Whamcloud - gitweb
merge b_devel into HEAD. Includes:
[fs/lustre-release.git] / lustre / llite / super25.c
index 30b582c..f296d10 100644 (file)
@@ -3,10 +3,22 @@
  *
  * Lustre Light Super operations
  *
- * This code is issued under the GNU General Public License.
- * See the file COPYING in this distribution
+ *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
  *
- * Copryright (C) 2002 Cluster File Systems, Inc.
+ *   This file is part of Lustre, http://www.lustre.org.
+ *
+ *   Lustre is free software; you can redistribute it and/or
+ *   modify it under the terms of version 2 of the GNU General Public
+ *   License as published by the Free Software Foundation.
+ *
+ *   Lustre 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 Lustre; if not, write to the Free Software
+ *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
 #include <linux/lprocfs_status.h>
 
 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
+#include <asm/statfs.h>
 kmem_cache_t *ll_file_data_slab;
 extern struct address_space_operations ll_aops;
 extern struct address_space_operations ll_dir_aops;
 struct super_operations ll_super_operations;
 
+/* /proc/lustre/llite root that tracks llite mount points */
+struct proc_dir_entry *proc_lustre_fs_root = NULL;
+/* lproc_llite.c */
+extern int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
+                                       struct super_block *sb,
+                                       char *osc, char *mdc);
+
 extern int ll_init_inodecache(void);
 extern void ll_destroy_inodecache(void);
 extern int ll_recover(struct recovd_data *, int);
@@ -34,7 +54,7 @@ extern int ll_commitcbd_setup(struct ll_sb_info *);
 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
 int ll_read_inode2(struct inode *inode, void *opaque);
 
-extern void ll_proc_namespace(struct super_block* sb, char* osc, char* mdc)
+extern int ll_proc_namespace(struct super_block* sb, char* osc, char* mdc);
 
 static char *ll_read_opt(const char *opt, char *data)
 {
@@ -43,14 +63,14 @@ static char *ll_read_opt(const char *opt, char *data)
         ENTRY;
 
         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
-        if ( strncmp(opt, data, strlen(opt)) )
+        if (strncmp(opt, data, strlen(opt)))
                 RETURN(NULL);
-        if ( (value = strchr(data, '=')) == NULL )
+        if ((value = strchr(data, '=')) == NULL)
                 RETURN(NULL);
 
         value++;
         OBD_ALLOC(retval, strlen(value) + 1);
-        if ( !retval ) {
+        if (!retval) {
                 CERROR("out of memory!\n");
                 RETURN(NULL);
         }
@@ -65,7 +85,7 @@ static int ll_set_opt(const char *opt, char *data, int fl)
         ENTRY;
 
         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
-        if ( strncmp(opt, data, strlen(opt)) )
+        if (strncmp(opt, data, strlen(opt)))
                 RETURN(0);
         else
                 RETURN(fl);
@@ -84,10 +104,11 @@ static void ll_options(char *options, char **ost, char **mds, int *flags)
 
         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
-                if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
-                     (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
-                     (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
-                      ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
+                if ((!*ost && (*ost = ll_read_opt("osc", this_char)))||
+                    (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
+                    (!(*flags & LL_SBI_NOLCK) &&
+                     ((*flags) = (*flags) |
+                      ll_set_opt("nolock", this_char, LL_SBI_NOLCK))))
                         continue;
         }
         EXIT;
@@ -112,19 +133,18 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
         struct ptlrpc_connection *mdc_conn;
         struct ll_read_inode2_cookie lic;
         class_uuid_t uuid;
+        struct obd_uuid param_uuid;
 
         ENTRY;
-        MOD_INC_USE_COUNT;
 
         OBD_ALLOC(sbi, sizeof(*sbi));
-        if (!sbi) {
-                MOD_DEC_USE_COUNT;
+        if (!sbi)
                 RETURN(-ENOMEM);
-        }
 
         INIT_LIST_HEAD(&sbi->ll_conn_chain);
+        INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
         generate_random_uuid(uuid);
-        class_uuid_unparse(uuid, sbi->ll_sb_uuid);
+        class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 
         sb->s_fs_info = sbi;
 
@@ -140,22 +160,21 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
                 GOTO(out_free, sb = NULL);
         }
 
-        obd = class_uuid2obd(mdc);
+        strncpy(param_uuid.uuid, mdc, sizeof(param_uuid.uuid));
+        obd = class_uuid2obd(&param_uuid);
         if (!obd) {
                 CERROR("MDC %s: not setup or attached\n", mdc);
                 GOTO(out_free, sb = NULL);
         }
 
-        err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
+        err = obd_connect(&sbi->ll_mdc_conn, obd, &sbi->ll_sb_uuid,
                           ptlrpc_recovd, ll_recover);
         if (err) {
                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
                 GOTO(out_free, sb = NULL);
         }
 
-#warning Peter: is this the right place to raise the connection level?
         mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
-        mdc_conn->c_level = LUSTRE_CONN_FULL;
         list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
 
         obd = class_uuid2obd(osc);
@@ -164,7 +183,7 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
                 GOTO(out_mdc, sb = NULL);
         }
 
-        err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
+        err = obd_connect(&sbi->ll_osc_conn, obd, &sbi->ll_sb_uuid,
                           ptlrpc_recovd, ll_recover);
         if (err) {
                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
@@ -180,7 +199,7 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
         sbi->ll_rootino = rootfid.id;
 
         memset(&osfs, 0, sizeof(osfs));
-        err = mdc_statfs(&sbi->ll_mdc_conn, &osfs);
+        err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
         sb->s_blocksize = osfs.os_bsize;
         sb->s_blocksize_bits = log2(osfs.os_bsize);
         sb->s_magic = LL_SUPER_MAGIC;
@@ -214,6 +233,7 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
 
         if (root) {
                 sb->s_root = d_alloc_root(root);
+                root->i_state &= ~(I_LOCK | I_NEW);
         } else {
                 CERROR("lustre_lite: bad iget4 for root\n");
                 GOTO(out_cdb, sb = NULL);
@@ -221,7 +241,14 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
 
         ptlrpc_req_finished(request);
         request = NULL;
-        ll_proc_namespace(sb, osc, mdc)
+
+        if (proc_lustre_fs_root) {
+                err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
+                                                  osc, mdc);
+                if (err < 0)
+                        CERROR("could not register mount in /proc/lustre");
+        }
+
 out_dev:
         if (mdc)
                 OBD_FREE(mdc, strlen(mdc) + 1);
@@ -240,7 +267,6 @@ out_mdc:
 out_free:
         OBD_FREE(sbi, sizeof(*sbi));
 
-        MOD_DEC_USE_COUNT;
         goto out_dev;
 } /* ll_fill_super */
 
@@ -253,6 +279,7 @@ struct super_block * ll_get_sb(struct file_system_type *fs_type,
 static void ll_put_super(struct super_block *sb)
 {
         struct ll_sb_info *sbi = ll_s2sbi(sb);
+        struct list_head *tmp, *next;
         struct ll_fid rootfid;
         ENTRY;
 
@@ -268,34 +295,65 @@ static void ll_put_super(struct super_block *sb)
          */
         mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
 
-        lprocfs_dereg_mnt(sbi->ll_proc_root);
+        if (sbi->ll_proc_root) {
+                lprocfs_remove(sbi->ll_proc_root);
         sbi->ll_proc_root = NULL;
+        }
 
         obd_disconnect(&sbi->ll_mdc_conn);
+
+        spin_lock(&dcache_lock);
+        list_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list){
+                struct dentry *dentry = list_entry(tmp, struct dentry, d_hash);
+                shrink_dcache_parent(dentry);
+        }
+        spin_unlock(&dcache_lock);
+
         OBD_FREE(sbi, sizeof(*sbi));
 
-        MOD_DEC_USE_COUNT;
         EXIT;
 } /* ll_put_super */
 
 static void ll_clear_inode(struct inode *inode)
 {
+        struct ll_sb_info *sbi = ll_i2sbi(inode);
+        struct ll_inode_info *lli = ll_i2info(inode);
+        int rc;
         ENTRY;
 
-        if (atomic_read(&inode->i_count) == 0) {
-                struct ll_inode_info *lli = ll_i2info(inode);
-                char *symlink_name = lli->lli_symlink_name;
+#warning "Is there a reason we don't do this in 2.5, but we do in 2.4?"
+#if 0
+        rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
+        if (rc < 0) {
+                CERROR("mdc_cancel_unused: %d\n", rc);
+                /* XXX FIXME do something dramatic */
+        }
 
-                if (lli->lli_smd)
-                        obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
-                if (symlink_name) {
-                        OBD_FREE(symlink_name, strlen(symlink_name) + 1);
-                        lli->lli_symlink_name = NULL;
+        if (lli->lli_smd) {
+                rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
+                if (rc < 0) {
+                        CERROR("obd_cancel_unused: %d\n", rc);
+                        /* XXX FIXME do something dramatic */
                 }
         }
+#endif
+
+        if (atomic_read(&inode->i_count) != 0)
+                CERROR("clearing in-use inode %lu: count = %d\n",
+                       inode->i_ino, atomic_read(&inode->i_count));
+
+        if (lli->lli_smd)
+                obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
+
+        if (lli->lli_symlink_name) {
+                OBD_FREE(lli->lli_symlink_name,strlen(lli->lli_symlink_name)+1);
+                lli->lli_symlink_name = NULL;
+        }
+
         EXIT;
 }
 
+#if 0
 static void ll_delete_inode(struct inode *inode)
 {
         ENTRY;
@@ -304,8 +362,9 @@ static void ll_delete_inode(struct inode *inode)
                 struct obdo *oa;
                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 
+                /* mcreate with no open */
                 if (!lsm)
-                        GOTO(out, -EINVAL);
+                        GOTO(out, 0);
 
                 if (lsm->lsm_object_id == 0) {
                         CERROR("This really happens\n");
@@ -319,17 +378,19 @@ static void ll_delete_inode(struct inode *inode)
 
                 oa->o_id = lsm->lsm_object_id;
                 oa->o_mode = inode->i_mode;
-                oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
+                oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
 
                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
                 obdo_free(oa);
-                CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
-                       lsm->lsm_object_id, err);
+                if (err)
+                        CDEBUG(D_SUPER, "obd destroy objid "LPX64" error %d\n",
+                               lsm->lsm_object_id, err);
         }
 out:
         clear_inode(inode);
         EXIT;
 }
+#endif
 
 /* like inode_setattr, but doesn't mark the inode dirty */
 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
@@ -367,18 +428,42 @@ int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 {
         struct ptlrpc_request *request = NULL;
         struct ll_sb_info *sbi = ll_i2sbi(inode);
-        int err;
+        int err = 0;
 
         ENTRY;
 
         /* change incore inode */
         ll_attr2inode(inode, attr, do_trunc);
 
-        err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
-        if (err)
-                CERROR("mdc_setattr fails (%d)\n", err);
-
-        ptlrpc_req_finished(request);
+        /* Don't send size changes to MDS to avoid "fast EA" problems, and
+         * also avoid a pointless RPC (we get file size from OST anyways).
+         */
+        attr->ia_valid &= ~ATTR_SIZE;
+        if (attr->ia_valid) {
+                err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
+                                  &request);
+                if (err)
+                        CERROR("mdc_setattr fails: err = %d\n", err);
+
+                ptlrpc_req_finished(request);
+                if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
+                        struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+                        struct obdo oa;
+                        int err2;
+
+                        CDEBUG(D_ERROR, "setting mtime on OST\n");
+                        oa.o_id = lsm->lsm_object_id;
+                        oa.o_mode = S_IFREG;
+                        oa.o_valid = OBD_MD_FLID |OBD_MD_FLTYPE |OBD_MD_FLMTIME;
+                        oa.o_mtime = attr->ia_mtime.tv_sec;
+                        err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
+                        if (err2) {
+                                CERROR("obd_setattr fails: rc=%d\n", err);
+                                if (!err)
+                                        err = err2;
+                        }
+                }
+        }
 
         RETURN(err);
 }
@@ -441,16 +526,22 @@ out:
         RETURN(rc);
 }
 
-void ll_update_inode(struct inode *inode, struct mds_body *body)
+void ll_update_inode(struct inode *inode, struct mds_body *body,
+                     struct lov_mds_md *lmm)
 {
+        struct ll_inode_info *lli = ll_i2info(inode);
+
+        if (lmm != NULL)
+                obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lmm);
+
         if (body->valid & OBD_MD_FLID)
                 inode->i_ino = body->ino;
         if (body->valid & OBD_MD_FLATIME)
-                inode->i_atime = body->atime;
+                inode->i_atime.tv_sec = body->atime;
         if (body->valid & OBD_MD_FLMTIME)
-                inode->i_mtime = body->mtime;
+                inode->i_mtime.tv_sec = body->mtime;
         if (body->valid & OBD_MD_FLCTIME)
-                inode->i_ctime = body->ctime;
+                inode->i_ctime.tv_sec = body->ctime;
         if (body->valid & OBD_MD_FLMODE)
                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
         if (body->valid & OBD_MD_FLTYPE)
@@ -469,6 +560,8 @@ void ll_update_inode(struct inode *inode, struct mds_body *body)
                 inode->i_rdev = to_kdev_t(body->rdev);
         if (body->valid & OBD_MD_FLSIZE)
                 inode->i_size = body->size;
+        if (body->valid & OBD_MD_FLBLOCKS)
+                inode->i_blocks = body->blocks;
 }
 
 int ll_read_inode2(struct inode *inode, void *opaque)
@@ -480,23 +573,32 @@ int ll_read_inode2(struct inode *inode, void *opaque)
         ENTRY;
 
         sema_init(&lli->lli_open_sem, 1);
+        lli->flags = 0;
+        init_MUTEX(&lli->lli_getattr_sem);
+        /* these are 2.4 only, but putting them here for consistency.. */
+        spin_lock_init(&lli->lli_read_extent_lock);
+        INIT_LIST_HEAD(&lli->lli_read_extents);
 
-        /* core attributes first */
-        ll_update_inode(inode, body);
-
-        //if (body->valid & OBD_MD_FLEASIZE)
         LASSERT(!lli->lli_smd);
-        if (lic && lic->lic_lmm)
-                obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lic->lic_lmm);
+
+        /* core attributes first */
+        ll_update_inode(inode, body, lic ? lic->lic_lmm : NULL);
 
         /* Get the authoritative file size */
         if (lli->lli_smd && S_ISREG(inode->i_mode)) {
-                rc = ll_file_size(inode, lli->lli_smd);
-                if (rc) {
-                        CERROR("ll_file_size: %d\n", rc);
-                        /* FIXME: need to somehow prevent inode creation */
-                        LBUG();
+                struct ll_file_data *fd = file->private_data;
+                struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
+                struct lustre_handle lockh = {0, 0};
+
+                LASSERT(lli->lli_smd->lsm_object_id != 0);
+
+                rc = ll_extent_lock(fd, inode, lsm, LCK_PR, &extent, &lockh);
+                if (err != ELDLM_OK && err != ELDLM_MATCHED) {
+                        ll_clear_inode(inode);
                         make_bad_inode(inode);
+                } else {
+                        l_extent_unlock(fd, inode, lsm, LCK_PR, &extent,
+                                        &lockh);
                 }
         }
 
@@ -552,7 +654,7 @@ void ll_umount_begin(struct super_block *sb)
 
                 spin_lock(&conn->c_lock);
                 conn->c_flags |= CONN_INVALID;
-                invalidate_request_list(&conn->c_sending_head);
+                /*invalidate_request_list(&conn->c_sending_head);*/
                 invalidate_request_list(&conn->c_delayed_head);
                 spin_unlock(&conn->c_lock);
         }
@@ -572,6 +674,7 @@ static struct inode *ll_alloc_inode(struct super_block *sb)
 
         memset(lli, 0, (char *)&lli->lli_vfs_inode - (char *)lli);
         sema_init(&lli->lli_open_sem, 1);
+        init_MUTEX(&lli->lli_size_valid_sem);
 
         return &lli->lli_vfs_inode;
 }
@@ -615,23 +718,25 @@ struct super_operations ll_super_operations =
         alloc_inode: ll_alloc_inode,
         destroy_inode: ll_destroy_inode,
         clear_inode: ll_clear_inode,
-        delete_inode: ll_delete_inode,
+//        delete_inode: ll_delete_inode,
         put_super: ll_put_super,
         statfs: ll_statfs,
         umount_begin: ll_umount_begin
 };
 
+
 struct file_system_type lustre_lite_fs_type = {
         .owner  = THIS_MODULE,
         .name =   "lustre_lite",
         .get_sb = ll_get_sb,
-        .kill_sb = kill_litter_super,
+        .kill_sb = kill_anon_super,
 };
 
 static int __init init_lustre_lite(void)
 {
         int rc;
-        printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
+        printk(KERN_INFO "Lustre Lite Client File System; "
+               "info@clusterfs.com\n");
         rc = ll_init_inodecache();
         if (rc)
                 return -ENOMEM;
@@ -642,6 +747,10 @@ static int __init init_lustre_lite(void)
                 ll_destroy_inodecache();
                 return -ENOMEM;
         }
+
+        proc_lustre_fs_root = proc_lustre_root ?
+                              proc_mkdir("llite", proc_lustre_root) : NULL;
+
         return register_filesystem(&lustre_lite_fs_type);
 }
 
@@ -650,10 +759,14 @@ static void __exit exit_lustre_lite(void)
         unregister_filesystem(&lustre_lite_fs_type);
         ll_destroy_inodecache();
         kmem_cache_destroy(ll_file_data_slab);
+        if (proc_lustre_fs_root) {
+                lprocfs_remove(proc_lustre_fs_root);
+                proc_lustre_fs_root = NULL;
+        }
 }
 
 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
-MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
+MODULE_DESCRIPTION("Lustre Lite Client File System");
 MODULE_LICENSE("GPL");
 
 module_init(init_lustre_lite);