Memory Leak in vboxdrv

[Larry Finger]

I detected a small memory leak in the openSUSE version that I maintain. To see if the problem had been fixed, I installed test build revision 143492. It has the same problem.

The kmemleak report is as follows:

localhost:cat > /sys/kernel/debug/kmemleak
unreferenced object 0xffff88827c95b380 (size 128):
  comm "EMT-0", pid 4550, jiffies 4294918407 (age 94.516s)
  hex dump (first 32 bytes):
    19 01 09 18 01 00 00 80 58 00 00 00 58 00 00 00  ........X...X...
    10 12 61 19 58 00 00 00 08 00 00 00 00 00 00 00  ..a.X...........
  backtrace:
    [<000000008fe681f2>] rtR0MemAllocEx+0xc5/0x130 [vboxdrv]
    [<00000000064e6217>] VBoxHost_RTMemAllocZTag+0x2a/0x70 [vboxdrv]
    [<00000000523d21fa>] rtR0MemObjNew+0x37/0x70 [vboxdrv]
    [<00000000df0eec39>] rtR0MemObjNativeMapKernel+0x57/0x140 [vboxdrv]
    [<000000008419dc13>] VBoxHost_RTR0MemObjMapKernelExTag+0x12d/0x1f0 [vboxdrv]
    [<00000000ec3fe3f7>] SUPR0PageMapKernel+0x1e2/0x270 [vboxdrv]
unreferenced object 0xffff88827c95b600 (size 128):
  comm "EMT-0", pid 4550, jiffies 4294918412 (age 94.496s)
  hex dump (first 32 bytes):
    19 01 09 18 01 00 00 80 58 00 00 00 58 00 00 00  ........X...X...
    10 12 61 19 58 00 00 00 08 00 00 00 00 00 00 00  ..a.X...........
  backtrace:
    [<000000008fe681f2>] rtR0MemAllocEx+0xc5/0x130 [vboxdrv]
    [<00000000064e6217>] VBoxHost_RTMemAllocZTag+0x2a/0x70 [vboxdrv]
    [<00000000523d21fa>] rtR0MemObjNew+0x37/0x70 [vboxdrv]
    [<00000000df0eec39>] rtR0MemObjNativeMapKernel+0x57/0x140 [vboxdrv]
    [<000000008419dc13>] VBoxHost_RTR0MemObjMapKernelExTag+0x12d/0x1f0 [vboxdrv]
    [<00000000ec3fe3f7>] SUPR0PageMapKernel+0x1e2/0x270 [vboxdrv]
unreferenced object 0xffff88827c95b680 (size 128):
  comm "EMT-0", pid 4550, jiffies 4294918412 (age 94.496s)
  hex dump (first 32 bytes):
    19 01 09 18 01 00 00 80 58 00 00 00 58 00 00 00  ........X...X...
    10 12 61 19 58 00 00 00 08 00 00 00 00 00 00 00  ..a.X...........
  backtrace:
    [<000000008fe681f2>] rtR0MemAllocEx+0xc5/0x130 [vboxdrv]
    [<00000000064e6217>] VBoxHost_RTMemAllocZTag+0x2a/0x70 [vboxdrv]
    [<00000000523d21fa>] rtR0MemObjNew+0x37/0x70 [vboxdrv]
    [<00000000df0eec39>] rtR0MemObjNativeMapKernel+0x57/0x140 [vboxdrv]
    [<000000008419dc13>] VBoxHost_RTR0MemObjMapKernelExTag+0x12d/0x1f0 [vboxdrv]
    [<00000000ec3fe3f7>] SUPR0PageMapKernel+0x1e2/0x270 [vboxdrv]
localhost:~ # 

To trigger this leak, load the GUI, start and stop a VM, and then close the GUI. Until the final step, the kmemleak scan is clean. Note, this is a real leak. If vboxdrv is unloaded, the backtrace info is lost. The first leaked block is reported as:

unreferenced object 0xffff88827c95b380 (size 128):
  comm "EMT-0", pid 4550, jiffies 4294918407 (age 1095.172s)
  hex dump (first 32 bytes):
    19 01 09 18 01 00 00 80 58 00 00 00 58 00 00 00  ........X...X...
    10 12 61 19 58 00 00 00 08 00 00 00 00 00 00 00  ..a.X...........
  backtrace:
    [<000000008fe681f2>] 0xffffffffa0c9fb05
    [<00000000064e6217>] 0xffffffffa0c9d82a
    [<00000000523d21fa>] 0xffffffffa0c9edd7
    [<00000000df0eec39>] 0xffffffffa0ca1067
    [<000000008419dc13>] 0xffffffffa0c9e55d
    [<00000000ec3fe3f7>] 0xffffffffa0c8c1b2

[aeichner]

Thanks for the report, this should be fixed in the next maintenance release. The attached patch fixes the issue as well, would be great if you could try it out.