Skip to contentSkip to conversation

io_uring: fixes for provided buffer ring

PATCH Applied Dylan Yudaken 4 participants 10 messages Jun 13, 2022 – Jun 14, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover io_uring: fixes for provided buffer ring

This fixes two problems in the new provided buffer ring feature. One is a simple arithmetic bug (I think this came out from a refactor). The other is due to type differences between head & tail, which causes it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

I will send test cases for liburing shortly.

One question might be if we should change the type of ring_entries to uint16_t in struct io_uring_buf_reg?

Dylan Yudaken (3):
io_uring: fix index calculation
io_uring: fix types in provided buffer ring
io_uring: limit size of provided buffer ring

 fs/io_uring.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

base-commit: f2906aa863381afb0015a9eb7fefad885d4e5a56

Signature · 1 line
2.30.2
#1 lore raw

PATCH 1/3 io_uring: fix index calculation

When indexing into a provided buffer ring, do not subtract 1 from the index.

Fixes c7fb19428d67 ("io_uring: add support for ring mapped supplied buffers")
---
 fs/io_uring.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/io_uring.c +1 -1
index 3aab4182fd89..9cf9aff51b70 100644
@@ -3888,7 +3888,7 @@ static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
 		buf = &br->bufs[head];
 	} else {
 		int off = head & (IO_BUFFER_LIST_BUF_PER_PAGE - 1);
-		int index = head / IO_BUFFER_LIST_BUF_PER_PAGE - 1;
+		int index = head / IO_BUFFER_LIST_BUF_PER_PAGE;
 		buf = page_address(bl->buf_pages[index]);
 		buf += off;
 	}
Signature · 1 line
2.30.2
#2 lore raw

PATCH 2/3 io_uring: fix types in provided buffer ring

In reply to Dylan Yudaken

The type of head needs to match that of tail in order for rollover and comparisons to work correctly.

Without this change the comparison of tail to head might incorrectly allow io_uring to use a buffer that userspace had not given it.

Fixes c7fb19428d67 ("io_uring: add support for ring mapped supplied buffers")
---
 fs/io_uring.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/io_uring.c +3 -3
index 9cf9aff51b70..6eea18e8330c 100644
@@ -298,8 +298,8 @@ struct io_buffer_list {
 	/* below is for ring provided buffers */
 	__u16 buf_nr_pages;
 	__u16 nr_entries;
-	__u32 head;
-	__u32 mask;
+	__u16 head;
+	__u16 mask;
 };
 
 struct io_buffer {
@@ -3876,7 +3876,7 @@ static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
 {
 	struct io_uring_buf_ring *br = bl->buf_ring;
 	struct io_uring_buf *buf;
-	__u32 head = bl->head;
+	__u16 head = bl->head;
 
 	if (unlikely(smp_load_acquire(&br->tail) == head)) {
 		io_ring_submit_unlock(req->ctx, issue_flags);
Signature · 1 line
2.30.2
#3 lore raw

PATCH 3/3 io_uring: limit size of provided buffer ring

In reply to Dylan Yudaken

The type of head and tail do not allow more than 2^15 entries in a provided buffer ring, so do not allow this.
At 2^16 while each entry can be indexed, there is no way to disambiguate full vs empty.

---
 fs/io_uring.c | 4 ++++
 1 file changed, 4 insertions(+)
fs/io_uring.c +4 -0
index 6eea18e8330c..85b116ddfd2a 100644
@@ -13002,6 +13002,10 @@ static int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 	if (!is_power_of_2(reg.ring_entries))
 		return -EINVAL;
 
+	/* cannot disambiguate full vs empty due to head/tail size */
+	if (reg.ring_entries >= 65536)
+		return -EINVAL;
+
 	if (unlikely(reg.bgid < BGID_ARRAY && !ctx->io_bl)) {
 		int ret = io_init_bl_list(ctx);
 		if (ret)
Signature · 1 line
2.30.2
Reviewed-by on the v1 series
#4 lore raw
In reply to Dylan Yudaken
Quoted 8 lines · Dylan Yudaken

This fixes two problems in the new provided buffer ring feature. One is a simple arithmetic bug (I think this came out from a refactor). The other is due to type differences between head & tail, which causes it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

Quoted 5 lines

I will send test cases for liburing shortly.

One question might be if we should change the type of ring_entries to uint16_t in struct io_uring_buf_reg?

Why not? 5.19 is just rc2 now. So we can assume there is no users using it right now I think?

#5 lore raw
Quoted 13 lines · Hao Xu
Quoted 8 lines · Dylan Yudaken

This fixes two problems in the new provided buffer ring feature. One is a simple arithmetic bug (I think this came out from a refactor). The other is due to type differences between head & tail, which causes it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

Quoted 1 line
Quoted 7 lines
Quoted 4 lines

I will send test cases for liburing shortly.

One question might be if we should change the type of ring_entries to uint16_t in struct io_uring_buf_reg?

Why not? 5.19 is just rc2 now. So we can assume there is no users using it right now I think?

It's fine to change, but might be better if we want to extend it in the future. Do other pbuf bits allow more than 2^16 buffers?

Signature · 1 line
Pavel Begunkov
#6 lore raw
Quoted 23 lines · Pavel Begunkov
Quoted 22 lines · Hao Xu
Quoted 11 lines · Dylan Yudaken

This fixes two problems in the new provided buffer ring feature. One
is a simple arithmetic bug (I think this came out from a refactor).
The other is due to type differences between head & tail, which causes
it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

Quoted 6 lines

I will send test cases for liburing shortly.

One question might be if we should change the type of ring_entries
to uint16_t in struct io_uring_buf_reg?

Quoted 7 lines
Quoted 3 lines

Why not? 5.19 is just rc2 now. So we can assume there is no users using
it right now I think?

It's fine to change, but might be better if we want to extend it in the future. Do other pbuf bits allow more than 2^16 buffers?

I guess with

+ if (reg.ring_entries >= 65536)
+ return -EINVAL;

it doesn't matter either way. we can always use those bits later if we need?

#7 lore raw
Quoted 24 lines · Dylan Yudaken
Quoted 23 lines · Pavel Begunkov
Quoted 22 lines · Hao Xu
Quoted 11 lines · Dylan Yudaken

This fixes two problems in the new provided buffer ring feature. One
is a simple arithmetic bug (I think this came out from a refactor).
The other is due to type differences between head & tail, which causes
it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

Quoted 6 lines

I will send test cases for liburing shortly.

One question might be if we should change the type of ring_entries
to uint16_t in struct io_uring_buf_reg?

Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

Why not? 5.19 is just rc2 now. So we can assume there is no users using
it right now I think?

It's fine to change, but might be better if we want to extend it in the future. Do other pbuf bits allow more than 2^16 buffers?

might be better to leave it u32 *

Quoted 7 lines

I guess with

(quoted diff) +2 -0
+	if (reg.ring_entries >= 65536)
+		return -EINVAL;

it doesn't matter either way. we can always use those bits later if we need?

I think so as well.

I was also wondering whether pbufs can potentially allow >16 bits, but taking a quick look IORING_CQE_BUFFER_SHIFT=16, so we only have 16 bits in cqe::flags for indexes we return.

Signature · 1 line
Pavel Begunkov
#8 lore raw
Quoted 40 lines · Pavel Begunkov
Quoted 24 lines · Dylan Yudaken
Quoted 23 lines · Pavel Begunkov
Quoted 22 lines · Hao Xu
Quoted 11 lines · Dylan Yudaken

This fixes two problems in the new provided buffer ring feature. One
is a simple arithmetic bug (I think this came out from a refactor).
The other is due to type differences between head & tail, which causes
it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

Quoted 6 lines

I will send test cases for liburing shortly.

One question might be if we should change the type of ring_entries
to uint16_t in struct io_uring_buf_reg?

Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

Why not? 5.19 is just rc2 now. So we can assume there is no users using
it right now I think?

It's fine to change, but might be better if we want to extend it in the future. Do other pbuf bits allow more than 2^16 buffers?

might be better to leave it u32 *

Quoted 5 lines

I guess with

+    if (reg.ring_entries >= 65536)
+        return -EINVAL;

Quoted 9 lines
Quoted 2 lines

it doesn't matter either way. we can always use those bits later if we need?

I think so as well.

I was also wondering whether pbufs can potentially allow >16 bits, but taking a quick look IORING_CQE_BUFFER_SHIFT=16, so we only have 16 bits in cqe::flags for indexes we return.

Yea, the 16 bits return index in cqe->flags is a hard limit for pbuf ring feature, but I do think it's ok since 1<<16 is already big

maintainer Applied
#9 lore raw
In reply to Dylan Yudaken
Quoted 10 lines · Dylan Yudaken

This fixes two problems in the new provided buffer ring feature. One is a simple arithmetic bug (I think this came out from a refactor). The other is due to type differences between head & tail, which causes it to sometimes reuse an old buffer incorrectly.

Patch 1&2 fix bugs
Patch 3 limits the size of the ring as it's not possible to address more entries with 16 bit head/tail

[...]

Applied, thanks!

[1/3] io_uring: fix index calculation

      commit: 97da4a537924d87e2261773f3ac9365abb191fc9
[2/3] io_uring: fix types in provided buffer ring
      commit: c6e9fa5c0ab811f4bec36a96337f4b1bb77d142c
[3/3] io_uring: limit size of provided buffer ring
      commit: f9437ac0f851cea2374d53594f52fbbefdd977bd

Best regards,

Signature · 1 line
Jens Axboe