net: qlogic: Fix error paths in ql_alloc_large_buffers()
Subsystems (3)
PATCH 1/1 net: qlogic: Fix error paths in ql_alloc_large_buffers()
ql_alloc_large_buffers() has the usual RX buffer allocation loop where it allocates skbs and maps them for DMA. It also treats failure as a fatal error.
There are (at least) three bugs in the error paths:
1. ql_free_large_buffers() assumes that the lrg_buf[] entry for the first buffer that couldn't be allocated will have .skb == NULL. But the qla_buf[] array is not zero-initialised.
2. ql_free_large_buffers() DMA-unmaps all skbs in lrg_buf[]. This is incorrect for the last allocated skb, if DMA mapping failed.
3. Commit 1acb8f2a7a9f ("net: qlogic: Fix memory leak in ql_alloc_large_buffers") added a direct call to dev_kfree_skb_any() after the skb is recorded in lrg_buf[], so ql_free_large_buffers() will double-free it.
The bugs are somewhat inter-twined, so fix them all at once:
* Clear each entry in qla_buf[] before attempting to allocate an skb for it. This goes half-way to fixing bug 1.
* Set the .skb field only after the skb is DMA-mapped. This fixes the rest.
---
This is compile-tested only, unfortunately. But if it's correct, it needs to go to stable.
Ben.
drivers/net/ethernet/qlogic/qla3xxx.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/net/ethernet/qlogic/qla3xxx.c +4 -4
@@ -2756,6 +2756,9 @@ static int ql_alloc_large_buffers(struct ql3_adapter *qdev)int err;for (i =0 ; i < qdev->num_large_buffers; i++) { + lrg_buf_cb = &qdev->lrg_buf[i]; +memset (lrg_buf_cb,0 ,sizeof (struct ql_rcv_buf_cb )); + skb =netdev_alloc_skb (qdev->ndev, qdev->lrg_buffer_len);if (unlikely (!skb)) {
@@ -2766,11 +2769,7 @@ static int ql_alloc_large_buffers(struct ql3_adapter *qdev)ql_free_large_buffers (qdev);return -ENOMEM ; }else { - - lrg_buf_cb = &qdev->lrg_buf[i]; -memset (lrg_buf_cb,0 ,sizeof (struct ql_rcv_buf_cb )); lrg_buf_cb->index = i; - lrg_buf_cb->skb = skb;/* * We save some space to copy the ethhdr from first * buffer
@@ -2792,6 +2791,7 @@ static int ql_alloc_large_buffers(struct ql3_adapter *qdev)return -ENOMEM ; } + lrg_buf_cb->skb = skb;dma_unmap_addr_set (lrg_buf_cb, mapaddr, map); dma_unmap_len_set(lrg_buf_cb, maplen, qdev->lrg_buffer_len -
David Miller
maintainerFrom: Ben Hutchings <[email protected]>
Date: Tue, 17 Dec 2019 01:57:40 +0000
Quoted 25 lines
ql_alloc_large_buffers()has the usual RX buffer allocation loop where it allocates skbs and maps them for DMA. It also treats failure as a fatal error.There are (at least) three bugs in the error paths:
1.
ql_free_large_buffers()assumes that the lrg_buf[] entry for the first buffer that couldn't be allocated will have .skb == NULL. But the qla_buf[] array is not zero-initialised.2.
ql_free_large_buffers()DMA-unmaps all skbs in lrg_buf[]. This is incorrect for the last allocated skb, if DMA mapping failed.3. Commit
1acb8f2a7a9f("net: qlogic: Fix memory leak inql_alloc_large_buffers") added a direct call todev_kfree_skb_any()after the skb is recorded in lrg_buf[], soql_free_large_buffers()will double-free it.The bugs are somewhat inter-twined, so fix them all at once:
* Clear each entry in qla_buf[] before attempting to allocate an skb for it. This goes half-way to fixing bug 1.
* Set the .skb field only after the skb is DMA-mapped. This fixes the rest.
Quoted 4 lines
Fixes 1357bfcf7106 ("qla3xxx: Dynamically size the rx buffer queue ...")Fixes 0f8ab89e825f ("qla3xxx: Check return code from pci_map_single() ...")Fixes 1acb8f2a7a9f ("net: qlogic: Fix memory leak in ql_alloc_large_buffers")Signed-off-by Ben Hutchings <[email protected]>
I've been over this a few times, applied and queued up for -stable.
Thanks.