[PATCH io_uring-7.1 v2 0/5] follow up zcrx fixes

[PATCH io_uring-7.1 v2 0/5] follow up zcrx fixes

[Pavel Begunkov]

Follow up fixes for the recent update flagged by review.

v2: reject REG_NODEV + ->rx_buf_size

Pavel Begunkov (5):
  io_uring/zcrx: reject REG_NODEV with large rx_buf_size
  io_uring/zcrx: don't use mark0 for allocating xarray
  io_uring/zcrx: don't clear not allocated niovs
  io_uring/zcrx: use dma_len for chunk size calculation
  io_uring/zcrx: use correct mmap off constants

 io_uring/zcrx.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

-- 
2.53.0

[PATCH io_uring-7.1 v2 1/5] io_uring/zcrx: reject REG_NODEV with large rx_buf_size

[Pavel Begunkov]

The copy fallback path doesn't care about the actual niov size and only
uses first PAGE_SIZE bytes, and any additional space will be wasted.
Since ZCRX_REG_NODEV solely relies on the copy path, it doesn't make
sense to support non-standard rx_buf_len. Reject it for now, and
re-enable once improved.

Fixes: c11728021d5cd ("io_uring/zcrx: implement device-less mode for zcrx")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/zcrx.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index f94f74d0f566..1ce867c68446 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -449,6 +449,8 @@ static int io_zcrx_create_area(struct io_zcrx_ifq *ifq,
 			return -EINVAL;
 		buf_size_shift = ilog2(reg->rx_buf_len);
 	}
+	if (!ifq->dev && buf_size_shift != PAGE_SHIFT)
+		return -EOPNOTSUPP;
 
 	ret = -ENOMEM;
 	area = kzalloc_obj(*area);
@@ -462,7 +464,7 @@ static int io_zcrx_create_area(struct io_zcrx_ifq *ifq,
 	if (ifq->dev)
 		area->is_mapped = true;
 
-	if (buf_size_shift > io_area_max_shift(&area->mem)) {
+	if (ifq->dev && buf_size_shift > io_area_max_shift(&area->mem)) {
 		ret = -ERANGE;
 		goto err;
 	}
-- 
2.53.0

[PATCH io_uring-7.1 v2 2/5] io_uring/zcrx: don't use mark0 for allocating xarray

[Pavel Begunkov]

XA_MARK_0 is not compatible with xarray allocating entries, use
XA_MARK_1.

Fixes: fda90d43f4fac ("io_uring/zcrx: return back two step unregistration")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/zcrx.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index 1ce867c68446..dede892bdda9 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -929,12 +929,12 @@ int io_register_zcrx(struct io_ring_ctx *ctx,
 
 static inline bool is_zcrx_entry_marked(struct io_ring_ctx *ctx, unsigned long id)
 {
-	return xa_get_mark(&ctx->zcrx_ctxs, id, XA_MARK_0);
+	return xa_get_mark(&ctx->zcrx_ctxs, id, XA_MARK_1);
 }
 
 static inline void set_zcrx_entry_mark(struct io_ring_ctx *ctx, unsigned long id)
 {
-	xa_set_mark(&ctx->zcrx_ctxs, id, XA_MARK_0);
+	xa_set_mark(&ctx->zcrx_ctxs, id, XA_MARK_1);
 }
 
 void io_terminate_zcrx(struct io_ring_ctx *ctx)
-- 
2.53.0

[PATCH io_uring-7.1 v2 3/5] io_uring/zcrx: don't clear not allocated niovs

[Pavel Begunkov]

Now that area->is_mapped is set earlier before niovs array is allocated,
io_zcrx_free_area -> io_zcrx_unmap_area in an error path can try to
clear dma addresses for unallocated niovs, fix it.

Fixes: 8c0cab0b7bf76 ("always dma map in advance")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/zcrx.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index dede892bdda9..d9174cb31a44 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -289,8 +289,10 @@ static void io_zcrx_unmap_area(struct io_zcrx_ifq *ifq,
 		return;
 	area->is_mapped = false;
 
-	for (i = 0; i < area->nia.num_niovs; i++)
-		net_mp_niov_set_dma_addr(&area->nia.niovs[i], 0);
+	if (area->nia.niovs) {
+		for (i = 0; i < area->nia.num_niovs; i++)
+			net_mp_niov_set_dma_addr(&area->nia.niovs[i], 0);
+	}
 
 	if (area->mem.is_dmabuf) {
 		io_release_dmabuf(&area->mem);
-- 
2.53.0

[PATCH io_uring-7.1 v2 4/5] io_uring/zcrx: use dma_len for chunk size calculation

[Pavel Begunkov]

Buffers are now dma-mapped earlier and we can sg_dma_len(), otherwise,
since it's walking with for_each_sgtable_dma_sg(), it might wrongfully
reject some configurations. As a bonus, it'd now be able to use larger
chunks if dma addresses are coalesced e.g by iommu.

Fixes: 8c0cab0b7bf76 ("always dma map in advance")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/zcrx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index d9174cb31a44..e1d2e1f1b766 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -63,7 +63,7 @@ static int io_area_max_shift(struct io_zcrx_mem *mem)
 	unsigned i;
 
 	for_each_sgtable_dma_sg(sgt, sg, i)
-		shift = min(shift, __ffs(sg->length));
+		shift = min(shift, __ffs(sg_dma_len(sg)));
 	return shift;
 }
 
-- 
2.53.0

[PATCH io_uring-7.1 v2 5/5] io_uring/zcrx: use correct mmap off constants

[Pavel Begunkov]

zcrx was using IORING_OFF_PBUF_SHIFT during first iterations, but there
is now a separate constant it should use. Both are 16 so it doesn't
change anything, but improve it for the future.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/zcrx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index e1d2e1f1b766..df8dae724be4 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -386,7 +386,7 @@ static int io_allocate_rbuf_ring(struct io_ring_ctx *ctx,
 		return -EINVAL;
 
 	mmap_offset = IORING_MAP_OFF_ZCRX_REGION;
-	mmap_offset += id << IORING_OFF_PBUF_SHIFT;
+	mmap_offset += id << IORING_OFF_ZCRX_SHIFT;
 
 	ret = io_create_region(ctx, &ifq->rq_region, rd, mmap_offset);
 	if (ret < 0)
-- 
2.53.0