VBoxGuestLib

Timestamp:

Apr 21, 2009 8:57:38 PM (16 years ago)

Author:

vboxsync

svn:sync-xref-src-repo-rev:

46239

Message:

Additions/common and linux: don'\t try to lock zero-length buffers when we do HGCM copies

File:

trunk/src/VBox/Additions/common/VBoxGuestLib/SysHlp.cpp

-              r15606
+              r19091
     int rc = VINF_SUCCESS;
+    /* Ugly edge case - zero size buffers shouldn't be locked. */
+    if (u32Size == 0)
+    {
+        *ppvCtx = NIL_RTR0MEMOBJ;
+        return VINF_SUCCESS;
+    }
 #ifdef RT_OS_WINDOWS
     PMDL pMdl = IoAllocateMdl (pv, u32Size, FALSE, FALSE, NULL);
 …
     /** @todo r=frank: Linux: pv is at least in some cases, e.g. with VBoxMapFolder,
      *  an R0 address -- the memory was allocated with kmalloc(). I don't know
+     *  if this is true in any case. */
+     *  if this is true in any case.
+     * r=michael: on Linux, we sometimes have R3 addresses (e.g. shared
+     *  clipboard) and sometimes R0 (e.g. shared folders).  We really ought
+     *  to have two separate paths here - at any rate, Linux R0 shouldn't
+     *  end up calling this API.  In practice, Linux R3 does it's own thing
+     *  before winding up in the R0 path - which calls this stub API.
+     */
     NOREF(ppvCtx);
     NOREF(pv);
 …
     NOREF(u32Size);
+    /* Ugly edge case - zero size buffers aren't be locked. */
+    if (pvCtx == NIL_RTR0MEMOBJ)
+        return;
 #ifdef RT_OS_WINDOWS
     PMDL pMdl = (PMDL)pvCtx;

Note: See TracChangeset for help on using the changeset viewer.