All of lore.kernel.org
 help / color / mirror / Atom feed
From: Jens Axboe <axboe@kernel.dk>
To: io-uring@vger.kernel.org
Cc: jannh@google.com, Jens Axboe <axboe@kernel.dk>
Subject: [PATCH 2/3] io_uring/register: document io_register_resize_rings() shared mem usage
Date: Wed, 15 Jan 2025 09:06:02 -0700	[thread overview]
Message-ID: <20250115160801.43369-3-axboe@kernel.dk> (raw)
In-Reply-To: <20250115160801.43369-1-axboe@kernel.dk>

It can be a bit hard to tell which parts of io_register_resize_rings()
are operating on shared memory, and which ones are not. And anything
reading or writing to those regions should really use the read/write
once primitives.

Hence add those, ensuring sanity in how this memory is accessed, and
helping document the shared nature of it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/register.c | 44 ++++++++++++++++++++++++++------------------
 1 file changed, 26 insertions(+), 18 deletions(-)

diff --git a/io_uring/register.c b/io_uring/register.c
index 5880eb75ae44..ffcbc840032e 100644
--- a/io_uring/register.c
+++ b/io_uring/register.c
@@ -449,10 +449,18 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 	if (IS_ERR(n.rings))
 		return PTR_ERR(n.rings);
 
-	n.rings->sq_ring_mask = p.sq_entries - 1;
-	n.rings->cq_ring_mask = p.cq_entries - 1;
-	n.rings->sq_ring_entries = p.sq_entries;
-	n.rings->cq_ring_entries = p.cq_entries;
+	/*
+	 * At this point n.rings is shared with userspace, just like o.rings
+	 * is as well. While we don't expect userspace to modify it while
+	 * a resize is in progress, and it's most likely that userspace will
+	 * shoot itself in the foot if it does, we can't always assume good
+	 * intent... Use read/write once helpers from here on to indicate the
+	 * shared nature of it.
+	 */
+	WRITE_ONCE(n.rings->sq_ring_mask, p.sq_entries - 1);
+	WRITE_ONCE(n.rings->cq_ring_mask, p.cq_entries - 1);
+	WRITE_ONCE(n.rings->sq_ring_entries, p.sq_entries);
+	WRITE_ONCE(n.rings->cq_ring_entries, p.cq_entries);
 
 	if (copy_to_user(arg, &p, sizeof(p))) {
 		io_register_free_rings(&p, &n);
@@ -509,20 +517,20 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 	 * rings can't hold what is already there, then fail the operation.
 	 */
 	n.sq_sqes = ptr;
-	tail = o.rings->sq.tail;
-	if (tail - o.rings->sq.head > p.sq_entries)
+	tail = READ_ONCE(o.rings->sq.tail);
+	if (tail - READ_ONCE(o.rings->sq.head) > p.sq_entries)
 		goto overflow;
-	for (i = o.rings->sq.head; i < tail; i++) {
+	for (i = READ_ONCE(o.rings->sq.head); i < tail; i++) {
 		unsigned src_head = i & (ctx->sq_entries - 1);
 		unsigned dst_head = i & (p.sq_entries - 1);
 
 		n.sq_sqes[dst_head] = o.sq_sqes[src_head];
 	}
-	n.rings->sq.head = o.rings->sq.head;
-	n.rings->sq.tail = o.rings->sq.tail;
+	WRITE_ONCE(n.rings->sq.head, READ_ONCE(o.rings->sq.head));
+	WRITE_ONCE(n.rings->sq.tail, READ_ONCE(o.rings->sq.tail));
 
-	tail = o.rings->cq.tail;
-	if (tail - o.rings->cq.head > p.cq_entries) {
+	tail = READ_ONCE(o.rings->cq.tail);
+	if (tail - READ_ONCE(o.rings->cq.head) > p.cq_entries) {
 overflow:
 		/* restore old rings, and return -EOVERFLOW via cleanup path */
 		ctx->rings = o.rings;
@@ -531,21 +539,21 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 		ret = -EOVERFLOW;
 		goto out;
 	}
-	for (i = o.rings->cq.head; i < tail; i++) {
+	for (i = READ_ONCE(o.rings->cq.head); i < tail; i++) {
 		unsigned src_head = i & (ctx->cq_entries - 1);
 		unsigned dst_head = i & (p.cq_entries - 1);
 
 		n.rings->cqes[dst_head] = o.rings->cqes[src_head];
 	}
-	n.rings->cq.head = o.rings->cq.head;
-	n.rings->cq.tail = o.rings->cq.tail;
+	WRITE_ONCE(n.rings->cq.head, READ_ONCE(o.rings->cq.head));
+	WRITE_ONCE(n.rings->cq.tail, READ_ONCE(o.rings->cq.tail));
 	/* invalidate cached cqe refill */
 	ctx->cqe_cached = ctx->cqe_sentinel = NULL;
 
-	n.rings->sq_dropped = o.rings->sq_dropped;
-	n.rings->sq_flags = o.rings->sq_flags;
-	n.rings->cq_flags = o.rings->cq_flags;
-	n.rings->cq_overflow = o.rings->cq_overflow;
+	WRITE_ONCE(n.rings->sq_dropped, READ_ONCE(o.rings->sq_dropped));
+	WRITE_ONCE(n.rings->sq_flags, READ_ONCE(o.rings->sq_flags));
+	WRITE_ONCE(n.rings->cq_flags, READ_ONCE(o.rings->cq_flags));
+	WRITE_ONCE(n.rings->cq_overflow, READ_ONCE(o.rings->cq_overflow));
 
 	/* all done, store old pointers and assign new ones */
 	if (!(ctx->flags & IORING_SETUP_NO_SQARRAY))
-- 
2.47.1


  parent reply	other threads:[~2025-01-15 16:08 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-01-15 16:06 [PATCHSET 0/3] Ring resizing fixes Jens Axboe
2025-01-15 16:06 ` [PATCH 1/3] io_uring/register: use stable SQ/CQ ring data during resize Jens Axboe
2025-01-15 16:06 ` Jens Axboe [this message]
2025-01-15 16:06 ` [PATCH 3/3] io_uring/register: cache old SQ/CQ head reading for copies 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 \
    --in-reply-to=20250115160801.43369-3-axboe@kernel.dk \
    --to=axboe@kernel.dk \
    --cc=io-uring@vger.kernel.org \
    --cc=jannh@google.com \
    /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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.