct futex_waitv __user *uwaitv,
-			     unsigned int nr_futexes)
+int futex_parse_waitv(struct futex_vector *futexv,
+		      struct futex_waitv __user *uwaitv,
+		      unsigned int nr_futexes, futex_wake_fn *wake,
+		      void *wake_data)
 {
 	struct futex_waitv aux;
 	unsigned int i;
@@ -214,6 +217,8 @@ static int futex_parse_waitv(struct futex_vector *futexv,
 		futexv[i].w.val = aux.val;
 		futexv[i].w.uaddr = aux.uaddr;
 		futexv[i].q = futex_q_init;
+		futexv[i].q.wake = wake;
+		futexv[i].q.wake_data = wake_data;
 	}
 
 	return 0;
@@ -306,7 +311,8 @@ SYSCALL_DEFINE5(futex_waitv, struct futex_waitv __user *, waiters,
 		goto destroy_timer;
 	}
 
-	ret = futex_parse_waitv(futexv, waiters, nr_futexes);
+	ret = futex_parse_waitv(futexv, waiters, nr_futexes, futex_wake_mark,
+				NULL);
 	if (!ret)
 		ret = futex_wait_multiple(futexv, nr_futexes, timeout ? &to : NULL);
 
@@ -421,7 +427,7 @@ SYSCALL_DEFINE4(futex_requeue,
 	if (!waiters)
 		return -EINVAL;
 
-	ret = futex_parse_waitv(futexes, waiters, 2);
+	ret = futex_parse_waitv(futexes, waiters, 2, futex_wake_mark, NULL);
 	if (ret)
 		return ret;
 
-- 
2.40.1

[PATCH 7/8] futex: make the vectored futex operations available

[Jens Axboe]

Rename unqueue_multiple() as futex_unqueue_multiple(), and make both
that and futex_wait_multiple_setup() available for external users. This
is in preparation for wiring up vectored waits in io_uring.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    |  5 +++++
 kernel/futex/waitwake.c | 10 +++++-----
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index 6b6a6b3da103..8b195d06f4e8 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -366,6 +366,11 @@ extern int futex_parse_waitv(struct futex_vector *futexv,
 			     unsigned int nr_futexes, futex_wake_fn *wake,
 			     void *wake_data);
 
+extern int futex_wait_multiple_setup(struct futex_vector *vs, int count,
+				     int *woken);
+
+extern int futex_unqueue_multiple(struct futex_vector *v, int count);
+
 extern int futex_wait_multiple(struct futex_vector *vs, unsigned int count,
 			       struct hrtimer_sleeper *to);
 
diff --git a/kernel/futex/waitwake.c b/kernel/futex/waitwake.c
index 6fcf5f723719..61b112897a84 100644
--- a/kernel/futex/waitwake.c
+++ b/kernel/futex/waitwake.c
@@ -372,7 +372,7 @@ void futex_wait_queue(struct futex_hash_bucket *hb, struct futex_q *q,
 }
 
 /**
- * unqueue_multiple - Remove various futexes from their hash bucket
+ * futex_unqueue_multiple - Remove various futexes from their hash bucket
  * @v:	   The list of futexes to unqueue
  * @count: Number of futexes in the list
  *
@@ -382,7 +382,7 @@ void futex_wait_queue(struct futex_hash_bucket *hb, struct futex_q *q,
  *  - >=0 - Index of the last futex that was awoken;
  *  - -1  - No futex was awoken
  */
-static int unqueue_multiple(struct futex_vector *v, int count)
+int futex_unqueue_multiple(struct futex_vector *v, int count)
 {
 	int ret = -1, i;
 
@@ -410,7 +410,7 @@ static int unqueue_multiple(struct futex_vector *v, int count)
  *  -  0 - Success
  *  - <0 - -EFAULT, -EWOULDBLOCK or -EINVAL
  */
-static int futex_wait_multiple_setup(struct futex_vector *vs, int count, int *woken)
+int futex_wait_multiple_setup(struct futex_vector *vs, int count, int *woken)
 {
 	struct futex_hash_bucket *hb;
 	bool retry = false;
@@ -472,7 +472,7 @@ static int futex_wait_multiple_setup(struct futex_vector *vs, int count, int *wo
 		 * was woken, we don't return error and return this index to
 		 * userspace
 		 */
-		*woken = unqueue_multiple(vs, i);
+		*woken = futex_unqueue_multiple(vs, i);
 		if (*woken >= 0)
 			return 1;
 
@@ -557,7 +557,7 @@ int futex_wait_multiple(struct futex_vector *vs, unsigned int count,
 
 		__set_current_state(TASK_RUNNING);
 
-		ret = unqueue_multiple(vs, count);
+		ret = futex_unqueue_multiple(vs, count);
 		if (ret >= 0)
 			return ret;
 
-- 
2.40.1

[PATCH 8/8] io_uring: add support for vectored futex waits

[Jens Axboe]

This adds support for IORING_OP_FUTEX_WAITV, which allows registering a
notification for a number of futexes at once. If one of the futexes are
woken, then the request will complete with the index of the futex that got
woken as the result. This is identical to what the normal vectored futex
waitv operation does.

Use like IORING_OP_FUTEX_WAIT, except sqe->addr must now contain a
pointer to a struct futex_waitv array, and sqe->off must now contain the
number of elements in that array. As flags are passed in the futex_vector
array, and likewise for the value and futex address(es), sqe->addr2
and sqe->addr3 are also reserved for IORING_OP_FUTEX_WAITV.

For cancelations, FUTEX_WAITV does not rely on the futex_unqueue()
return value as we're dealing with multiple futexes. Instead, a separate
per io_uring request atomic is used to claim ownership of the request.

Waiting on N futexes could be done with IORING_OP_FUTEX_WAIT as well,
but that punts a lot of the work to the application:

1) Application would need to submit N IORING_OP_FUTEX_WAIT requests,
   rather than just a single IORING_OP_FUTEX_WAITV.

2) When one futex is woken, application would need to cancel the
   remaining N-1 requests that didn't trigger.

While this is of course doable, having a single vectored futex wait
makes for much simpler application code.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/uapi/linux/io_uring.h |   1 +
 io_uring/futex.c              | 169 ++++++++++++++++++++++++++++++++--
 io_uring/futex.h              |   2 +
 io_uring/opdef.c              |  11 +++
 4 files changed, 174 insertions(+), 9 deletions(-)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 04f9fba38d4b..92be89a871fc 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -246,6 +246,7 @@ enum io_uring_op {
 	IORING_OP_WAITID,
 	IORING_OP_FUTEX_WAIT,
 	IORING_OP_FUTEX_WAKE,
+	IORING_OP_FUTEX_WAITV,
 
 	/* this goes last, obviously */
 	IORING_OP_LAST,
diff --git a/io_uring/futex.c b/io_uring/futex.c
index eb4406ac46fb..3c3575303c3d 100644
--- a/io_uring/futex.c
+++ b/io_uring/futex.c
@@ -14,10 +14,16 @@
 
 struct io_futex {
 	struct file	*file;
-	u32 __user	*uaddr;
+	union {
+		u32 __user			*uaddr;
+		struct futex_waitv __user	*uwaitv;
+	};
 	unsigned long	futex_val;
 	unsigned long	futex_mask;
+	unsigned long	futexv_owned;
 	u32		futex_flags;
+	unsigned int	futex_nr;
+	bool		futexv_unqueued;
 };
 
 struct io_futex_data {
@@ -44,6 +50,13 @@ void io_futex_cache_free(struct io_ring_ctx *ctx)
 	io_alloc_cache_free(&ctx->futex_cache, io_futex_cache_entry_free);
 }
 
+static void __io_futex_complete(struct io_kiocb *req, struct io_tw_state *ts)
+{
+	req->async_data = NULL;
+	hlist_del_init(&req->hash_node);
+	io_req_task_complete(req, ts);
+}
+
 static void io_futex_complete(struct io_kiocb *req, struct io_tw_state *ts)
 {
 	struct io_futex_data *ifd = req->async_data;
@@ -52,22 +65,56 @@ static void io_futex_complete(struct io_kiocb *req, struct io_tw_state *ts)
 	io_tw_lock(ctx, ts);
 	if (!io_alloc_cache_put(&ctx->futex_cache, &ifd->cache))
 		kfree(ifd);
-	req->async_data = NULL;
-	hlist_del_init(&req->hash_node);
-	io_req_task_complete(req, ts);
+	__io_futex_complete(req, ts);
 }
 
-static bool __io_futex_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req)
+static void io_futexv_complete(struct io_kiocb *req, struct io_tw_state *ts)
 {
-	struct io_futex_data *ifd = req->async_data;
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+	struct futex_vector *futexv = req->async_data;
 
-	/* futex wake already done or in progress */
-	if (!futex_unqueue(&ifd->q))
+	io_tw_lock(req->ctx, ts);
+
+	if (!iof->futexv_unqueued) {
+		int res;
+
+		res = futex_unqueue_multiple(futexv, iof->futex_nr);
+		if (res != -1)
+			io_req_set_res(req, res, 0);
+	}
+
+	kfree(req->async_data);
+	req->flags &= ~REQ_F_ASYNC_DATA;
+	__io_futex_complete(req, ts);
+}
+
+static bool io_futexv_claim(struct io_futex *iof)
+{
+	if (test_bit(0, &iof->futexv_owned) ||
+	    test_and_set_bit_lock(0, &iof->futexv_owned))
 		return false;
+	return true;
+}
+
+static bool __io_futex_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req)
+{
+	/* futex wake already done or in progress */
+	if (req->opcode == IORING_OP_FUTEX_WAIT) {
+		struct io_futex_data *ifd = req->async_data;
+
+		if (!futex_unqueue(&ifd->q))
+			return false;
+		req->io_task_work.func = io_futex_complete;
+	} else {
+		struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+
+		if (!io_futexv_claim(iof))
+			return false;
+		req->io_task_work.func = io_futexv_complete;
+	}
 
 	hlist_del_init(&req->hash_node);
 	io_req_set_res(req, -ECANCELED, 0);
-	req->io_task_work.func = io_futex_complete;
 	io_req_task_work_add(req);
 	return true;
 }
@@ -147,6 +194,55 @@ int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	return 0;
 }
 
+static void io_futex_wakev_fn(struct wake_q_head *wake_q, struct futex_q *q)
+{
+	struct io_kiocb *req = q->wake_data;
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+
+	if (!io_futexv_claim(iof))
+		return;
+	if (unlikely(!__futex_wake_mark(q)))
+		return;
+
+	io_req_set_res(req, 0, 0);
+	req->io_task_work.func = io_futexv_complete;
+	io_req_task_work_add(req);
+}
+
+int io_futexv_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+	struct futex_vector *futexv;
+	int ret;
+
+	/* No flags or mask supported for waitv */
+	if (unlikely(sqe->fd || sqe->buf_index || sqe->file_index ||
+		     sqe->addr2 || sqe->futex_flags || sqe->addr3))
+		return -EINVAL;
+
+	iof->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	iof->futex_nr = READ_ONCE(sqe->len);
+	if (!iof->futex_nr || iof->futex_nr > FUTEX_WAITV_MAX)
+		return -EINVAL;
+
+	futexv = kcalloc(iof->futex_nr, sizeof(*futexv), GFP_KERNEL);
+	if (!futexv)
+		return -ENOMEM;
+
+	ret = futex_parse_waitv(futexv, iof->uwaitv, iof->futex_nr,
+				io_futex_wakev_fn, req);
+	if (ret) {
+		kfree(futexv);
+		return ret;
+	}
+
+	iof->futexv_owned = 0;
+	iof->futexv_unqueued = 0;
+	req->flags |= REQ_F_ASYNC_DATA;
+	req->async_data = futexv;
+	return 0;
+}
+
 static void io_futex_wake_fn(struct wake_q_head *wake_q, struct futex_q *q)
 {
 	struct io_futex_data *ifd = container_of(q, struct io_futex_data, q);
@@ -171,6 +267,61 @@ static struct io_futex_data *io_alloc_ifd(struct io_ring_ctx *ctx)
 	return kmalloc(sizeof(struct io_futex_data), GFP_NOWAIT);
 }
 
+int io_futexv_wait(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
+	struct futex_vector *futexv = req->async_data;
+	struct io_ring_ctx *ctx = req->ctx;
+	int ret, woken = -1;
+
+	io_ring_submit_lock(ctx, issue_flags);
+
+	ret = futex_wait_multiple_setup(futexv, iof->futex_nr, &woken);
+
+	/*
+	 * Error case, ret is < 0. Mark the request as failed.
+	 */
+	if (unlikely(ret < 0)) {
+		io_ring_submit_unlock(ctx, issue_flags);
+		req_set_fail(req);
+		io_req_set_res(req, ret, 0);
+		kfree(futexv);
+		req->async_data = NULL;
+		req->flags &= ~REQ_F_ASYNC_DATA;
+		return IOU_OK;
+	}
+
+	/*
+	 * 0 return means that we successfully setup the waiters, and that
+	 * nobody triggered a wakeup while we were doing so. If the wakeup
+	 * happened post setup, the task_work will be run post this issue and
+	 * under the submission lock. 1 means We got woken while setting up,
+	 * let that side do the completion. Note that
+	 * futex_wait_multiple_setup() will have unqueued all the futexes in
+	 * this case. Mark us as having done that already, since this is
+	 * different from normal wakeup.
+	 */
+	if (!ret) {
+		/*
+		 * If futex_wait_multiple_setup() returns 0 for a
+		 * successful setup, then the task state will not be
+		 * runnable. This is fine for the sync syscall, as
+		 * it'll be blocking unless we already got one of the
+		 * futexes woken, but it obviously won't work for an
+		 * async invocation. Mark us runnable again.
+		 */
+		__set_current_state(TASK_RUNNING);
+		hlist_add_head(&req->hash_node, &ctx->futex_list);
+	} else {
+		iof->futexv_unqueued = 1;
+		if (woken != -1)
+			io_req_set_res(req, woken, 0);
+	}
+
+	io_ring_submit_unlock(ctx, issue_flags);
+	return IOU_ISSUE_SKIP_COMPLETE;
+}
+
 int io_futex_wait(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
diff --git a/io_uring/futex.h b/io_uring/futex.h
index ddc9e0d73c52..0847e9e8a127 100644
--- a/io_uring/futex.h
+++ b/io_uring/futex.h
@@ -3,7 +3,9 @@
 #include "cancel.h"
 
 int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_futexv_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe);
 int io_futex_wait(struct io_kiocb *req, unsigned int issue_flags);
+int io_futexv_wait(struct io_kiocb *req, unsigned int issue_flags);
 int io_futex_wake(struct io_kiocb *req, unsigned int issue_flags);
 
 #if defined(CONFIG_FUTEX)
diff --git a/io_uring/opdef.c b/io_uring/opdef.c
index 31a3a421e94d..25a3515a177c 100644
--- a/io_uring/opdef.c
+++ b/io_uring/opdef.c
@@ -459,6 +459,14 @@ const struct io_issue_def io_issue_defs[] = {
 		.issue			= io_futex_wake,
 #else
 		.prep			= io_eopnotsupp_prep,
+#endif
+	},
+	[IORING_OP_FUTEX_WAITV] = {
+#if defined(CONFIG_FUTEX)
+		.prep			= io_futexv_prep,
+		.issue			= io_futexv_wait,
+#else
+		.prep			= io_eopnotsupp_prep,
 #endif
 	},
 };
@@ -693,6 +701,9 @@ const struct io_cold_def io_cold_defs[] = {
 	[IORING_OP_FUTEX_WAKE] = {
 		.name			= "FUTEX_WAKE",
 	},
+	[IORING_OP_FUTEX_WAITV] = {
+		.name			= "FUTEX_WAITV",
+	},
 };
 
 const char *io_uring_get_opcode(u8 opcode)
-- 
2.40.1

Re: [PATCHSET v6] Add io_uring futex/futexv support

[Peter Zijlstra]

On Thu, Sep 28, 2023 at 11:25:09AM -0600, Jens Axboe wrote:

>  include/linux/io_uring_types.h |   5 +
>  include/uapi/linux/io_uring.h  |   4 +
>  io_uring/Makefile              |   1 +
>  io_uring/cancel.c              |   5 +
>  io_uring/cancel.h              |   4 +
>  io_uring/futex.c               | 386 +++++++++++++++++++++++++++++++++
>  io_uring/futex.h               |  36 +++
>  io_uring/io_uring.c            |   7 +
>  io_uring/opdef.c               |  34 +++
>  kernel/futex/futex.h           |  20 ++
>  kernel/futex/requeue.c         |   3 +-
>  kernel/futex/syscalls.c        |  18 +-
>  kernel/futex/waitwake.c        |  49 +++--
>  13 files changed, 545 insertions(+), 27 deletions(-)

Thanks for bearing with us on the futex2 thing!

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>

Re: [PATCHSET v6] Add io_uring futex/futexv support

[Jens Axboe]

On 9/29/23 1:53 AM, Peter Zijlstra wrote:
> On Thu, Sep 28, 2023 at 11:25:09AM -0600, Jens Axboe wrote:
> 
>>  include/linux/io_uring_types.h |   5 +
>>  include/uapi/linux/io_uring.h  |   4 +
>>  io_uring/Makefile              |   1 +
>>  io_uring/cancel.c              |   5 +
>>  io_uring/cancel.h              |   4 +
>>  io_uring/futex.c               | 386 +++++++++++++++++++++++++++++++++
>>  io_uring/futex.h               |  36 +++
>>  io_uring/io_uring.c            |   7 +
>>  io_uring/opdef.c               |  34 +++
>>  kernel/futex/futex.h           |  20 ++
>>  kernel/futex/requeue.c         |   3 +-
>>  kernel/futex/syscalls.c        |  18 +-
>>  kernel/futex/waitwake.c        |  49 +++--
>>  13 files changed, 545 insertions(+), 27 deletions(-)
> 
> Thanks for bearing with us on the futex2 thing!
> 
> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>

Thanks Peter! Going with the futex2 interface was the right choice, the
old one was kinda wonky anyway. New one is definitely cleaner.

-- 
Jens Axboe

[PATCHSET v5] Add io_uring futex/futexv support

[Jens Axboe]

Hi,

This patchset adds support for first futex wake and wait, and then
futexv.

For both wait/wake/waitv, we support the bitset variant, as the
"normal" variants can be easily implemented on top of that.

PI and requeue are not supported through io_uring, just the above
mentioned parts. This may change in the future, but in the spirit
of keeping this small (and based on what people have been asking for),
this is what we currently have.

When I did these patches, I forgot that Pavel had previously posted a
futex variant for io_uring. The major thing that had been holding me
back from people asking about futexes and io_uring, is that I wanted
to do this what I consider the right way - no usage of io-wq or thread
offload, an actually async implementation that is efficient to use
and don't rely on a blocking thread for futex wait/waitv. This is what
this patchset attempts to do, while being minimally invasive on the
futex side. I believe the diffstat reflects that.

As far as I can recall, the first request for futex support with
io_uring came from Andres Freund, working on postgres. His aio rework
of postgres was one of the early adopters of io_uring, and futex
support was a natural extension for that. This is relevant from both
a usability point of view, as well as for effiency and performance.
In Andres's words, for the former:

"Futex wait support in io_uring makes it a lot easier to avoid deadlocks
in concurrent programs that have their own buffer pool: Obviously pages in
the application buffer pool have to be locked during IO. If the initiator
of IO A needs to wait for a held lock B, the holder of lock B might wait
for the IO A to complete.  The ability to wait for a lock and IO
completions at the same time provides an efficient way to avoid such
deadlocks."

and in terms of effiency, even without unlocking the full potential yet,
Andres says:

"Futex wake support in io_uring is useful because it allows for more
efficient directed wakeups.  For some "locks" postgres has queues
implemented in userspace, with wakeup logic that cannot easily be
implemented with FUTEX_WAKE_BITSET on a single "futex word" (imagine
waiting for journal flushes to have completed up to a certain point). Thus
a "lock release" sometimes need to wake up many processes in a row.  A
quick-and-dirty conversion to doing these wakeups via io_uring lead to a
3% throughput increase, with 12% fewer context switches, albeit in a
fairly extreme workload."

Some basic io_uring futex support and test cases are available in the
liburing 'futex' branch:

https://git.kernel.dk/cgit/liburing/log/?h=futex

testing all of the variants. I originally wrote this code about a
month ago and Andres has been using it with postgres, and I'm not
aware of any bugs in it. That's not to say it's perfect, obviously,
and I welcome some feedback so we can move this forward and hash out
any potential issues.

In terms of testing, there's a functionality and beat-up test case
in liburing, and I've run all the ltp futex test cases as well to
ensure we didn't inadvertently break anything. It's also been in
linux-next for a long time and haven't heard any complaints.

 include/linux/io_uring_types.h |   5 +
 include/uapi/linux/io_uring.h  |   4 +
 io_uring/Makefile              |   1 +
 io_uring/cancel.c              |   5 +
 io_uring/cancel.h              |   4 +
 io_uring/futex.c               | 376 +++++++++++++++++++++++++++++++++
 io_uring/futex.h               |  36 ++++
 io_uring/io_uring.c            |   7 +
 io_uring/opdef.c               |  34 +++
 kernel/futex/futex.h           |  20 ++
 kernel/futex/requeue.c         |   3 +-
 kernel/futex/syscalls.c        |  18 +-
 kernel/futex/waitwake.c        |  49 +++--
 13 files changed, 535 insertions(+), 27 deletions(-)

You can also find the code here:

https://git.kernel.dk/cgit/linux/log/?h=io_uring-futex

V5:
- Rebase on PeterZ's futex2 changes. Pulled in the tip locking/core
  branch.
- Shuffle order of some io_uring patchsets, which changed the value
  of the futex opcodes.

V4:
- Refactor the prep setup so it's fully independent between the vectoed
  and non-vectored futex handling.
- Ensure we -EINVAL any futex/futexv wait/waitv that specifies unused
  fields.
- Fix a comment typo
- Update the patches from Peter.
- Fix two kerneldoc warnings
- Add a prep patch moving FUTEX2_MASK to futex.h

-- 
Jens Axboe

[PATCH 2/8] futex: factor out the futex wake handling

[Jens Axboe]

In preparation for having another waker that isn't futex_wake_mark(),
add a wake handler in futex_q. No extra data is associated with the
handler outside of struct futex_q itself. futex_wake_mark() is defined as
the standard wakeup helper, now set through futex_q_init like other
defaults.

Normal sync futex waiting relies on wake_q holding tasks that should
be woken up. This is what futex_wake_mark() does, it'll unqueue the
futex and add the associated task to the wake queue. For async usage of
futex waiting, rather than having tasks sleeping on the futex, we'll
need to deal with a futex wake differently. For the planned io_uring
case, that means posting a completion event for the task in question.
Having a definable wake handler can help support that use case.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    | 5 +++++
 kernel/futex/requeue.c  | 3 ++-
 kernel/futex/waitwake.c | 6 +++---
 3 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index a173a9d501e1..547f509b2c87 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -139,11 +139,15 @@ struct futex_pi_state {
 	union futex_key key;
 } __randomize_layout;
 
+struct futex_q;
+typedef void (futex_wake_fn)(struct wake_q_head *wake_q, struct futex_q *q);
+
 /**
  * struct futex_q - The hashed futex queue entry, one per waiting task
  * @list:		priority-sorted list of tasks waiting on this futex
  * @task:		the task waiting on the futex
  * @lock_ptr:		the hash bucket lock
+ * @wake:		the wake handler for this queue
  * @key:		the key the futex is hashed on
  * @pi_state:		optional priority inheritance state
  * @rt_waiter:		rt_waiter storage for use with requeue_pi
@@ -168,6 +172,7 @@ struct futex_q {
 
 	struct task_struct *task;
 	spinlock_t *lock_ptr;
+	futex_wake_fn *wake;
 	union futex_key key;
 	struct futex_pi_state *pi_state;
 	struct rt_mutex_waiter *rt_waiter;
diff --git a/kernel/futex/requeue.c b/kernel/futex/requeue.c
index a0a79954f506..9dc789399a1a 100644
--- a/kernel/futex/requeue.c
+++ b/kernel/futex/requeue.c
@@ -58,6 +58,7 @@ enum {
 
 const struct futex_q futex_q_init = {
 	/* list gets initialized in futex_queue()*/
+	.wake		= futex_wake_mark,
 	.key		= FUTEX_KEY_INIT,
 	.bitset		= FUTEX_BITSET_MATCH_ANY,
 	.requeue_state	= ATOMIC_INIT(Q_REQUEUE_PI_NONE),
@@ -593,7 +594,7 @@ int futex_requeue(u32 __user *uaddr1, unsigned int flags1,
 		/* Plain futexes just wake or requeue and are done */
 		if (!requeue_pi) {
 			if (++task_count <= nr_wake)
-				futex_wake_mark(&wake_q, this);
+				this->wake(&wake_q, this);
 			else
 				requeue_futex(this, hb1, hb2, &key2);
 			continue;
diff --git a/kernel/futex/waitwake.c b/kernel/futex/waitwake.c
index 37860f794bf7..35c6a637a4bb 100644
--- a/kernel/futex/waitwake.c
+++ b/kernel/futex/waitwake.c
@@ -177,7 +177,7 @@ int futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
 			if (!(this->bitset & bitset))
 				continue;
 
-			futex_wake_mark(&wake_q, this);
+			this->wake(&wake_q, this);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -292,7 +292,7 @@ int futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 				ret = -EINVAL;
 				goto out_unlock;
 			}
-			futex_wake_mark(&wake_q, this);
+			this->wake(&wake_q, this);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -306,7 +306,7 @@ int futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 					ret = -EINVAL;
 					goto out_unlock;
 				}
-				futex_wake_mark(&wake_q, this);
+				this->wake(&wake_q, this);
 				if (++op_ret >= nr_wake2)
 					break;
 			}
-- 
2.40.1

[PATCHSET v2 0/8] Add io_uring futex/futexv support

[Jens Axboe]

Hi,

This patchset adds support for first futex wake and wait, and then
futexv. Patches 1..3 are just prep patches, patch 4 adds the wait
and wake support for io_uring, and then patches 5..7 are again prep
patches to end up with futexv support in patch 8.

For both wait/wake/waitv, we support the bitset variant, as the
"normal" variants can be easily implemented on top of that.

PI and requeue are not supported through io_uring, just the above
mentioned parts. This may change in the future, but in the spirit
of keeping this small (and based on what people have been asking for),
this is what we currently have.

When I did these patches, I forgot that Pavel had previously posted a
futex variant for io_uring. The major thing that had been holding me
back from people asking about futexes and io_uring, is that I wanted
to do this what I consider the right way - no usage of io-wq or thread
offload, an actually async implementation that is efficient to use
and don't rely on a blocking thread for futex wait/waitv. This is what
this patchset attempts to do, while being minimally invasive on the
futex side. I believe the diffstat reflects that.

As far as I can recall, the first request for futex support with
io_uring came from Andres Freund, working on postgres. His aio rework
of postgres was one of the early adopters of io_uring, and futex
support was a natural extension for that. This is relevant from both
a usability point of view, as well as for effiency and performance.
In Andres's words, for the former:

"Futex wait support in io_uring makes it a lot easier to avoid deadlocks
in concurrent programs that have their own buffer pool: Obviously pages in
the application buffer pool have to be locked during IO. If the initiator
of IO A needs to wait for a held lock B, the holder of lock B might wait
for the IO A to complete.  The ability to wait for a lock and IO
completions at the same time provides an efficient way to avoid such
deadlocks."

and in terms of effiency, even without unlocking the full potential yet,
Andres says:

"Futex wake support in io_uring is useful because it allows for more
efficient directed wakeups.  For some "locks" postgres has queues
implemented in userspace, with wakeup logic that cannot easily be
implemented with FUTEX_WAKE_BITSET on a single "futex word" (imagine
waiting for journal flushes to have completed up to a certain point). Thus
a "lock release" sometimes need to wake up many processes in a row.  A
quick-and-dirty conversion to doing these wakeups via io_uring lead to a
3% throughput increase, with 12% fewer context switches, albeit in a
fairly extreme workload."

Some basic io_uring futex support and test cases are available in the
liburing 'futex' branch:

https://git.kernel.dk/cgit/liburing/log/?h=futex

testing all of the variants. I originally wrote this code about a
month ago and Andres has been using it with postgres, and I'm not
aware of any bugs in it. That's not to say it's perfect, obviously,
and I welcome some feedback so we can move this forward and hash out
any potential issues.

 include/linux/io_uring_types.h |   3 +
 include/uapi/linux/io_uring.h  |   4 +
 io_uring/Makefile              |   4 +-
 io_uring/cancel.c              |   5 +
 io_uring/cancel.h              |   4 +
 io_uring/futex.c               | 376 +++++++++++++++++++++++++++++++++
 io_uring/futex.h               |  36 ++++
 io_uring/io_uring.c            |   5 +
 io_uring/opdef.c               |  35 ++-
 kernel/futex/futex.h           |  33 +++
 kernel/futex/requeue.c         |   3 +-
 kernel/futex/syscalls.c        |  27 ++-
 kernel/futex/waitwake.c        |  49 +++--
 13 files changed, 548 insertions(+), 36 deletions(-)

You can also find the code here:

https://git.kernel.dk/cgit/linux/log/?h=io_uring-futex

V2:
- Abstract out __futex_wake_mark() helper. Use it both in the
  futex and io_uring code. This also fixes a missing WARN_ON
  on the io_uring side.
- Have futex_op_to_flags() unconditionally clear flags to
  zero rather than do that in both callers.
- Remove comment on needing to open-code futex_queue(),
  and associated hunk doing that. This was a leftover
  from an earlier version.
- Expand the commit message logs in various patches.

-- 
Jens Axboe

[PATCH 2/8] futex: factor out the futex wake handling

[Jens Axboe]

In preparation for having another waker that isn't futex_wake_mark(),
add a wake handler in futex_q. No extra data is associated with the
handler outside of struct futex_q itself. futex_wake_mark() is defined as
the standard wakeup helper, now set through futex_q_init like other
defaults.

Normal sync futex waiting relies on wake_q holding tasks that should
be woken up. This is what futex_wake_mark() does, it'll unqueue the
futex and add the associated task to the wake queue. For async usage of
futex waiting, rather than having tasks sleeping on the futex, we'll
need to deal with a futex wake differently. For the planned io_uring
case, that means posting a completion event for the task in question.
Having a definable wake handler can help support that use case.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 kernel/futex/futex.h    | 4 ++++
 kernel/futex/requeue.c  | 3 ++-
 kernel/futex/waitwake.c | 6 +++---
 3 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/kernel/futex/futex.h b/kernel/futex/futex.h
index b8f454792304..54b7de21ef00 100644
--- a/kernel/futex/futex.h
+++ b/kernel/futex/futex.h
@@ -69,6 +69,9 @@ struct futex_pi_state {
 	union futex_key key;
 } __randomize_layout;
 
+struct futex_q;
+typedef void (futex_wake_fn)(struct wake_q_head *wake_q, struct futex_q *q);
+
 /**
  * struct futex_q - The hashed futex queue entry, one per waiting task
  * @list:		priority-sorted list of tasks waiting on this futex
@@ -98,6 +101,7 @@ struct futex_q {
 
 	struct task_struct *task;
 	spinlock_t *lock_ptr;
+	futex_wake_fn *wake;
 	union futex_key key;
 	struct futex_pi_state *pi_state;
 	struct rt_mutex_waiter *rt_waiter;
diff --git a/kernel/futex/requeue.c b/kernel/futex/requeue.c
index cba8b1a6a4cc..e892bc6c41d8 100644
--- a/kernel/futex/requeue.c
+++ b/kernel/futex/requeue.c
@@ -58,6 +58,7 @@ enum {
 
 const struct futex_q futex_q_init = {
 	/* list gets initialized in futex_queue()*/
+	.wake		= futex_wake_mark,
 	.key		= FUTEX_KEY_INIT,
 	.bitset		= FUTEX_BITSET_MATCH_ANY,
 	.requeue_state	= ATOMIC_INIT(Q_REQUEUE_PI_NONE),
@@ -591,7 +592,7 @@ int futex_requeue(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 		/* Plain futexes just wake or requeue and are done */
 		if (!requeue_pi) {
 			if (++task_count <= nr_wake)
-				futex_wake_mark(&wake_q, this);
+				this->wake(&wake_q, this);
 			else
 				requeue_futex(this, hb1, hb2, &key2);
 			continue;
diff --git a/kernel/futex/waitwake.c b/kernel/futex/waitwake.c
index ba01b9408203..3471af87cb7d 100644
--- a/kernel/futex/waitwake.c
+++ b/kernel/futex/waitwake.c
@@ -174,7 +174,7 @@ int futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
 			if (!(this->bitset & bitset))
 				continue;
 
-			futex_wake_mark(&wake_q, this);
+			this->wake(&wake_q, this);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -289,7 +289,7 @@ int futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 				ret = -EINVAL;
 				goto out_unlock;
 			}
-			futex_wake_mark(&wake_q, this);
+			this->wake(&wake_q, this);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -303,7 +303,7 @@ int futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
 					ret = -EINVAL;
 					goto out_unlock;
 				}
-				futex_wake_mark(&wake_q, this);
+				this->wake(&wake_q, this);
 				if (++op_ret >= nr_wake2)
 					break;
 			}
-- 
2.40.1