Whamcloud - gitweb
Merge b_md to HEAD for 0.5.19 release.
[fs/lustre-release.git] / lustre / llite / super25.c
index 83867eb..557d715 100644 (file)
 #include <linux/version.h>
 #include <linux/lustre_lite.h>
 #include <linux/lustre_ha.h>
-#include <linux/obd_lov.h>
 #include <linux/lustre_dlm.h>
 #include <linux/init.h>
 #include <linux/fs.h>
+#include <linux/lprocfs_status.h>
 
 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 kmem_cache_t *ll_file_data_slab;
@@ -34,6 +34,8 @@ 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)
+
 static char *ll_read_opt(const char *opt, char *data)
 {
         char *value;
@@ -84,7 +86,7 @@ static void ll_options(char *options, char **ost, char **mds, int *flags)
                 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) | 
+                     (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
                         continue;
         }
@@ -106,24 +108,18 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
         int err;
         struct ll_fid rootfid;
         struct obd_statfs osfs;
-        __u64 last_committed;
-        __u64 last_xid;
         struct ptlrpc_request *request = NULL;
         struct ptlrpc_connection *mdc_conn;
         struct ll_read_inode2_cookie lic;
         class_uuid_t 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);
-        sbi->ll_mount_epoch = 0;
         generate_random_uuid(uuid);
         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 
@@ -154,9 +150,7 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
                 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);
@@ -172,21 +166,16 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
                 GOTO(out_mdc, sb = NULL);
         }
 
-        /* XXX: need to store the last_* values somewhere */
-        err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
-                            &last_xid, &request);
-        ptlrpc_req_finished(request);
+        err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
         if (err) {
                 CERROR("cannot mds_connect: rc = %d\n", err);
                 GOTO(out_mdc, sb = NULL);
         }
-        CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
+        CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
         sbi->ll_rootino = rootfid.id;
 
         memset(&osfs, 0, sizeof(osfs));
-        request = NULL;
-        err = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
-        ptlrpc_req_finished(request);
+        err = mdc_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;
@@ -195,7 +184,6 @@ static int ll_fill_super(struct super_block *sb, void *data, int silent)
         sb->s_op = &ll_super_operations;
 
         /* make root inode */
-        request = NULL;
         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
         if (err) {
@@ -227,7 +215,8 @@ 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)
 out_dev:
         if (mdc)
                 OBD_FREE(mdc, strlen(mdc) + 1);
@@ -246,49 +235,78 @@ out_mdc:
 out_free:
         OBD_FREE(sbi, sizeof(*sbi));
 
-        MOD_DEC_USE_COUNT;
         goto out_dev;
 } /* ll_fill_super */
 
-struct super_block * ll_get_sb(struct file_system_type *fs_type, 
-                                   int flags, char *devname, void * data)
+struct super_block * ll_get_sb(struct file_system_type *fs_type,
+                               int flags, char *devname, void * data)
 {
-       return get_sb_nodev(fs_type, flags, data, ll_fill_super);
+        return get_sb_nodev(fs_type, flags, data, ll_fill_super);
 }
 
 static void ll_put_super(struct super_block *sb)
 {
         struct ll_sb_info *sbi = ll_s2sbi(sb);
-
+        struct ll_fid rootfid;
         ENTRY;
+
         list_del(&sbi->ll_conn_chain);
         ll_commitcbd_cleanup(sbi);
         obd_disconnect(&sbi->ll_osc_conn);
+
+        /* NULL request to force sync on the MDS, and get the last_committed
+         * value to flush remaining RPCs from the pending queue on client.
+         *
+         * XXX This should be an mdc_sync() call to sync the whole MDS fs,
+         *     which we can call for other reasons as well.
+         */
+        mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
+
+        lprocfs_dereg_mnt(sbi->ll_proc_root);
+        sbi->ll_proc_root = NULL;
+
         obd_disconnect(&sbi->ll_mdc_conn);
         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);
-                struct lov_stripe_md *lsm = lli->lli_smd;
-                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 (lsm) {
-                        OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
-                        lli->lli_smd = NULL;
-                }
-                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;
 }
 
@@ -300,8 +318,9 @@ 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");
@@ -314,14 +333,14 @@ static void ll_delete_inode(struct inode *inode)
                         GOTO(out, -ENOMEM);
 
                 oa->o_id = lsm->lsm_object_id;
-                oa->o_easize = ll_mds_easize(inode->i_sb);
                 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);
@@ -364,18 +383,24 @@ 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);
+        /* 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, &request);
+                if (err)
+                        CERROR("mdc_setattr fails (%d)\n", err);
 
-        ptlrpc_req_finished(request);
+                ptlrpc_req_finished(request);
+        }
 
         RETURN(err);
 }
@@ -392,16 +417,14 @@ int ll_setattr(struct dentry *de, struct iattr *attr)
 
 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 {
-        struct ptlrpc_request *request = NULL;
         struct ll_sb_info *sbi = ll_s2sbi(sb);
         struct obd_statfs osfs;
         int rc;
         ENTRY;
 
         memset(sfs, 0, sizeof(*sfs));
-        rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
+        rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
         statfs_unpack(sfs, &osfs);
-        ptlrpc_req_finished(request);
         if (rc)
                 CERROR("mdc_statfs fails: rc = %d\n", rc);
         else
@@ -475,44 +498,26 @@ int ll_read_inode2(struct inode *inode, void *opaque)
         struct ll_read_inode2_cookie *lic = opaque;
         struct mds_body *body = lic->lic_body;
         struct ll_inode_info *lli = ll_i2info(inode);
+        int rc = 0;
         ENTRY;
-        
+
         sema_init(&lli->lli_open_sem, 1);
-        lli->lli_mount_epoch = ll_i2sbi(inode)->ll_mount_epoch;
-        
+
         /* core attributes first */
         ll_update_inode(inode, body);
 
-        //if (body->valid & OBD_MD_FLEASIZE)
-        if (lic && lic->lic_lmm) {
-                struct lov_mds_md *lmm = lic->lic_lmm;
-                int size;
-
-                if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
-                        CERROR("Striping metadata size error %ld\n",
-                               inode->i_ino);
-                        LBUG();
-                }
-                size = ll_ost_easize(inode->i_sb);
-                OBD_ALLOC(lli->lli_smd, size);
-                if (!lli->lli_smd) {
-                        CERROR("No memory for %d\n", size);
-                        LBUG();
-                }
-                lov_unpackmd(lli->lli_smd, lmm);
-        } else {
-                lli->lli_smd = NULL;
-        }
+        LASSERT(!lli->lli_smd);
+        if (lic && lic->lic_lmm)
+                obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lic->lic_lmm);
 
         /* Get the authoritative file size */
-        if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
-                int rc;
-
+        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();
+                        ll_clear_inode(inode);
+                        make_bad_inode(inode);
+                        RETURN(rc);
                 }
         }
 
@@ -532,18 +537,19 @@ int ll_read_inode2(struct inode *inode, void *opaque)
                 inode->i_op = &ll_fast_symlink_inode_operations;
                 EXIT;
         } else {
-                init_special_inode(inode, inode->i_mode, 
+                init_special_inode(inode, inode->i_mode,
                                    kdev_t_to_nr(inode->i_rdev));
                 EXIT;
         }
-        return 0;
+
+        return rc;
 }
 
 static inline void invalidate_request_list(struct list_head *req_list)
 {
         struct list_head *tmp, *n;
         list_for_each_safe(tmp, n, req_list) {
-                struct ptlrpc_request *req = 
+                struct ptlrpc_request *req =
                         list_entry(tmp, struct ptlrpc_request, rq_list);
                 CERROR("invalidating req xid %d op %d to %s:%d\n",
                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
@@ -560,7 +566,7 @@ void ll_umount_begin(struct super_block *sb)
         struct list_head *ctmp;
 
         ENTRY;
-       
+
         list_for_each(ctmp, &sbi->ll_conn_chain) {
                 struct ptlrpc_connection *conn;
                 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
@@ -580,47 +586,46 @@ static kmem_cache_t *ll_inode_cachep;
 
 static struct inode *ll_alloc_inode(struct super_block *sb)
 {
-       struct ll_inode_info *lli;
-       lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
-       if (!lli)
-               return NULL;
+        struct ll_inode_info *lli;
+        lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
+        if (!lli)
+                return NULL;
 
-       memset(lli, 0, sizeof(*lli));
+        memset(lli, 0, (char *)&lli->lli_vfs_inode - (char *)lli);
         sema_init(&lli->lli_open_sem, 1);
-        lli->lli_mount_epoch = ll_i2sbi(&lli->lli_vfs_inode)->ll_mount_epoch;
 
-       return &lli->lli_vfs_inode;
+        return &lli->lli_vfs_inode;
 }
 
 static void ll_destroy_inode(struct inode *inode)
 {
-       kmem_cache_free(ll_inode_cachep, LL_I(inode));
+        kmem_cache_free(ll_inode_cachep, ll_i2info(inode));
 }
 
 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 {
-       struct ll_inode_info *lli = foo;
+        struct ll_inode_info *lli = foo;
 
-       if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
-           SLAB_CTOR_CONSTRUCTOR)
-               inode_init_once(&lli->lli_vfs_inode);
+        if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+            SLAB_CTOR_CONSTRUCTOR)
+                inode_init_once(&lli->lli_vfs_inode);
 }
+
 int ll_init_inodecache(void)
 {
-       ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
-                                            sizeof(struct ll_inode_info),
-                                            0, SLAB_HWCACHE_ALIGN,
-                                            init_once, NULL);
-       if (ll_inode_cachep == NULL)
-               return -ENOMEM;
-       return 0;
+        ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
+                                            sizeof(struct ll_inode_info),
+                                            0, SLAB_HWCACHE_ALIGN,
+                                            init_once, NULL);
+        if (ll_inode_cachep == NULL)
+                return -ENOMEM;
+        return 0;
 }
 
 void ll_destroy_inodecache(void)
 {
-       if (kmem_cache_destroy(ll_inode_cachep))
-               CERROR("ll_inode_cache: not all structures were freed\n");
+        if (kmem_cache_destroy(ll_inode_cachep))
+                CERROR("ll_inode_cache: not all structures were freed\n");
 }
 
 
@@ -639,7 +644,7 @@ struct super_operations ll_super_operations =
 
 struct file_system_type lustre_lite_fs_type = {
         .owner  = THIS_MODULE,
-        .name =   "lustre_lite", 
+        .name =   "lustre_lite",
         .get_sb = ll_get_sb,
         .kill_sb = kill_litter_super,
 };
@@ -649,12 +654,12 @@ static int __init init_lustre_lite(void)
         int rc;
         printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
         rc = ll_init_inodecache();
-        if (rc) 
+        if (rc)
                 return -ENOMEM;
         ll_file_data_slab = kmem_cache_create("ll_file_data",
                                               sizeof(struct ll_file_data), 0,
                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
-        if (ll_file_data_slab == NULL) { 
+        if (ll_file_data_slab == NULL) {
                 ll_destroy_inodecache();
                 return -ENOMEM;
         }