From: Peter Dinda <pdinda@northwestern.edu>
Date: Wed, 13 Mar 2013 15:49:36 +0000 (-0500)
Subject: Support for optional full /dev/mem access even if host kernel has strict devmem
X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=commitdiff_plain;h=afdee70966b757149da1c1a47c5d0043a37336a1;p=palacios.releases.git

Support for optional full /dev/mem access even if host kernel has strict devmem

This adds a new option:

  insmod v3vee.ko allow_devmem=1

This will enable host user space programs to mmap any host physical memory using
/dev/mem even if the host kernel has strict /dev/mem limitations.
---

diff --git a/linux_module/Makefile b/linux_module/Makefile
index 3b8f699..a7eeb53 100644
--- a/linux_module/Makefile
+++ b/linux_module/Makefile
@@ -17,6 +17,7 @@ v3vee-y := 	palacios-stubs.o \
 		main.o \
 		vm.o \
 		mm.o \
+		allow_devmem.o \
 		util-queue.o \
 	        util-hashtable.o \
 		linux-exts.o
diff --git a/linux_module/allow_devmem.c b/linux_module/allow_devmem.c
new file mode 100644
index 0000000..696f98c
--- /dev/null
+++ b/linux_module/allow_devmem.c
@@ -0,0 +1,119 @@
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/kprobes.h>
+
+#include "palacios.h"
+#include "allow_devmem.h"
+
+/*
+  The purpose of this component is to disable and reenable
+  strict devmem control.
+
+  On kernels compiled with CONFIG_STRICT_DEVMEM, /dev/mem access
+  from user space is limited to the first 1 MB and to non-memory
+  regions (ie, devices).  For various purposes in Palacios,
+  for example linux_usr/v3_guest_mem.[ch], we want to allow 
+  host user space access to guest memory via mmap.   
+
+  This borrows from Dave Anderson @ Red Hat's implementation
+*/
+
+#define DEVMEM_CHECK_FUNC "devmem_is_allowed"
+
+static enum { NOTRUN=0, OPEN, SET, NOTSET } devmem_state = NOTRUN;
+
+
+// this is invoked after return from devmem_is_allowed()
+static int devmem_ret_handler(struct kretprobe_instance *ri, 
+			      struct pt_regs *regs)
+{
+  regs->ax = 1;  // yes, it's allowed - "ax"  now means "rax"
+  return 0;
+}
+
+static struct kretprobe devmem_kretprobe = {
+  .handler = devmem_ret_handler,
+  .maxactive = 20 // up to 20 at a time
+};
+
+
+int palacios_allow_devmem(void)
+{
+#ifndef CONFIG_STRICT_DEVMEM
+  INFO("System already has open /dev/mem - doing nothing\n");
+  devmem_state = OPEN;
+  return 0;
+#else
+  switch (devmem_state) { 
+  case NOTRUN:
+  case NOTSET: {
+    int rc;
+
+    devmem_kretprobe.kp.symbol_name = DEVMEM_CHECK_FUNC;
+    
+    rc = register_kretprobe(&devmem_kretprobe);
+    
+    if (rc<0) {
+      ERROR("register_kretprobe failed, returned %d\n", rc);
+      return -1;
+    }
+    
+    devmem_state=SET;
+    INFO("/dev/mem is now enabled (probe at %p)\n",devmem_kretprobe.kp.addr);
+
+    return 0;
+  }
+
+    break;
+  default:
+    // already set
+    return 0;
+    break;
+  }
+#endif    
+}
+
+
+int palacios_restore_devmem(void) 
+{
+#ifndef CONFIG_STRICT_DEVMEM
+  INFO("System already has open /dev/mem - doing nothing\n");
+  devmem_state = OPEN;
+  return 0;
+#else 
+  switch (devmem_state) { 
+  case NOTRUN: 
+    ERROR("Ignoring disable of dev mem\n");
+    return 0;
+  case NOTSET: 
+    INFO("/dev/mem not explicitly enabled, ignoring restore request\n");
+    return 0;
+    break;
+
+  case SET: {
+
+    unregister_kretprobe(&devmem_kretprobe);
+    
+    if (devmem_kretprobe.nmissed>0) { 
+      ERROR("Note: missed %d instances of %s\n", 
+	    devmem_kretprobe.nmissed, DEVMEM_CHECK_FUNC);
+    }
+    
+    devmem_state=NOTSET;
+
+    INFO("Restored strict /dev/mem access\n");
+    return 0;
+  }
+
+    break;
+  default:
+    // already set
+    return 0;
+    break;
+  }
+
+  return 0;
+#endif
+}
+
+
diff --git a/linux_module/allow_devmem.h b/linux_module/allow_devmem.h
new file mode 100644
index 0000000..3b7c229
--- /dev/null
+++ b/linux_module/allow_devmem.h
@@ -0,0 +1,9 @@
+#ifndef _ALLOW_DEVMEM
+#define _ALLOW_DEVMEM
+
+
+
+int palacios_allow_devmem(void);
+int palacios_restore_devmem(void);
+
+#endif
diff --git a/linux_module/main.c b/linux_module/main.c
index 296d01b..5c9ffa2 100644
--- a/linux_module/main.c
+++ b/linux_module/main.c
@@ -26,11 +26,11 @@
 #include "palacios.h"
 #include "mm.h"
 #include "vm.h"
+#include "allow_devmem.h"
 
 #include "linux-exts.h"
 
 
-
 MODULE_LICENSE("GPL");
 
 // Module parameter
@@ -39,11 +39,16 @@ int cpu_list_len = 0;
 module_param_array(cpu_list, int, &cpu_list_len, 0644);
 MODULE_PARM_DESC(cpu_list, "Comma-delimited list of CPUs that Palacios will run on");
 
+static int allow_devmem = 0;
+module_param(allow_devmem, int, 0);
+MODULE_PARM_DESC(allow_devmem, "Allow general user-space /dev/mem access even if kernel is strict");
+
 // Palacios options parameter
 static char *options;
 module_param(options, charp, 0);
 MODULE_PARM_DESC(options, "Generic options to internal Palacios modules");
 
+
 int mod_allocs = 0;
 int mod_frees = 0;
 
@@ -332,6 +337,9 @@ static int show_mem(char * buf, char ** start, off_t off, int count,
 }
 
 
+
+
+
 static int __init v3_init(void) {
     dev_t dev = MKDEV(0, 0); // We dynamicallly assign the major number
     int ret = 0;
@@ -339,6 +347,10 @@ static int __init v3_init(void) {
 
     palacios_init_mm();
 
+    if (allow_devmem) {
+      palacios_allow_devmem();
+    }
+
     // Initialize Palacios
     palacios_vmm_init(options);
 
@@ -460,6 +472,10 @@ static void __exit v3_exit(void) {
 
     deinit_lnx_extensions();
 
+    if (allow_devmem) {
+      palacios_restore_devmem();
+    }
+
     palacios_deinit_mm();
 
     remove_proc_entry("v3-guests", dir);