thebeanogamer / rpms / qemu-kvm

Forked from rpms/qemu-kvm 5 months ago
Clone
016a62
From 155e3ab0c8a7886781755ba44849c77048815025 Mon Sep 17 00:00:00 2001
016a62
From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
016a62
Date: Wed, 15 Jan 2020 12:07:41 +0100
016a62
Subject: [PATCH 3/7] usbredir: Prevent recursion in usbredir_write
016a62
MIME-Version: 1.0
016a62
Content-Type: text/plain; charset=UTF-8
016a62
Content-Transfer-Encoding: 8bit
016a62
016a62
RH-Author: Dr. David Alan Gilbert <dgilbert@redhat.com>
016a62
Message-id: <20200115120742.19583-2-dgilbert@redhat.com>
016a62
Patchwork-id: 93352
016a62
O-Subject: [RHEL-8.2.0 qemu-kvm PATCH 1/2] usbredir: Prevent recursion in usbredir_write
016a62
Bugzilla: 1752320
016a62
RH-Acked-by: Gerd Hoffmann <kraxel@redhat.com>
016a62
RH-Acked-by: Philippe Mathieu-Daudé <philmd@redhat.com>
016a62
RH-Acked-by: Peter Xu <peterx@redhat.com>
016a62
016a62
From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
016a62
016a62
I've got a case where usbredir_write manages to call back into itself
016a62
via spice; this patch causes the recursion to fail (0 bytes) the write;
016a62
this seems to avoid the deadlock I was previously seeing.
016a62
016a62
I can't say I fully understand the interaction of usbredir and spice;
016a62
but there are a few similar guards in spice and usbredir
016a62
to catch other cases especially onces also related to spice_server_char_device_wakeup
016a62
016a62
This case seems to be triggered by repeated migration+repeated
016a62
reconnection of the viewer; but my debugging suggests the migration
016a62
finished before this hits.
016a62
016a62
The backtrace of the hang looks like:
016a62
  reds_handle_ticket
016a62
  reds_handle_other_links
016a62
  reds_channel_do_link
016a62
  red_channel_connect
016a62
  spicevmc_connect
016a62
  usbredir_create_parser
016a62
  usbredirparser_do_write
016a62
  usbredir_write
016a62
  qemu_chr_fe_write
016a62
  qemu_chr_write
016a62
  qemu_chr_write_buffer
016a62
  spice_chr_write
016a62
  spice_server_char_device_wakeup
016a62
  red_char_device_wakeup
016a62
  red_char_device_write_to_device
016a62
  vmc_write
016a62
  usbredirparser_do_write
016a62
  usbredir_write
016a62
  qemu_chr_fe_write
016a62
  qemu_chr_write
016a62
  qemu_chr_write_buffer
016a62
  qemu_mutex_lock_impl
016a62
016a62
and we fail as we land through qemu_chr_write_buffer's lock
016a62
twice.
016a62
016a62
Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1752320
016a62
016a62
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
016a62
Message-Id: <20191218113012.13331-1-dgilbert@redhat.com>
016a62
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
016a62
(cherry picked from commit 394642a8d3742c885e397d5bb5ee0ec40743cdc6)
016a62
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
016a62
---
016a62
 hw/usb/redirect.c | 9 +++++++++
016a62
 1 file changed, 9 insertions(+)
016a62
016a62
diff --git a/hw/usb/redirect.c b/hw/usb/redirect.c
016a62
index 65a9196..3dcf76e 100644
016a62
--- a/hw/usb/redirect.c
016a62
+++ b/hw/usb/redirect.c
016a62
@@ -107,6 +107,7 @@ struct USBRedirDevice {
016a62
     /* Properties */
016a62
     CharBackend cs;
016a62
     bool enable_streams;
016a62
+    bool in_write;
016a62
     uint8_t debug;
016a62
     int32_t bootindex;
016a62
     char *filter_str;
016a62
@@ -284,6 +285,13 @@ static int usbredir_write(void *priv, uint8_t *data, int count)
016a62
         return 0;
016a62
     }
016a62
 
016a62
+    /* Recursion check */
016a62
+    if (dev->in_write) {
016a62
+        DPRINTF("usbredir_write recursion\n");
016a62
+        return 0;
016a62
+    }
016a62
+    dev->in_write = true;
016a62
+
016a62
     r = qemu_chr_fe_write(&dev->cs, data, count);
016a62
     if (r < count) {
016a62
         if (!dev->watch) {
016a62
@@ -294,6 +302,7 @@ static int usbredir_write(void *priv, uint8_t *data, int count)
016a62
             r = 0;
016a62
         }
016a62
     }
016a62
+    dev->in_write = false;
016a62
     return r;
016a62
 }
016a62
 
016a62
-- 
016a62
1.8.3.1
016a62