Skip to contentSkip to conversation

uring_cmd copy avoidance

PATCH Superseded Jens Axboe 3 participants 10 messages Jun 6, 2025 – Jun 8, 2025 io_uring

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Jens Axboe

author maintainer
#0 lore raw

v3 cover uring_cmd copy avoidance

Hi,

Currently uring_cmd unconditionally copies the SQE at prep time, as it has no other choice - the SQE data must remain stable after submit. This can lead to excessive memory bandwidth being used for that copy, as passthrough will often use 128b SQEs, and efficiency concerns as those copies will potentially use quite a lot of CPU cycles as well.

As a quick test, running the current -git kernel on a box with 23 NVMe drives doing passthrough IO, memcpy() is the highest cycle user at 9.05%, which is all off the uring_cmd prep path. The test case is a 512b random read, which runs at 91-92M IOPS.

With these patches, memcpy() is gone from the profiles, and it runs at 98-99M IOPS, or about 7-8% faster.

Before:

IOPS=91.12M, BW=44.49GiB/s, IOS/call=32/32
IOPS=91.16M, BW=44.51GiB/s, IOS/call=32/32
IOPS=91.18M, BW=44.52GiB/s, IOS/call=31/32
IOPS=91.92M, BW=44.88GiB/s, IOS/call=32/32
IOPS=91.88M, BW=44.86GiB/s, IOS/call=32/32
IOPS=91.82M, BW=44.83GiB/s, IOS/call=32/31
IOPS=91.52M, BW=44.69GiB/s, IOS/call=32/32

with the top perf report -g --no-children being:

+ 9.07% io_uring [kernel.kallsyms] [k] memcpy

and after:

# bash run-peak-pass.sh
[...]
IOPS=99.30M, BW=48.49GiB/s, IOS/call=32/32
IOPS=99.27M, BW=48.47GiB/s, IOS/call=31/32
IOPS=99.60M, BW=48.63GiB/s, IOS/call=32/32
IOPS=99.68M, BW=48.67GiB/s, IOS/call=32/31
IOPS=99.80M, BW=48.73GiB/s, IOS/call=31/32
IOPS=99.84M, BW=48.75GiB/s, IOS/call=32/32

with memcpy not even in profiles. If you do the actual math of 100M requests per second, and 128b of copying per IOP, then it's almost 12GB/sec of reduced memory bandwidth.

Even for lower IOPS production testing, Caleb reports that memcpy() overhead is in the realm of 1.1% of CPU time.

v3 cleans up the ->sqe_copy() handling, relying solely on the IO_URING_F_INLINE flag. And the core will handle it now, rejecting commands if they should not be copied. This will returns in an -EFAULT failure.

I think this approach is saner, and in fact it can be extended to reduce over-eager copies in other spots. For now I just did uring_cmd, and verified that the memcpy's are still gone from my test.

Can also be found here:

https://git.kernel.dk/cgit/linux/log/?h=uring_cmd.2

 include/linux/io_uring_types.h |  2 ++
 io_uring/io_uring.c            | 37 ++++++++++++++++++++++-------
 io_uring/opdef.c               |  1 +
 io_uring/opdef.h               |  1 +
 io_uring/uring_cmd.c           | 43 +++++++++++++++-------------------
 io_uring/uring_cmd.h           |  1 +
 6 files changed, 53 insertions(+), 32 deletions(-)
Signature · 1 line
Jens Axboe
author maintainer
#1 lore raw

PATCH v3 1/4 io_uring: add IO_URING_F_INLINE issue flag

Set when the execution of the request is done inline from the system call itself. Any deferred issue will never have this flag set.

---
 include/linux/io_uring_types.h |  2 ++
 io_uring/io_uring.c            | 12 +++++++-----
 2 files changed, 9 insertions(+), 5 deletions(-)
include/linux/io_uring_types.h +2 -0
index 2922635986f5..054c43c02c96 100644
@@ -26,6 +26,8 @@ enum io_uring_cmd_flags {
 	IO_URING_F_MULTISHOT		= 4,
 	/* executed by io-wq */
 	IO_URING_F_IOWQ			= 8,
+	/* executed inline from syscall */
+	IO_URING_F_INLINE		= 16,
 	/* int's last bit, sign checks are usually faster than a bit test */
 	IO_URING_F_NONBLOCK		= INT_MIN,
 
io_uring/io_uring.c +7 -5
index cf759c172083..0f9f6a173e66 100644
@@ -147,7 +147,7 @@ static bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
 					 bool cancel_all,
 					 bool is_sqpoll_thread);
 
-static void io_queue_sqe(struct io_kiocb *req);
+static void io_queue_sqe(struct io_kiocb *req, unsigned int extra_flags);
 static void __io_req_caches_free(struct io_ring_ctx *ctx);
 
 static __read_mostly DEFINE_STATIC_KEY_FALSE(io_key_has_sqarray);
@@ -1377,7 +1377,7 @@ void io_req_task_submit(struct io_kiocb *req, io_tw_token_t tw)
 	else if (req->flags & REQ_F_FORCE_ASYNC)
 		io_queue_iowq(req);
 	else
-		io_queue_sqe(req);
+		io_queue_sqe(req, 0);
 }
 
 void io_req_task_queue_fail(struct io_kiocb *req, int ret)
@@ -1957,12 +1957,14 @@ static void io_queue_async(struct io_kiocb *req, int ret)
 	}
 }
 
-static inline void io_queue_sqe(struct io_kiocb *req)
+static inline void io_queue_sqe(struct io_kiocb *req, unsigned int extra_flags)
 	__must_hold(&req->ctx->uring_lock)
 {
+	unsigned int issue_flags = IO_URING_F_NONBLOCK |
+				   IO_URING_F_COMPLETE_DEFER | extra_flags;
 	int ret;
 
-	ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
+	ret = io_issue_sqe(req, issue_flags);
 
 	/*
 	 * We async punt it if the file wasn't marked NOWAIT, or if the file
@@ -2218,7 +2220,7 @@ static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
 		return 0;
 	}
 
-	io_queue_sqe(req);
+	io_queue_sqe(req, IO_URING_F_INLINE);
 	return 0;
 }
 
Signature · 1 line
2.49.0
author maintainer
#3 lore raw

PATCH v3 2/4 io_uring: add struct io_cold_def->sqe_copy() method

In reply to Jens Axboe

Will be called by the core of io_uring, if inline issue is not going to be tried for a request. Opcodes can define this handler to defer copying of SQE data that should remain stable.

Only called if IO_URING_F_INLINE is set. If it isn't set, then there's a bug in the core handling of this, and -EFAULT will be returned instead to terminate the request. This will trigger a WARN_ON_ONCE(). Don't expect this to ever trigger, and down the line this can be removed.

---
 io_uring/io_uring.c | 25 ++++++++++++++++++++++---
 io_uring/opdef.h    |  1 +
 2 files changed, 23 insertions(+), 3 deletions(-)
io_uring/io_uring.c +22 -3
index 0f9f6a173e66..9799a31a2b29 100644
@@ -1935,14 +1935,31 @@ struct file *io_file_get_normal(struct io_kiocb *req, int fd)
 	return file;
 }
 
-static void io_queue_async(struct io_kiocb *req, int ret)
+static int io_req_sqe_copy(struct io_kiocb *req, unsigned int issue_flags)
+{
+	const struct io_cold_def *def = &io_cold_defs[req->opcode];
+
+	if (!def->sqe_copy)
+		return 0;
+	if (WARN_ON_ONCE(!(issue_flags & IO_URING_F_INLINE)))
+		return -EFAULT;
+	def->sqe_copy(req);
+	return 0;
+}
+
+static void io_queue_async(struct io_kiocb *req, unsigned int issue_flags, int ret)
 	__must_hold(&req->ctx->uring_lock)
 {
 	if (ret != -EAGAIN || (req->flags & REQ_F_NOWAIT)) {
+fail:
 		io_req_defer_failed(req, ret);
 		return;
 	}
 
+	ret = io_req_sqe_copy(req, issue_flags);
+	if (unlikely(ret))
+		goto fail;
+
 	switch (io_arm_poll_handler(req, 0)) {
 	case IO_APOLL_READY:
 		io_kbuf_recycle(req, 0);
@@ -1971,7 +1988,7 @@ static inline void io_queue_sqe(struct io_kiocb *req, unsigned int extra_flags)
 	 * doesn't support non-blocking read/write attempts
 	 */
 	if (unlikely(ret))
-		io_queue_async(req, ret);
+		io_queue_async(req, issue_flags, ret);
 }
 
 static void io_queue_sqe_fallback(struct io_kiocb *req)
@@ -1986,6 +2003,8 @@ static void io_queue_sqe_fallback(struct io_kiocb *req)
 		req->flags |= REQ_F_LINK;
 		io_req_defer_failed(req, req->cqe.res);
 	} else {
+		/* can't fail with IO_URING_F_INLINE */
+		io_req_sqe_copy(req, IO_URING_F_INLINE);
 		if (unlikely(req->ctx->drain_active))
 			io_drain_req(req);
 		else
@@ -2201,7 +2220,7 @@ static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
 		link->last = req;
 
 		if (req->flags & IO_REQ_LINK_FLAGS)
-			return 0;
+			return io_req_sqe_copy(req, IO_URING_F_INLINE);
 		/* last request of the link, flush it */
 		req = link->head;
 		link->head = NULL;
io_uring/opdef.h +1 -0
index 719a52104abe..c2f0907ed78c 100644
@@ -38,6 +38,7 @@ struct io_issue_def {
 struct io_cold_def {
 	const char		*name;
 
+	void (*sqe_copy)(struct io_kiocb *);
 	void (*cleanup)(struct io_kiocb *);
 	void (*fail)(struct io_kiocb *);
 };
Signature · 1 line
2.49.0
#4 lore raw

Re: [PATCH 2/4] io_uring: add struct io_cold_def->sqe_copy() method

Quoted 22 lines · Jens Axboe

Will be called by the core of io_uring, if inline issue is not going to be tried for a request. Opcodes can define this handler to defer copying of SQE data that should remain stable.

Only called if IO_URING_F_INLINE is set. If it isn't set, then there's a bug in the core handling of this, and -EFAULT will be returned instead to terminate the request. This will trigger a WARN_ON_ONCE(). Don't expect this to ever trigger, and down the line this can be removed.

---
 io_uring/io_uring.c | 25 ++++++++++++++++++++++---
 io_uring/opdef.h    |  1 +
 2 files changed, 23 insertions(+), 3 deletions(-)
io_uring/io_uring.c +0 -0
index 0f9f6a173e66..9799a31a2b29 100644
@@ -1935,14 +1935,31 @@ struct file *io_file_get_normal(struct io_kiocb *req, int fd)
        return file;

io_uring/io_uring.c struct file *io_file_get_normal(struct io_kiocb *req, int fd) · line 1944

Quoted 10 lines
io_uring/io_uring.c +7 -1
 }
 
-static void io_queue_async(struct io_kiocb *req, int ret)
+static int io_req_sqe_copy(struct io_kiocb *req, unsigned int issue_flags)
+{
+       const struct io_cold_def *def = &io_cold_defs[req->opcode];
+
+       if (!def->sqe_copy)
+               return 0;
+       if (WARN_ON_ONCE(!(issue_flags & IO_URING_F_INLINE)))

I'm pretty confident that every initial async path under io_submit_sqe() will call io_req_sqe_copy(). But I'm not positive that io_req_sqe_copy() won't get called *additional* times from non-inline contexts. One example scenario:
- io_submit_sqe() calls io_queue_sqe()
- io_issue_sqe() returns -EAGAIN, so io_queue_sqe() calls io_queue_async()
- io_queue_async() calls io_req_sqe_copy() in inline context
- io_queue_async() calls io_arm_poll_handler(), which returns IO_APOLL_READY, so io_req_task_queue() is called
- Some other I/O to the file (possibly on a different task) clears the ready poll events
- io_req_task_submit() calls io_queue_sqe() in task work context
- io_issue_sqe() returns -EAGAIN again, so io_queue_async() is called
- io_queue_async() calls io_req_sqe_copy() a second time in non-inline (task work) context

If this is indeed possible, then I think we may need to relax this check so it only verifies that IO_URING_F_INLINE is set *the first time* io_req_sqe_copy() is called for a given req. (Or just remove the IO_URING_F_INLINE check entirely.)

Quoted 40 lines
io_uring/io_uring.c +14 -1
+               return -EFAULT;
+       def->sqe_copy(req);
+       return 0;
+}
+
+static void io_queue_async(struct io_kiocb *req, unsigned int issue_flags, int ret)
        __must_hold(&req->ctx->uring_lock)
 {
        if (ret != -EAGAIN || (req->flags & REQ_F_NOWAIT)) {
+fail:
                io_req_defer_failed(req, ret);
                return;
        }
 
+       ret = io_req_sqe_copy(req, issue_flags);
+       if (unlikely(ret))
+               goto fail;
+
        switch (io_arm_poll_handler(req, 0)) {
        case IO_APOLL_READY:
                io_kbuf_recycle(req, 0);
@@ -1971,7 +1988,7 @@ static inline void io_queue_sqe(struct io_kiocb *req, unsigned int extra_flags)
         * doesn't support non-blocking read/write attempts
         */
        if (unlikely(ret))
-               io_queue_async(req, ret);
+               io_queue_async(req, issue_flags, ret);
 }
 
 static void io_queue_sqe_fallback(struct io_kiocb *req)
@@ -1986,6 +2003,8 @@ static void io_queue_sqe_fallback(struct io_kiocb *req)
                req->flags |= REQ_F_LINK;
                io_req_defer_failed(req, req->cqe.res);
        } else {
+               /* can't fail with IO_URING_F_INLINE */
+               io_req_sqe_copy(req, IO_URING_F_INLINE);
                if (unlikely(req->ctx->drain_active))
                        io_drain_req(req);
                else

io_uring/io_uring.c static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, · line 2223

Quoted 5 lines
io_uring/io_uring.c +1 -1
@@ -2201,7 +2220,7 @@ static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
                link->last = req;
 
                if (req->flags & IO_REQ_LINK_FLAGS)
-                       return 0;
+                       return io_req_sqe_copy(req, IO_URING_F_INLINE);

I still think this misses the last req in a linked chain, which will be issued async but won't have IO_REQ_LINK_FLAGS set. Am I missing something?

Best,
Caleb

Quoted 18 lines
io_uring/io_uring.c +0 -0
                /* last request of the link, flush it */
                req = link->head;
                link->head = NULL;
io_uring/opdef.h +1 -1
index 719a52104abe..c2f0907ed78c 100644
@@ -38,6 +38,7 @@ struct io_issue_def {
 struct io_cold_def {
        const char              *name;
 
+       void (*sqe_copy)(struct io_kiocb *);
        void (*cleanup)(struct io_kiocb *);
        void (*fail)(struct io_kiocb *);
 };
--

2.49.0

author maintainer
#5 lore raw

Re: [PATCH 2/4] io_uring: add struct io_cold_def->sqe_copy() method

Quoted 50 lines · Caleb Sander Mateos
Quoted 22 lines · On Fri, Jun 6, 2025 at 2:56?PM Jens Axboe <[email protected]> wrote:

Will be called by the core of io_uring, if inline issue is not going to be tried for a request. Opcodes can define this handler to defer copying of SQE data that should remain stable.

Only called if IO_URING_F_INLINE is set. If it isn't set, then there's a bug in the core handling of this, and -EFAULT will be returned instead to terminate the request. This will trigger a WARN_ON_ONCE(). Don't expect this to ever trigger, and down the line this can be removed.

---
 io_uring/io_uring.c | 25 ++++++++++++++++++++++---
 io_uring/opdef.h    |  1 +
 2 files changed, 23 insertions(+), 3 deletions(-)
io_uring/io_uring.c +0 -0
index 0f9f6a173e66..9799a31a2b29 100644
@@ -1935,14 +1935,31 @@ struct file *io_file_get_normal(struct io_kiocb *req, int fd)
        return file;

io_uring/io_uring.c struct file *io_file_get_normal(struct io_kiocb *req, int fd) · line 1944

Quoted 10 lines
io_uring/io_uring.c +7 -1
 }
 
-static void io_queue_async(struct io_kiocb *req, int ret)
+static int io_req_sqe_copy(struct io_kiocb *req, unsigned int issue_flags)
+{
+       const struct io_cold_def *def = &io_cold_defs[req->opcode];
+
+       if (!def->sqe_copy)
+               return 0;
+       if (WARN_ON_ONCE(!(issue_flags & IO_URING_F_INLINE)))

I'm pretty confident that every initial async path under io_submit_sqe() will call io_req_sqe_copy(). But I'm not positive that io_req_sqe_copy() won't get called *additional* times from non-inline contexts. One example scenario:
- io_submit_sqe() calls io_queue_sqe()
- io_issue_sqe() returns -EAGAIN, so io_queue_sqe() calls io_queue_async()
- io_queue_async() calls io_req_sqe_copy() in inline context
- io_queue_async() calls io_arm_poll_handler(), which returns IO_APOLL_READY, so io_req_task_queue() is called
- Some other I/O to the file (possibly on a different task) clears the ready poll events
- io_req_task_submit() calls io_queue_sqe() in task work context
- io_issue_sqe() returns -EAGAIN again, so io_queue_async() is called
- io_queue_async() calls io_req_sqe_copy() a second time in non-inline (task work) context

Quoted 4 lines

If this is indeed possible, then I think we may need to relax this check so it only verifies that IO_URING_F_INLINE is set *the first time* io_req_sqe_copy() is called for a given req. (Or just remove the IO_URING_F_INLINE check entirely.)

Yes, the check is a bit eager indeed. I've added a flag for this, so that we only go through the IO_URING_F_INLINE check and ->sqe_copy() callback once.

Quoted 42 lines
Quoted 42 lines
io_uring/io_uring.c +14 -1
+               return -EFAULT;
+       def->sqe_copy(req);
+       return 0;
+}
+
+static void io_queue_async(struct io_kiocb *req, unsigned int issue_flags, int ret)
        __must_hold(&req->ctx->uring_lock)
 {
        if (ret != -EAGAIN || (req->flags & REQ_F_NOWAIT)) {
+fail:
                io_req_defer_failed(req, ret);
                return;
        }
 
+       ret = io_req_sqe_copy(req, issue_flags);
+       if (unlikely(ret))
+               goto fail;
+
        switch (io_arm_poll_handler(req, 0)) {
        case IO_APOLL_READY:
                io_kbuf_recycle(req, 0);
@@ -1971,7 +1988,7 @@ static inline void io_queue_sqe(struct io_kiocb *req, unsigned int extra_flags)
         * doesn't support non-blocking read/write attempts
         */
        if (unlikely(ret))
-               io_queue_async(req, ret);
+               io_queue_async(req, issue_flags, ret);
 }
 
 static void io_queue_sqe_fallback(struct io_kiocb *req)
@@ -1986,6 +2003,8 @@ static void io_queue_sqe_fallback(struct io_kiocb *req)
                req->flags |= REQ_F_LINK;
                io_req_defer_failed(req, req->cqe.res);
        } else {
+               /* can't fail with IO_URING_F_INLINE */
+               io_req_sqe_copy(req, IO_URING_F_INLINE);
                if (unlikely(req->ctx->drain_active))
                        io_drain_req(req);
                else
@@ -2201,7 +2220,7 @@ static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
                link->last = req;
 
Quoted 7 lines

io_uring/io_uring.c static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, · line 2223

Quoted 3 lines
io_uring/io_uring.c +1 -1
                if (req->flags & IO_REQ_LINK_FLAGS)
-                       return 0;
+                       return io_req_sqe_copy(req, IO_URING_F_INLINE);

I still think this misses the last req in a linked chain, which will be issued async but won't have IO_REQ_LINK_FLAGS set. Am I missing something?

Indeed, we need to call this before that section. Fixed that up too, thanks.

Signature · 1 line
Jens Axboe
author maintainer
#6 lore raw

PATCH v3 3/4 io_uring/uring_cmd: get rid of io_uring_cmd_prep_setup()

In reply to Jens Axboe

It's a pretty pointless helper, just allocates and copies data. Fold it into io_uring_cmd_prep().

---
 io_uring/uring_cmd.c | 34 +++++++++++++---------------------
 1 file changed, 13 insertions(+), 21 deletions(-)
io_uring/uring_cmd.c +13 -21
index 929cad6ee326..e204f4941d72 100644
@@ -181,8 +181,7 @@ void io_uring_cmd_done(struct io_uring_cmd *ioucmd, ssize_t ret, u64 res2,
 }
 EXPORT_SYMBOL_GPL(io_uring_cmd_done);
 
-static int io_uring_cmd_prep_setup(struct io_kiocb *req,
-				   const struct io_uring_sqe *sqe)
+int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
 	struct io_async_cmd *ac;
@@ -190,6 +189,18 @@ static int io_uring_cmd_prep_setup(struct io_kiocb *req,
 	/* see io_uring_cmd_get_async_data() */
 	BUILD_BUG_ON(offsetof(struct io_async_cmd, data) != 0);
 
+	if (sqe->__pad1)
+		return -EINVAL;
+
+	ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags);
+	if (ioucmd->flags & ~IORING_URING_CMD_MASK)
+		return -EINVAL;
+
+	if (ioucmd->flags & IORING_URING_CMD_FIXED)
+		req->buf_index = READ_ONCE(sqe->buf_index);
+
+	ioucmd->cmd_op = READ_ONCE(sqe->cmd_op);
+
 	ac = io_uring_alloc_async_data(&req->ctx->cmd_cache, req);
 	if (!ac)
 		return -ENOMEM;
@@ -207,25 +218,6 @@ static int io_uring_cmd_prep_setup(struct io_kiocb *req,
 	return 0;
 }
 
-int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
-{
-	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
-
-	if (sqe->__pad1)
-		return -EINVAL;
-
-	ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags);
-	if (ioucmd->flags & ~IORING_URING_CMD_MASK)
-		return -EINVAL;
-
-	if (ioucmd->flags & IORING_URING_CMD_FIXED)
-		req->buf_index = READ_ONCE(sqe->buf_index);
-
-	ioucmd->cmd_op = READ_ONCE(sqe->cmd_op);
-
-	return io_uring_cmd_prep_setup(req, sqe);
-}
-
 int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
Signature · 1 line
2.49.0
author maintainer
#8 lore raw

PATCH v3 4/4 io_uring/uring_cmd: implement ->sqe_copy() to avoid unnecessary copies

In reply to Jens Axboe

uring_cmd currently copies the full SQE at prep time, just in case it needs it to be stable. However, for inline completions or requests that get queued up on the device side, there's no need to ever copy the SQE. This is particularly important, as various use cases of uring_cmd will be using 128b sized SQEs.

Opt in to using ->sqe_copy() to let the core of io_uring decide when to copy SQEs, rather than to it upfront unconditionally.

This provides two checks to see if ioucmd->sqe is still valid:

1) IO_URING_F_INLINE must be set, indicating the ->sqe_copy() call is happening inline from the syscal submitting the IO. As long as we're in that context, the SQE cannot have been reused.

2) If the SQE being passed in is NULL, then we're off the task_work submission path. This check could be combined with IO_URING_F_INLINE, but it'd require an additional branch-and-check in SQE queueing.

If either of these aren't true and the SQE hasn't been copied already, then fail the request with -EFAULT and trigger a WARN_ON_ONCE() to indicate that there's a bug to figure out. With that, it should not be possible to ever reuse an SQE outside of the direct syscall path.

---
 io_uring/opdef.c     |  1 +
 io_uring/uring_cmd.c | 21 ++++++++++++---------
 io_uring/uring_cmd.h |  1 +
 3 files changed, 14 insertions(+), 9 deletions(-)
io_uring/opdef.c +1 -0
index 6e0882b051f9..287f9a23b816 100644
@@ -759,6 +759,7 @@ const struct io_cold_def io_cold_defs[] = {
 	},
 	[IORING_OP_URING_CMD] = {
 		.name			= "URING_CMD",
+		.sqe_copy		= io_uring_cmd_sqe_copy,
 		.cleanup		= io_uring_cmd_cleanup,
 	},
 	[IORING_OP_SEND_ZC] = {
io_uring/uring_cmd.c +12 -9
index e204f4941d72..a99dc2f9c4b5 100644
@@ -205,17 +205,20 @@ int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	if (!ac)
 		return -ENOMEM;
 	ac->data.op_data = NULL;
+	ioucmd->sqe = sqe;
+	return 0;
+}
+
+void io_uring_cmd_sqe_copy(struct io_kiocb *req)
+{
+	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
+	struct io_async_cmd *ac = req->async_data;
 
-	/*
-	 * Unconditionally cache the SQE for now - this is only needed for
-	 * requests that go async, but prep handlers must ensure that any
-	 * sqe data is stable beyond prep. Since uring_cmd is special in
-	 * that it doesn't read in per-op data, play it safe and ensure that
-	 * any SQE data is stable beyond prep. This can later get relaxed.
-	 */
-	memcpy(ac->sqes, sqe, uring_sqe_size(req->ctx));
+	/* already copied, nothing to do */
+	if (ioucmd->sqe == ac->sqes)
+		return;
+	memcpy(ac->sqes, ioucmd->sqe, uring_sqe_size(req->ctx));
 	ioucmd->sqe = ac->sqes;
-	return 0;
 }
 
 int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
io_uring/uring_cmd.h +1 -0
index e6a5142c890e..a6dad47afc6b 100644
@@ -11,6 +11,7 @@ struct io_async_cmd {
 
 int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags);
 int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+void io_uring_cmd_sqe_copy(struct io_kiocb *req);
 void io_uring_cmd_cleanup(struct io_kiocb *req);
 
 bool io_uring_try_cancel_uring_cmd(struct io_ring_ctx *ctx,
Signature · 1 line
2.49.0
Reviewed-by on v3 4/4
#9 lore raw

Re: [PATCH 4/4] io_uring/uring_cmd: implement ->sqe_copy() to avoid unnecessary copies

Quoted 20 lines · Jens Axboe

uring_cmd currently copies the full SQE at prep time, just in case it needs it to be stable. However, for inline completions or requests that get queued up on the device side, there's no need to ever copy the SQE. This is particularly important, as various use cases of uring_cmd will be using 128b sized SQEs.

Opt in to using ->sqe_copy() to let the core of io_uring decide when to copy SQEs, rather than to it upfront unconditionally.

This provides two checks to see if ioucmd->sqe is still valid:

1) IO_URING_F_INLINE must be set, indicating the ->sqe_copy() call is happening inline from the syscal submitting the IO. As long as we're in that context, the SQE cannot have been reused.

2) If the SQE being passed in is NULL, then we're off the task_work submission path. This check could be combined with IO_URING_F_INLINE, but it'd require an additional branch-and-check in SQE queueing.

Quoted 6 lines

If either of these aren't true and the SQE hasn't been copied already, then fail the request with -EFAULT and trigger a WARN_ON_ONCE() to indicate that there's a bug to figure out. With that, it should not be possible to ever reuse an SQE outside of the direct syscall path.

With the parts about IO_URING_F_INLINE and NULL SQE removed from the commit message,