Whamcloud - gitweb
LU-8508 nodemap: improve object handling in cache saving
[fs/lustre-release.git] / lustre / mgs / mgs_fs.c
index 586487f..00de194 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  * Lustre Management Server (MGS) filesystem interface code
  *
  * Author: Nathan Rutman <nathan@clusterfs.com>
+ * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MGS
 
-#include <linux/module.h>
-#include <linux/kmod.h>
-#include <linux/version.h>
-#include <linux/sched.h>
-#include <linux/mount.h>
-#include <obd_class.h>
-#include <obd_support.h>
-#include <lustre_disk.h>
-#include <lustre_lib.h>
-#include <lustre_fsfilt.h>
-#include <libcfs/list.h>
+#include <lustre_fid.h>
 #include "mgs_internal.h"
 
-static int mgs_export_stats_init(struct obd_device *obd, struct obd_export *exp,
-                                 void *localdata)
+/**
+ * Initialize MGS per-export statistics.
+ *
+ * This function sets up procfs entries for various MGS export counters. These
+ * counters are for per-client statistics tracked on the server.
+ *
+ * \param[in] obd      OBD device
+ * \param[in] exp      OBD export
+ * \param[in] localdata        NID of client
+ *
+ * \retval             0 if successful
+ * \retval             negative value on error
+ */
+int mgs_export_stats_init(struct obd_device *obd, struct obd_export *exp,
+                         void *localdata)
 {
-        lnet_nid_t *client_nid = localdata;
-        int rc, newnid;
-
-        rc = lprocfs_exp_setup(exp, client_nid, &newnid);
-        if (rc) {
-                /* Mask error for already created
-                 * /proc entries */
-                if (rc == -EALREADY)
-                        rc = 0;
-                return rc;
-        }
-
-        if ((obd->md_stats == NULL) &&
-            (rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST)))
-                return rc;
-        if (newnid) {
-                /* Always add in ldlm_stats */
-                exp->exp_nid_stats->nid_ldlm_stats =
-                        lprocfs_alloc_stats(LDLM_LAST_OPC - LDLM_FIRST_OPC, 
-                                            LPROCFS_STATS_FLAG_NOPERCPU);
-                if (exp->exp_nid_stats->nid_ldlm_stats == NULL)
-                        return -ENOMEM;
-                lprocfs_init_ldlm_stats(exp->exp_nid_stats->nid_ldlm_stats);
-                rc = lprocfs_register_stats(exp->exp_nid_stats->nid_proc,
-                                            "ldlm_stats",
-                                            exp->exp_nid_stats->nid_ldlm_stats);
-        }
-        return rc;
+       lnet_nid_t *client_nid = localdata;
+       struct nid_stat *stats;
+       int rc;
+       ENTRY;
+
+       rc = lprocfs_exp_setup(exp, client_nid);
+       if (rc != 0)
+               /* Mask error for already created /proc entries */
+               RETURN(rc == -EALREADY ? 0 : rc);
+
+       stats = exp->exp_nid_stats;
+       stats->nid_stats = lprocfs_alloc_stats(NUM_OBD_STATS + LPROC_MGS_LAST,
+                                               LPROCFS_STATS_FLAG_NOPERCPU);
+       if (stats->nid_stats == NULL)
+               RETURN(-ENOMEM);
+
+       lprocfs_init_ops_stats(LPROC_MGS_LAST, stats->nid_stats);
+
+       mgs_stats_counter_init(stats->nid_stats);
+
+       rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
+       if (rc != 0) {
+               lprocfs_free_stats(&stats->nid_stats);
+               GOTO(out, rc);
+       }
+
+       rc = lprocfs_nid_ldlm_stats_init(stats);
+       if (rc != 0)
+               GOTO(out, rc);
+
+out:
+       RETURN(rc);
 }
 
 /**
@@ -97,9 +102,9 @@ static int mgs_export_stats_init(struct obd_device *obd, struct obd_export *exp,
  * crash all connections are treated as new connections.
  */
 int mgs_client_add(struct obd_device *obd, struct obd_export *exp,
-                   void *localdata)
+                         void *localdata)
 {
-        return mgs_export_stats_init(obd, exp, localdata);
+        return 0;
 }
 
 /* Remove client export data from the MGS */
@@ -108,155 +113,138 @@ int mgs_client_free(struct obd_export *exp)
         return 0;
 }
 
-/* Same as mds_fid2dentry */
-/* Look up an entry by inode number. */
-/* this function ONLY returns valid dget'd dentries with an initialized inode
-   or errors */
-static struct dentry *mgs_fid2dentry(struct mgs_obd *mgs,
-                                     __u64 ino, __u32 gen)
-{
-        char fid_name[32];
-        struct inode *inode;
-        struct dentry *result;
-        ENTRY;
-
-        CDEBUG(D_DENTRY, "--> mgs_fid2dentry: ino/gen %lu/%u, sb %p\n",
-               (unsigned long)ino, gen, mgs->mgs_sb);
-
-        if (ino == 0)
-                RETURN(ERR_PTR(-ESTALE));
-
-        snprintf(fid_name, sizeof(fid_name), "0x%lx", (unsigned long)ino);
-
-        /* under ext3 this is neither supposed to return bad inodes nor NULL
-           inodes. */
-        result = ll_lookup_one_len(fid_name, mgs->mgs_fid_de, strlen(fid_name));
-        if (IS_ERR(result))
-                RETURN(result);
-
-        inode = result->d_inode;
-        if (!inode)
-                RETURN(ERR_PTR(-ENOENT));
-
-        if (inode->i_generation == 0 || inode->i_nlink == 0) {
-                LCONSOLE_WARN("Found inode with zero generation or link -- this"
-                              " may indicate disk corruption (inode: %lu, link:"
-                              " %lu, count: %d)\n", inode->i_ino,
-                              (unsigned long)inode->i_nlink,
-                              atomic_read(&inode->i_count));
-                l_dput(result);
-                RETURN(ERR_PTR(-ENOENT));
-        }
-
-        if (gen && inode->i_generation != gen) {
-                /* we didn't find the right inode.. */
-                CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
-                       "count: %d, generation %u/%u\n", inode->i_ino,
-                       (unsigned long)inode->i_nlink, atomic_read(&inode->i_count),
-                       inode->i_generation, gen);
-                l_dput(result);
-                RETURN(ERR_PTR(-ENOENT));
-        }
-
-        RETURN(result);
-}
-
-static struct dentry *mgs_lvfs_fid2dentry(__u64 id, __u32 gen,
-                                          __u64 gr, void *data)
-{
-        struct obd_device *obd = data;
-        return mgs_fid2dentry(&obd->u.mgs, id, gen);
-}
-
-struct lvfs_callback_ops mgs_lvfs_ops = {
-        l_fid2dentry:     mgs_lvfs_fid2dentry,
-};
-
-int mgs_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
+int mgs_fs_setup(const struct lu_env *env, struct mgs_device *mgs)
 {
-        struct mgs_obd *mgs = &obd->u.mgs;
-        struct lvfs_run_ctxt saved;
-        struct dentry *dentry;
-        int rc;
-        ENTRY;
-
-        /* FIXME what's this?  Do I need it? */
-        rc = cleanup_group_info();
-        if (rc)
-                RETURN(rc);
-
-        mgs->mgs_vfsmnt = mnt;
-        mgs->mgs_sb = mnt->mnt_root->d_inode->i_sb;
-
-        fsfilt_setup(obd, mgs->mgs_sb);
-
-        OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
-        obd->obd_lvfs_ctxt.pwdmnt = mnt;
-        obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
-        obd->obd_lvfs_ctxt.fs = get_ds();
-        obd->obd_lvfs_ctxt.cb_ops = mgs_lvfs_ops;
-
-        push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-
-        /* Setup the configs dir */
-        dentry = simple_mkdir(current->fs->pwd, mnt, MOUNT_CONFIGS_DIR, 0777, 1);
-        if (IS_ERR(dentry)) {
-                rc = PTR_ERR(dentry);
-                CERROR("cannot create %s directory: rc = %d\n",
-                       MOUNT_CONFIGS_DIR, rc);
-                GOTO(err_pop, rc);
-        }
-        mgs->mgs_configs_dir = dentry;
-
-        /* Need the iopen dir for fid2dentry, required by
-           LLOG_ORIGIN_HANDLE_READ_HEADER */
-        dentry = lookup_one_len("__iopen__", current->fs->pwd,
-                                strlen("__iopen__"));
-        if (IS_ERR(dentry)) {
-                rc = PTR_ERR(dentry);
-                CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
-                GOTO(err_configs, rc);
-        }
-        mgs->mgs_fid_de = dentry;
-        if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
-                rc = -ENOENT;
-                CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
-                GOTO(err_fid, rc);
-        }
-
-err_pop:
-        pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-        return rc;
-err_fid:
-        dput(mgs->mgs_fid_de);
-err_configs:
-        dput(mgs->mgs_configs_dir);
-        goto err_pop;
+       struct lu_fid            fid;
+       struct dt_object        *o;
+       struct lu_fid            rfid;
+       struct dt_object        *root;
+       struct dt_object        *nm_config_file_obj;
+       struct nm_config_file   *nm_config_file;
+       int                      rc;
+
+       ENTRY;
+
+       OBD_SET_CTXT_MAGIC(&mgs->mgs_obd->obd_lvfs_ctxt);
+       mgs->mgs_obd->obd_lvfs_ctxt.dt = mgs->mgs_bottom;
+
+       /* XXX: fix when support for N:1 layering is implemented */
+       LASSERT(mgs->mgs_dt_dev.dd_lu_dev.ld_site);
+       mgs->mgs_dt_dev.dd_lu_dev.ld_site->ls_top_dev =
+               &mgs->mgs_dt_dev.dd_lu_dev;
+
+       /* Setup the configs dir */
+       fid.f_seq = FID_SEQ_LOCAL_NAME;
+       fid.f_oid = 1;
+       fid.f_ver = 0;
+       rc = local_oid_storage_init(env, mgs->mgs_bottom, &fid, &mgs->mgs_los);
+       if (rc)
+               GOTO(out, rc);
+
+       rc = dt_root_get(env, mgs->mgs_bottom, &rfid);
+       if (rc)
+               GOTO(out_los, rc);
+
+       root = dt_locate_at(env, mgs->mgs_bottom, &rfid,
+                           &mgs->mgs_dt_dev.dd_lu_dev, NULL);
+       if (unlikely(IS_ERR(root)))
+               GOTO(out_los, rc = PTR_ERR(root));
+
+       o = local_file_find_or_create(env, mgs->mgs_los, root,
+                                     MOUNT_CONFIGS_DIR,
+                                     S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO);
+       if (IS_ERR(o))
+               GOTO(out_root, rc = PTR_ERR(o));
+
+       if (!dt_try_as_dir(env, o)) {
+               lu_object_put(env, &o->do_lu);
+               GOTO(out_root, rc = -ENOTDIR);
+       }
+
+       mgs->mgs_configs_dir = o;
+
+       /* colocated MDT will cache config in target root dir */
+       nm_config_file_obj = local_index_find_or_create(env, mgs->mgs_los,
+                                                       mgs->mgs_configs_dir,
+                                                       LUSTRE_NODEMAP_NAME,
+                                                       S_IFREG | S_IRUGO |
+                                                                 S_IWUSR,
+                                                       &dt_nodemap_features);
+       if (IS_ERR(nm_config_file_obj))
+               GOTO(out_configs, rc = PTR_ERR(nm_config_file_obj));
+
+       if (nm_config_file_obj->do_index_ops == NULL) {
+               rc = nm_config_file_obj->do_ops->do_index_try(env,
+                                                       nm_config_file_obj,
+                                                       &dt_nodemap_features);
+               if (rc < 0) {
+                       lu_object_put(env, &nm_config_file_obj->do_lu);
+                       GOTO(out_configs, rc);
+               }
+       }
+       nm_config_file = nm_config_file_register_mgs(env, nm_config_file_obj,
+                                                    mgs->mgs_los);
+       lu_object_put(env, &nm_config_file_obj->do_lu);
+       if (IS_ERR(nm_config_file)) {
+               CERROR("%s: error loading nodemap config file, file must be "
+                      "removed via ldiskfs: rc = %ld\n",
+                      mgs->mgs_obd->obd_name, PTR_ERR(nm_config_file));
+               GOTO(out_configs, rc = PTR_ERR(nm_config_file));
+       }
+       mgs->mgs_obd->u.obt.obt_nodemap_config_file = nm_config_file;
+
+       /* create directory to store nid table versions */
+       o = local_file_find_or_create(env, mgs->mgs_los, root, MGS_NIDTBL_DIR,
+                                     S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO);
+       if (IS_ERR(o))
+               GOTO(out_nm, rc = PTR_ERR(o));
+
+       mgs->mgs_nidtbl_dir = o;
+
+out_nm:
+       if (rc < 0) {
+               nm_config_file_deregister_mgs(env, nm_config_file);
+               mgs->mgs_obd->u.obt.obt_nodemap_config_file = NULL;
+       }
+out_configs:
+       if (rc < 0) {
+               lu_object_put(env, &mgs->mgs_configs_dir->do_lu);
+               mgs->mgs_configs_dir = NULL;
+       }
+out_root:
+       lu_object_put(env, &root->do_lu);
+out_los:
+       if (rc) {
+               local_oid_storage_fini(env, mgs->mgs_los);
+               mgs->mgs_los = NULL;
+       }
+out:
+       mgs->mgs_dt_dev.dd_lu_dev.ld_site->ls_top_dev = NULL;
+
+       return rc;
 }
 
-int mgs_fs_cleanup(struct obd_device *obd)
+int mgs_fs_cleanup(const struct lu_env *env, struct mgs_device *mgs)
 {
-        struct mgs_obd *mgs = &obd->u.mgs;
-        struct lvfs_run_ctxt saved;
-        int rc = 0;
-
-        class_disconnect_exports(obd); /* cleans up client info too */
-
-        push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-
-        if (mgs->mgs_configs_dir) {
-                /*CERROR("configs dir dcount=%d\n",
-                       atomic_read(&mgs->mgs_configs_dir->d_count));*/
-                l_dput(mgs->mgs_configs_dir);
-                mgs->mgs_configs_dir = NULL;
-        }
-
-        shrink_dcache_parent(mgs->mgs_fid_de);
-        /*CERROR("fid dir dcount=%d\n",
-               atomic_read(&mgs->mgs_fid_de->d_count));*/
-        dput(mgs->mgs_fid_de);
-
-        pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-
-        return rc;
+       if (mgs->mgs_configs_dir) {
+               lu_object_put(env, &mgs->mgs_configs_dir->do_lu);
+               mgs->mgs_configs_dir = NULL;
+       }
+       if (mgs->mgs_nidtbl_dir) {
+               lu_object_put(env, &mgs->mgs_nidtbl_dir->do_lu);
+               mgs->mgs_nidtbl_dir = NULL;
+       }
+       if (mgs->mgs_obd->u.obt.obt_nodemap_config_file != NULL) {
+               struct nm_config_file *ncf = mgs->mgs_obd->u.obt.obt_nodemap_config_file;
+
+               nm_config_file_deregister_mgs(env, ncf);
+               mgs->mgs_obd->u.obt.obt_nodemap_config_file = NULL;
+       }
+
+       if (mgs->mgs_los) {
+               local_oid_storage_fini(env, mgs->mgs_los);
+               mgs->mgs_los = NULL;
+       }
+
+       return 0;
 }