[libvirt] [PATCH 09/23] Add support for disks with LXC

[Daniel P. Berrange]

From: "Daniel P. Berrange" <berrange at redhat.com>

Currently LXC guests can be given arbitrary pre-mounted
filesystems, however, for some usecases it is more appropriate
to provide block devices which the container can mount itself.
This first impl only allows for <disk type='block'>, in other
words exposing a host disk device to a container. Since LXC
does not have device namespace virtualization, we are cheating
a little bit. If the XML specifies /dev/sdc4 to be given to
the container as /dev/sda1, when we do the mknod /dev/sda1
in the container's /dev, we actually use the major:minor
number of /dev/sdc4, not /dev/sda1.

Signed-off-by: Daniel P. Berrange <berrange at redhat.com>
---
 src/lxc/lxc_cgroup.c    |  18 +++++++++
 src/lxc/lxc_container.c | 104 ++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 122 insertions(+)

diff --git a/src/lxc/lxc_cgroup.c b/src/lxc/lxc_cgroup.c
index 767ef26..0636869 100644
--- a/src/lxc/lxc_cgroup.c
+++ b/src/lxc/lxc_cgroup.c
@@ -332,6 +332,24 @@ static int virLXCCgroupSetupDeviceACL(virDomainDefPtr def,
         }
     }
 
+    for (i = 0 ; i < def->ndisks ; i++) {
+        if (def->disks[i]->type != VIR_DOMAIN_DISK_TYPE_BLOCK)
+            continue;
+
+        rc = virCgroupAllowDevicePath(cgroup,
+                                      def->disks[i]->src,
+                                      (def->disks[i]->readonly ?
+                                       VIR_CGROUP_DEVICE_READ :
+                                       VIR_CGROUP_DEVICE_RW) |
+                                      VIR_CGROUP_DEVICE_MKNOD);
+        if (rc != 0) {
+            virReportSystemError(-rc,
+                                 _("Unable to allow device %s for domain %s"),
+                                 def->disks[i]->src, def->name);
+            goto cleanup;
+        }
+    }
+
     for (i = 0 ; i < def->nfss ; i++) {
         if (def->fss[i]->type != VIR_DOMAIN_FS_TYPE_BLOCK)
             continue;
diff --git a/src/lxc/lxc_container.c b/src/lxc/lxc_container.c
index 3014564..be26de9 100644
--- a/src/lxc/lxc_container.c
+++ b/src/lxc/lxc_container.c
@@ -1211,6 +1211,106 @@ static int lxcContainerMountAllFS(virDomainDefPtr vmDef,
 }
 
 
+static int lxcContainerSetupDisk(virDomainDefPtr vmDef,
+                                 virDomainDiskDefPtr def,
+                                 const char *dstprefix,
+                                 virSecurityManagerPtr securityDriver)
+{
+    char *src = NULL;
+    char *dst = NULL;
+    int ret = -1;
+    struct stat sb;
+    mode_t mode;
+    char *tmpsrc = def->src;
+
+    if (def->type != VIR_DOMAIN_DISK_TYPE_BLOCK) {
+        virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
+                       _("Can't setup disk for non-block device"));
+        goto cleanup;
+    }
+    if (def->src == NULL) {
+        virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
+                       _("Can't setup disk without media"));
+        goto cleanup;
+    }
+
+    if (virAsprintf(&src, "%s/%s", dstprefix, def->src) < 0) {
+        virReportOOMError();
+        goto cleanup;
+    }
+
+    if (virAsprintf(&dst, "/dev/%s", def->dst) < 0) {
+        virReportOOMError();
+        goto cleanup;
+    }
+
+    if (stat(src, &sb) < 0) {
+        virReportSystemError(errno,
+                             _("Unable to access %s"), def->src);
+        goto cleanup;
+    }
+
+    if (!S_ISCHR(sb.st_mode) && !S_ISBLK(sb.st_mode)) {
+        virReportError(VIR_ERR_CONFIG_UNSUPPORTED,
+                       _("Disk source %s must be a character/block device"),
+                       def->src);
+        goto cleanup;
+    }
+
+    mode = 0700;
+    if (S_ISCHR(sb.st_mode))
+        mode |= S_IFCHR;
+    else
+        mode |= S_IFBLK;
+
+    /* Yes, the device name we're creating may not
+     * actually correspond to the major:minor number
+     * we're using, but we've no other option at this
+     * time. Just have to hope that containerized apps
+     * don't get upset that the major:minor is different
+     * to that normally implied by the device name
+     */
+    VIR_DEBUG("Creating dev %s (%d,%d) from %s",
+              dst, major(sb.st_rdev), minor(sb.st_rdev), src);
+    if (mknod(dst, mode, sb.st_rdev) < 0) {
+        virReportSystemError(errno,
+                             _("Unable to create device %s"),
+                             dst);
+        goto cleanup;
+    }
+    /* Labelling normally operates on src, but we need
+     * to actally label the dst here, so hack the config */
+    def->src = dst;
+    if (virSecurityManagerSetImageLabel(securityDriver, vmDef, def) < 0)
+        goto cleanup;
+
+    ret = 0;
+
+cleanup:
+    def->src = tmpsrc;
+    VIR_FREE(src);
+    VIR_FREE(dst);
+    return ret;
+}
+
+static int lxcContainerSetupAllDisks(virDomainDefPtr vmDef,
+                                     const char *dstprefix,
+                                     virSecurityManagerPtr securityDriver)
+{
+    size_t i;
+    VIR_DEBUG("Setting up disks %s", dstprefix);
+
+    for (i = 0 ; i < vmDef->ndisks ; i++) {
+        if (lxcContainerSetupDisk(vmDef, vmDef->disks[i],
+                                  dstprefix, securityDriver) < 0)
+            return -1;
+    }
+
+    VIR_DEBUG("Setup all disks");
+    return 0;
+}
+
+
 static int lxcContainerGetSubtree(const char *prefix,
                                   char ***mountsret,
                                   size_t *nmountsret)
@@ -1606,6 +1706,10 @@ static int lxcContainerSetupPivotRoot(virDomainDefPtr vmDef,
     if (lxcContainerMountAllFS(vmDef, "/.oldroot", true, sec_mount_options) < 0)
         goto cleanup;
 
+    /* Sets up any extra disks from guest config */
+    if (lxcContainerSetupAllDisks(vmDef, "/.oldroot", securityDriver) < 0)
+        goto cleanup;
+
     /* Gets rid of all remaining mounts from host OS, including /.oldroot itself */
     if (lxcContainerUnmountSubtree("/.oldroot", true) < 0)
         goto cleanup;
-- 
1.8.0.1