r24302: add a module that overloads the file_id_create VFS function
authorStefan Metzmacher <metze@samba.org>
Fri, 10 Aug 2007 09:05:18 +0000 (09:05 +0000)
committerGerald (Jerry) Carter <jerry@samba.org>
Wed, 10 Oct 2007 17:29:25 +0000 (12:29 -0500)
and alters the device id depending on the configured algorithm.

The algorithm is configured via "fileid:algorithm":
- "fsname" (default) uses a uint64 hash over the mount point
- "fsid" uses the fsid returned from statfs()

This is needed for "clustering = yes" on some clusterfilesystems

metze

source/Makefile.in
source/configure.in
source/modules/vfs_fileid.c [new file with mode: 0644]

index 4ffebdac35ae40c7caee9e6aabf4daf93a19adaf..36b5df1beddc39aa4b3053b6c2661adfad71d305 100644 (file)
@@ -479,6 +479,7 @@ VFS_COMMIT_OBJ = modules/vfs_commit.o
 VFS_GPFS_OBJ = modules/vfs_gpfs.o modules/gpfs.o modules/nfs4_acls.o
 VFS_NOTIFY_FAM_OBJ = modules/vfs_notify_fam.o
 VFS_READAHEAD_OBJ = modules/vfs_readahead.o
+VFS_FILEID_OBJ = modules/vfs_fileid.o
 
 PLAINTEXT_AUTH_OBJ = auth/pampass.o auth/pass_check.o
 
@@ -1681,6 +1682,10 @@ bin/readahead.@SHLIBEXT@: $(BINARY_PREREQS) $(VFS_READAHEAD_OBJ)
        @echo "Building plugin $@"
        @$(SHLD_MODULE) $(VFS_READAHEAD_OBJ)
 
+bin/fileid.@SHLIBEXT@: $(BINARY_PREREQS) $(VFS_FILEID_OBJ)
+       @echo "Building plugin $@"
+       @$(SHLD_MODULE) $(VFS_FILEID_OBJ)
+
 #########################################################
 ## IdMap NSS plugins
 
index d1ef464202e534cb6c5736ba0809bb4f00c8908d..4029a0a10c0f3efa84e8ac593f5abccf3d81a01e 100644 (file)
@@ -6236,6 +6236,38 @@ AC_ARG_WITH(python,
   esac ])
 AC_SUBST(PYTHON)
 
+
+# Checks for the vfs_fileid module
+# Start
+AC_CHECK_FUNC(getmntent)
+
+AC_CHECK_HEADERS(sys/statfs.h)
+
+AC_MSG_CHECKING([vfs_fileid: checking for statfs() and struct statfs.f_fsid)])
+AC_CACHE_VAL(vfsfileid_cv_statfs,[
+            AC_TRY_RUN([
+               #include <sys/types.h>
+               #include <sys/statfs.h>
+               int main(void)
+               {
+                       struct statfs fsd;
+                       fsid_t fsid = fsd.f_fsid;
+                       return statfs (".", &fsd);
+               }],
+               vfsfileid_cv_statfs=yes,
+               vfsfileid_cv_statfs=no,
+               vfsfileid_cv_statfs=cross)
+])
+AC_MSG_RESULT($vfsfileid_cv_statfs)
+
+if test x"$ac_cv_func_getmntent" = x"yes" -a \
+       x"$vfsfileid_cv_statfs" = x"yes"; then
+       default_shared_modules="$default_shared_modules vfs_fileid"
+fi
+# End
+# Checks for the vfs_fileid module
+
+
 for i in `echo $default_static_modules | sed -e 's/,/ /g'`
 do
        eval MODULE_DEFAULT_$i=STATIC
@@ -6356,6 +6388,7 @@ SMB_MODULE(vfs_commit, \$(VFS_COMMIT_OBJ), "bin/commit.$SHLIBEXT", VFS)
 SMB_MODULE(vfs_gpfs, \$(VFS_GPFS_OBJ), "bin/gpfs.$SHLIBEXT", VFS)
 SMB_MODULE(vfs_notify_fam, \$(VFS_NOTIFY_FAM_OBJ), "bin/notify_fam.$SHLIBEXT", VFS)
 SMB_MODULE(vfs_readahead, \$(VFS_READAHEAD_OBJ), "bin/readahead.$SHLIBEXT", VFS)
+SMB_MODULE(vfs_fileid, \$(VFS_FILEID_OBJ), "bin/fileid.$SHLIBEXT", VFS)
 
 SMB_SUBSYSTEM(VFS,smbd/vfs.o)
 
diff --git a/source/modules/vfs_fileid.c b/source/modules/vfs_fileid.c
new file mode 100644 (file)
index 0000000..5954a10
--- /dev/null
@@ -0,0 +1,288 @@
+/*
+ * VFS module to alter the algorithm to calculate
+ * the struct file_id used as key for the share mode
+ * and byte range locking db's.
+ *
+ * Copyright (C) 2007, Stefan Metzmacher
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "includes.h"
+
+static int vfs_fileid_debug_level = DBGC_VFS;
+
+#undef DBGC_CLASS
+#define DBGC_CLASS vfs_fileid_debug_level
+
+struct fileid_mount_entry {
+       SMB_DEV_T device;
+       const char *mnt_fsname;
+       fsid_t fsid;
+       uint64_t devid;
+};
+
+struct fileid_handle_data {
+       uint64_t (*device_mapping_fn)(struct fileid_handle_data *data,
+                                     SMB_DEV_T dev);
+       unsigned num_mount_entries;
+       struct fileid_mount_entry *mount_entries;
+};
+
+/* load all the mount entries from the mtab */
+static void fileid_load_mount_entries(struct fileid_handle_data *data)
+{
+       FILE *f;
+       struct mntent *m;
+
+       data->num_mount_entries = 0;
+       TALLOC_FREE(data->mount_entries);
+
+       f = setmntent("/etc/mtab", "r");
+       if (!f) return;
+
+       while ((m = getmntent(f))) {
+               struct stat st;
+               struct statfs sfs;
+               struct fileid_mount_entry *cur;
+
+               if (stat(m->mnt_dir, &st) != 0) continue;
+               if (statfs(m->mnt_dir, &sfs) != 0) continue;
+
+               if (strncmp(m->mnt_fsname, "/dev/", 5) == 0) {
+                       m->mnt_fsname += 5;
+               }
+
+               data->mount_entries = TALLOC_REALLOC_ARRAY(data,
+                                                          data->mount_entries,
+                                                          struct fileid_mount_entry,
+                                                          data->num_mount_entries+1);
+               if (data->mount_entries == NULL) {
+                       goto nomem;
+               }
+
+               cur = &data->mount_entries[data->num_mount_entries];
+               cur->device     = st.st_dev;
+               cur->mnt_fsname = talloc_strdup(data->mount_entries,
+                                               m->mnt_fsname);
+               if (!cur->mnt_fsname) goto nomem;
+               cur->fsid       = sfs.f_fsid;
+               cur->devid      = (uint64_t)-1;
+
+               data->num_mount_entries++;
+       }
+       endmntent(f);
+       return;
+       
+nomem:
+       if (f) endmntent(f);
+
+       data->num_mount_entries = 0;
+       TALLOC_FREE(data->mount_entries);
+
+       return;
+}
+
+/* find a mount entry given a dev_t */
+static struct fileid_mount_entry *fileid_find_mount_entry(struct fileid_handle_data *data,
+                                                         SMB_DEV_T dev)
+{
+       int i;
+
+       if (data->num_mount_entries == 0) {
+               fileid_load_mount_entries(data);
+       }
+       for (i=0;i<data->num_mount_entries;i++) {
+               if (data->mount_entries[i].device == dev) {
+                       return &data->mount_entries[i];
+               }
+       }
+       /* 2nd pass after reloading */
+       fileid_load_mount_entries(data);
+       for (i=0;i<data->num_mount_entries;i++) {
+               if (data->mount_entries[i].device == dev) {
+                       return &data->mount_entries[i];
+               }
+       }       
+       return NULL;
+}
+
+
+/* a 64 bit hash, based on the one in tdb */
+static uint64_t fileid_uint64_hash(const uint8_t *s, size_t len)
+{
+       uint64_t value; /* Used to compute the hash value.  */
+       uint32_t i;     /* Used to cycle through random values. */
+
+       /* Set the initial value from the key size. */
+       for (value = 0x238F13AFLL * len, i=0; i < len; i++)
+               value = (value + (s[i] << (i*5 % 24)));
+
+       return (1103515243LL * value + 12345LL);
+}
+
+/* a device mapping using a fsname */
+static uint64_t fileid_device_mapping_fsname(struct fileid_handle_data *data,
+                                            SMB_DEV_T dev)
+{
+       struct fileid_mount_entry *m;
+
+       m = fileid_find_mount_entry(data, dev);
+       if (!m) return dev;
+
+       if (m->devid == (uint64_t)-1) {
+               m->devid = fileid_uint64_hash((uint8_t *)m->mnt_fsname,
+                                             strlen(m->mnt_fsname));
+       }
+
+       return m->devid;
+}
+
+/* device mapping functions using a fsid */
+static uint64_t fileid_device_mapping_fsid(struct fileid_handle_data *data,
+                                          SMB_DEV_T dev)
+{
+       struct fileid_mount_entry *m;
+
+       m = fileid_find_mount_entry(data, dev);
+       if (!m) return dev;
+
+       if (m->devid == (uint64_t)-1) {
+               if (sizeof(fsid_t) > sizeof(uint64_t)) {
+                       m->devid = fileid_uint64_hash((uint8_t *)&m->fsid,
+                                                     sizeof(m->fsid));
+               } else {
+                       union {
+                               uint64_t ret;
+                               fsid_t fsid;
+                       } u;
+                       ZERO_STRUCT(u);
+                       u.fsid = m->fsid;
+                       m->devid = u.ret;
+               }
+       }
+
+       return m->devid;
+}
+
+static int fileid_connect(struct vfs_handle_struct *handle,
+                         const char *service, const char *user)
+{
+       struct fileid_handle_data *data;
+       const char *algorithm;
+
+       data = talloc_zero(handle->conn->mem_ctx, struct fileid_handle_data);
+       if (!data) {
+               DEBUG(0, ("talloc_zero() failed\n"));
+               return -1;
+       }
+
+       data->device_mapping_fn = fileid_device_mapping_fsid;
+       algorithm = lp_parm_const_string(SNUM(handle->conn),
+                                        "fileid", "algorithm",
+                                        "fsname");
+       if (strcmp("fsname", algorithm) == 0) {
+               data->device_mapping_fn = fileid_device_mapping_fsname;
+       } else if (strcmp("fsid", algorithm) == 0) {
+               data->device_mapping_fn = fileid_device_mapping_fsid;
+       } else {
+               DEBUG(0,("fileid_connect(): unknown algorithm[%s]\n", algorithm));
+               return -1;
+       }
+
+       SMB_VFS_HANDLE_SET_DATA(handle, data, NULL,
+                               struct fileid_handle_data,
+                               return -1);
+
+       DEBUG(10, ("fileid_connect(): connect to service[%s] with algorithm[%s]\n",
+               service, algorithm));
+
+       return SMB_VFS_NEXT_CONNECT(handle, service, user);
+}
+
+static void fileid_disconnect(struct vfs_handle_struct *handle)
+{
+       DEBUG(10,("fileid_disconnect() connect to service[%s].\n",
+               lp_servicename(SNUM(handle->conn))));
+
+       SMB_VFS_NEXT_DISCONNECT(handle);
+}
+
+static struct file_id fileid_file_id_create(struct vfs_handle_struct *handle,
+                                           SMB_DEV_T dev, SMB_INO_T inode)
+{
+       struct fileid_handle_data *data;
+       struct file_id id;
+
+       ZERO_STRUCT(id);
+
+       SMB_VFS_HANDLE_GET_DATA(handle, data,
+                               struct fileid_handle_data,
+                               return id);
+
+       id.devid        = data->device_mapping_fn(data, dev);
+       id.inode        = inode;
+
+       return id;
+}
+
+static vfs_op_tuple fileid_ops[] = {
+
+       /* Disk operations */
+       {
+               SMB_VFS_OP(fileid_connect),
+               SMB_VFS_OP_CONNECT,
+               SMB_VFS_LAYER_TRANSPARENT
+       },
+       {
+               SMB_VFS_OP(fileid_disconnect),
+               SMB_VFS_OP_DISCONNECT,
+               SMB_VFS_LAYER_TRANSPARENT
+       },
+
+       /* File operations */
+       {
+               SMB_VFS_OP(fileid_file_id_create),
+               SMB_VFS_OP_FILE_ID_CREATE,
+               SMB_VFS_LAYER_OPAQUE
+       },
+
+       /* End marker */
+       {
+               SMB_VFS_OP(NULL),
+               SMB_VFS_OP_NOOP,
+               SMB_VFS_LAYER_NOOP
+       }
+};
+
+NTSTATUS vfs_fileid_init(void);
+NTSTATUS vfs_fileid_init(void)
+{
+       NTSTATUS ret;
+
+       ret = smb_register_vfs(SMB_VFS_INTERFACE_VERSION, "fileid", fileid_ops);
+       if (!NT_STATUS_IS_OK(ret)) {
+               return ret;
+       }
+
+       vfs_fileid_debug_level = debug_add_class("fileid");
+       if (vfs_fileid_debug_level == -1) {
+               vfs_fileid_debug_level = DBGC_VFS;
+               DEBUG(0, ("vfs_fileid: Couldn't register custom debugging class!\n"));
+       } else {
+               DEBUG(10, ("vfs_fileid: Debug class number of 'fileid': %d\n", vfs_fileid_debug_level));
+       }
+
+       return ret;
+}