Re: [PATCH V2] IB/uverbs: Fix race between uverbs_close and remove_one

On Mon, Mar 7, 2016 at 4:44 PM, Yishai Hadas [off-list ref] wrote:

On 3/7/2016 11:44 AM, Devesh Sharma wrote:

quoted

Fixes: 35d4a0b63dc0 ("IB/uverbs: Fix race between ib_uverbs_open and
remove_one")

It fixes 036b10635739 (IB/uverbs: Enable device removal when there are
active user space applications) and not the commit that you pointed on.

quoted

While testing ocrdma for disassociate_ucontext support following
kernel panic was seen:

BUG: unable to handle kernel paging request at ffffffffa07ccd7a
[67139.981020] IP: [<ffffffffa07ccd7a>] 0xffffffffa07ccd7a
[67139.987185] PGD 19c5067 PUD 19c6063 PMD 469d08067 PTE 0
[67139.993370] Oops: 0010 [#1] SMP
[67140.257286] Call Trace:
[67140.260665]  [<ffffffff810c16a0>] ? prepare_to_wait_event+0xf0/0xf0
[67140.268337]  [<ffffffffa04cabc3>] ? ib_dereg_mr+0x23/0x30 [ib_core]
[67140.276009]  [<ffffffffa03ee5f0>] ?
ib_uverbs_cleanup_ucontext+0x320/0x440 [ib_uverbs]
[67140.285550]  [<ffffffffa03ee9e9>] ? ib_uverbs_close+0x59/0xb0
[ib_uverbs]
[67140.293807]  [<ffffffff811ff744>] ? __fput+0xe4/0x210
[67140.300132]  [<ffffffff811ff8ae>] ? ____fput+0xe/0x10
[67140.306457]  [<ffffffff8109b697>] ? task_work_run+0x77/0x90
[67140.313388]  [<ffffffff81081af2>] ? do_exit+0x2d2/0xab0
[67140.319910]  [<ffffffff8108234f>] ? do_group_exit+0x3f/0xa0
[67140.326821]  [<ffffffff8108d54c>] ? get_signal+0x1cc/0x5e0
[67140.333635]  [<ffffffff81017387>] ? do_signal+0x37/0x660
[67140.340257]  [<ffffffffa021669a>] ? ucma_write+0x7a/0xc0 [rdma_ucm]
[67140.347949]  [<ffffffff81079c37>] ? exit_to_usermode_loop+0x59/0xa2
[67140.355651]  [<ffffffff81003bad>] ? syscall_return_slowpath+0x8d/0xa0
[67140.363554]  [<ffffffff81680dcc>] ? int_ret_from_sys_call+0x25/0x8f
[67140.371259] Code:  Bad RIP value.
[67140.375678] RIP  [<ffffffffa07ccd7a>] 0xffffffffa07ccd7a
[67140.382314]  RSP <ffff880867727ad8>
[67140.386894] CR2: ffffffffa07ccd7a
[67140.393737] ---[ end trace 807b4472c30412d0 ]---
[67141.682413] Kernel panic - not syncing: Fatal exception
[67141.682431] Kernel Offset: disabled
[67141.733934] ---[ end Kernel panic - not syncing: Fatal exception

Root Cause:

During rmmod <vendor-driver> "ib_uverbs_close()" context is
still running, while "ib_uverbs_remove_one()" context completes and
ends up freeing ib_dev pointer, thus causing a Kernel Panic.

Kernel Panic -> kernel panic.

Will fix this in V3.

quoted

This patch fixes the race. ib_uverbs_close validates dev->ib_dev against
NULL
inside an srcu lock. If it is NULL, it waits for a completion and drops
the srcu
else continues with the normal flow.

Need to fix this description as of the expected code change, see below.
Please also describe why/how it solves the problem and not what it does.

Sure, will do that.

quoted

CC: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Devesh Sharma <devesh.sharma-dY08KVG/lbpWk0Htik3J/w@public.gmane.org>
---
  drivers/infiniband/core/uverbs.h      |  1 +
  drivers/infiniband/core/uverbs_main.c | 18 ++++++++++++++++--
  2 files changed, 17 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/core/uverbs.h

b/drivers/infiniband/core/uverbs.h
index 612ccfd..94a7339 100644

--- a/drivers/infiniband/core/uverbs.h
+++ b/drivers/infiniband/core/uverbs.h

@@ -121,6 +121,7 @@ struct ib_uverbs_file {
        struct ib_event_handler                 event_handler;
        struct ib_uverbs_event_file            *async_file;
        struct list_head                        list;
+       struct completion                       fcomp;
        int                                     is_closed;
  };

diff --git a/drivers/infiniband/core/uverbs_main.c

b/drivers/infiniband/core/uverbs_main.c
index 39680ae..9531168 100644

--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c

@@ -928,6 +928,7 @@ static int ib_uverbs_open(struct inode *inode, struct

file *filp)
        file->async_file = NULL;
        kref_init(&file->ref);
        mutex_init(&file->mutex);
+       init_completion(&file->fcomp);

        filp->private_data = file;
        kobject_get(&dev->kobj);

@@ -954,21 +955,33 @@ static int ib_uverbs_close(struct inode *inode,

struct file *filp)
        struct ib_uverbs_file *file = filp->private_data;
        struct ib_uverbs_device *dev = file->device;
        struct ib_ucontext *ucontext = NULL;
+       struct ib_device *ib_dev;
+       int srcu_key;

-       mutex_lock(&file->device->lists_mutex);
+       srcu_key = srcu_read_lock(&dev->disassociate_srcu);
+       ib_dev = srcu_dereference(dev->ib_dev,
+                                 &dev->disassociate_srcu);
+       if (!ib_dev)

You need to free the sruc lock, wait for completion then go to out.
Doing in the opposite order as you did below, might end up with a dead-lock
in case ib_uverbs_free_hw_resources is waiting for the
synchronize_srcu(disassociate_srcu) and can't mark the file as completed.

Agreed. Will fix it.

quoted

+               goto out;
+
+       mutex_lock(&dev->lists_mutex);
        ucontext = file->ucontext;
        file->ucontext = NULL;
        if (!file->is_closed) {
                list_del(&file->list);
                file->is_closed = 1;
        }
-       mutex_unlock(&file->device->lists_mutex);
+       mutex_unlock(&dev->lists_mutex);
        if (ucontext)
                ib_uverbs_cleanup_ucontext(file, ucontext);

        if (file->async_file)
                kref_put(&file->async_file->ref,
ib_uverbs_release_event_file);

+       complete(&file->fcomp);

No need to do that in that flow, only in above flow.

Agreed.

quoted

+out:
+       wait_for_completion(&file->fcomp);
+       srcu_read_unlock(&dev->disassociate_srcu, srcu_key);

See above, might end-up with a dead-lock, should be done above in opposite
order.

Will fix in next version.

quoted

        kref_put(&file->ref, ib_uverbs_release_file);
        kobject_put(&dev->kobj);

@@ -1199,6 +1212,7 @@ static void ib_uverbs_free_hw_resources(struct

ib_uverbs_device *uverbs_dev,
                }

                mutex_lock(&uverbs_dev->lists_mutex);
+               complete(&file->fcomp);
                kref_put(&file->ref, ib_uverbs_release_file);
        }

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html