Whamcloud - gitweb
LU-12514 LNet: get entropy from nid when nid set.
[fs/lustre-release.git] / lustre / llite / super25.c
index db0646c..4dc69e3 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
+#define D_MOUNT (D_SUPER | D_CONFIG/*|D_WARNING */)
+
 #include <linux/module.h>
 #include <linux/types.h>
 #include <linux/version.h>
@@ -43,6 +40,7 @@
 #include <lustre_dlm.h>
 #include <linux/init.h>
 #include <linux/fs.h>
+#include <linux/random.h>
 #include <lprocfs_status.h>
 #include "llite_internal.h"
 #include "vvp_internal.h"
@@ -52,7 +50,6 @@ static struct kmem_cache *ll_inode_cachep;
 static struct inode *ll_alloc_inode(struct super_block *sb)
 {
        struct ll_inode_info *lli;
-       ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_ALLOC_INODE, 1);
        OBD_SLAB_ALLOC_PTR_GFP(lli, ll_inode_cachep, GFP_NOFS);
        if (lli == NULL)
                return NULL;
@@ -61,11 +58,11 @@ static struct inode *ll_alloc_inode(struct super_block *sb)
        return &lli->lli_vfs_inode;
 }
 
-#ifdef HAVE_INODE_I_RCU
 static void ll_inode_destroy_callback(struct rcu_head *head)
 {
        struct inode *inode = container_of(head, struct inode, i_rcu);
        struct ll_inode_info *ptr = ll_i2info(inode);
+       llcrypt_free_inode(inode);
        OBD_SLAB_FREE_PTR(ptr, ll_inode_cachep);
 }
 
@@ -73,43 +70,160 @@ static void ll_destroy_inode(struct inode *inode)
 {
        call_rcu(&inode->i_rcu, ll_inode_destroy_callback);
 }
-#else
-static void ll_destroy_inode(struct inode *inode)
+
+static int ll_drop_inode(struct inode *inode)
 {
-       struct ll_inode_info *ptr = ll_i2info(inode);
-       OBD_SLAB_FREE_PTR(ptr, ll_inode_cachep);
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+       int drop;
+
+       if (!sbi->ll_inode_cache_enabled)
+               return 1;
+
+       drop = generic_drop_inode(inode);
+       if (!drop)
+               drop = llcrypt_drop_inode(inode);
+
+       return drop;
 }
-#endif
 
 /* exported operations */
-struct super_operations lustre_super_operations =
+const struct super_operations lustre_super_operations =
 {
-        .alloc_inode   = ll_alloc_inode,
-        .destroy_inode = ll_destroy_inode,
-#ifdef HAVE_SBOPS_EVICT_INODE
-        .evict_inode   = ll_delete_inode,
+       .alloc_inode   = ll_alloc_inode,
+       .destroy_inode = ll_destroy_inode,
+       .drop_inode    = ll_drop_inode,
+       .evict_inode   = ll_delete_inode,
+       .put_super     = ll_put_super,
+       .statfs        = ll_statfs,
+       .umount_begin  = ll_umount_begin,
+       .remount_fs    = ll_remount_fs,
+       .show_options  = ll_show_options,
+};
+
+/**
+ * This is the entry point for the mount call into Lustre.
+ * This is called when a client is mounted, and this is
+ * where we start setting things up.
+ *
+ * @lmd2data data Mount options (e.g. -o flock,abort_recov)
+ */
+static int lustre_fill_super(struct super_block *sb, void *lmd2_data,
+                            int silent)
+{
+       struct lustre_mount_data *lmd;
+       struct lustre_sb_info *lsi;
+       int rc;
+
+       ENTRY;
+
+       CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
+
+       lsi = lustre_init_lsi(sb);
+       if (!lsi)
+               RETURN(-ENOMEM);
+       lmd = lsi->lsi_lmd;
+
+       /*
+        * Disable lockdep during mount, because mount locking patterns are
+        * 'special'.
+        */
+       lockdep_off();
+
+       /*
+        * LU-639: the OBD cleanup of last mount may not finish yet, wait here.
+        */
+       obd_zombie_barrier();
+
+       /* Figure out the lmd from the mount options */
+       if (lmd_parse(lmd2_data, lmd)) {
+               lustre_put_lsi(sb);
+               GOTO(out, rc = -EINVAL);
+       }
+
+       if (!lmd_is_client(lmd)) {
+#ifdef HAVE_SERVER_SUPPORT
+#if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(2, 15, 51, 0)
+               static bool printed;
+
+               if (!printed) {
+                       LCONSOLE_WARN("%s: mounting server target with '-t lustre' deprecated, use '-t lustre_tgt'\n",
+                                     lmd->lmd_profile);
+                       printed = true;
+               }
+#endif
+               rc = server_fill_super(sb);
 #else
-        .clear_inode   = ll_clear_inode,
-        .delete_inode  = ll_delete_inode,
+               rc = -ENODEV;
+               CERROR("%s: This is client-side-only module, cannot handle server mount: rc = %d\n",
+                      lmd->lmd_profile, rc);
+               lustre_put_lsi(sb);
 #endif
-        .put_super     = ll_put_super,
-        .statfs        = ll_statfs,
-        .umount_begin  = ll_umount_begin,
-        .remount_fs    = ll_remount_fs,
-        .show_options  = ll_show_options,
-};
+               GOTO(out, rc);
+       }
 
+       CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
+       /* filename encryption is disabled by default */
+       lsi->lsi_flags &= ~LSI_FILENAME_ENC;
+       rc = lustre_start_mgc(sb);
+       if (rc) {
+               lustre_common_put_super(sb);
+               GOTO(out, rc);
+       }
+       /* Connect and start */
+       rc = ll_fill_super(sb);
+       /* ll_file_super will call lustre_common_put_super on failure,
+        * which takes care of the module reference.
+        *
+        * If error happens in fill_super() call, @lsi will be killed there.
+        * This is why we do not put it here.
+        */
+out:
+       if (rc) {
+               CERROR("llite: Unable to mount %s: rc = %d\n",
+                      s2lsi(sb) ? lmd->lmd_dev : "<unknown>", rc);
+       } else {
+               CDEBUG(D_SUPER, "%s: Mount complete\n",
+                      lmd->lmd_dev);
+       }
+       lockdep_on();
+       return rc;
+}
 
-void lustre_register_client_process_config(int (*cpc)(struct lustre_cfg *lcfg));
+/***************** FS registration ******************/
+static struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
+                                  const char *devname, void *data)
+{
+       return mount_nodev(fs_type, flags, data, lustre_fill_super);
+}
+
+static void lustre_kill_super(struct super_block *sb)
+{
+       struct lustre_sb_info *lsi = s2lsi(sb);
+
+       if (lsi && !IS_SERVER(lsi))
+               ll_kill_super(sb);
+
+       kill_anon_super(sb);
+}
+
+/** Register the "lustre" fs type
+ */
+static struct file_system_type lustre_fs_type = {
+       .owner          = THIS_MODULE,
+       .name           = "lustre",
+       .mount          = lustre_mount,
+       .kill_sb        = lustre_kill_super,
+       .fs_flags       = FS_RENAME_DOES_D_MOVE,
+};
+MODULE_ALIAS_FS("lustre");
 
 static int __init lustre_init(void)
 {
-       struct proc_dir_entry *entry;
-       lnet_process_id_t lnet_id;
-       struct timeval tv;
-       int i, rc, seed[2];
+       int rc;
+       unsigned long lustre_inode_cache_flags;
 
-       CLASSERT(sizeof(LUSTRE_VOLATILE_HDR) == LUSTRE_VOLATILE_HDR_LEN + 1);
+       BUILD_BUG_ON(sizeof(LUSTRE_VOLATILE_HDR) !=
+                    LUSTRE_VOLATILE_HDR_LEN + 1);
 
        /* print an address of _any_ initialized kernel symbol from this
         * module, to allow debugging with gdb that doesn't support data
@@ -117,9 +231,15 @@ static int __init lustre_init(void)
        CDEBUG(D_INFO, "Lustre client module (%p).\n",
               &lustre_super_operations);
 
+       lustre_inode_cache_flags = SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT |
+                                  SLAB_MEM_SPREAD;
+#ifdef SLAB_ACCOUNT
+       lustre_inode_cache_flags |= SLAB_ACCOUNT;
+#endif
+
        ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
                                            sizeof(struct ll_inode_info),
-                                           0, SLAB_HWCACHE_ALIGN, NULL);
+                                           0, lustre_inode_cache_flags, NULL);
        if (ll_inode_cachep == NULL)
                GOTO(out_cache, rc = -ENOMEM);
 
@@ -129,34 +249,19 @@ static int __init lustre_init(void)
        if (ll_file_data_slab == NULL)
                GOTO(out_cache, rc = -ENOMEM);
 
-       entry = lprocfs_register("llite", proc_lustre_root, NULL, NULL);
-       if (IS_ERR(entry)) {
-               rc = PTR_ERR(entry);
-               CERROR("cannot register '/proc/fs/lustre/llite': rc = %d\n",
-                      rc);
-               GOTO(out_cache, rc);
-       }
-
-       proc_lustre_fs_root = entry;
-
-       cfs_get_random_bytes(seed, sizeof(seed));
-
-       /* Nodes with small feet have little entropy. The NID for this
-        * node gives the most entropy in the low bits. */
-       for (i = 0;; i++) {
-               if (LNetGetId(i, &lnet_id) == -ENOENT)
-                       break;
-
-               if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND)
-                       seed[0] ^= LNET_NIDADDR(lnet_id.nid);
-       }
+       pcc_inode_slab = kmem_cache_create("ll_pcc_inode",
+                                          sizeof(struct pcc_inode), 0,
+                                          SLAB_HWCACHE_ALIGN, NULL);
+       if (pcc_inode_slab == NULL)
+               GOTO(out_cache, rc = -ENOMEM);
 
-       do_gettimeofday(&tv);
-       cfs_srand(tv.tv_sec ^ seed[0], tv.tv_usec ^ seed[1]);
+       rc = llite_tunables_register();
+       if (rc)
+               GOTO(out_cache, rc);
 
        rc = vvp_global_init();
        if (rc != 0)
-               GOTO(out_proc, rc);
+               GOTO(out_tunables, rc);
 
        cl_inode_fini_env = cl_env_alloc(&cl_inode_fini_refcheck,
                                         LCT_REMEMBER | LCT_NOREF);
@@ -169,42 +274,46 @@ static int __init lustre_init(void)
        if (rc != 0)
                GOTO(out_inode_fini_env, rc);
 
-       lustre_register_client_fill_super(ll_fill_super);
-       lustre_register_kill_super_cb(ll_kill_super);
-       lustre_register_client_process_config(ll_process_config);
+       rc = register_filesystem(&lustre_fs_type);
+       if (rc)
+               GOTO(out_xattr, rc);
 
        RETURN(0);
 
+out_xattr:
+       ll_xattr_fini();
 out_inode_fini_env:
        cl_env_put(cl_inode_fini_env, &cl_inode_fini_refcheck);
 out_vvp:
        vvp_global_fini();
-out_proc:
-       lprocfs_remove(&proc_lustre_fs_root);
+out_tunables:
+       llite_tunables_unregister();
 out_cache:
-       if (ll_inode_cachep != NULL)
-               kmem_cache_destroy(ll_inode_cachep);
-
-       if (ll_file_data_slab != NULL)
-               kmem_cache_destroy(ll_file_data_slab);
-
+       kmem_cache_destroy(ll_inode_cachep);
+       kmem_cache_destroy(ll_file_data_slab);
+       kmem_cache_destroy(pcc_inode_slab);
        return rc;
 }
 
 static void __exit lustre_exit(void)
 {
-       lustre_register_client_fill_super(NULL);
-       lustre_register_kill_super_cb(NULL);
-       lustre_register_client_process_config(NULL);
+       unregister_filesystem(&lustre_fs_type);
 
-       lprocfs_remove(&proc_lustre_fs_root);
+       llite_tunables_unregister();
 
        ll_xattr_fini();
        cl_env_put(cl_inode_fini_env, &cl_inode_fini_refcheck);
        vvp_global_fini();
 
+       /*
+        * Make sure all delayed rcu free inodes are flushed before we
+        * destroy cache.
+        */
+       rcu_barrier();
+
        kmem_cache_destroy(ll_inode_cachep);
        kmem_cache_destroy(ll_file_data_slab);
+       kmem_cache_destroy(pcc_inode_slab);
 }
 
 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");