Re: [libvirt] PATCH: Support storage copy on write volumes

22 Jan 2009

"Daniel P. Berrange" <berrange@redhat.com> wrote:
...
This is a follow up on Miloslav's proposal to add copy on write
support to the storage APIs, changing the XML to that described
here:
http://www.redhat.com/archives/libvir-list/2009-January/msg00231.html
In addition to the original QCOW/VMDK support, I have done an impl which
can extract the backing store for LVM volumes
This looks fine, but I'd feel a lot better about it
if there were a few tests of the new bits.
A few questions/suggestions:
...

diff --git a/src/storage_backend.c b/src/storage_backend.c
...
@@ -235,38 +263,38 @@ virStorageBackendUpdateVolInfoFD(virConn
                 continue;
             if (memcmp(buffer+disk_types[i].offset, &disk_types[i].magic,
                 disk_types[i].length) == 0) {
-                vol->target.format = disk_types[i].part_table_type;
+                target->format = disk_types[i].part_table_type;
                 break;
             }
         }
     }
-    vol->target.perms.mode = sb.st_mode;
-    vol->target.perms.uid = sb.st_uid;
-    vol->target.perms.gid = sb.st_gid;
+    target->perms.mode = sb.st_mode & (0x200-1);
How about S_IRWXUGO:

       target->perms.mode = sb.st_mode & S_IRWXUGO;
...
+    target->perms.uid = sb.st_uid;
+    target->perms.gid = sb.st_gid;
...
diff --git a/src/storage_backend_fs.c b/src/storage_backend_fs.c
...
+static int
+cowGetBackingStore(virConnectPtr conn,
+                   char **res,
+                   const unsigned char *buf,
+                   size_t buf_size)
+{
+    size_t len;
+
+    *res = NULL;
+    if (buf_size < 4+4+1024)
+        return BACKING_STORE_INVALID;
+    if (buf[4+4] == '\0') /* cow_header_v2.backing_file[0] */
+        return BACKING_STORE_OK;
+
+    len = 1024;
+    if (VIR_ALLOC_N(*res, len + 1) < 0) {
+        virStorageReportError(conn, VIR_ERR_NO_MEMORY, _("backing store path"));
+        return BACKING_STORE_ERROR;
+    }
+    memcpy(*res, buf + 4+4, len); /* cow_header_v2.backing_file */
+    (*res)[len] = '\0';
+    if (VIR_REALLOC_N(*res, strlen(*res) + 1) < 0) {
Is this just-copied 1024-byte block of data guaranteed
not to contain any NUL bytes?  Or maybe you just want that
NUL-terminated string?
...
+        /* Ignore failure */
+    }
+    return BACKING_STORE_OK;
+}
...
@@ -846,17 +1081,36 @@ virStorageBackendFileSystemVolCreate(vir
                                   vol->target.format);
             return -1;
         }
+        if (vol->backingStore.path) {
+            if ((backingType = virStorageVolFormatFileSystemTypeToString(vol->backingStore.format)) == NULL) {
+                virStorageReportError(conn, VIR_ERR_INTERNAL_ERROR,
+                                      _("unknown storage vol backing store type %d"),
+                                      vol->backingStore.format);
+                return -1;
+            }
+            if (access(vol->backingStore.path, R_OK) != 0) {
+                virReportSystemError(conn, errno,
+                                     _("inaccessible backing store volume %s"),
+                                     vol->backingStore.path);
+                return -1;
+            }
+        }
/* Size in KB */
         snprintf(size, sizeof(size), "%llu", vol->capacity/1024);
-        imgargv[0] = QEMU_IMG;
-        imgargv[1] = "create";
-        imgargv[2] = "-f";
-        imgargv[3] = type;
-        imgargv[4] = vol->target.path;
-        imgargv[5] = size;
-        imgargv[6] = NULL;
+        i = 0;
+        imgargv[i++] = QEMU_IMG;
+        imgargv[i++] = "create";
+        imgargv[i++] = "-f";
+        imgargv[i++] = type;
+        if (vol->backingStore.path != NULL) {
+            imgargv[i++] = "-b";
+            imgargv[i++] = vol->backingStore.path;
+        }
+        imgargv[i++] = vol->target.path;
+        imgargv[i++] = size;
+        imgargv[i++] = NULL;
Add an assertion like

   assert (i < ARRAY_CARDINALITY (imgargv));

Otherwise, it's far too easy to forget to update the "9"
above when adding a new option here.

...
...
diff --git a/src/storage_backend_iscsi.c b/src/storage_backend_iscsi.c
--- a/src/storage_backend_iscsi.c
+++ b/src/storage_backend_iscsi.c
@@ -226,7 +226,11 @@ virStorageBackendISCSINewLun(virConnectP
VIR_FREE(devpath);
-    if (virStorageBackendUpdateVolInfoFD(conn, vol, fd, 1) < 0)
+    if (virStorageBackendUpdateVolTargetInfoFD(conn,
+                                               &vol->target,
+                                               fd,
+                                               &vol->allocation,
+                                               &vol->capacity) < 0)
         goto cleanup;
/* XXX use unique iSCSI id instead */
diff --git a/src/storage_backend_logical.c b/src/storage_backend_logical.c
--- a/src/storage_backend_logical.c
+++ b/src/storage_backend_logical.c
@@ -116,8 +116,22 @@ virStorageBackendLogicalMakeVol(virConne
         strcat(vol->target.path, vol->name);
     }
+    if (groups[1] && !STREQ(groups[1], "")) {
+        if (VIR_ALLOC_N(vol->backingStore.path, strlen(pool->def->target.path) +
+                        1 + strlen(groups[1]) + 1) < 0) {
+            virStorageReportError(conn, VIR_ERR_NO_MEMORY, "%s", _("volume"));
+            return -1;
+        }
+        strcpy(vol->backingStore.path, pool->def->target.path);
+        strcat(vol->backingStore.path, "/");
+        strcat(vol->backingStore.path, groups[1]);
How about using virAsprintf in place of VIR_ALLOC_N + strcpy + strcat?

    

Re: [libvirt] PATCH: Support storage copy on write volumes

Jim Meyering