public inbox for [email protected]
 help / color / mirror / Atom feed
From: Jens Axboe <[email protected]>
To: [email protected]
Cc: Pavel Begunkov <[email protected]>, Jens Axboe <[email protected]>
Subject: [PATCH 2/8] io_uring: allocate memory for overflowed CQEs
Date: Fri, 19 Mar 2021 14:35:10 -0600	[thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>

From: Pavel Begunkov <[email protected]>

Instead of using a request itself for overflowed CQE stashing, allocate a
separate entry. The disadvantage is that the allocation may fail and it
will be accounted as lost (see rings->cq_overflow), so we lose reliability
in case of memory pressure if the application is driving the CQ ring into
overflow. However, it opens a way for for multiple CQEs per an SQE and
even generating SQE-less CQEs.

Signed-off-by: Pavel Begunkov <[email protected]>
[axboe: use GFP_ATOMIC | __GFP_ACCOUNT]
Signed-off-by: Jens Axboe <[email protected]>
---
 fs/io_uring.c | 101 +++++++++++++++++++++++---------------------------
 1 file changed, 46 insertions(+), 55 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 28bccb5ad4e6..a91033cc62c3 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -203,6 +203,11 @@ struct io_mapped_ubuf {
 
 struct io_ring_ctx;
 
+struct io_overflow_cqe {
+	struct io_uring_cqe cqe;
+	struct list_head list;
+};
+
 struct io_rsrc_put {
 	struct list_head list;
 	union {
@@ -1413,41 +1418,33 @@ static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
 }
 
 /* Returns true if there are no backlogged entries after the flush */
-static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
-				       struct task_struct *tsk,
-				       struct files_struct *files)
+static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
 {
 	struct io_rings *rings = ctx->rings;
-	struct io_kiocb *req, *tmp;
-	struct io_uring_cqe *cqe;
 	unsigned long flags;
 	bool all_flushed, posted;
-	LIST_HEAD(list);
 
 	if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
 		return false;
 
 	posted = false;
 	spin_lock_irqsave(&ctx->completion_lock, flags);
-	list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
-		if (!io_match_task(req, tsk, files))
-			continue;
+	while (!list_empty(&ctx->cq_overflow_list)) {
+		struct io_uring_cqe *cqe = io_get_cqring(ctx);
+		struct io_overflow_cqe *ocqe;
 
-		cqe = io_get_cqring(ctx);
 		if (!cqe && !force)
 			break;
-
-		list_move(&req->compl.list, &list);
-		if (cqe) {
-			WRITE_ONCE(cqe->user_data, req->user_data);
-			WRITE_ONCE(cqe->res, req->result);
-			WRITE_ONCE(cqe->flags, req->compl.cflags);
-		} else {
-			ctx->cached_cq_overflow++;
+		ocqe = list_first_entry(&ctx->cq_overflow_list,
+					struct io_overflow_cqe, list);
+		if (cqe)
+			memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
+		else
 			WRITE_ONCE(ctx->rings->cq_overflow,
-				   ctx->cached_cq_overflow);
-		}
+				   ++ctx->cached_cq_overflow);
 		posted = true;
+		list_del(&ocqe->list);
+		kfree(ocqe);
 	}
 
 	all_flushed = list_empty(&ctx->cq_overflow_list);
@@ -1462,19 +1459,10 @@ static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
 	spin_unlock_irqrestore(&ctx->completion_lock, flags);
 	if (posted)
 		io_cqring_ev_posted(ctx);
-
-	while (!list_empty(&list)) {
-		req = list_first_entry(&list, struct io_kiocb, compl.list);
-		list_del(&req->compl.list);
-		io_put_req(req);
-	}
-
 	return all_flushed;
 }
 
-static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
-				     struct task_struct *tsk,
-				     struct files_struct *files)
+static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
 {
 	bool ret = true;
 
@@ -1482,7 +1470,7 @@ static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
 		/* iopoll syncs against uring_lock, not completion_lock */
 		if (ctx->flags & IORING_SETUP_IOPOLL)
 			mutex_lock(&ctx->uring_lock);
-		ret = __io_cqring_overflow_flush(ctx, force, tsk, files);
+		ret = __io_cqring_overflow_flush(ctx, force);
 		if (ctx->flags & IORING_SETUP_IOPOLL)
 			mutex_unlock(&ctx->uring_lock);
 	}
@@ -1543,29 +1531,33 @@ static void __io_cqring_fill_event(struct io_kiocb *req, long res,
 		WRITE_ONCE(cqe->user_data, req->user_data);
 		WRITE_ONCE(cqe->res, res);
 		WRITE_ONCE(cqe->flags, cflags);
-	} else if (ctx->cq_overflow_flushed ||
-		   atomic_read(&req->task->io_uring->in_idle)) {
-		/*
-		 * If we're in ring overflow flush mode, or in task cancel mode,
-		 * then we cannot store the request for later flushing, we need
-		 * to drop it on the floor.
-		 */
-		ctx->cached_cq_overflow++;
-		WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
-	} else {
+		return;
+	}
+	if (!ctx->cq_overflow_flushed &&
+	    !atomic_read(&req->task->io_uring->in_idle)) {
+		struct io_overflow_cqe *ocqe;
+
+		ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
+		if (!ocqe)
+			goto overflow;
 		if (list_empty(&ctx->cq_overflow_list)) {
 			set_bit(0, &ctx->sq_check_overflow);
 			set_bit(0, &ctx->cq_check_overflow);
 			ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
 		}
-		if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
-			io_clean_op(req);
-
-		req->result = res;
-		req->compl.cflags = cflags;
-		req_ref_get(req);
-		list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
+		ocqe->cqe.user_data = req->user_data;
+		ocqe->cqe.res = res;
+		ocqe->cqe.flags = cflags;
+		list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
+		return;
 	}
+overflow:
+	/*
+	 * If we're in ring overflow flush mode, or in task cancel mode,
+	 * or cannot allocate an overflow entry, then we need to drop it
+	 * on the floor.
+	 */
+	WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow);
 }
 
 static void io_cqring_fill_event(struct io_kiocb *req, long res)
@@ -2410,7 +2402,7 @@ static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
 		 * already triggered a CQE (eg in error).
 		 */
 		if (test_bit(0, &ctx->cq_check_overflow))
-			__io_cqring_overflow_flush(ctx, false, NULL, NULL);
+			__io_cqring_overflow_flush(ctx, false);
 		if (io_cqring_events(ctx))
 			break;
 
@@ -6580,7 +6572,7 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
 
 	/* if we have a backlog and couldn't flush it all, return BUSY */
 	if (test_bit(0, &ctx->sq_check_overflow)) {
-		if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
+		if (!__io_cqring_overflow_flush(ctx, false))
 			return -EBUSY;
 	}
 
@@ -6873,7 +6865,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	int ret;
 
 	do {
-		io_cqring_overflow_flush(ctx, false, NULL, NULL);
+		io_cqring_overflow_flush(ctx, false);
 		if (io_cqring_events(ctx) >= min_events)
 			return 0;
 		if (!io_run_task_work())
@@ -6905,7 +6897,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	trace_io_uring_cqring_wait(ctx, min_events);
 	do {
 		/* if we can't even flush overflow, don't wait for more */
-		if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) {
+		if (!io_cqring_overflow_flush(ctx, false)) {
 			ret = -EBUSY;
 			break;
 		}
@@ -8578,7 +8570,7 @@ static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
 	/* if force is set, the ring is going away. always drop after that */
 	ctx->cq_overflow_flushed = 1;
 	if (ctx->rings)
-		__io_cqring_overflow_flush(ctx, true, NULL, NULL);
+		__io_cqring_overflow_flush(ctx, true);
 	xa_for_each(&ctx->personalities, index, creds)
 		io_unregister_personality(ctx, index);
 	mutex_unlock(&ctx->uring_lock);
@@ -8736,7 +8728,6 @@ static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
 		ret |= io_kill_timeouts(ctx, task, files);
 		ret |= io_run_task_work();
 		ret |= io_run_ctx_fallback(ctx);
-		io_cqring_overflow_flush(ctx, true, task, files);
 		if (!ret)
 			break;
 		cond_resched();
@@ -9153,7 +9144,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 	 */
 	ret = 0;
 	if (ctx->flags & IORING_SETUP_SQPOLL) {
-		io_cqring_overflow_flush(ctx, false, NULL, NULL);
+		io_cqring_overflow_flush(ctx, false);
 
 		ret = -EOWNERDEAD;
 		if (unlikely(ctx->sq_data->thread == NULL)) {
-- 
2.31.0


  parent reply	other threads:[~2021-03-19 20:36 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-03-19 20:35 [PATCHSET v2 0/8] Poll improvements Jens Axboe
2021-03-19 20:35 ` [PATCH 1/8] io_uring: mask in error/nval/hangup consistently for poll Jens Axboe
2021-03-19 20:35 ` Jens Axboe [this message]
2021-03-19 20:35 ` [PATCH 3/8] io_uring: include cflags in completion trace event Jens Axboe
2021-03-19 20:35 ` [PATCH 4/8] io_uring: add multishot mode for IORING_OP_POLL_ADD Jens Axboe
2021-03-19 20:35 ` [PATCH 5/8] io_uring: abstract out helper for removing poll waitqs/hashes Jens Axboe
2021-03-19 20:35 ` [PATCH 6/8] io_uring: terminate multishot poll for CQ ring overflow Jens Axboe
2021-03-19 20:35 ` [PATCH 7/8] io_uring: abstract out a io_poll_find_helper() Jens Axboe
2021-03-19 20:35 ` [PATCH 8/8] io_uring: allow events and user_data update of running poll requests Jens Axboe

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox