From 32927e281c004c352c0b88d2975f5511e5b7cf8e Mon Sep 17 00:00:00 2001 From: Mark Bloch Date: Tue, 20 Mar 2018 15:45:37 +0200 Subject: IB/mlx5: Don't clean uninitialized UMR resources In case we failed to create UMR resources, mark them as invalid so we won't try to destroy them on the unwind path. Add the relevant checks to destroy_umrc_res(), this is done for the unlikely event ib_register_device() or create_umr_res() err out and we try to destroy invalid objects. Fixes: 42cea83f9524 ("IB/mlx5: Fix cleanup order on unload") Signed-off-by: Mark Bloch Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/mlx5/main.c | 12 +++++++++--- drivers/infiniband/hw/mlx5/mr.c | 3 +++ 2 files changed, 12 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index da091de4e69d..7f8bda3a2005 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -3448,9 +3448,12 @@ static void destroy_umrc_res(struct mlx5_ib_dev *dev) if (err) mlx5_ib_warn(dev, "mr cache cleanup failed\n"); - mlx5_ib_destroy_qp(dev->umrc.qp); - ib_free_cq(dev->umrc.cq); - ib_dealloc_pd(dev->umrc.pd); + if (dev->umrc.qp) + mlx5_ib_destroy_qp(dev->umrc.qp); + if (dev->umrc.cq) + ib_free_cq(dev->umrc.cq); + if (dev->umrc.pd) + ib_dealloc_pd(dev->umrc.pd); } enum { @@ -3552,12 +3555,15 @@ static int create_umr_res(struct mlx5_ib_dev *dev) error_4: mlx5_ib_destroy_qp(qp); + dev->umrc.qp = NULL; error_3: ib_free_cq(cq); + dev->umrc.cq = NULL; error_2: ib_dealloc_pd(pd); + dev->umrc.pd = NULL; error_0: kfree(attr); diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index c51c602f06d6..3e0b3f0238d6 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -739,6 +739,9 @@ int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev) { int i; + if (!dev->cache.wq) + return 0; + dev->cache.stopped = 1; flush_workqueue(dev->cache.wq); -- cgit From 103140eccbd2e3d56bcd366eba3e657f5d358bf5 Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Wed, 21 Mar 2018 09:00:29 +0200 Subject: RDMA/restrack: Move restrack_clean to be symmetrical to restrack_init The fact that resource tracking commit 02d8883f520e ("RDMA/restrack: Add general infrastructure to track RDMA resources") was added immediately after commit 16c1975f1032 ("IB/mlx5: Create profile infrastructure to add and remove stages") caused us to miss the fact that PD and CQ are created after ib_register_device, but released after ib_unregister_device() and not before as it is expected from normal flow. Fix introduced in commit 42cea83f9524 ("IB/mlx5: Fix cleanup order on unload") revealed this fact, so this patch is needed to avoid from restrack warnings It fixes resource tracking warnings during shutdown. [ 43.473906] CPU: 5 PID: 3016 Comm: modprobe Not tainted 4.16.0-rc5-for-linust-perf-2018-03-19_07-01-58-14 #1 [ 43.473907] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu2 04/01/2014 [ 43.473919] RIP: 0010:rdma_restrack_clean+0x25/0x30 [ib_core] [ 43.473921] RSP: 0018:ffffc9000267be48 EFLAGS: 00010282 [ 43.473924] RAX: 0000000000000000 RBX: ffff88033c690070 RCX: 0000000180080006 [ 43.473925] RDX: ffff88035ce922e0 RSI: ffffea000cf1a200 RDI: ffff88033c6907c8 [ 43.473926] RBP: ffff88033c690070 R08: ffff88033c689000 R09: 0000000180080006 [ 43.473927] R10: 000000003c68a001 R11: ffff88033c689000 R12: ffff88033c690000 [ 43.473929] R13: ffff88033c69005c R14: 0000000000000000 R15: 0000000000000000 [ 43.473932] FS: 00007f5928359740(0000) GS:ffff88036c540000(0000) knlGS:0000000000000000 [ 43.473933] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 43.473935] CR2: 00007ffffc760cc8 CR3: 000000035620c000 CR4: 00000000000006e0 [ 43.473940] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 43.473941] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 43.473942] Call Trace: [ 43.473969] ib_unregister_device+0xf5/0x190 [ib_core] [ 43.474000] __mlx5_ib_remove+0x2e/0x40 [mlx5_ib] [ 43.474098] mlx5_remove_device+0xf5/0x120 [mlx5_core] [ 43.474132] mlx5_unregister_interface+0x37/0x90 [mlx5_core] [ 43.474142] mlx5_ib_cleanup+0xc/0x16a [mlx5_ib] [ 43.474152] SyS_delete_module+0x159/0x260 [ 43.474159] do_syscall_64+0x61/0x110 [ 43.474165] entry_SYSCALL_64_after_hwframe+0x3d/0xa2 [ 43.474168] RIP: 0033:0x7f59278466b7 [ 43.474170] RSP: 002b:00007ffffc763e38 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0 [ 43.474172] RAX: ffffffffffffffda RBX: 000000000130d590 RCX: 00007f59278466b7 [ 43.474173] RDX: 0000000000000000 RSI: 0000000000000800 RDI: 000000000130d5f8 [ 43.474175] RBP: 0000000000000000 R08: 00007f5927b0b060 R09: 00007f59278b6a40 [ 43.474176] R10: 00007ffffc763bc0 R11: 0000000000000202 R12: 0000000000000000 [ 43.474177] R13: 0000000000000001 R14: 000000000130d5f8 R15: 0000000000000000 [ 43.474179] Code: 84 00 00 00 00 00 0f 1f 44 00 00 48 83 c7 28 31 c0 eb 0c 48 83 c0 08 48 3d 00 08 00 00 74 0f 48 8d 14 07 48 8b 12 48 85 d2 74 e8 <0f> 0b c3 f3 c3 66 0f 1f 44 00 00 0f 1f 44 00 00 53 48 8b 47 28 [ 43.474221] ---[ end trace e89771e2250ffc23 ]--- Fixes: 42cea83f9524 ("IB/mlx5: Fix cleanup order on unload") Reviewed-by: Mark Bloch Signed-off-by: Parav Pandit Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/device.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c index bb065c9449be..b7459cf524e4 100644 --- a/drivers/infiniband/core/device.c +++ b/drivers/infiniband/core/device.c @@ -290,6 +290,7 @@ void ib_dealloc_device(struct ib_device *device) { WARN_ON(device->reg_state != IB_DEV_UNREGISTERED && device->reg_state != IB_DEV_UNINITIALIZED); + rdma_restrack_clean(&device->res); put_device(&device->dev); } EXPORT_SYMBOL(ib_dealloc_device); @@ -600,8 +601,6 @@ void ib_unregister_device(struct ib_device *device) } up_read(&lists_rwsem); - rdma_restrack_clean(&device->res); - ib_device_unregister_rdmacg(device); ib_device_unregister_sysfs(device); -- cgit From 5f3e3b85cc0a5eae1c46d72e47d3de7bf208d9e2 Mon Sep 17 00:00:00 2001 From: Chien Tin Tung Date: Wed, 21 Mar 2018 13:09:25 -0500 Subject: RDMA/ucma: Correct option size check using optlen The option size check is using optval instead of optlen causing the set option call to fail. Use the correct field, optlen, for size check. Fixes: 6a21dfc0d0db ("RDMA/ucma: Limit possible option size") Signed-off-by: Chien Tin Tung Signed-off-by: Shiraz Saleem Reviewed-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/ucma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index e5a1e7d81326..60449d611fb4 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -1307,7 +1307,7 @@ static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf, if (IS_ERR(ctx)) return PTR_ERR(ctx); - if (unlikely(cmd.optval > KMALLOC_MAX_SIZE)) + if (unlikely(cmd.optlen > KMALLOC_MAX_SIZE)) return -EINVAL; optval = memdup_user((void __user *) (unsigned long) cmd.optval, -- cgit From c3594f22302cca5e924e47ec1cc8edd265708f41 Mon Sep 17 00:00:00 2001 From: "Kalderon, Michal" Date: Wed, 21 Mar 2018 14:51:50 +0200 Subject: RDMA/qedr: fix QP's ack timeout configuration QPs that were configured with ack timeout value lower than 1 msec will not implement re-transmission timeout. This means that if a packet / ACK were dropped, the QP will not retransmit this packet. This can lead to an application hang. Fixes: cecbcddf6 ("qedr: Add support for QP verbs") Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/qedr/verbs.c | 25 +++++++++++++++---------- 1 file changed, 15 insertions(+), 10 deletions(-) diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c index 875b17272d65..716fae063122 100644 --- a/drivers/infiniband/hw/qedr/verbs.c +++ b/drivers/infiniband/hw/qedr/verbs.c @@ -2086,18 +2086,23 @@ int qedr_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, SET_FIELD(qp_params.modify_flags, QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT, 1); - qp_params.ack_timeout = attr->timeout; - if (attr->timeout) { - u32 temp; - - temp = 4096 * (1UL << attr->timeout) / 1000 / 1000; - /* FW requires [msec] */ - qp_params.ack_timeout = temp; - } else { - /* Infinite */ + /* The received timeout value is an exponent used like this: + * "12.7.34 LOCAL ACK TIMEOUT + * Value representing the transport (ACK) timeout for use by + * the remote, expressed as: 4.096 * 2^timeout [usec]" + * The FW expects timeout in msec so we need to divide the usec + * result by 1000. We'll approximate 1000~2^10, and 4.096 ~ 2^2, + * so we get: 2^2 * 2^timeout / 2^10 = 2^(timeout - 8). + * The value of zero means infinite so we use a 'max_t' to make + * sure that sub 1 msec values will be configured as 1 msec. + */ + if (attr->timeout) + qp_params.ack_timeout = + 1 << max_t(int, attr->timeout - 8, 0); + else qp_params.ack_timeout = 0; - } } + if (attr_mask & IB_QP_RETRY_CNT) { SET_FIELD(qp_params.modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT, 1); -- cgit From b15606f47b89b0b09936d7f45b59ba6275527041 Mon Sep 17 00:00:00 2001 From: "Kalderon, Michal" Date: Wed, 21 Mar 2018 14:51:51 +0200 Subject: RDMA/qedr: Fix rc initialization on CNQ allocation failure Return code wasn't set properly when CNQ allocation failed. This only affect error message logging, currently user will receive an error message that says the qedr driver load failed with rc '0', instead of ENOMEM Fixes: ec72fce4 ("qedr: Add support for RoCE HW init") Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/qedr/main.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c index db4bf97c0e15..0ffb9b93e22d 100644 --- a/drivers/infiniband/hw/qedr/main.c +++ b/drivers/infiniband/hw/qedr/main.c @@ -833,7 +833,8 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev, dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev); if (!dev->num_cnq) { - DP_ERR(dev, "not enough CNQ resources.\n"); + DP_ERR(dev, "Failed. At least one CNQ is required.\n"); + rc = -ENOMEM; goto init_err; } -- cgit From caf61b1b8b88ccf1451f7321a176393797e8d292 Mon Sep 17 00:00:00 2001 From: "Kalderon, Michal" Date: Wed, 21 Mar 2018 14:51:52 +0200 Subject: RDMA/qedr: Fix QP state initialization race Once the FW is transitioned to error, FLUSH cqes can be received. We want the driver to be aware of the fact that QP is already in error. Without this fix, a user may see false error messages in the dmesg log, mentioning that a FLUSH cqe was received while QP is not in error state. Fixes: cecbcddf ("qedr: Add support for QP verbs") Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/qedr/verbs.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c index 716fae063122..419a158e8fca 100644 --- a/drivers/infiniband/hw/qedr/verbs.c +++ b/drivers/infiniband/hw/qedr/verbs.c @@ -1841,14 +1841,15 @@ static void qedr_reset_qp_hwq_info(struct qedr_qp_hwq_info *qph) static int qedr_update_qp_state(struct qedr_dev *dev, struct qedr_qp *qp, + enum qed_roce_qp_state cur_state, enum qed_roce_qp_state new_state) { int status = 0; - if (new_state == qp->state) + if (new_state == cur_state) return 0; - switch (qp->state) { + switch (cur_state) { case QED_ROCE_QP_STATE_RESET: switch (new_state) { case QED_ROCE_QP_STATE_INIT: @@ -1955,6 +1956,7 @@ int qedr_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, struct qedr_dev *dev = get_qedr_dev(&qp->dev->ibdev); const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr); enum ib_qp_state old_qp_state, new_qp_state; + enum qed_roce_qp_state cur_state; int rc = 0; DP_DEBUG(dev, QEDR_MSG_QP, @@ -2175,13 +2177,25 @@ int qedr_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, qp->dest_qp_num = attr->dest_qp_num; } + cur_state = qp->state; + + /* Update the QP state before the actual ramrod to prevent a race with + * fast path. Modifying the QP state to error will cause the device to + * flush the CQEs and while polling the flushed CQEs will considered as + * a potential issue if the QP isn't in error state. + */ + if ((attr_mask & IB_QP_STATE) && qp->qp_type != IB_QPT_GSI && + !udata && qp_params.new_state == QED_ROCE_QP_STATE_ERR) + qp->state = QED_ROCE_QP_STATE_ERR; + if (qp->qp_type != IB_QPT_GSI) rc = dev->ops->rdma_modify_qp(dev->rdma_ctx, qp->qed_qp, &qp_params); if (attr_mask & IB_QP_STATE) { if ((qp->qp_type != IB_QPT_GSI) && (!udata)) - rc = qedr_update_qp_state(dev, qp, qp_params.new_state); + rc = qedr_update_qp_state(dev, qp, cur_state, + qp_params.new_state); qp->state = qp_params.new_state; } -- cgit From 9137108cc3d64ade13e753108ec611a0daed16a0 Mon Sep 17 00:00:00 2001 From: Jason Gunthorpe Date: Thu, 22 Mar 2018 14:04:23 -0600 Subject: RDMA/rdma_cm: Fix use after free race with process_one_req process_one_req() can race with rdma_addr_cancel(): CPU0 CPU1 ==== ==== process_one_work() debug_work_deactivate(work); process_one_req() rdma_addr_cancel() mutex_lock(&lock); set_timeout(&req->work,..); __queue_work() debug_work_activate(work); mutex_unlock(&lock); mutex_lock(&lock); [..] list_del(&req->list); mutex_unlock(&lock); [..] // ODEBUG explodes since the work is still queued. kfree(req); Causing ODEBUG to detect the use after free: ODEBUG: free active (active state 0) object type: work_struct hint: process_one_req+0x0/0x6c0 include/net/dst.h:165 WARNING: CPU: 0 PID: 79 at lib/debugobjects.c:291 debug_print_object+0x166/0x220 lib/debugobjects.c:288 kvm: emulating exchange as write Kernel panic - not syncing: panic_on_warn set ... CPU: 0 PID: 79 Comm: kworker/u4:3 Not tainted 4.16.0-rc6+ #361 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Workqueue: ib_addr process_one_req Call Trace: __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x24d lib/dump_stack.c:53 panic+0x1e4/0x41c kernel/panic.c:183 __warn+0x1dc/0x200 kernel/panic.c:547 report_bug+0x1f4/0x2b0 lib/bug.c:186 fixup_bug.part.11+0x37/0x80 arch/x86/kernel/traps.c:178 fixup_bug arch/x86/kernel/traps.c:247 [inline] do_error_trap+0x2d7/0x3e0 arch/x86/kernel/traps.c:296 do_invalid_op+0x1b/0x20 arch/x86/kernel/traps.c:315 invalid_op+0x1b/0x40 arch/x86/entry/entry_64.S:986 RIP: 0010:debug_print_object+0x166/0x220 lib/debugobjects.c:288 RSP: 0000:ffff8801d966f210 EFLAGS: 00010086 RAX: dffffc0000000008 RBX: 0000000000000003 RCX: ffffffff815acd6e RDX: 0000000000000000 RSI: 1ffff1003b2cddf2 RDI: 0000000000000000 RBP: ffff8801d966f250 R08: 0000000000000000 R09: 1ffff1003b2cddc8 R10: ffffed003b2cde71 R11: ffffffff86f39a98 R12: 0000000000000001 R13: ffffffff86f15540 R14: ffffffff86408700 R15: ffffffff8147c0a0 __debug_check_no_obj_freed lib/debugobjects.c:745 [inline] debug_check_no_obj_freed+0x662/0xf1f lib/debugobjects.c:774 kfree+0xc7/0x260 mm/slab.c:3799 process_one_req+0x2e7/0x6c0 drivers/infiniband/core/addr.c:592 process_one_work+0xc47/0x1bb0 kernel/workqueue.c:2113 worker_thread+0x223/0x1990 kernel/workqueue.c:2247 kthread+0x33c/0x400 kernel/kthread.c:238 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:406 Fixes: 5fff41e1f89d ("IB/core: Fix race condition in resolving IP to MAC") Reported-by: Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/addr.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c index 9183d148d644..c3e6811bb1bd 100644 --- a/drivers/infiniband/core/addr.c +++ b/drivers/infiniband/core/addr.c @@ -586,6 +586,15 @@ static void process_one_req(struct work_struct *_work) list_del(&req->list); mutex_unlock(&lock); + /* + * Although the work will normally have been canceled by the + * workqueue, it can still be requeued as long as it is on the + * req_list, so it could have been requeued before we grabbed &lock. + * We need to cancel it after it is removed from req_list to really be + * sure it is safe to free. + */ + cancel_delayed_work(&req->work); + req->callback(req->status, (struct sockaddr *)&req->src_addr, req->addr, req->context); put_client(req->client); -- cgit From 4b658d1bbc16605330694bb3ef2570c465ef383d Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Sun, 25 Mar 2018 11:23:55 +0300 Subject: RDMA/ucma: Check that device is connected prior to access it Add missing check that device is connected prior to access it. [ 55.358652] BUG: KASAN: null-ptr-deref in rdma_init_qp_attr+0x4a/0x2c0 [ 55.359389] Read of size 8 at addr 00000000000000b0 by task qp/618 [ 55.360255] [ 55.360432] CPU: 1 PID: 618 Comm: qp Not tainted 4.16.0-rc1-00071-gcaf61b1b8b88 #91 [ 55.361693] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.0-0-g63451fca13-prebuilt.qemu-project.org 04/01/2014 [ 55.363264] Call Trace: [ 55.363833] dump_stack+0x5c/0x77 [ 55.364215] kasan_report+0x163/0x380 [ 55.364610] ? rdma_init_qp_attr+0x4a/0x2c0 [ 55.365238] rdma_init_qp_attr+0x4a/0x2c0 [ 55.366410] ucma_init_qp_attr+0x111/0x200 [ 55.366846] ? ucma_notify+0xf0/0xf0 [ 55.367405] ? _get_random_bytes+0xea/0x1b0 [ 55.367846] ? urandom_read+0x2f0/0x2f0 [ 55.368436] ? kmem_cache_alloc_trace+0xd2/0x1e0 [ 55.369104] ? refcount_inc_not_zero+0x9/0x60 [ 55.369583] ? refcount_inc+0x5/0x30 [ 55.370155] ? rdma_create_id+0x215/0x240 [ 55.370937] ? _copy_to_user+0x4f/0x60 [ 55.371620] ? mem_cgroup_commit_charge+0x1f5/0x290 [ 55.372127] ? _copy_from_user+0x5e/0x90 [ 55.372720] ucma_write+0x174/0x1f0 [ 55.373090] ? ucma_close_id+0x40/0x40 [ 55.373805] ? __lru_cache_add+0xa8/0xd0 [ 55.374403] __vfs_write+0xc4/0x350 [ 55.374774] ? kernel_read+0xa0/0xa0 [ 55.375173] ? fsnotify+0x899/0x8f0 [ 55.375544] ? fsnotify_unmount_inodes+0x170/0x170 [ 55.376689] ? __fsnotify_update_child_dentry_flags+0x30/0x30 [ 55.377522] ? handle_mm_fault+0x174/0x320 [ 55.378169] vfs_write+0xf7/0x280 [ 55.378864] SyS_write+0xa1/0x120 [ 55.379270] ? SyS_read+0x120/0x120 [ 55.379643] ? mm_fault_error+0x180/0x180 [ 55.380071] ? task_work_run+0x7d/0xd0 [ 55.380910] ? __task_pid_nr_ns+0x120/0x140 [ 55.381366] ? SyS_read+0x120/0x120 [ 55.381739] do_syscall_64+0xeb/0x250 [ 55.382143] entry_SYSCALL_64_after_hwframe+0x21/0x86 [ 55.382841] RIP: 0033:0x7fc2ef803e99 [ 55.383227] RSP: 002b:00007fffcc5f3be8 EFLAGS: 00000217 ORIG_RAX: 0000000000000001 [ 55.384173] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fc2ef803e99 [ 55.386145] RDX: 0000000000000057 RSI: 0000000020000080 RDI: 0000000000000003 [ 55.388418] RBP: 00007fffcc5f3c00 R08: 0000000000000000 R09: 0000000000000000 [ 55.390542] R10: 0000000000000000 R11: 0000000000000217 R12: 0000000000400480 [ 55.392916] R13: 00007fffcc5f3cf0 R14: 0000000000000000 R15: 0000000000000000 [ 55.521088] Code: e5 4d 1e ff 48 89 df 44 0f b6 b3 b8 01 00 00 e8 65 50 1e ff 4c 8b 2b 49 8d bd b0 00 00 00 e8 56 50 1e ff 41 0f b6 c6 48 c1 e0 04 <49> 03 85 b0 00 00 00 48 8d 78 08 48 89 04 24 e8 3a 4f 1e ff 48 [ 55.525980] RIP: rdma_init_qp_attr+0x52/0x2c0 RSP: ffff8801e2c2f9d8 [ 55.532648] CR2: 00000000000000b0 [ 55.534396] ---[ end trace 70cee64090251c0b ]--- Fixes: 75216638572f ("RDMA/cma: Export rdma cm interface to userspace") Fixes: d541e45500bd ("IB/core: Convert ah_attr from OPA to IB when copying to user") Reported-by: Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/ucma.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 60449d611fb4..0811f58c670f 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -1166,6 +1166,11 @@ static ssize_t ucma_init_qp_attr(struct ucma_file *file, if (IS_ERR(ctx)) return PTR_ERR(ctx); + if (!ctx->cm_id->device) { + ret = -EINVAL; + goto out; + } + resp.qp_attr_mask = 0; memset(&qp_attr, 0, sizeof qp_attr); qp_attr.qp_state = cmd.qp_state; -- cgit From c8d3bcbfc5eab3f01cf373d039af725f3b488813 Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Sun, 25 Mar 2018 11:39:05 +0300 Subject: RDMA/ucma: Check that device exists prior to accessing it Ensure that device exists prior to accessing its properties. Reported-by: Fixes: 75216638572f ("RDMA/cma: Export rdma cm interface to userspace") Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/ucma.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 0811f58c670f..21585055cf32 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -1336,7 +1336,7 @@ static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf, { struct rdma_ucm_notify cmd; struct ucma_context *ctx; - int ret; + int ret = -EINVAL; if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; @@ -1345,7 +1345,9 @@ static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf, if (IS_ERR(ctx)) return PTR_ERR(ctx); - ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event); + if (ctx->cm_id->device) + ret = rdma_notify(ctx->cm_id, (enum ib_event_type)cmd.event); + ucma_put_ctx(ctx); return ret; } -- cgit From 38759d6175d338fbf9282c8ea2b51f3b7ab9bc98 Mon Sep 17 00:00:00 2001 From: Colin Ian King Date: Mon, 26 Mar 2018 16:10:18 +0100 Subject: RDMA/hns: ensure for-loop actually iterates and free's buffers The current for-loop zeros variable i and only loops once, hence not all the buffers are free'd. Fix this by setting i correctly. Detected by CoverityScan, CID#1463415 ("Operands don't affect result") Fixes: a5073d6054f7 ("RDMA/hns: Add eq support of hip08") Signed-off-by: Colin Ian King Reviewed-by: Yixian Liu Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index db2ff352d75f..ec638778661c 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -4383,7 +4383,7 @@ err_dma_alloc_buf: eq->l0_dma = 0; if (mhop_num == 1) - for (i -= i; i >= 0; i--) + for (i -= 1; i >= 0; i--) dma_free_coherent(dev, buf_chk_sz, eq->buf[i], eq->buf_dma[i]); else if (mhop_num == 2) { -- cgit From 84652aefb347297aa08e91e283adf7b18f77c2d5 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Wed, 28 Mar 2018 11:27:22 -0700 Subject: RDMA/ucma: Introduce safer rdma_addr_size() variants There are several places in the ucma ABI where userspace can pass in a sockaddr but set the address family to AF_IB. When that happens, rdma_addr_size() will return a size bigger than sizeof struct sockaddr_in6, and the ucma kernel code might end up copying past the end of a buffer not sized for a struct sockaddr_ib. Fix this by introducing new variants int rdma_addr_size_in6(struct sockaddr_in6 *addr); int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr); that are type-safe for the types used in the ucma ABI and return 0 if the size computed is bigger than the size of the type passed in. We can use these new variants to check what size userspace has passed in before copying any addresses. Reported-by: Signed-off-by: Roland Dreier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/core/addr.c | 16 ++++++++++++++++ drivers/infiniband/core/ucma.c | 34 +++++++++++++++++----------------- include/rdma/ib_addr.h | 2 ++ 3 files changed, 35 insertions(+), 17 deletions(-) diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c index c3e6811bb1bd..cb1d2ab13c66 100644 --- a/drivers/infiniband/core/addr.c +++ b/drivers/infiniband/core/addr.c @@ -207,6 +207,22 @@ int rdma_addr_size(struct sockaddr *addr) } EXPORT_SYMBOL(rdma_addr_size); +int rdma_addr_size_in6(struct sockaddr_in6 *addr) +{ + int ret = rdma_addr_size((struct sockaddr *) addr); + + return ret <= sizeof(*addr) ? ret : 0; +} +EXPORT_SYMBOL(rdma_addr_size_in6); + +int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr) +{ + int ret = rdma_addr_size((struct sockaddr *) addr); + + return ret <= sizeof(*addr) ? ret : 0; +} +EXPORT_SYMBOL(rdma_addr_size_kss); + static struct rdma_addr_client self; void rdma_addr_register_client(struct rdma_addr_client *client) diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 21585055cf32..d933336d7e01 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -632,6 +632,9 @@ static ssize_t ucma_bind_ip(struct ucma_file *file, const char __user *inbuf, if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; + if (!rdma_addr_size_in6(&cmd.addr)) + return -EINVAL; + ctx = ucma_get_ctx(file, cmd.id); if (IS_ERR(ctx)) return PTR_ERR(ctx); @@ -645,22 +648,21 @@ static ssize_t ucma_bind(struct ucma_file *file, const char __user *inbuf, int in_len, int out_len) { struct rdma_ucm_bind cmd; - struct sockaddr *addr; struct ucma_context *ctx; int ret; if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - addr = (struct sockaddr *) &cmd.addr; - if (cmd.reserved || !cmd.addr_size || (cmd.addr_size != rdma_addr_size(addr))) + if (cmd.reserved || !cmd.addr_size || + cmd.addr_size != rdma_addr_size_kss(&cmd.addr)) return -EINVAL; ctx = ucma_get_ctx(file, cmd.id); if (IS_ERR(ctx)) return PTR_ERR(ctx); - ret = rdma_bind_addr(ctx->cm_id, addr); + ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr); ucma_put_ctx(ctx); return ret; } @@ -670,23 +672,22 @@ static ssize_t ucma_resolve_ip(struct ucma_file *file, int in_len, int out_len) { struct rdma_ucm_resolve_ip cmd; - struct sockaddr *src, *dst; struct ucma_context *ctx; int ret; if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - src = (struct sockaddr *) &cmd.src_addr; - dst = (struct sockaddr *) &cmd.dst_addr; - if (!rdma_addr_size(src) || !rdma_addr_size(dst)) + if (!rdma_addr_size_in6(&cmd.src_addr) || + !rdma_addr_size_in6(&cmd.dst_addr)) return -EINVAL; ctx = ucma_get_ctx(file, cmd.id); if (IS_ERR(ctx)) return PTR_ERR(ctx); - ret = rdma_resolve_addr(ctx->cm_id, src, dst, cmd.timeout_ms); + ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr, + (struct sockaddr *) &cmd.dst_addr, cmd.timeout_ms); ucma_put_ctx(ctx); return ret; } @@ -696,24 +697,23 @@ static ssize_t ucma_resolve_addr(struct ucma_file *file, int in_len, int out_len) { struct rdma_ucm_resolve_addr cmd; - struct sockaddr *src, *dst; struct ucma_context *ctx; int ret; if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - src = (struct sockaddr *) &cmd.src_addr; - dst = (struct sockaddr *) &cmd.dst_addr; - if (cmd.reserved || (cmd.src_size && (cmd.src_size != rdma_addr_size(src))) || - !cmd.dst_size || (cmd.dst_size != rdma_addr_size(dst))) + if (cmd.reserved || + (cmd.src_size && (cmd.src_size != rdma_addr_size_kss(&cmd.src_addr))) || + !cmd.dst_size || (cmd.dst_size != rdma_addr_size_kss(&cmd.dst_addr))) return -EINVAL; ctx = ucma_get_ctx(file, cmd.id); if (IS_ERR(ctx)) return PTR_ERR(ctx); - ret = rdma_resolve_addr(ctx->cm_id, src, dst, cmd.timeout_ms); + ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr, + (struct sockaddr *) &cmd.dst_addr, cmd.timeout_ms); ucma_put_ctx(ctx); return ret; } @@ -1433,7 +1433,7 @@ static ssize_t ucma_join_ip_multicast(struct ucma_file *file, join_cmd.response = cmd.response; join_cmd.uid = cmd.uid; join_cmd.id = cmd.id; - join_cmd.addr_size = rdma_addr_size((struct sockaddr *) &cmd.addr); + join_cmd.addr_size = rdma_addr_size_in6(&cmd.addr); if (!join_cmd.addr_size) return -EINVAL; @@ -1452,7 +1452,7 @@ static ssize_t ucma_join_multicast(struct ucma_file *file, if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - if (!rdma_addr_size((struct sockaddr *)&cmd.addr)) + if (!rdma_addr_size_kss(&cmd.addr)) return -EINVAL; return ucma_process_join(file, &cmd, out_len); diff --git a/include/rdma/ib_addr.h b/include/rdma/ib_addr.h index d656809f1217..415e09960017 100644 --- a/include/rdma/ib_addr.h +++ b/include/rdma/ib_addr.h @@ -130,6 +130,8 @@ void rdma_copy_addr(struct rdma_dev_addr *dev_addr, const unsigned char *dst_dev_addr); int rdma_addr_size(struct sockaddr *addr); +int rdma_addr_size_in6(struct sockaddr_in6 *addr); +int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr); int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid, const union ib_gid *dgid, -- cgit