Skip to contentSkip to conversation

io_uring: poll: remove unnecessary req->ref set

PATCH In review Hao Xu 2 participants 10 messages Jun 11, 2022 – Jun 13, 2022 io_uring

lore
Share & tools
lore

Hao Xu

author
#0 lore raw

PATCH 1/6 io_uring: poll: remove unnecessary req->ref set

From: Hao Xu <[email protected]>

We now don't need to set req->refcount for poll requests since the reworked poll code ensures no request release race.

---
 io_uring/poll.c | 1 -
 1 file changed, 1 deletion(-)
io_uring/poll.c +0 -1
index 0df5eca93b16..73584c4e3e9b 100644
@@ -683,7 +683,6 @@ int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	if ((flags & IORING_POLL_ADD_MULTI) && (req->flags & REQ_F_CQE_SKIP))
 		return -EINVAL;
 
-	io_req_set_refcount(req);
 	req->apoll_events = poll->events = io_poll_parse_events(sqe, flags);
 	return 0;
 }
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH 2/6 io_uring: openclose: fix bug of closing wrong fixed file

From: Hao Xu <[email protected]>

Don't update ret until fixed file is closed, otherwise the file slot becomes the error code.

Fixes a7c41b4687f5 ("io_uring: let IORING_OP_FILES_UPDATE support choosing fixed file slots")
---
 io_uring/rsrc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/rsrc.c +1 -1
index d78e7f2ea91f..cf8c85d1fb59 100644
@@ -705,8 +705,8 @@ static int io_files_update_with_index_alloc(struct io_kiocb *req,
 		if (ret < 0)
 			break;
 		if (copy_to_user(&fds[done], &ret, sizeof(ret))) {
-			ret = -EFAULT;
 			__io_close_fixed(req, issue_flags, ret);
+			ret = -EFAULT;
 			break;
 		}
 	}
Signature · 1 line
2.25.1
#2 lore raw

Re: [PATCH 5.19 2/6] io_uring: openclose: fix bug of closing wrong fixed file

io_uring/rsrc.c

Quoted 12 lines · Hao Xu

From: Hao Xu <[email protected]>

Don't update ret until fixed file is closed, otherwise the file slot becomes the error code.

Fixes a7c41b4687f5 ("io_uring: let IORING_OP_FILES_UPDATE support choosing fixed file slots")
---
  io_uring/rsrc.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/rsrc.c +0 -0

Something strange is going on here, io_uring/rsrc.c was only queued for 5.20, but it's marked 5.19, weird.

Quoted 13 lines

index d78e7f2ea91f..cf8c85d1fb59 100644

io_uring/rsrc.c +1 -1
@@ -705,8 +705,8 @@ static int io_files_update_with_index_alloc(struct io_kiocb *req,
  		if (ret < 0)
  			break;
  		if (copy_to_user(&fds[done], &ret, sizeof(ret))) {
-			ret = -EFAULT;
  			__io_close_fixed(req, issue_flags, ret);
+			ret = -EFAULT;
  			break;
  		}
  	}
Signature · 1 line
Pavel Begunkov
#3 lore raw

Re: [PATCH 5.19 2/6] io_uring: openclose: fix bug of closing wrong fixed file

In reply to Hao Xu
Quoted 4 lines · Hao Xu

From: Hao Xu <[email protected]>

Don't update ret until fixed file is closed, otherwise the file slot becomes the error code.

I rebased and queued this and 6/6, will send them out together later, thanks

https://github.com/isilence/linux/tree/io_uring/io_uring-5.19

Quoted 21 lines
Fixes a7c41b4687f5 ("io_uring: let IORING_OP_FILES_UPDATE support choosing fixed file slots")
---
  io_uring/rsrc.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/rsrc.c +1 -1
index d78e7f2ea91f..cf8c85d1fb59 100644
@@ -705,8 +705,8 @@ static int io_files_update_with_index_alloc(struct io_kiocb *req,
  		if (ret < 0)
  			break;
  		if (copy_to_user(&fds[done], &ret, sizeof(ret))) {
-			ret = -EFAULT;
  			__io_close_fixed(req, issue_flags, ret);
+			ret = -EFAULT;
  			break;
  		}
  	}
Signature · 1 line
Pavel Begunkov
author
#4 lore raw

Re: [PATCH 5.19 2/6] io_uring: openclose: fix bug of closing wrong fixed file

Quoted 11 lines · Pavel Begunkov
Quoted 4 lines · Hao Xu

From: Hao Xu <[email protected]>

Don't update ret until fixed file is closed, otherwise the file slot becomes the error code.

I rebased and queued this and 6/6, will send them out together later, thanks

https://github.com/isilence/linux/tree/io_uring/io_uring-5.19

Thanks for rebasing it, was busy with something yesterday.

Thanks,
Hao

author
#5 lore raw

PATCH 3/6 io_uring: openclose: fix bug of unexpected return value in IORING_CLOSE_FD_AND_FILE_SLOT mode

In reply to Hao Xu

From: Hao Xu <[email protected]>

In IORING_CLOSE_FD_AND_FILE_SLOT mode, if we successfully close fixed file but fails in close->fd >= fdt->max_fds check, cqe-res = 0 is returned, which should be -EBADF.

Fixes a7c41b4687f5 ("io_uring: let IORING_OP_FILES_UPDATE support choosing fixed file slots")
---
 io_uring/openclose.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
io_uring/openclose.c +2 -1
index b35bf5f66dd9..4eb1f23e028a 100644
@@ -248,7 +248,7 @@ int io_close(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_close *close = io_kiocb_to_cmd(req);
 	struct fdtable *fdt;
 	struct file *file;
-	int ret = -EBADF;
+	int ret;
 
 	if (close->file_slot) {
 		ret = io_close_fixed(req, issue_flags);
@@ -256,6 +256,7 @@ int io_close(struct io_kiocb *req, unsigned int issue_flags)
 			goto err;
 	}
 
+	ret = -EBADF;
 	spin_lock(&files->file_lock);
 	fdt = files_fdtable(files);
 	if (close->fd >= fdt->max_fds) {
Signature · 1 line
2.25.1
author
#6 lore raw

PATCH 4/6 io_uring: openclose: support separate return value for IORING_CLOSE_FD_AND_FILE_SLOT

In reply to Hao Xu

From: Hao Xu <[email protected]>

In IORING_CLOSE_FD_AND_FILE_SLOT mode, we just stop and return error code if either of fixed or normal file close fails. But we can actually continue to do the close even one of them fails. What we need to do is put the two result in two place: for normal file close, put the result in cqe->res like the previous behaviour, while for fixed file close, put it in cqe->flags. Users should check both member to get the status of fixed and normal file close.

---
 io_uring/openclose.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
io_uring/openclose.c +6 -3
index 4eb1f23e028a..da930081c03c 100644
@@ -248,12 +248,15 @@ int io_close(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_close *close = io_kiocb_to_cmd(req);
 	struct fdtable *fdt;
 	struct file *file;
-	int ret;
+	int ret, ret2;
 
 	if (close->file_slot) {
 		ret = io_close_fixed(req, issue_flags);
-		if (ret || !(close->flags & IORING_CLOSE_FD_AND_FILE_SLOT))
+		if (!(close->flags & IORING_CLOSE_FD_AND_FILE_SLOT))
 			goto err;
+		else
+			ret2 = ret;
+
 	}
 
 	ret = -EBADF;
@@ -286,6 +289,6 @@ int io_close(struct io_kiocb *req, unsigned int issue_flags)
 err:
 	if (ret < 0)
 		req_set_fail(req);
-	io_req_set_res(req, ret, 0);
+	io_req_set_res(req, ret, ret2);
 	return IOU_OK;
 }
Signature · 1 line
2.25.1
author
#7 lore raw

PATCH 5/6 io_uring: remove duplicate kbuf release

In reply to Hao Xu

From: Hao Xu <[email protected]>

__io_req_complete_put() is called by __io_req_complete_post, and before that we already put kbuf, no need to do that in __io_req_complete_put() again.

---
 io_uring/io_uring.c | 6 ------
 1 file changed, 6 deletions(-)
io_uring/io_uring.c +0 -6
index 1572ebe3cff1..a94ddd0ba507 100644
@@ -1245,12 +1245,6 @@ static void __io_req_complete_put(struct io_kiocb *req)
 			}
 		}
 		io_req_put_rsrc(req);
-		/*
-		 * Selected buffer deallocation in io_clean_op() assumes that
-		 * we don't hold ->completion_lock. Clean them here to avoid
-		 * deadlocks.
-		 */
-		io_put_kbuf_comp(req);
 		io_dismantle_req(req);
 		io_put_task(req->task, 1);
 		wq_list_add_head(&req->comp_list, &ctx->locked_free_list);
Signature · 1 line
2.25.1
author
#8 lore raw

PATCH 6/6 io_uring: kbuf: fix bug of not consuming ring buffer in partial io case

In reply to Hao Xu

From: Hao Xu <[email protected]>

When we use ring-mapped provided buffer, we should consume it before arm poll if partial io has been done. Otherwise the buffer may be used by other requests and thus we lost the data.

Fixes c7fb19428d67 ("io_uring: add support for ring mapped supplied buffers")
---
 io_uring/kbuf.c |  9 +++++++--
 io_uring/kbuf.h | 10 ++++++++--
 2 files changed, 15 insertions(+), 4 deletions(-)
io_uring/kbuf.c +7 -2
index d2b2b4728381..0680b63af1d2 100644
@@ -49,8 +49,13 @@ void __io_kbuf_recycle(struct io_kiocb *req, unsigned issue_flags)
 	 */
 	if (req->flags & REQ_F_BUFFER_RING) {
 		if (req->buf_list) {
-			req->buf_index = req->buf_list->bgid;
-			req->flags &= ~REQ_F_BUFFER_RING;
+			if (req->flags & REQ_F_PARTIAL_IO) {
+				req->buf_list->head++;
+				req->buf_list = NULL;
+			} else {
+				req->buf_index = req->buf_list->bgid;
+				req->flags &= ~REQ_F_BUFFER_RING;
+			}
 		}
 		return;
 	}
io_uring/kbuf.h +8 -2
index b58d9d20c97e..9ecb175e60a9 100644
@@ -58,8 +58,14 @@ static inline void io_kbuf_recycle(struct io_kiocb *req, unsigned issue_flags)
 {
 	if (!(req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING)))
 		return;
-	/* don't recycle if we already did IO to this buffer */
-	if (req->flags & REQ_F_PARTIAL_IO)
+	/*
+	 * For legacy provided buffer mode, don't recycle if we already did
+	 * IO to this buffer. For ring-mapped provided buffer mode, we should
+	 * increment ring->head to explicitly monopolize the buffer to avoid
+	 * multiple use.
+	 */
+	if ((req->flags & REQ_F_BUFFER_SELECTED) &&
+	    (req->flags & REQ_F_PARTIAL_IO))
 		return;
 	__io_kbuf_recycle(req, issue_flags);
 }
Signature · 1 line
2.25.1
#9 lore raw
In reply to Hao Xu
Quoted 4 lines · Hao Xu

From: Hao Xu <[email protected]>

We now don't need to set req->refcount for poll requests since the reworked poll code ensures no request release race.

Nice, I took this one into the 5.20 branch, cheers

https://github.com/isilence/linux/tree/io_uring/for-5.20

Quoted 18 lines
---
  io_uring/poll.c | 1 -
  1 file changed, 1 deletion(-)
io_uring/poll.c +0 -1
index 0df5eca93b16..73584c4e3e9b 100644
@@ -683,7 +683,6 @@ int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
  	if ((flags & IORING_POLL_ADD_MULTI) && (req->flags & REQ_F_CQE_SKIP))
  		return -EINVAL;
 
-	io_req_set_refcount(req);
  	req->apoll_events = poll->events = io_poll_parse_events(sqe, flags);
  	return 0;
  }
Signature · 1 line
Pavel Begunkov