[RFC] kni: properly translate pa2va for cloned mbuf

Message ID 201909101917.x8AJH281027807@lectura.cs.arizona.edu
State New
Delegated to: Thomas Monjalon
Headers show
Series
  • [RFC] kni: properly translate pa2va for cloned mbuf
Related show

Checks

Context Check Description
ci/Intel-compilation success Compilation OK

Commit Message

Junxiao Shi Sept. 10, 2019, 6:58 p.m.
Previously, KNI kernel module uses the difference between m->buf_addr
and m->buf_iova to calculate userspace virtual address from physical
address. This works for direct mbufs, but does not work for indirect
(cloned) mbufs that come from another mempool. Transmitting a cloned
mbuf may cause segmentation fault in userspace.

Now, userspace KNI library writes the virtual address of each mbuf
in m->userdata field, and KNI kernel module uses this field to restore
virtual address before putting mbuf into free_q. This approach works
for both direct and indirect mbufs.

NOTE TO REVIEWER - DO NOT MERGE
The idea of this change is at https://bugs.dpdk.org/show_bug.cgi?id=183#c4
Test case is at https://bugs.dpdk.org/show_bug.cgi?id=183#c5
I only modified kni_net_rx_normal function.
If this approach is acceptable, I will modify kni_net_rx_lo_fifo,
kni_net_rx_lo_fifo_skb, and kni_fifo_trans_pa2va(rx_q) as well.

Bugzilla ID: 183

Signed-off-by: Junxiao Shi <git@mail1.yoursunny.com>
---
 kernel/linux/kni/kni_net.c                        | 4 ++--
 lib/librte_eal/linux/eal/include/rte_kni_common.h | 3 ++-
 lib/librte_kni/rte_kni.c                          | 8 ++++++--
 3 files changed, 10 insertions(+), 5 deletions(-)

Patch

diff --git a/kernel/linux/kni/kni_net.c b/kernel/linux/kni/kni_net.c
index 7bd3a9f1e..b34ed2ed4 100644
--- a/kernel/linux/kni/kni_net.c
+++ b/kernel/linux/kni/kni_net.c
@@ -341,7 +341,7 @@  kni_net_rx_normal(struct kni_dev *kni)
 		kva = pa2kva(kni->pa[i]);
 		len = kva->pkt_len;
 		data_kva = kva2data_kva(kva);
-		kni->va[i] = pa2va(kni->pa[i], kva);
+		kni->va[i] = kva->va;
 
 		skb = netdev_alloc_skb(dev, len);
 		if (!skb) {
@@ -367,7 +367,7 @@  kni_net_rx_normal(struct kni_dev *kni)
 				kva = pa2kva(kva->next);
 				data_kva = kva2data_kva(kva);
 				/* Convert physical address to virtual address */
-				prev_kva->next = pa2va(prev_kva->next, kva);
+				prev_kva->next = kva->va;
 			}
 		}
 
diff --git a/lib/librte_eal/linux/eal/include/rte_kni_common.h b/lib/librte_eal/linux/eal/include/rte_kni_common.h
index 37d9ee8f0..bc92c0067 100644
--- a/lib/librte_eal/linux/eal/include/rte_kni_common.h
+++ b/lib/librte_eal/linux/eal/include/rte_kni_common.h
@@ -84,7 +84,8 @@  struct rte_kni_mbuf {
 	uint16_t data_len;      /**< Amount of data in segment buffer. */
 
 	/* fields on second cache line */
-	char pad3[8] __attribute__((__aligned__(RTE_CACHE_LINE_MIN_SIZE)));
+	void *va __attribute__((__aligned__(RTE_CACHE_LINE_MIN_SIZE)));
+	         /**< Virtual address of this mbuf in userspace (overwrites userdata). */
 	void *pool;
 	void *next;             /**< Physical address of next mbuf in kernel. */
 };
diff --git a/lib/librte_kni/rte_kni.c b/lib/librte_kni/rte_kni.c
index 4b51fb4fe..96a6f6af2 100644
--- a/lib/librte_kni/rte_kni.c
+++ b/lib/librte_kni/rte_kni.c
@@ -352,13 +352,15 @@  va2pa(struct rte_mbuf *m)
 static void *
 va2pa_all(struct rte_mbuf *mbuf)
 {
-	void *phy_mbuf = va2pa(mbuf);
+	void *phy_mbuf = (void*)rte_mempool_virt2iova(mbuf);
 	struct rte_mbuf *next = mbuf->next;
 	while (next) {
-		mbuf->next = va2pa(next);
+		mbuf->userdata = mbuf;
+		mbuf->next = (void*)rte_mempool_virt2iova(next);
 		mbuf = next;
 		next = mbuf->next;
 	}
+	mbuf->userdata = mbuf;
 	return phy_mbuf;
 }
 
@@ -609,6 +611,8 @@  kni_allocate_mbufs(struct rte_kni *kni)
 			 offsetof(struct rte_kni_mbuf, buf_addr));
 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, next) !=
 			 offsetof(struct rte_kni_mbuf, next));
+	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, userdata) !=
+			 offsetof(struct rte_kni_mbuf, va));
 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) !=
 			 offsetof(struct rte_kni_mbuf, data_off));
 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_len) !=