-/* -*- 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, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Wang Di <wangdi@clusterfs.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MDS
#include <linux/module.h>
-#include <linux/jbd.h>
+#include <ldiskfs/ldiskfs_jbd2.h>
#include <obd.h>
#include <obd_class.h>
#include <lustre_ver.h>
#include <obd_support.h>
#include <lprocfs_status.h>
-#include <lustre_disk.h>
#include <lustre_fid.h>
-#include <linux/ldiskfs_fs.h>
+#include <ldiskfs/ldiskfs.h>
#include <lustre_mds.h>
#include <lustre/lustre_idl.h>
#include <lustre_disk.h> /* for changelogs */
mdd->mdd_child = lu2dt_dev(next);
/* Prepare transactions callbacks. */
- mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
+ mdd->mdd_txn_cb.dtc_txn_start = NULL;
mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
- mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb;
+ mdd->mdd_txn_cb.dtc_txn_commit = NULL;
mdd->mdd_txn_cb.dtc_cookie = mdd;
mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
mdd->mdd_atime_diff = MAX_ATIME_DIFF;
+ /* sync permission changes */
+ mdd->mdd_sync_permission = 1;
rc = mdd_procfs_init(mdd, name);
RETURN(rc);
struct mdd_device *m, struct lustre_cfg *cfg)
{
ENTRY;
+ mdd_lfsck_cleanup(env, m);
mdd_changelog_fini(env, m);
dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
- mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
- mdd_object_put(env, m->mdd_dot_lustre);
+ if (m->mdd_dot_lustre_objs.mdd_obf)
+ mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
+ if (m->mdd_dot_lustre)
+ mdd_object_put(env, m->mdd_dot_lustre);
if (m->mdd_obd_dev)
mdd_fini_obd(env, m, cfg);
orph_index_fini(env, m);
+ if (m->mdd_capa != NULL) {
+ lu_object_put(env, &m->mdd_capa->do_lu);
+ m->mdd_capa = NULL;
+ }
/* remove upcall device*/
md_upcall_fini(&m->mdd_md_dev);
EXIT;
}
-static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
- void *data)
+static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
{
struct mdd_device *mdd = (struct mdd_device *)data;
struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
CDEBUG(D_INFO,
"seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
- hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
- rec->cr_type, rec->cr_namelen, rec->cr_name,
+ hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
+ rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
llh->lgh_id.lgl_oid);
- mdd->mdd_cl.mc_index = rec->cr_index;
+ mdd->mdd_cl.mc_index = rec->cr.cr_index;
RETURN(LLOG_PROC_BREAK);
}
-static int changelog_user_init_cb(struct llog_handle *llh,
- struct llog_rec_hdr *hdr, void *data)
+static int changelog_user_init_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
{
struct mdd_device *mdd = (struct mdd_device *)data;
struct llog_changelog_user_rec *rec =
" in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
- spin_lock(&mdd->mdd_cl.mc_user_lock);
+ cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
mdd->mdd_cl.mc_lastuser = rec->cur_id;
- spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
RETURN(LLOG_PROC_BREAK);
}
return -EINVAL;
}
- rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
+ rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
+ changelog_init_cb, mdd);
llog_ctxt_put(ctxt);
if (rc < 0) {
return -EINVAL;
}
- rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
- mdd);
+ rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
+ changelog_user_init_cb, mdd);
llog_ctxt_put(ctxt);
if (rc < 0) {
int rc;
mdd->mdd_cl.mc_index = 0;
- spin_lock_init(&mdd->mdd_cl.mc_lock);
- cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
+ cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
mdd->mdd_cl.mc_starttime = cfs_time_current_64();
mdd->mdd_cl.mc_flags = 0; /* off by default */
mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
- spin_lock_init(&mdd->mdd_cl.mc_user_lock);
+ cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
mdd->mdd_cl.mc_lastuser = 0;
rc = mdd_changelog_llog_init(mdd);
mdd2obd_dev(mdd)->obd_name);
rc = -ESRCH;
} else {
- spin_lock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
mdd->mdd_cl.mc_flags |= CLM_ON;
- spin_unlock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
rc = mdd_changelog_write_header(mdd, CLM_START);
}
} else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
rc = mdd_changelog_write_header(mdd, CLM_FINI);
- spin_lock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
mdd->mdd_cl.mc_flags &= ~CLM_ON;
- spin_unlock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
}
return rc;
}
+static __u64 cl_time(void) {
+ cfs_fs_time_t time;
+
+ cfs_fs_time_current(&time);
+ return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
+}
+
/** Add a changelog entry \a rec to the changelog llog
* \param mdd
* \param rec
struct llog_ctxt *ctxt;
int rc;
- if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
- return 0;
-
- rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
+ rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
/* llog_lvfs_write_rec sets the llog tail len */
rec->cr_hdr.lrh_type = CHANGELOG_REC;
- rec->cr_time = cfs_time_current_64();
- spin_lock(&mdd->mdd_cl.mc_lock);
+ rec->cr.cr_time = cl_time();
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
/* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
but as long as the MDD transactions are ordered correctly for e.g.
rename conflicts, I don't think this should matter. */
- rec->cr_index = ++mdd->mdd_cl.mc_index;
- spin_unlock(&mdd->mdd_cl.mc_lock);
+ rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
if (ctxt == NULL)
return -ENXIO;
/* nested journal transaction */
- rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
+ rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
llog_ctxt_put(ctxt);
- cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
-
return rc;
}
+/** Add a changelog_ext entry \a rec to the changelog llog
+ * \param mdd
+ * \param rec
+ * \param handle - currently ignored since llogs start their own transaction;
+ * this will hopefully be fixed in llog rewrite
+ * \retval 0 ok
+ */
+int mdd_changelog_ext_llog_write(struct mdd_device *mdd,
+ struct llog_changelog_ext_rec *rec,
+ struct thandle *handle)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt;
+ int rc;
+
+ rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
+ /* llog_lvfs_write_rec sets the llog tail len */
+ rec->cr_hdr.lrh_type = CHANGELOG_REC;
+ rec->cr.cr_time = cl_time();
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
+ /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
+ * but as long as the MDD transactions are ordered correctly for e.g.
+ * rename conflicts, I don't think this should matter. */
+ rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ if (ctxt == NULL)
+ return -ENXIO;
+
+ /* nested journal transaction */
+ rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
+ llog_ctxt_put(ctxt);
+
+ return rc;
+}
+
/** Remove entries with indicies up to and including \a endrec from the
* changelog
* \param mdd
* \param endrec
* \retval 0 ok
*/
-int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
+int mdd_changelog_llog_cancel(const struct lu_env *env,
+ struct mdd_device *mdd, long long endrec)
{
struct obd_device *obd = mdd2obd_dev(mdd);
struct llog_ctxt *ctxt;
if (ctxt == NULL)
return -ENXIO;
- spin_lock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
cur = (long long)mdd->mdd_cl.mc_index;
- spin_unlock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
if (endrec > cur)
endrec = cur;
time. In case of crash, we just restart with old log so we're
allright. */
if (endrec == cur) {
+ /* XXX: transaction is started by llog itself */
rc = mdd_changelog_write_header(mdd, CLM_PURGE);
if (rc)
goto out;
changed since the last purge) */
mdd->mdd_cl.mc_starttime = cfs_time_current_64();
- rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
+ /* XXX: transaction is started by llog itself */
+ rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
out:
llog_ctxt_put(ctxt);
return rc;
if (rec == NULL)
RETURN(-ENOMEM);
- rec->cr_flags = CLF_VERSION;
- rec->cr_type = CL_MARK;
- rec->cr_namelen = len;
- memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
+ rec->cr.cr_flags = CLF_VERSION;
+ rec->cr.cr_type = CL_MARK;
+ rec->cr.cr_namelen = len;
+ memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
/* Status and action flags */
- rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
+ rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
- rc = mdd_changelog_llog_write(mdd, rec, NULL);
+ /* XXX: transaction is started by llog itself */
+ rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
+ mdd_changelog_llog_write(mdd, rec, NULL) : 0;
/* assume on or off event; reset repeat-access time */
- mdd->mdd_cl.mc_starttime = rec->cr_time;
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
OBD_FREE(rec, reclen);
RETURN(rc);
memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
- mdd_root_dir_name, mdd_dot_lustre_name,
+ mdd_root_dir_name, dot_lustre_name,
fid, &dt_directory_features);
/* .lustre dir may be already present */
if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
rc = PTR_ERR(mdo);
CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
- mdd_dot_lustre_name, PFID(fid), rc);
- RETURN(rc);
+ dot_lustre_name, PFID(fid), rc);
+ return rc;
}
if (!IS_ERR(mdo))
return 0;
}
-static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
- struct md_attr *ma)
+
+static int dot_lustre_mdd_permission(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *cobj,
+ struct md_attr *attr, int mask)
{
- struct mdd_object *mdd_obj = md2mdd_obj(obj);
+ if (mask & ~(MAY_READ | MAY_EXEC))
+ return -EPERM;
+ else
+ return 0;
+}
- return mdd_attr_get_internal_locked(env, mdd_obj, ma);
+static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf,
+ const char *name)
+{
+ return 0;
}
-static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
- const struct md_attr *ma)
+static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
+{
+ return 0;
+}
+
+static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_buf *buf, const char *name,
+ int fl)
{
return -EPERM;
}
-static int dot_lustre_xattr_get(const struct lu_env *env,
- struct md_object *obj, struct lu_buf *buf,
- const char *name)
+static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
+ struct md_object *obj,
+ const char *name)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_readlink(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
{
return 0;
}
+static int dot_lustre_mdd_object_create(const struct lu_env *env,
+ struct md_object *obj,
+ const struct md_op_spec *spec,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_ref_add(const struct lu_env *env,
+ struct md_object *obj,
+ const struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_ref_del(const struct lu_env *env,
+ struct md_object *obj,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
int flags)
{
return 0;
}
-static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
- char *path, int pathlen, __u64 *recno, int *linkno)
-{
- return -ENOSYS;
-}
-
-static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
- struct md_attr *ma)
+static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
+ struct md_attr *ma, int mode)
{
struct mdd_object *mdd_obj = md2mdd_obj(obj);
return 0;
}
+static int dot_lustre_mdd_object_sync(const struct lu_env *env,
+ struct md_object *obj)
+{
+ return -ENOSYS;
+}
+
+static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
+ char *path, int pathlen, __u64 *recno, int *linkno)
+{
+ return -ENOSYS;
+}
+
+static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
+ struct lov_mds_md *lmm, struct ldlm_extent *extent,
+ struct lustre_handle *lockh)
+{
+ return -ENOSYS;
+}
+
+static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
+ struct lov_mds_md *lmm, struct lustre_handle *lockh)
+{
+ return -ENOSYS;
+}
+
static struct md_object_operations mdd_dot_lustre_obj_ops = {
- .moo_attr_get = dot_lustre_attr_get,
- .moo_attr_set = dot_lustre_attr_set,
- .moo_xattr_get = dot_lustre_xattr_get,
- .moo_open = dot_lustre_mdd_open,
- .moo_close = dot_lustre_close,
- .moo_readpage = mdd_readpage,
- .moo_path = dot_lustre_path
+ .moo_permission = dot_lustre_mdd_permission,
+ .moo_attr_get = mdd_attr_get,
+ .moo_attr_set = mdd_attr_set,
+ .moo_xattr_get = dot_lustre_mdd_xattr_get,
+ .moo_xattr_list = dot_lustre_mdd_xattr_list,
+ .moo_xattr_set = dot_lustre_mdd_xattr_set,
+ .moo_xattr_del = dot_lustre_mdd_xattr_del,
+ .moo_readpage = mdd_readpage,
+ .moo_readlink = dot_lustre_mdd_readlink,
+ .moo_object_create = dot_lustre_mdd_object_create,
+ .moo_ref_add = dot_lustre_mdd_ref_add,
+ .moo_ref_del = dot_lustre_mdd_ref_del,
+ .moo_open = dot_lustre_mdd_open,
+ .moo_close = dot_lustre_mdd_close,
+ .moo_capa_get = mdd_capa_get,
+ .moo_object_sync = dot_lustre_mdd_object_sync,
+ .moo_path = dot_lustre_mdd_path,
+ .moo_file_lock = dot_file_lock,
+ .moo_file_unlock = dot_file_unlock,
};
-static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
- const struct lu_name *lname, struct lu_fid *f,
- struct md_op_spec *spec)
+
+static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
+ const struct lu_name *lname, struct lu_fid *f,
+ struct md_op_spec *spec)
{
if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
*f = LU_OBF_FID;
return 0;
}
-static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
- const struct lu_name *lname,
- struct md_object *child, struct md_op_spec *spec,
- struct md_attr* ma)
+static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
+ struct md_object *obj,
+ mdl_mode_t mode)
+{
+ return MDL_MINMODE;
+}
+
+static int dot_lustre_mdd_create(const struct lu_env *env,
+ struct md_object *pobj,
+ const struct lu_name *lname,
+ struct md_object *child,
+ struct md_op_spec *spec,
+ struct md_attr* ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_create_data(const struct lu_env *env,
+ struct md_object *p,
+ struct md_object *o,
+ const struct md_op_spec *spec,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_rename(const struct lu_env *env,
+ struct md_object *src_pobj,
+ struct md_object *tgt_pobj,
+ const struct lu_fid *lf,
+ const struct lu_name *lsname,
+ struct md_object *tobj,
+ const struct lu_name *ltname,
+ struct md_attr *ma)
{
return -EPERM;
}
-static int dot_lustre_rename(const struct lu_env *env,
- struct md_object *src_pobj,
- struct md_object *tgt_pobj,
- const struct lu_fid *lf,
- const struct lu_name *lsname,
- struct md_object *tobj,
- const struct lu_name *ltname, struct md_attr *ma)
+static int dot_lustre_mdd_link(const struct lu_env *env,
+ struct md_object *tgt_obj,
+ struct md_object *src_obj,
+ const struct lu_name *lname,
+ struct md_attr *ma)
{
return -EPERM;
}
-static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
- struct md_object *src_obj,
- const struct lu_name *lname, struct md_attr *ma)
+static int dot_lustre_mdd_unlink(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *cobj,
+ const struct lu_name *lname,
+ struct md_attr *ma)
{
return -EPERM;
}
-static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
- struct md_object *cobj, const struct lu_name *lname,
- struct md_attr *ma)
+static int dot_lustre_mdd_name_insert(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_name *lname,
+ const struct lu_fid *fid,
+ const struct md_attr *ma)
{
return -EPERM;
}
+static int dot_lustre_mdd_name_remove(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_name *lname,
+ const struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *tobj,
+ const struct lu_fid *fid,
+ const struct lu_name *lname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+
static struct md_dir_operations mdd_dot_lustre_dir_ops = {
- .mdo_lookup = dot_lustre_lookup,
- .mdo_create = dot_lustre_create,
- .mdo_rename = dot_lustre_rename,
- .mdo_link = dot_lustre_link,
- .mdo_unlink = dot_lustre_unlink,
+ .mdo_is_subdir = mdd_is_subdir,
+ .mdo_lookup = dot_lustre_mdd_lookup,
+ .mdo_lock_mode = dot_lustre_mdd_lock_mode,
+ .mdo_create = dot_lustre_mdd_create,
+ .mdo_create_data = dot_lustre_mdd_create_data,
+ .mdo_rename = dot_lustre_mdd_rename,
+ .mdo_link = dot_lustre_mdd_link,
+ .mdo_unlink = dot_lustre_mdd_unlink,
+ .mdo_name_insert = dot_lustre_mdd_name_insert,
+ .mdo_name_remove = dot_lustre_mdd_name_remove,
+ .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
};
static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
/* "fid" is a virtual object and hence does not have any "real"
* attributes. So we reuse attributes of .lustre for "fid" dir */
ma->ma_need |= MA_INODE;
- rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
+ rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
if (rc)
return rc;
ma->ma_valid |= MA_INODE;
return 0;
if (ma->ma_need & MA_LOV_DEF) {
- rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
- &ma->ma_lmm_size);
+ rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
if (rc > 0) {
+ ma->ma_lmm_size = rc;
ma->ma_valid |= MA_LOV;
rc = 0;
}
return -EPERM;
}
+static int obf_xattr_list(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
+{
+ return 0;
+}
+
static int obf_xattr_get(const struct lu_env *env,
struct md_object *obj, struct lu_buf *buf,
const char *name)
return 0;
}
+static int obf_xattr_set(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_buf *buf, const char *name,
+ int fl)
+{
+ return -EPERM;
+}
+
+static int obf_xattr_del(const struct lu_env *env,
+ struct md_object *obj,
+ const char *name)
+{
+ return -EPERM;
+}
+
static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
int flags)
{
}
static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
- struct md_attr *ma)
+ struct md_attr *ma, int mode)
{
struct mdd_object *mdd_obj = md2mdd_obj(obj);
}
static struct md_object_operations mdd_obf_obj_ops = {
- .moo_attr_get = obf_attr_get,
- .moo_attr_set = obf_attr_set,
- .moo_xattr_get = obf_xattr_get,
- .moo_open = obf_mdd_open,
- .moo_close = obf_mdd_close,
- .moo_readpage = obf_mdd_readpage,
- .moo_path = obf_path
+ .moo_attr_get = obf_attr_get,
+ .moo_attr_set = obf_attr_set,
+ .moo_xattr_list = obf_xattr_list,
+ .moo_xattr_get = obf_xattr_get,
+ .moo_xattr_set = obf_xattr_set,
+ .moo_xattr_del = obf_xattr_del,
+ .moo_open = obf_mdd_open,
+ .moo_close = obf_mdd_close,
+ .moo_readpage = obf_mdd_readpage,
+ .moo_path = obf_path
};
/**
while (*name == '[')
name++;
- sscanf(name, SFID, &(f->f_seq), &(f->f_oid),
- &(f->f_ver));
+ sscanf(name, SFID, RFID(f));
if (!fid_is_sane(f)) {
- CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
- (__u64)1, 2, 0);
+ CWARN("%s: bad FID format [%s], should be "DFID"\n",
+ mdd->mdd_obd_dev->obd_name, lname->ln_name,
+ (__u64)FID_SEQ_NORMAL, 1, 0);
GOTO(out, rc = -EINVAL);
}
+ if (!fid_is_norm(f)) {
+ CWARN("%s: "DFID" is invalid, sequence should be "
+ ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f),
+ (__u64)FID_SEQ_NORMAL);
+ GOTO(out, rc = -EINVAL);
+ }
+
/* Check if object with this fid exists */
child = mdd_object_find(env, mdd, f);
if (child == NULL)
return rc;
dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
- mdd_dot_lustre_name, fid);
+ dot_lustre_name, fid);
if (IS_ERR(dt_dot_lustre)) {
rc = PTR_ERR(dt_dot_lustre);
GOTO(out, rc);
rc = mdd_init_obd(env, m, cfg);
if (rc) {
- CERROR("lov init error %d \n", rc);
+ CERROR("lov init error %d\n", rc);
GOTO(out, rc);
}
- rc = mdd_txn_init_credits(env, m);
- if (rc)
- break;
mdd_changelog_init(env, m);
break;
case LCFG_CLEANUP:
+ lu_dev_del_linkage(d->ld_site, d);
mdd_device_shutdown(env, m, cfg);
default:
rc = next->ld_ops->ldo_process_config(env, next, cfg);
ENTRY;
LASSERT(mds->mds_lov_objids != NULL);
- rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
+ rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
mds->mds_lov_objids, NULL);
}
#endif
/* Call that with obd_recovering = 1 just to update objids */
- obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
+ obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
/* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
+ cfs_spin_lock(&obd->obd_dev_lock);
obd->obd_recovering = 0;
+ cfs_spin_unlock(&obd->obd_dev_lock);
obd->obd_type->typ_dt_ops->o_postrecov(obd);
/* XXX: orphans handling. */
struct mdd_device *mdd = lu2mdd_dev(cdev);
struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
struct dt_object *root;
+ struct lu_fid fid;
int rc;
ENTRY;
GOTO(out, rc);
}
+ /* we use capa file to declare llog changes,
+ * will be fixed with new llog in 2.3 */
+ root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
+ if (IS_ERR(root))
+ GOTO(out, rc = PTR_ERR(root));
+
+ mdd->mdd_capa = root;
+ rc = mdd_lfsck_setup(env, mdd);
+
+ GOTO(out, rc);
+
out:
- RETURN(rc);
+ return rc;
}
const struct lu_device_operations mdd_lu_ops = {
* No permission check is needed.
*/
static int mdd_statfs(const struct lu_env *env, struct md_device *m,
- struct kstatfs *sfs)
+ struct obd_statfs *sfs)
{
struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
int rc;
int rc;
ENTRY;
+ /* need barrier for mds_capa_keys access. */
+ cfs_down_write(&mds->mds_notify_lock);
mds->mds_capa_keys = keys;
+ cfs_up_write(&mds->mds_notify_lock);
+
rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
timeout, alg, keys);
RETURN(rc);
struct md_device *m,
struct lustre_capa_key *key)
{
+ struct mds_capa_info info = { .uuid = NULL, .capa = key };
struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
- struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
+ struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
int rc;
ENTRY;
- rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
- sizeof(*key), key, NULL);
+ rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
+ KEY_CAPA_KEY, sizeof(info), &info, NULL);
RETURN(rc);
}
struct lu_device *next = &m->mdd_child->dd_lu_dev;
ENTRY;
- LASSERT(atomic_read(&lu->ld_ref) == 0);
+ LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
md_device_fini(&m->mdd_md_dev);
OBD_FREE_PTR(m);
RETURN(next);
}
EXPORT_SYMBOL(md_capainfo);
+/*
+ * context key constructor/destructor:
+ * mdd_quota_key_init, mdd_quota_key_fini
+ */
+LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
+
+struct lu_context_key mdd_quota_key = {
+ .lct_tags = LCT_SESSION,
+ .lct_init = mdd_quota_key_init,
+ .lct_fini = mdd_quota_key_fini
+};
+
+struct md_quota *md_quota(const struct lu_env *env)
+{
+ LASSERT(env->le_ses != NULL);
+ return lu_context_key_get(env->le_ses, &mdd_quota_key);
+}
+EXPORT_SYMBOL(md_quota);
+
static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
{
struct llog_ctxt *ctxt;
rec->cur_hdr.lrh_len = sizeof(*rec);
rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
- spin_lock(&mdd->mdd_cl.mc_user_lock);
+ cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
- spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
CERROR("Maximum number of changelog users exceeded!\n");
GOTO(out, rc = -EOVERFLOW);
}
*id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
rec->cur_endrec = mdd->mdd_cl.mc_index;
- spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
- rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
+ rc = llog_add(NULL, ctxt, &rec->cur_hdr, NULL, NULL, 0);
CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
out:
RETURN(rc);
}
+int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
+ struct thandle *handle)
+{
+ int rc;
+
+
+ /* XXX: this is a temporary solution to declare llog changes
+ * will be fixed in 2.3 with new llog implementation */
+
+ LASSERT(mdd->mdd_capa);
+
+ /* the llog record could be canceled either by modifying
+ * the plain llog's header or by destroying the llog itself
+ * when this record is the last one in it, it can't be known
+ * here, but the catlog's header will also be modified for
+ * the second case, then the first case can be covered and
+ * is no need to declare it */
+
+ /* destroy empty plain log */
+ rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
+ if (rc)
+ return rc;
+
+ /* record the catlog's header if an empty plain log was destroyed */
+ rc = dt_declare_record_write(env, mdd->mdd_capa,
+ sizeof(struct llog_logid_rec), 0, handle);
+ return rc;
+}
+
struct mdd_changelog_user_data {
__u64 mcud_endrec; /**< purge record for this user */
__u64 mcud_minrec; /**< lowest changelog recno still referenced */
__u32 mcud_minid; /**< user id with lowest rec reference */
__u32 mcud_usercount;
int mcud_found:1;
+ struct mdd_device *mcud_mdd;
};
#define MCUD_UNREGISTER -1LL
* 1. Find the smallest record everyone is willing to purge
* 2. Update the last purgeable record for this user
*/
-static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
- struct llog_rec_hdr *hdr, void *data)
+static int mdd_changelog_user_purge_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
{
- struct llog_changelog_user_rec *rec;
- struct mdd_changelog_user_data *mcud =
- (struct mdd_changelog_user_data *)data;
- int rc;
+ struct llog_changelog_user_rec *rec;
+ struct mdd_changelog_user_data *mcud = data;
+ int rc;
+
ENTRY;
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
/* Special case: unregister this user */
if (mcud->mcud_endrec == MCUD_UNREGISTER) {
struct llog_cookie cookie;
+ void *th;
+ struct mdd_device *mdd = mcud->mcud_mdd;
+
cookie.lgc_lgl = llh->lgh_id;
cookie.lgc_index = hdr->lrh_index;
- rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
- 1, &cookie);
+
+ /* XXX This is a workaround for the deadlock of changelog
+ * adding vs. changelog cancelling. LU-81. */
+ th = mdd_trans_create(env, mdd);
+ if (IS_ERR(th)) {
+ CERROR("Cannot get thandle\n");
+ RETURN(-ENOMEM);
+ }
+
+ rc = mdd_declare_llog_cancel(env, mdd, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = mdd_trans_start(env, mdd, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
+ 1, &cookie);
if (rc == 0)
mcud->mcud_usercount--;
+
+stop:
+ mdd_trans_stop(env, mdd, 0, th);
RETURN(rc);
}
/* hdr+1 is loc of data */
hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
- rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
- hdr->lrh_index);
+ rc = llog_write_rec(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
+ hdr->lrh_index);
RETURN(rc);
}
-static int mdd_changelog_user_purge(struct mdd_device *mdd, int id,
+static int mdd_changelog_user_purge(const struct lu_env *env,
+ struct mdd_device *mdd, int id,
long long endrec)
{
struct mdd_changelog_user_data data;
int rc;
ENTRY;
- CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec);
+ CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
data.mcud_id = id;
data.mcud_minid = 0;
data.mcud_minrec = 0;
data.mcud_usercount = 0;
data.mcud_endrec = endrec;
- spin_lock(&mdd->mdd_cl.mc_lock);
+ data.mcud_mdd = mdd;
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
endrec = mdd->mdd_cl.mc_index;
- spin_unlock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
if ((data.mcud_endrec == 0) ||
((data.mcud_endrec > endrec) &&
(data.mcud_endrec != MCUD_UNREGISTER)))
return -ENXIO;
LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
- rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
- (void *)&data, 0, 0);
+ rc = llog_cat_process(env, ctxt->loc_handle,
+ mdd_changelog_user_purge_cb, (void *)&data,
+ 0, 0);
if ((rc >= 0) && (data.mcud_minrec > 0)) {
CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
", referenced by "CHANGELOG_USER_PREFIX"%d\n",
data.mcud_minrec, data.mcud_minid);
- rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
+ rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
} else {
CWARN("Could not determine changelog records to purge; rc=%d\n",
rc);
/** mdd_iocontrol
* May be called remotely from mdt_iocontrol_handle or locally from
- * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce or
- * swab an obd_ioctl_data format (but local ioctl handler does).
+ * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
+ * an obd_ioctl_data format (but local ioctl handler does).
* \param cmd - ioc
* \param len - data len
* \param karg - ioctl data, in kernel space
mdd = lu2mdd_dev(&m->md_lu_dev);
/* Doesn't use obd_ioctl_data */
- if (cmd == OBD_IOC_CHANGELOG_CLEAR) {
+ switch (cmd) {
+ case OBD_IOC_CHANGELOG_CLEAR: {
struct changelog_setinfo *cs = karg;
- if (len != sizeof(*cs)) {
- CERROR("Bad changelog_clear ioctl size %d\n", len);
- RETURN(-EINVAL);
- }
- rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno);
+ rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
+ cs->cs_recno);
RETURN(rc);
}
+ case OBD_IOC_GET_MNTOPT: {
+ mntopt_t *mntopts = (mntopt_t *)karg;
+ *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
+ RETURN(0);
+ }
+ case OBD_IOC_START_LFSCK: {
+ struct lfsck_start *start = karg;
+ struct md_lfsck *lfsck = &mdd->mdd_lfsck;
+
+ /* Return the kernel service version. */
+ /* XXX: version can be used for compatibility in the future. */
+ start->ls_version = lfsck->ml_version;
+ rc = mdd_lfsck_start(env, lfsck, start);
+ RETURN(rc);
+ }
+ case OBD_IOC_STOP_LFSCK: {
+ rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
+ RETURN(rc);
+ }
+ }
/* Below ioctls use obd_ioctl_data */
if (len != sizeof(*data)) {
rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
break;
case OBD_IOC_CHANGELOG_DEREG:
- rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1,
+ rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
MCUD_UNREGISTER);
break;
default:
- rc = -EOPNOTSUPP;
+ rc = -ENOTTY;
}
RETURN (rc);
}
/* type constructor/destructor: mdd_type_init, mdd_type_fini */
-LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
+LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
+ &mdd_quota_key);
const struct md_device_operations mdd_ops = {
.mdo_statfs = mdd_statfs,
.llod_feat = &dt_directory_features,
};
+static struct lu_local_obj_desc llod_lfsck_bookmark = {
+ .llod_name = lfsck_bookmark_name,
+ .llod_oid = LFSCK_BOOKMARK_OID,
+ .llod_is_index = 0,
+};
+
static int __init mdd_mod_init(void)
{
- struct lprocfs_static_vars lvars;
- lprocfs_mdd_init_vars(&lvars);
+ struct lprocfs_static_vars lvars;
+ lprocfs_mdd_init_vars(&lvars);
- llo_local_obj_register(&llod_capa_key);
- llo_local_obj_register(&llod_mdd_orphan);
- llo_local_obj_register(&llod_mdd_root);
+ llo_local_obj_register(&llod_capa_key);
+ llo_local_obj_register(&llod_mdd_orphan);
+ llo_local_obj_register(&llod_mdd_root);
+ llo_local_obj_register(&llod_lfsck_bookmark);
- return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
- LUSTRE_MDD_NAME, &mdd_device_type);
+ return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
+ LUSTRE_MDD_NAME, &mdd_device_type);
}
static void __exit mdd_mod_exit(void)
{
- llo_local_obj_unregister(&llod_capa_key);
- llo_local_obj_unregister(&llod_mdd_orphan);
- llo_local_obj_unregister(&llod_mdd_root);
+ llo_local_obj_unregister(&llod_capa_key);
+ llo_local_obj_unregister(&llod_mdd_orphan);
+ llo_local_obj_unregister(&llod_mdd_root);
+ llo_local_obj_unregister(&llod_lfsck_bookmark);
- class_unregister_type(LUSTRE_MDD_NAME);
+ class_unregister_type(LUSTRE_MDD_NAME);
}
MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");