volser: add more logs for failures during restore

[openafs.git] / src / volser / dumpstuff.c

diff --git a/src/volser/dumpstuff.c b/src/volser/dumpstuff.c
index 8c01aa0..546b4f6 100644 (file)
--- a/src/volser/dumpstuff.c
+++ b/src/volser/dumpstuff.c
@@ -265,12 +265,17 @@ ReadString(struct iod *iodp, char *to, int maxa)
     }
 }
 
-static void
-ReadByteString(struct iod *iodp, byte * to,
-              int size)
+static int
+ReadByteString(struct iod *iodp, byte * to, int size)
 {
-    while (size--)
-       *to++ = iod_getc(iodp);
+    int nbytes = 0;
+    int c;
+
+    while (size-- > 0 && (c = iod_getc(iodp)) != EOF) {
+       *to++ = c;
+       nbytes++;
+    }
+    return nbytes;
 }
 
 /*
@@ -1093,7 +1098,11 @@ DumpVnode(struct iod *iodp, struct VnodeDiskObject *v, VolumeId volid,
     if (!code)
        code = DumpInt32(iodp, 's', v->serverModifyTime);
     if (v->type == vDirectory) {
-       acl_HtonACL(VVnodeDiskACL(v));
+       code = acl_HtonACL(VVnodeDiskACL(v));
+       if (code) {
+           Log("DumpVnode: Skipping invalid acl vnode %u (volume %"AFS_VOLID_FMT")\n",
+                vnodeNumber, afs_printable_VolumeId_lu(volid));
+       }
        if (!code)
            code =
                DumpByteString(iodp, 'A', (byte *) VVnodeDiskACL(v),
@@ -1238,6 +1247,7 @@ RestoreVolume(struct rx_call *call, Volume * avp, int incremental,
     int s1 = 0, s2 = 0, delo = 0, tdelo;
     int tag;
     VolumeDiskData saved_header;
+    afs_uint32 uptime, crtime;
 
     iod_Init(iodp, call);
 
@@ -1255,8 +1265,11 @@ RestoreVolume(struct rx_call *call, Volume * avp, int incremental,
        Log("1 Volser: RestoreVolume: Volume header missing from dump; not restored\n");
        return VOLSERREAD_DUMPERROR;
     }
-    if (ReadVolumeHeader(iodp, &vol) == VOLSERREAD_DUMPERROR)
+    if (ReadVolumeHeader(iodp, &vol) == VOLSERREAD_DUMPERROR) {
+       Log("1 Volser: RestoreVolume: Error reading volume header (id: %u); aborted\n",
+           V_id(vp));
        return VOLSERREAD_DUMPERROR;
+    }
 
     if (!delo)
        delo = ProcessIndex(vp, vLarge, &b1, &s1, 0);
@@ -1278,6 +1291,8 @@ RestoreVolume(struct rx_call *call, Volume * avp, int incremental,
     tdelo = delo;
     while (1) {
        if (ReadVnodes(iodp, vp, 0, b1, s1, b2, s2, tdelo)) {
+           Log("1 Volser: RestoreVolume: Error reading vnodes (id: %u); aborted\n",
+               V_id(vp));
            error = VOLSERREAD_DUMPERROR;
            goto clean;
        }
@@ -1286,6 +1301,8 @@ RestoreVolume(struct rx_call *call, Volume * avp, int incremental,
            break;
 
        if (ReadVolumeHeader(iodp, &vol) == VOLSERREAD_DUMPERROR) {
+           Log("1 Volser: RestoreVolume: Error reading volume header (id: %u); aborted\n",
+               V_id(vp));
            error = VOLSERREAD_DUMPERROR;
            goto out;
        }
@@ -1325,6 +1342,8 @@ RestoreVolume(struct rx_call *call, Volume * avp, int incremental,
         * prevent it from getting overwritten. */
        vol.needsSalvaged = V_needsSalvaged(vp);
     }
+    crtime = V_creationDate(vp);
+    uptime = V_updateDate(vp);
     CopyVolumeHeader(&vol, &V_disk(vp));
     V_destroyMe(vp) = 0;
     VUpdateVolume(&vupdate, vp);
@@ -1332,6 +1351,20 @@ RestoreVolume(struct rx_call *call, Volume * avp, int incremental,
        Log("1 Volser: RestoreVolume: Unable to rewrite volume header; restore aborted\n");
        error = VOLSERREAD_DUMPERROR;
        goto out;
+    } else {
+       /*
+        * If the volume was not a new empty volume and the restored dump was
+        * older than the volume in question, this is probably a mistake, and
+        * may mean the resulting volume is corrupted. Log the following message
+        * to give a clue as to why this volume suddenly looks strange or corrupt.
+        */
+       if ((crtime != uptime) && (uptime > V_updateDate(vp))) {
+           Log("1 Volser: RestoreVolume: volume %s (%u) appears to have been partially or "
+               "completely restored to an earlier version (updateDate went from %u to %u). "
+               "This is allowed, but may indicate a mistake in whatever tool is restoring "
+               "this volume. If this volume appears corrupted, this is probably why.\n",
+               V_name(vp), V_id(vp), uptime, V_updateDate(vp));
+       }
     }
   out:
     /* Free the malloced space above */
@@ -1359,6 +1392,7 @@ ReadVnodes(struct iod *iodp, Volume * vp, int incremental,
     FdHandle_t *fdP;
     Inode nearInode AFS_UNUSED;
     afs_int32 critical = 0;
+    int nbytes;
 
     tag = iod_getc(iodp);
     V_pref(vp, nearInode);
@@ -1424,9 +1458,18 @@ ReadVnodes(struct iod *iodp, Volume * vp, int incremental,
                    return VOLSERREAD_DUMPERROR;
                break;
            case 'A':
-               ReadByteString(iodp, (byte *) VVnodeDiskACL(vnode),
+               nbytes = ReadByteString(iodp, (byte *) VVnodeDiskACL(vnode),
                               VAclDiskSize(vnode));
-               acl_NtohACL(VVnodeDiskACL(vnode));
+               if (nbytes != VAclDiskSize(vnode)) {
+                   Log("ReadVnodes: could not read acl for vnode %lu in dump.\n",
+                        (unsigned long)vnodeNumber);
+                   return VOLSERREAD_DUMPERROR;
+               }
+               if (acl_NtohACL(VVnodeDiskACL(vnode)) != 0) {
+                   Log("ReadVnodes: invalid acl for vnode %lu in dump.\n",
+                        (unsigned long)vnodeNumber);
+                   return VOLSERREAD_DUMPERROR;
+               }
                break;
            case 'h':
            case 'f':{