From 8c83d39cc730378bbac64d67a551897b203a606e Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Sat, 15 Jan 2022 18:02:33 -0500 Subject: IB/hfi1: Fix panic with larger ipoib send_queue_size When the ipoib send_queue_size is increased from the default the following panic happens: RIP: 0010:hfi1_ipoib_drain_tx_ring+0x45/0xf0 [hfi1] Code: 31 e4 eb 0f 8b 85 c8 02 00 00 41 83 c4 01 44 39 e0 76 60 8b 8d cc 02 00 00 44 89 e3 be 01 00 00 00 d3 e3 48 03 9d c0 02 00 00 83 18 01 00 00 00 00 00 00 48 8b bb 30 01 00 00 e8 25 af a7 e0 RSP: 0018:ffffc9000798f4a0 EFLAGS: 00010286 RAX: 0000000000008000 RBX: ffffc9000aa0f000 RCX: 000000000000000f RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000000 RBP: ffff88810ff08000 R08: ffff88889476d900 R09: 0000000000000101 R10: 0000000000000000 R11: ffffc90006590ff8 R12: 0000000000000200 R13: ffffc9000798fba8 R14: 0000000000000000 R15: 0000000000000001 FS: 00007fd0f79cc3c0(0000) GS:ffff88885fb00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: ffffc9000aa0f118 CR3: 0000000889c84001 CR4: 00000000001706e0 Call Trace: hfi1_ipoib_napi_tx_disable+0x45/0x60 [hfi1] hfi1_ipoib_dev_stop+0x18/0x80 [hfi1] ipoib_ib_dev_stop+0x1d/0x40 [ib_ipoib] ipoib_stop+0x48/0xc0 [ib_ipoib] __dev_close_many+0x9e/0x110 __dev_change_flags+0xd9/0x210 dev_change_flags+0x21/0x60 do_setlink+0x31c/0x10f0 ? __nla_validate_parse+0x12d/0x1a0 ? __nla_parse+0x21/0x30 ? inet6_validate_link_af+0x5e/0xf0 ? cpumask_next+0x1f/0x20 ? __snmp6_fill_stats64.isra.53+0xbb/0x140 ? __nla_validate_parse+0x47/0x1a0 __rtnl_newlink+0x530/0x910 ? pskb_expand_head+0x73/0x300 ? __kmalloc_node_track_caller+0x109/0x280 ? __nla_put+0xc/0x20 ? cpumask_next_and+0x20/0x30 ? update_sd_lb_stats.constprop.144+0xd3/0x820 ? _raw_spin_unlock_irqrestore+0x25/0x37 ? __wake_up_common_lock+0x87/0xc0 ? kmem_cache_alloc_trace+0x3d/0x3d0 rtnl_newlink+0x43/0x60 The issue happens when the shift that should have been a function of the txq item size mistakenly used the ring size. Fix by using the item size. Cc: stable@vger.kernel.org Fixes: d47dfc2b00e6 ("IB/hfi1: Remove cache and embed txreq in ring") Link: https://lore.kernel.org/r/1642287756-182313-2-git-send-email-mike.marciniszyn@cornelisnetworks.com Reviewed-by: Dennis Dalessandro Signed-off-by: Mike Marciniszyn Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hfi1/ipoib_tx.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/hfi1/ipoib_tx.c b/drivers/infiniband/hw/hfi1/ipoib_tx.c index f4010890309f..bf62956c8667 100644 --- a/drivers/infiniband/hw/hfi1/ipoib_tx.c +++ b/drivers/infiniband/hw/hfi1/ipoib_tx.c @@ -731,7 +731,7 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv) goto free_txqs; txq->tx_ring.max_items = tx_ring_size; - txq->tx_ring.shift = ilog2(tx_ring_size); + txq->tx_ring.shift = ilog2(tx_item_size); txq->tx_ring.avail = hfi1_ipoib_ring_hwat(txq); netif_tx_napi_add(dev, &txq->napi, -- cgit v1.2.3 From b1151b74ff68cc83c2a8e1a618efe7d056e4f237 Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Sat, 15 Jan 2022 18:02:34 -0500 Subject: IB/hfi1: Fix alloc failure with larger txqueuelen The following allocation with large txqueuelen will result in the following warning: Call Trace: __alloc_pages_nodemask+0x283/0x2c0 kmalloc_large_node+0x3c/0xa0 __kmalloc_node+0x22a/0x2f0 hfi1_ipoib_txreq_init+0x19f/0x330 [hfi1] hfi1_ipoib_setup_rn+0xd3/0x1a0 [hfi1] rdma_init_netdev+0x5a/0x80 [ib_core] ipoib_intf_init+0x6c/0x350 [ib_ipoib] ipoib_intf_alloc+0x5c/0xc0 [ib_ipoib] ipoib_add_one+0xbe/0x300 [ib_ipoib] add_client_context+0x12c/0x1a0 [ib_core] ib_register_client+0x147/0x190 [ib_core] ipoib_init_module+0xdd/0x132 [ib_ipoib] do_one_initcall+0x46/0x1c3 do_init_module+0x5a/0x220 load_module+0x14c5/0x17f0 __do_sys_init_module+0x13b/0x180 do_syscall_64+0x5b/0x1a0 entry_SYSCALL_64_after_hwframe+0x65/0xca For ipoib, the txqueuelen is modified with the module parameter send_queue_size. Fix by changing to use kv versions of the same allocator to handle the large allocations. The allocation embeds a hdr struct that is dma mapped. Change that struct to a pointer to a kzalloced struct. Cc: stable@vger.kernel.org Fixes: d99dc602e2a5 ("IB/hfi1: Add functions to transmit datagram ipoib packets") Link: https://lore.kernel.org/r/1642287756-182313-3-git-send-email-mike.marciniszyn@cornelisnetworks.com Reviewed-by: Dennis Dalessandro Signed-off-by: Mike Marciniszyn Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hfi1/ipoib.h | 2 +- drivers/infiniband/hw/hfi1/ipoib_tx.c | 36 ++++++++++++++++++++++++----------- 2 files changed, 26 insertions(+), 12 deletions(-) diff --git a/drivers/infiniband/hw/hfi1/ipoib.h b/drivers/infiniband/hw/hfi1/ipoib.h index 909122934246..aec60d4888eb 100644 --- a/drivers/infiniband/hw/hfi1/ipoib.h +++ b/drivers/infiniband/hw/hfi1/ipoib.h @@ -55,7 +55,7 @@ union hfi1_ipoib_flow { */ struct ipoib_txreq { struct sdma_txreq txreq; - struct hfi1_sdma_header sdma_hdr; + struct hfi1_sdma_header *sdma_hdr; int sdma_status; int complete; struct hfi1_ipoib_dev_priv *priv; diff --git a/drivers/infiniband/hw/hfi1/ipoib_tx.c b/drivers/infiniband/hw/hfi1/ipoib_tx.c index bf62956c8667..d6bbdb8fcb50 100644 --- a/drivers/infiniband/hw/hfi1/ipoib_tx.c +++ b/drivers/infiniband/hw/hfi1/ipoib_tx.c @@ -122,7 +122,7 @@ static void hfi1_ipoib_free_tx(struct ipoib_txreq *tx, int budget) dd_dev_warn(priv->dd, "%s: Status = 0x%x pbc 0x%llx txq = %d sde = %d\n", __func__, tx->sdma_status, - le64_to_cpu(tx->sdma_hdr.pbc), tx->txq->q_idx, + le64_to_cpu(tx->sdma_hdr->pbc), tx->txq->q_idx, tx->txq->sde->this_idx); } @@ -231,7 +231,7 @@ static int hfi1_ipoib_build_tx_desc(struct ipoib_txreq *tx, { struct hfi1_devdata *dd = txp->dd; struct sdma_txreq *txreq = &tx->txreq; - struct hfi1_sdma_header *sdma_hdr = &tx->sdma_hdr; + struct hfi1_sdma_header *sdma_hdr = tx->sdma_hdr; u16 pkt_bytes = sizeof(sdma_hdr->pbc) + (txp->hdr_dwords << 2) + tx->skb->len; int ret; @@ -256,7 +256,7 @@ static void hfi1_ipoib_build_ib_tx_headers(struct ipoib_txreq *tx, struct ipoib_txparms *txp) { struct hfi1_ipoib_dev_priv *priv = tx->txq->priv; - struct hfi1_sdma_header *sdma_hdr = &tx->sdma_hdr; + struct hfi1_sdma_header *sdma_hdr = tx->sdma_hdr; struct sk_buff *skb = tx->skb; struct hfi1_pportdata *ppd = ppd_from_ibp(txp->ibp); struct rdma_ah_attr *ah_attr = txp->ah_attr; @@ -483,7 +483,7 @@ static int hfi1_ipoib_send_dma_single(struct net_device *dev, if (likely(!ret)) { tx_ok: trace_sdma_output_ibhdr(txq->priv->dd, - &tx->sdma_hdr.hdr, + &tx->sdma_hdr->hdr, ib_is_sc5(txp->flow.sc5)); hfi1_ipoib_check_queue_depth(txq); return NETDEV_TX_OK; @@ -547,7 +547,7 @@ static int hfi1_ipoib_send_dma_list(struct net_device *dev, hfi1_ipoib_check_queue_depth(txq); trace_sdma_output_ibhdr(txq->priv->dd, - &tx->sdma_hdr.hdr, + &tx->sdma_hdr->hdr, ib_is_sc5(txp->flow.sc5)); if (!netdev_xmit_more()) @@ -683,7 +683,8 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv) { struct net_device *dev = priv->netdev; u32 tx_ring_size, tx_item_size; - int i; + struct hfi1_ipoib_circ_buf *tx_ring; + int i, j; /* * Ring holds 1 less than tx_ring_size @@ -701,7 +702,9 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv) for (i = 0; i < dev->num_tx_queues; i++) { struct hfi1_ipoib_txq *txq = &priv->txqs[i]; + struct ipoib_txreq *tx; + tx_ring = &txq->tx_ring; iowait_init(&txq->wait, 0, hfi1_ipoib_flush_txq, @@ -725,14 +728,19 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv) priv->dd->node); txq->tx_ring.items = - kcalloc_node(tx_ring_size, tx_item_size, - GFP_KERNEL, priv->dd->node); + kvzalloc_node(array_size(tx_ring_size, tx_item_size), + GFP_KERNEL, priv->dd->node); if (!txq->tx_ring.items) goto free_txqs; txq->tx_ring.max_items = tx_ring_size; txq->tx_ring.shift = ilog2(tx_item_size); txq->tx_ring.avail = hfi1_ipoib_ring_hwat(txq); + tx_ring = &txq->tx_ring; + for (j = 0; j < tx_ring_size; j++) + hfi1_txreq_from_idx(tx_ring, j)->sdma_hdr = + kzalloc_node(sizeof(*tx->sdma_hdr), + GFP_KERNEL, priv->dd->node); netif_tx_napi_add(dev, &txq->napi, hfi1_ipoib_poll_tx_ring, @@ -746,7 +754,10 @@ free_txqs: struct hfi1_ipoib_txq *txq = &priv->txqs[i]; netif_napi_del(&txq->napi); - kfree(txq->tx_ring.items); + tx_ring = &txq->tx_ring; + for (j = 0; j < tx_ring_size; j++) + kfree(hfi1_txreq_from_idx(tx_ring, j)->sdma_hdr); + kvfree(tx_ring->items); } kfree(priv->txqs); @@ -780,17 +791,20 @@ static void hfi1_ipoib_drain_tx_list(struct hfi1_ipoib_txq *txq) void hfi1_ipoib_txreq_deinit(struct hfi1_ipoib_dev_priv *priv) { - int i; + int i, j; for (i = 0; i < priv->netdev->num_tx_queues; i++) { struct hfi1_ipoib_txq *txq = &priv->txqs[i]; + struct hfi1_ipoib_circ_buf *tx_ring = &txq->tx_ring; iowait_cancel_work(&txq->wait); iowait_sdma_drain(&txq->wait); hfi1_ipoib_drain_tx_list(txq); netif_napi_del(&txq->napi); hfi1_ipoib_drain_tx_ring(txq); - kfree(txq->tx_ring.items); + for (j = 0; j < tx_ring->max_items; j++) + kfree(hfi1_txreq_from_idx(tx_ring, j)->sdma_hdr); + kvfree(tx_ring->items); } kfree(priv->txqs); -- cgit v1.2.3 From 5f8f55b92edd621f056bdf09e572092849fabd83 Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Sat, 15 Jan 2022 18:02:35 -0500 Subject: IB/hfi1: Fix AIP early init panic An early failure in hfi1_ipoib_setup_rn() can lead to the following panic: BUG: unable to handle kernel NULL pointer dereference at 00000000000001b0 PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI Workqueue: events work_for_cpu_fn RIP: 0010:try_to_grab_pending+0x2b/0x140 Code: 1f 44 00 00 41 55 41 54 55 48 89 d5 53 48 89 fb 9c 58 0f 1f 44 00 00 48 89 c2 fa 66 0f 1f 44 00 00 48 89 55 00 40 84 f6 75 77 48 0f ba 2b 00 72 09 31 c0 5b 5d 41 5c 41 5d c3 48 89 df e8 6c RSP: 0018:ffffb6b3cf7cfa48 EFLAGS: 00010046 RAX: 0000000000000246 RBX: 00000000000001b0 RCX: 0000000000000000 RDX: 0000000000000246 RSI: 0000000000000000 RDI: 00000000000001b0 RBP: ffffb6b3cf7cfa70 R08: 0000000000000f09 R09: 0000000000000001 R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000000 R13: ffffb6b3cf7cfa90 R14: ffffffff9b2fbfc0 R15: ffff8a4fdf244690 FS: 0000000000000000(0000) GS:ffff8a527f400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000000001b0 CR3: 00000017e2410003 CR4: 00000000007706f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace: __cancel_work_timer+0x42/0x190 ? dev_printk_emit+0x4e/0x70 iowait_cancel_work+0x15/0x30 [hfi1] hfi1_ipoib_txreq_deinit+0x5a/0x220 [hfi1] ? dev_err+0x6c/0x90 hfi1_ipoib_netdev_dtor+0x15/0x30 [hfi1] hfi1_ipoib_setup_rn+0x10e/0x150 [hfi1] rdma_init_netdev+0x5a/0x80 [ib_core] ? hfi1_ipoib_free_rdma_netdev+0x20/0x20 [hfi1] ipoib_intf_init+0x6c/0x350 [ib_ipoib] ipoib_intf_alloc+0x5c/0xc0 [ib_ipoib] ipoib_add_one+0xbe/0x300 [ib_ipoib] add_client_context+0x12c/0x1a0 [ib_core] enable_device_and_get+0xdc/0x1d0 [ib_core] ib_register_device+0x572/0x6b0 [ib_core] rvt_register_device+0x11b/0x220 [rdmavt] hfi1_register_ib_device+0x6b4/0x770 [hfi1] do_init_one.isra.20+0x3e3/0x680 [hfi1] local_pci_probe+0x41/0x90 work_for_cpu_fn+0x16/0x20 process_one_work+0x1a7/0x360 ? create_worker+0x1a0/0x1a0 worker_thread+0x1cf/0x390 ? create_worker+0x1a0/0x1a0 kthread+0x116/0x130 ? kthread_flush_work_fn+0x10/0x10 ret_from_fork+0x1f/0x40 The panic happens in hfi1_ipoib_txreq_deinit() because there is a NULL deref when hfi1_ipoib_netdev_dtor() is called in this error case. hfi1_ipoib_txreq_init() and hfi1_ipoib_rxq_init() are self unwinding so fix by adjusting the error paths accordingly. Other changes: - hfi1_ipoib_free_rdma_netdev() is deleted including the free_netdev() since the netdev core code deletes calls free_netdev() - The switch to the accelerated entrances is moved to the success path. Cc: stable@vger.kernel.org Fixes: d99dc602e2a5 ("IB/hfi1: Add functions to transmit datagram ipoib packets") Link: https://lore.kernel.org/r/1642287756-182313-4-git-send-email-mike.marciniszyn@cornelisnetworks.com Reviewed-by: Dennis Dalessandro Signed-off-by: Mike Marciniszyn Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hfi1/ipoib_main.c | 13 +++---------- 1 file changed, 3 insertions(+), 10 deletions(-) diff --git a/drivers/infiniband/hw/hfi1/ipoib_main.c b/drivers/infiniband/hw/hfi1/ipoib_main.c index e1a2b02bbd91..8306ed5c1b80 100644 --- a/drivers/infiniband/hw/hfi1/ipoib_main.c +++ b/drivers/infiniband/hw/hfi1/ipoib_main.c @@ -168,12 +168,6 @@ static void hfi1_ipoib_netdev_dtor(struct net_device *dev) free_percpu(dev->tstats); } -static void hfi1_ipoib_free_rdma_netdev(struct net_device *dev) -{ - hfi1_ipoib_netdev_dtor(dev); - free_netdev(dev); -} - static void hfi1_ipoib_set_id(struct net_device *dev, int id) { struct hfi1_ipoib_dev_priv *priv = hfi1_ipoib_priv(dev); @@ -211,24 +205,23 @@ static int hfi1_ipoib_setup_rn(struct ib_device *device, priv->port_num = port_num; priv->netdev_ops = netdev->netdev_ops; - netdev->netdev_ops = &hfi1_ipoib_netdev_ops; - ib_query_pkey(device, port_num, priv->pkey_index, &priv->pkey); rc = hfi1_ipoib_txreq_init(priv); if (rc) { dd_dev_err(dd, "IPoIB netdev TX init - failed(%d)\n", rc); - hfi1_ipoib_free_rdma_netdev(netdev); return rc; } rc = hfi1_ipoib_rxq_init(netdev); if (rc) { dd_dev_err(dd, "IPoIB netdev RX init - failed(%d)\n", rc); - hfi1_ipoib_free_rdma_netdev(netdev); + hfi1_ipoib_txreq_deinit(priv); return rc; } + netdev->netdev_ops = &hfi1_ipoib_netdev_ops; + netdev->priv_destructor = hfi1_ipoib_netdev_dtor; netdev->needs_free_netdev = true; -- cgit v1.2.3 From e5cce44aff3be9ad2cd52f63f35edbd706181d50 Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Sat, 15 Jan 2022 18:02:36 -0500 Subject: IB/hfi1: Fix tstats alloc and dealloc The tstats allocation is done in the accelerated ndo_init function but the allocation is not tested to succeed. The deallocation is not done in the accelerated ndo_uninit function. Resolve issues by testing for an allocation failure and adding the free_percpu in the uninit function. Fixes: aa0616a9bd52 ("IB/hfi1: switch to core handling of rx/tx byte/packet counters") Link: https://lore.kernel.org/r/1642287756-182313-5-git-send-email-mike.marciniszyn@cornelisnetworks.com Reviewed-by: Dennis Dalessandro Signed-off-by: Mike Marciniszyn Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hfi1/ipoib_main.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/hw/hfi1/ipoib_main.c b/drivers/infiniband/hw/hfi1/ipoib_main.c index 8306ed5c1b80..5d814afdf7f3 100644 --- a/drivers/infiniband/hw/hfi1/ipoib_main.c +++ b/drivers/infiniband/hw/hfi1/ipoib_main.c @@ -22,26 +22,35 @@ static int hfi1_ipoib_dev_init(struct net_device *dev) int ret; dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); + if (!dev->tstats) + return -ENOMEM; ret = priv->netdev_ops->ndo_init(dev); if (ret) - return ret; + goto out_ret; ret = hfi1_netdev_add_data(priv->dd, qpn_from_mac(priv->netdev->dev_addr), dev); if (ret < 0) { priv->netdev_ops->ndo_uninit(dev); - return ret; + goto out_ret; } return 0; +out_ret: + free_percpu(dev->tstats); + dev->tstats = NULL; + return ret; } static void hfi1_ipoib_dev_uninit(struct net_device *dev) { struct hfi1_ipoib_dev_priv *priv = hfi1_ipoib_priv(dev); + free_percpu(dev->tstats); + dev->tstats = NULL; + hfi1_netdev_remove_data(priv->dd, qpn_from_mac(priv->netdev->dev_addr)); priv->netdev_ops->ndo_uninit(dev); @@ -166,6 +175,7 @@ static void hfi1_ipoib_netdev_dtor(struct net_device *dev) hfi1_ipoib_rxq_deinit(priv->netdev); free_percpu(dev->tstats); + dev->tstats = NULL; } static void hfi1_ipoib_set_id(struct net_device *dev, int id) -- cgit v1.2.3 From d9e410ebbed9d091b97bdf45b8a3792e2878dc48 Mon Sep 17 00:00:00 2001 From: Maor Gottlieb Date: Tue, 18 Jan 2022 09:35:00 +0200 Subject: RDMA/cma: Use correct address when leaving multicast group In RoCE we should use cma_iboe_set_mgid() and not cma_set_mgid to generate the mgid, otherwise we will generate an IGMP for an incorrect address. Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join") Link: https://lore.kernel.org/r/913bc6783fd7a95fe71ad9454e01653ee6fb4a9a.1642491047.git.leonro@nvidia.com Signed-off-by: Maor Gottlieb Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/cma.c | 22 ++++++++++++---------- 1 file changed, 12 insertions(+), 10 deletions(-) diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 27a00ce2e101..c447526288f4 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -67,8 +67,8 @@ static const char * const cma_events[] = { [RDMA_CM_EVENT_TIMEWAIT_EXIT] = "timewait exit", }; -static void cma_set_mgid(struct rdma_id_private *id_priv, struct sockaddr *addr, - union ib_gid *mgid); +static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid, + enum ib_gid_type gid_type); const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event) { @@ -1846,17 +1846,19 @@ static void destroy_mc(struct rdma_id_private *id_priv, if (dev_addr->bound_dev_if) ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if); - if (ndev) { + if (ndev && !send_only) { + enum ib_gid_type gid_type; union ib_gid mgid; - cma_set_mgid(id_priv, (struct sockaddr *)&mc->addr, - &mgid); - - if (!send_only) - cma_igmp_send(ndev, &mgid, false); - - dev_put(ndev); + gid_type = id_priv->cma_dev->default_gid_type + [id_priv->id.port_num - + rdma_start_port( + id_priv->cma_dev->device)]; + cma_iboe_set_mgid((struct sockaddr *)&mc->addr, &mgid, + gid_type); + cma_igmp_send(ndev, &mgid, false); } + dev_put(ndev); cancel_work_sync(&mc->iboe_join.work); } -- cgit v1.2.3 From 36e8169ec973359f671f9ec7213547059cae972e Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Tue, 18 Jan 2022 09:35:01 +0200 Subject: RDMA/ucma: Protect mc during concurrent multicast leaves Partially revert the commit mentioned in the Fixes line to make sure that allocation and erasing multicast struct are locked. BUG: KASAN: use-after-free in ucma_cleanup_multicast drivers/infiniband/core/ucma.c:491 [inline] BUG: KASAN: use-after-free in ucma_destroy_private_ctx+0x914/0xb70 drivers/infiniband/core/ucma.c:579 Read of size 8 at addr ffff88801bb74b00 by task syz-executor.1/25529 CPU: 0 PID: 25529 Comm: syz-executor.1 Not tainted 5.16.0-rc7-syzkaller #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:88 [inline] dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106 print_address_description.constprop.0.cold+0x8d/0x320 mm/kasan/report.c:247 __kasan_report mm/kasan/report.c:433 [inline] kasan_report.cold+0x83/0xdf mm/kasan/report.c:450 ucma_cleanup_multicast drivers/infiniband/core/ucma.c:491 [inline] ucma_destroy_private_ctx+0x914/0xb70 drivers/infiniband/core/ucma.c:579 ucma_destroy_id+0x1e6/0x280 drivers/infiniband/core/ucma.c:614 ucma_write+0x25c/0x350 drivers/infiniband/core/ucma.c:1732 vfs_write+0x28e/0xae0 fs/read_write.c:588 ksys_write+0x1ee/0x250 fs/read_write.c:643 do_syscall_x64 arch/x86/entry/common.c:50 [inline] do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80 entry_SYSCALL_64_after_hwframe+0x44/0xae Currently the xarray search can touch a concurrently freeing mc as the xa_for_each() is not surrounded by any lock. Rather than hold the lock for a full scan hold it only for the effected items, which is usually an empty list. Fixes: 95fe51096b7a ("RDMA/ucma: Remove mc_list and rely on xarray") Link: https://lore.kernel.org/r/1cda5fabb1081e8d16e39a48d3a4f8160cea88b8.1642491047.git.leonro@nvidia.com Reported-by: syzbot+e3f96c43d19782dd14a7@syzkaller.appspotmail.com Suggested-by: Jason Gunthorpe Reviewed-by: Maor Gottlieb Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/ucma.c | 34 +++++++++++++++++++++++----------- 1 file changed, 23 insertions(+), 11 deletions(-) diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 2b72c4fa9550..9d6ac9dff39a 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -95,6 +95,7 @@ struct ucma_context { u64 uid; struct list_head list; + struct list_head mc_list; struct work_struct close_work; }; @@ -105,6 +106,7 @@ struct ucma_multicast { u64 uid; u8 join_state; + struct list_head list; struct sockaddr_storage addr; }; @@ -198,6 +200,7 @@ static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file) INIT_WORK(&ctx->close_work, ucma_close_id); init_completion(&ctx->comp); + INIT_LIST_HEAD(&ctx->mc_list); /* So list_del() will work if we don't do ucma_finish_ctx() */ INIT_LIST_HEAD(&ctx->list); ctx->file = file; @@ -484,19 +487,19 @@ err1: static void ucma_cleanup_multicast(struct ucma_context *ctx) { - struct ucma_multicast *mc; - unsigned long index; + struct ucma_multicast *mc, *tmp; - xa_for_each(&multicast_table, index, mc) { - if (mc->ctx != ctx) - continue; + xa_lock(&multicast_table); + list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) { + list_del(&mc->list); /* * At this point mc->ctx->ref is 0 so the mc cannot leave the * lock on the reader and this is enough serialization */ - xa_erase(&multicast_table, index); + __xa_erase(&multicast_table, mc->id); kfree(mc); } + xa_unlock(&multicast_table); } static void ucma_cleanup_mc_events(struct ucma_multicast *mc) @@ -1469,12 +1472,16 @@ static ssize_t ucma_process_join(struct ucma_file *file, mc->uid = cmd->uid; memcpy(&mc->addr, addr, cmd->addr_size); - if (xa_alloc(&multicast_table, &mc->id, NULL, xa_limit_32b, + xa_lock(&multicast_table); + if (__xa_alloc(&multicast_table, &mc->id, NULL, xa_limit_32b, GFP_KERNEL)) { ret = -ENOMEM; goto err_free_mc; } + list_add_tail(&mc->list, &ctx->mc_list); + xa_unlock(&multicast_table); + mutex_lock(&ctx->mutex); ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *)&mc->addr, join_state, mc); @@ -1500,8 +1507,11 @@ err_leave_multicast: mutex_unlock(&ctx->mutex); ucma_cleanup_mc_events(mc); err_xa_erase: - xa_erase(&multicast_table, mc->id); + xa_lock(&multicast_table); + list_del(&mc->list); + __xa_erase(&multicast_table, mc->id); err_free_mc: + xa_unlock(&multicast_table); kfree(mc); err_put_ctx: ucma_put_ctx(ctx); @@ -1569,15 +1579,17 @@ static ssize_t ucma_leave_multicast(struct ucma_file *file, mc = ERR_PTR(-EINVAL); else if (!refcount_inc_not_zero(&mc->ctx->ref)) mc = ERR_PTR(-ENXIO); - else - __xa_erase(&multicast_table, mc->id); - xa_unlock(&multicast_table); if (IS_ERR(mc)) { + xa_unlock(&multicast_table); ret = PTR_ERR(mc); goto out; } + list_del(&mc->list); + __xa_erase(&multicast_table, mc->id); + xa_unlock(&multicast_table); + mutex_lock(&mc->ctx->mutex); rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr); mutex_unlock(&mc->ctx->mutex); -- cgit v1.2.3 From a75badebfdc0b3823054bedf112edb54d6357c75 Mon Sep 17 00:00:00 2001 From: Dan Carpenter Date: Tue, 18 Jan 2022 12:11:04 +0300 Subject: RDMA/siw: Fix refcounting leak in siw_create_qp() The atomic_inc() needs to be paired with an atomic_dec() on the error path. Fixes: 514aee660df4 ("RDMA: Globally allocate and release QP memory") Link: https://lore.kernel.org/r/20220118091104.GA11671@kili Signed-off-by: Dan Carpenter Reviewed-by: Leon Romanovsky Reviewed-by: Bernard Metzler Signed-off-by: Jason Gunthorpe --- drivers/infiniband/sw/siw/siw_verbs.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/sw/siw/siw_verbs.c b/drivers/infiniband/sw/siw/siw_verbs.c index a3dd2cb6d5c9..54ef367b074a 100644 --- a/drivers/infiniband/sw/siw/siw_verbs.c +++ b/drivers/infiniband/sw/siw/siw_verbs.c @@ -313,7 +313,8 @@ int siw_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attrs, if (atomic_inc_return(&sdev->num_qp) > SIW_MAX_QP) { siw_dbg(base_dev, "too many QP's\n"); - return -ENOMEM; + rv = -ENOMEM; + goto err_atomic; } if (attrs->qp_type != IB_QPT_RC) { siw_dbg(base_dev, "only RC QP's supported\n"); -- cgit v1.2.3 From b856101a1774b5f1c8c99e8dfdef802856520732 Mon Sep 17 00:00:00 2001 From: Mark Zhang Date: Wed, 19 Jan 2022 10:37:55 +0200 Subject: IB/cm: Release previously acquired reference counter in the cm_id_priv In failure flow, the reference counter acquired was not released, and the following error was reported: drivers/infiniband/core/cm.c:3373 cm_lap_handler() warn: inconsistent refcounting 'cm_id_priv->refcount.refs.counter': Fixes: 7345201c3963 ("IB/cm: Improve the calling of cm_init_av_for_lap and cm_init_av_by_path") Link: https://lore.kernel.org/r/7615f23bbb5c5b66d03f6fa13e1c99d51dae6916.1642581448.git.leonro@nvidia.com Reported-by: Dan Carpenter Signed-off-by: Mark Zhang Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/cm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c index c903b74f46a4..35f0d5e7533d 100644 --- a/drivers/infiniband/core/cm.c +++ b/drivers/infiniband/core/cm.c @@ -3322,7 +3322,7 @@ static int cm_lap_handler(struct cm_work *work) ret = cm_init_av_by_path(param->alternate_path, NULL, &alt_av); if (ret) { rdma_destroy_ah_attr(&ah_attr); - return -EINVAL; + goto deref; } spin_lock_irq(&cm_id_priv->lock); -- cgit v1.2.3 From 4028bccb003cf67e46632dee7f97ddc5d7b6e685 Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Wed, 19 Jan 2022 04:28:09 -0500 Subject: IB/rdmavt: Validate remote_addr during loopback atomic tests The rdma-core test suite sends an unaligned remote address and expects a failure. ERROR: test_atomic_non_aligned_addr (tests.test_atomic.AtomicTest) The qib/hfi1 rc handling validates properly, but the test has the client and server on the same system. The loopback of these operations is a distinct code path. Fix by syntaxing the proposed remote address in the loopback code path. Fixes: 15703461533a ("IB/{hfi1, qib, rdmavt}: Move ruc_loopback to rdmavt") Link: https://lore.kernel.org/r/1642584489-141005-1-git-send-email-mike.marciniszyn@cornelisnetworks.com Reviewed-by: Dennis Dalessandro Signed-off-by: Mike Marciniszyn Signed-off-by: Jason Gunthorpe --- drivers/infiniband/sw/rdmavt/qp.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c index 3305f2744bfa..ae50b56e8913 100644 --- a/drivers/infiniband/sw/rdmavt/qp.c +++ b/drivers/infiniband/sw/rdmavt/qp.c @@ -3073,6 +3073,8 @@ do_write: case IB_WR_ATOMIC_FETCH_AND_ADD: if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC))) goto inv_err; + if (unlikely(wqe->atomic_wr.remote_addr & (sizeof(u64) - 1))) + goto inv_err; if (unlikely(!rvt_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64), wqe->atomic_wr.remote_addr, wqe->atomic_wr.rkey, -- cgit v1.2.3 From b43a76f423aa304037603fd6165c4a534d2c09a7 Mon Sep 17 00:00:00 2001 From: Bernard Metzler Date: Sun, 30 Jan 2022 18:08:15 +0100 Subject: RDMA/siw: Fix broken RDMA Read Fence/Resume logic. Code unconditionally resumed fenced SQ processing after next RDMA Read completion, even if other RDMA Read responses are still outstanding, or ORQ is full. Also adds comments for better readability of fence processing, and removes orq_get_tail() helper, which is not needed anymore. Fixes: 8b6a361b8c48 ("rdma/siw: receive path") Fixes: a531975279f3 ("rdma/siw: main include file") Link: https://lore.kernel.org/r/20220130170815.1940-1-bmt@zurich.ibm.com Reported-by: Jared Holzman Signed-off-by: Bernard Metzler Signed-off-by: Jason Gunthorpe --- drivers/infiniband/sw/siw/siw.h | 7 +------ drivers/infiniband/sw/siw/siw_qp_rx.c | 20 +++++++++++--------- 2 files changed, 12 insertions(+), 15 deletions(-) diff --git a/drivers/infiniband/sw/siw/siw.h b/drivers/infiniband/sw/siw/siw.h index 368959ae9a8c..df03d84c6868 100644 --- a/drivers/infiniband/sw/siw/siw.h +++ b/drivers/infiniband/sw/siw/siw.h @@ -644,14 +644,9 @@ static inline struct siw_sqe *orq_get_current(struct siw_qp *qp) return &qp->orq[qp->orq_get % qp->attrs.orq_size]; } -static inline struct siw_sqe *orq_get_tail(struct siw_qp *qp) -{ - return &qp->orq[qp->orq_put % qp->attrs.orq_size]; -} - static inline struct siw_sqe *orq_get_free(struct siw_qp *qp) { - struct siw_sqe *orq_e = orq_get_tail(qp); + struct siw_sqe *orq_e = &qp->orq[qp->orq_put % qp->attrs.orq_size]; if (READ_ONCE(orq_e->flags) == 0) return orq_e; diff --git a/drivers/infiniband/sw/siw/siw_qp_rx.c b/drivers/infiniband/sw/siw/siw_qp_rx.c index 60116f20653c..875ea6f1b04a 100644 --- a/drivers/infiniband/sw/siw/siw_qp_rx.c +++ b/drivers/infiniband/sw/siw/siw_qp_rx.c @@ -1153,11 +1153,12 @@ static int siw_check_tx_fence(struct siw_qp *qp) spin_lock_irqsave(&qp->orq_lock, flags); - rreq = orq_get_current(qp); - /* free current orq entry */ + rreq = orq_get_current(qp); WRITE_ONCE(rreq->flags, 0); + qp->orq_get++; + if (qp->tx_ctx.orq_fence) { if (unlikely(tx_waiting->wr_status != SIW_WR_QUEUED)) { pr_warn("siw: [QP %u]: fence resume: bad status %d\n", @@ -1165,10 +1166,12 @@ static int siw_check_tx_fence(struct siw_qp *qp) rv = -EPROTO; goto out; } - /* resume SQ processing */ + /* resume SQ processing, if possible */ if (tx_waiting->sqe.opcode == SIW_OP_READ || tx_waiting->sqe.opcode == SIW_OP_READ_LOCAL_INV) { - rreq = orq_get_tail(qp); + + /* SQ processing was stopped because of a full ORQ */ + rreq = orq_get_free(qp); if (unlikely(!rreq)) { pr_warn("siw: [QP %u]: no ORQE\n", qp_id(qp)); rv = -EPROTO; @@ -1181,15 +1184,14 @@ static int siw_check_tx_fence(struct siw_qp *qp) resume_tx = 1; } else if (siw_orq_empty(qp)) { + /* + * SQ processing was stopped by fenced work request. + * Resume since all previous Read's are now completed. + */ qp->tx_ctx.orq_fence = 0; resume_tx = 1; - } else { - pr_warn("siw: [QP %u]: fence resume: orq idx: %d:%d\n", - qp_id(qp), qp->orq_get, qp->orq_put); - rv = -EPROTO; } } - qp->orq_get++; out: spin_unlock_irqrestore(&qp->orq_lock, flags); -- cgit v1.2.3 From f3136c4ce7acf64bee43135971ca52a880572e32 Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Mon, 31 Jan 2022 11:45:26 +0200 Subject: RDMA/mlx4: Don't continue event handler after memory allocation failure MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The failure to allocate memory during MLX4_DEV_EVENT_PORT_MGMT_CHANGE event handler will cause skip the assignment logic, but ib_dispatch_event() will be called anyway. Fix it by calling to return instead of break after memory allocation failure. Fixes: 00f5ce99dc6e ("mlx4: Use port management change event instead of smp_snoop") Link: https://lore.kernel.org/r/12a0e83f18cfad4b5f62654f141e240d04915e10.1643622264.git.leonro@nvidia.com Signed-off-by: Leon Romanovsky Reviewed-by: Håkon Bugge Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/mlx4/main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 1c3d97229988..93b1650eacfa 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -3237,7 +3237,7 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, case MLX4_DEV_EVENT_PORT_MGMT_CHANGE: ew = kmalloc(sizeof *ew, GFP_ATOMIC); if (!ew) - break; + return; INIT_WORK(&ew->work, handle_port_mgmt_change_event); memcpy(&ew->ib_eqe, eqe, sizeof *eqe); -- cgit v1.2.3