* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_MDS
#include <linux/module.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_fid.h>
-#include <ldiskfs/ldiskfs.h>
#include <lustre_mds.h>
-#include <lustre/lustre_idl.h>
-#include <lustre_disk.h> /* for changelogs */
+#include <obd_support.h>
+#include <lu_object.h>
#include <lustre_param.h>
#include <lustre_fid.h>
static const char mdd_root_dir_name[] = "ROOT";
static const char mdd_obf_dir_name[] = "fid";
-static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
- const char *name, struct lu_device *next)
+/* Slab for MDD object allocation */
+cfs_mem_cache_t *mdd_object_kmem;
+
+static struct lu_kmem_descr mdd_caches[] = {
+ {
+ .ckd_cache = &mdd_object_kmem,
+ .ckd_name = "mdd_obj",
+ .ckd_size = sizeof(struct mdd_object)
+ },
+ {
+ .ckd_cache = NULL
+ }
+};
+
+static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
+ const char *nextdev)
{
- struct mdd_device *mdd = lu2mdd_dev(d);
- int rc;
- ENTRY;
+ struct obd_connect_data *data = NULL;
+ struct lu_device *lud = mdd2lu_dev(m);
+ struct obd_device *obd;
+ int rc;
+ ENTRY;
- mdd->mdd_child = lu2dt_dev(next);
+ LASSERT(m->mdd_child_exp == NULL);
+
+ OBD_ALLOC(data, sizeof(*data));
+ if (data == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ obd = class_name2obd(nextdev);
+ if (obd == NULL) {
+ CERROR("can't locate next device: %s\n", nextdev);
+ GOTO(out, rc = -ENOTCONN);
+ }
+
+ data->ocd_connect_flags = OBD_CONNECT_VERSION;
+ data->ocd_version = LUSTRE_VERSION_CODE;
+
+ rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
+ if (rc) {
+ CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
+ GOTO(out, rc);
+ }
+
+ lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
+ LASSERT(lud->ld_site);
+ m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
+ m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
+ lu_dev_add_linkage(lud->ld_site, lud);
+
+out:
+ if (data)
+ OBD_FREE(data, sizeof(*data));
+ RETURN(rc);
+}
- /* Prepare transactions callbacks. */
- 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 = 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;
+static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
+ struct lu_device_type *t, struct lustre_cfg *lcfg)
+{
+ int rc;
+ ENTRY;
+
+ mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
+ mdd->mdd_md_dev.md_ops = &mdd_ops;
+
+ rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
+ if (rc)
+ RETURN(rc);
+
+ mdd->mdd_atime_diff = MAX_ATIME_DIFF;
/* sync permission changes */
mdd->mdd_sync_permission = 1;
- rc = mdd_procfs_init(mdd, name);
+ dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
+
+ /* we are using service name but not mdd obd name
+ * for compatibility reasons.
+ * It is passed from MDT in lustre_cfg[2] buffer */
+ rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
+ if (rc < 0)
+ obd_disconnect(mdd->mdd_child_exp);
+
RETURN(rc);
}
struct lu_device *d)
{
struct mdd_device *mdd = lu2mdd_dev(d);
- struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
int rc;
+ if (d->ld_site)
+ lu_dev_del_linkage(d->ld_site, d);
+
rc = mdd_procfs_fini(mdd);
if (rc) {
CERROR("proc fini error %d \n", rc);
return ERR_PTR(rc);
}
- return next;
+ return NULL;
}
static void mdd_changelog_fini(const struct lu_env *env,
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);
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;
}
+ lu_site_purge(env, m->mdd_md_dev.md_lu_dev.ld_site, -1);
/* remove upcall device*/
md_upcall_fini(&m->mdd_md_dev);
+
+ if (m->mdd_child_exp)
+ obd_disconnect(m->mdd_child_exp);
+
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;
- ENTRY;
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
llh->lgh_id.lgl_oid);
mdd->mdd_cl.mc_index = rec->cr.cr_index;
- RETURN(LLOG_PROC_BREAK);
+ 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 =
(struct llog_changelog_user_rec *)hdr;
- ENTRY;
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
" in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
- cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
- mdd->mdd_cl.mc_lastuser = rec->cur_id;
- cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ spin_lock(&mdd->mdd_cl.mc_user_lock);
+ mdd->mdd_cl.mc_lastuser = rec->cur_id;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
+
+ return LLOG_PROC_BREAK;
+}
+
+static int llog_changelog_cancel_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
+ struct llog_cookie cookie;
+ long long endrec = *(long long *)data;
+ int rc;
+
+ ENTRY;
- RETURN(LLOG_PROC_BREAK);
+ /* This is always a (sub)log, not the catalog */
+ LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
+
+ if (rec->cr.cr_index > endrec)
+ /* records are in order, so we're done */
+ RETURN(LLOG_PROC_BREAK);
+
+ cookie.lgc_lgl = llh->lgh_id;
+ cookie.lgc_index = hdr->lrh_index;
+
+ /* cancel them one at a time. I suppose we could store up the cookies
+ * and cancel them all at once; probably more efficient, but this is
+ * done as a user call, so who cares... */
+ rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
+ &cookie);
+ RETURN(rc < 0 ? rc : 0);
+}
+
+static int llog_changelog_cancel(const struct lu_env *env,
+ struct llog_ctxt *ctxt,
+ struct lov_stripe_md *lsm, int count,
+ struct llog_cookie *cookies, int flags)
+{
+ struct llog_handle *cathandle = ctxt->loc_handle;
+ int rc;
+
+ ENTRY;
+
+ /* This should only be called with the catalog handle */
+ LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
+
+ rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
+ (void *)cookies, 0, 0);
+ if (rc >= 0)
+ /* 0 or 1 means we're done */
+ rc = 0;
+ else
+ CERROR("%s: cancel idx %u of catalog "LPX64" rc=%d\n",
+ ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
+ cathandle->lgh_id.lgl_oid, rc);
+
+ RETURN(rc);
}
+static struct llog_operations changelog_orig_logops;
-static int mdd_changelog_llog_init(struct mdd_device *mdd)
+int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
+
+static int mdd_changelog_llog_init(const struct lu_env *env,
+ struct mdd_device *mdd)
{
- struct obd_device *obd = mdd2obd_dev(mdd);
- struct llog_ctxt *ctxt;
- int rc;
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
+ int rc;
- /* Find last changelog entry number */
- ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
- if (ctxt == NULL) {
- CERROR("no changelog context\n");
- return -EINVAL;
- }
- if (!ctxt->loc_handle) {
- llog_ctxt_put(ctxt);
- return -EINVAL;
- }
+ OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
+ obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
+ rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
+ obd, &changelog_orig_logops);
+ if (rc) {
+ CERROR("%s: changelog llog setup failed: rc = %d\n",
+ obd->obd_name, rc);
+ RETURN(rc);
+ }
- rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ LASSERT(ctxt);
+
+ rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
+ CHANGELOG_CATALOG);
+ if (rc)
+ GOTO(out_cleanup, rc);
+
+ rc = llog_cat_init_and_process(env, ctxt->loc_handle);
+ if (rc)
+ GOTO(out_close, rc);
+
+ rc = llog_cat_reverse_process(env, ctxt->loc_handle,
changelog_init_cb, mdd);
- llog_ctxt_put(ctxt);
- if (rc < 0) {
- CERROR("changelog init failed: %d\n", rc);
- return rc;
- }
- CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
- mdd->mdd_cl.mc_index);
-
- /* Find last changelog user id */
- ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
- if (ctxt == NULL) {
- CERROR("no changelog user context\n");
- return -EINVAL;
- }
- if (!ctxt->loc_handle) {
- llog_ctxt_put(ctxt);
- return -EINVAL;
- }
+ if (rc < 0) {
+ CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
+ rc);
+ GOTO(out_close, rc);
+ }
- rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
- changelog_user_init_cb, mdd);
- llog_ctxt_put(ctxt);
+ CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
+ mdd->mdd_cl.mc_index);
- if (rc < 0) {
- CERROR("changelog user init failed: %d\n", rc);
- return rc;
- }
+ /* setup user changelog */
+ rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
+ obd, &changelog_orig_logops);
+ if (rc) {
+ CERROR("%s: changelog users llog setup failed: rc = %d\n",
+ obd->obd_name, rc);
+ GOTO(out_close, rc);
+ }
- /* If we have registered users, assume we want changelogs on */
- if (mdd->mdd_cl.mc_lastuser > 0)
- rc = mdd_changelog_on(mdd, 1);
+ uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
+ LASSERT(ctxt);
- return rc;
+ rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
+ CHANGELOG_USERS);
+ if (rc)
+ GOTO(out_ucleanup, rc);
+
+ uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
+ uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
+
+ rc = llog_cat_init_and_process(env, uctxt->loc_handle);
+ if (rc)
+ GOTO(out_uclose, rc);
+
+ rc = llog_cat_reverse_process(env, uctxt->loc_handle,
+ changelog_user_init_cb, mdd);
+ if (rc < 0) {
+ CERROR("%s: changelog user init failed: rc = %d\n",
+ obd->obd_name, rc);
+ GOTO(out_uclose, rc);
+ }
+
+ /* If we have registered users, assume we want changelogs on */
+ if (mdd->mdd_cl.mc_lastuser > 0) {
+ rc = mdd_changelog_on(env, mdd, 1);
+ if (rc < 0)
+ GOTO(out_uclose, rc);
+ }
+ llog_ctxt_put(ctxt);
+ llog_ctxt_put(uctxt);
+ RETURN(0);
+out_uclose:
+ llog_cat_close(env, uctxt->loc_handle);
+out_ucleanup:
+ llog_cleanup(env, uctxt);
+out_close:
+ llog_cat_close(env, ctxt->loc_handle);
+out_cleanup:
+ llog_cleanup(env, ctxt);
+ return rc;
}
static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
{
- int rc;
-
- mdd->mdd_cl.mc_index = 0;
- 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;
- cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
- mdd->mdd_cl.mc_lastuser = 0;
-
- rc = mdd_changelog_llog_init(mdd);
- if (rc) {
- CERROR("Changelog setup during init failed %d\n", rc);
- mdd->mdd_cl.mc_flags |= CLM_ERR;
- }
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ int rc;
+
+ mdd->mdd_cl.mc_index = 0;
+ 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);
+ mdd->mdd_cl.mc_lastuser = 0;
+
+ rc = mdd_changelog_llog_init(env, mdd);
+ if (rc) {
+ CERROR("%s: changelog setup during init failed: rc = %d\n",
+ obd->obd_name, rc);
+ mdd->mdd_cl.mc_flags |= CLM_ERR;
+ }
- return rc;
+ return rc;
}
-static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
+static void mdd_changelog_fini(const struct lu_env *env,
+ struct mdd_device *mdd)
{
- mdd->mdd_cl.mc_flags = 0;
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt;
+
+ mdd->mdd_cl.mc_flags = 0;
+
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ if (ctxt) {
+ llog_cat_close(env, ctxt->loc_handle);
+ llog_cleanup(env, ctxt);
+ }
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
+ if (ctxt) {
+ llog_cat_close(env, ctxt->loc_handle);
+ llog_cleanup(env, ctxt);
+ }
}
+int mdd_changelog_write_header(const struct lu_env *env,
+ struct mdd_device *mdd, int markerflags);
+
/* Start / stop recording */
-int mdd_changelog_on(struct mdd_device *mdd, int on)
+int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
{
int rc = 0;
mdd2obd_dev(mdd)->obd_name);
rc = -ESRCH;
} else {
- cfs_spin_lock(&mdd->mdd_cl.mc_lock);
- mdd->mdd_cl.mc_flags |= CLM_ON;
- 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);
- cfs_spin_lock(&mdd->mdd_cl.mc_lock);
- mdd->mdd_cl.mc_flags &= ~CLM_ON;
- 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
- * \param handle - currently ignored since llogs start their own transaction;
- * this will hopefully be fixed in llog rewrite
- * \retval 0 ok
- */
-int mdd_changelog_llog_write(struct mdd_device *mdd,
- struct llog_changelog_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(ctxt, &rec->cr_hdr, NULL, NULL, 0);
- llog_ctxt_put(ctxt);
-
- 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(ctxt, &rec->cr_hdr, NULL, NULL, 0);
- llog_ctxt_put(ctxt);
-
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ mdd->mdd_cl.mc_flags |= CLM_ON;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ rc = mdd_changelog_write_header(env, 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(env, mdd, CLM_FINI);
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ mdd->mdd_cl.mc_flags &= ~CLM_ON;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ }
return rc;
}
* \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;
- cfs_spin_lock(&mdd->mdd_cl.mc_lock);
- cur = (long long)mdd->mdd_cl.mc_index;
- cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ cur = (long long)mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
if (endrec > cur)
endrec = cur;
allright. */
if (endrec == cur) {
/* XXX: transaction is started by llog itself */
- rc = mdd_changelog_write_header(mdd, CLM_PURGE);
+ rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
if (rc)
goto out;
}
changed since the last purge) */
mdd->mdd_cl.mc_starttime = cfs_time_current_64();
- /* XXX: transaction is started by llog itself */
- rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
+ rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
out:
llog_ctxt_put(ctxt);
return rc;
* \param markerflags - CLM_*
* \retval 0 ok
*/
-int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
-{
- struct obd_device *obd = mdd2obd_dev(mdd);
- struct llog_changelog_rec *rec;
- int reclen;
- int len = strlen(obd->obd_name);
- int rc;
- ENTRY;
+int mdd_changelog_write_header(const struct lu_env *env,
+ struct mdd_device *mdd, int markerflags)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_changelog_rec *rec;
+ struct lu_buf *buf;
+ struct llog_ctxt *ctxt;
+ int reclen;
+ int len = strlen(obd->obd_name);
+ int rc;
+
+ ENTRY;
+
+ if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+ RETURN(0);
+ }
- reclen = llog_data_len(sizeof(*rec) + len);
- OBD_ALLOC(rec, reclen);
- if (rec == NULL)
- RETURN(-ENOMEM);
+ reclen = llog_data_len(sizeof(*rec) + len);
+ buf = mdd_buf_alloc(env, reclen);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+ rec = buf->lb_buf;
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.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
+ rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
+ rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
+ rec->cr_hdr.lrh_type = CHANGELOG_REC;
+ rec->cr.cr_time = cl_time();
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
- /* XXX: transaction is started by llog itself */
- rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
- mdd_changelog_llog_write(mdd, rec, NULL) : 0;
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ LASSERT(ctxt);
- /* assume on or off event; reset repeat-access time */
- mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+ rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
+ if (rc > 0)
+ rc = 0;
+ llog_ctxt_put(ctxt);
- OBD_FREE(rec, reclen);
- RETURN(rc);
+ /* assume on or off event; reset repeat-access time */
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+ RETURN(rc);
}
/**
return -EPERM;
}
+static int dot_lustre_mdd_swap_layouts(const struct lu_env *env,
+ struct md_object *obj1,
+ struct md_object *obj2,
+ __u64 flags)
+{
+ return -EPERM;
+}
+
static int dot_lustre_mdd_readlink(const struct lu_env *env,
struct md_object *obj, struct lu_buf *buf)
{
}
static struct md_object_operations mdd_dot_lustre_obj_ops = {
- .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,
+ .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_swap_layouts = dot_lustre_mdd_swap_layouts,
+ .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,
};
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;
- else
- return -ENOENT;
+ if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0) {
+ *f = LU_OBF_FID;
+ return 0;
+ }
- return 0;
+ return -ENOENT;
}
static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
struct md_attr *ma)
{
- int rc = 0;
-
- if (ma->ma_need & MA_INODE) {
- struct mdd_device *mdd = mdo2mdd(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 = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
- if (rc)
- return rc;
- ma->ma_valid |= MA_INODE;
- }
-
- /* "fid" directory does not have any striping information. */
- if (ma->ma_need & MA_LOV) {
- struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
- if (ma->ma_valid & MA_LOV)
- return 0;
-
- if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
- S_ISDIR(mdd_object_type(mdd_obj))))
- return 0;
-
- if (ma->ma_need & MA_LOV_DEF) {
- 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;
- }
- }
- }
+ struct mdd_device *mdd = mdo2mdd(obj);
- return rc;
+ /* "fid" is a virtual object and hence does not have any "real"
+ * attributes. So we reuse attributes of .lustre for "fid" dir */
+ return mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
}
static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
struct md_object *obj, struct lu_buf *buf,
const char *name)
{
- return 0;
+ struct mdd_device *mdd = mdo2mdd(obj);
+ struct mdd_object *root;
+ int rc = 0;
+
+ /*
+ * .lustre returns default striping which is 'stored'
+ * in the root
+ */
+ if (strcmp(name, XATTR_NAME_LOV) == 0) {
+ root = mdd_object_find(env, mdd, &mdd->mdd_local_root_fid);
+ if (IS_ERR(root))
+ return PTR_ERR(root);
+ rc = mdo_xattr_get(env, root, buf, name,
+ mdd_object_capa(env, md2mdd_obj(obj)));
+ mdd_object_put(env, root);
+ }
+
+ return rc;
}
static int obf_xattr_set(const struct lu_env *env,
sscanf(name, SFID, RFID(f));
if (!fid_is_sane(f)) {
CWARN("%s: bad FID format [%s], should be "DFID"\n",
- mdd->mdd_obd_dev->obd_name, lname->ln_name,
+ mdd2obd_dev(mdd)->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),
+ ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
(__u64)FID_SEQ_NORMAL);
GOTO(out, rc = -EINVAL);
}
struct dt_object *dt_dot_lustre;
struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
int rc;
+ ENTRY;
rc = create_dot_lustre_dir(env, m);
if (rc)
if (rc)
GOTO(out, rc);
dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
-
- rc = mdd_init_obd(env, m, cfg);
- if (rc) {
- CERROR("lov init error %d\n", rc);
- GOTO(out, rc);
- }
-
- mdd_changelog_init(env, m);
break;
case LCFG_CLEANUP:
+ mdd_lfsck_cleanup(env, m);
+ mdd_changelog_fini(env, m);
+ rc = next->ld_ops->ldo_process_config(env, next, cfg);
lu_dev_del_linkage(d->ld_site, d);
mdd_device_shutdown(env, m, cfg);
+ break;
default:
rc = next->ld_ops->ldo_process_config(env, next, cfg);
break;
RETURN(rc);
}
-#if 0
-static int mdd_lov_set_nextid(const struct lu_env *env,
- struct mdd_device *mdd)
-{
- struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
- int rc;
- ENTRY;
-
- LASSERT(mds->mds_lov_objids != NULL);
- 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);
-
- RETURN(rc);
-}
-
-static int mdd_cleanup_unlink_llog(const struct lu_env *env,
- struct mdd_device *mdd)
-{
- /* XXX: to be implemented! */
- return 0;
-}
-#endif
-
static int mdd_recovery_complete(const struct lu_env *env,
struct lu_device *d)
{
struct mdd_device *mdd = lu2mdd_dev(d);
- struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
- struct obd_device *obd = mdd2obd_dev(mdd);
+ struct lu_device *next;
int rc;
ENTRY;
LASSERT(mdd != NULL);
- LASSERT(obd != NULL);
-#if 0
- /* XXX: Do we need this in new stack? */
- rc = mdd_lov_set_nextid(env, mdd);
- if (rc) {
- CERROR("mdd_lov_set_nextid() failed %d\n",
- rc);
- RETURN(rc);
- }
-
- /* XXX: cleanup unlink. */
- rc = mdd_cleanup_unlink_llog(env, mdd);
- if (rc) {
- CERROR("mdd_cleanup_unlink_llog() failed %d\n",
- rc);
- RETURN(rc);
- }
-#endif
- /* Call that with obd_recovering = 1 just to update objids */
- 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);
+ next = &mdd->mdd_child->dd_lu_dev;
/* XXX: orphans handling. */
__mdd_orphan_cleanup(env, mdd);
RETURN(rc);
}
+static int mdd_find_or_create_root(const struct lu_env *env,
+ struct mdd_device *mdd)
+{
+ struct dt_object *root;
+ struct md_object *mroot;
+ struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
+ int rc = 0;
+
+ ENTRY;
+
+ /* Check if the "ROOT" entry exists already */
+ root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
+ fid);
+ if (!IS_ERR(root)) {
+ lu_object_put(env, &root->do_lu);
+ GOTO(out, rc = 0);
+ }
+
+ lu_root_fid(fid);
+ /* New Filesystem, create /ROOT */
+ mroot = llo_store_create_index(env, &mdd->mdd_md_dev, mdd->mdd_bottom,
+ "", mdd_root_dir_name, fid,
+ &dt_directory_features);
+ if (IS_ERR(mroot))
+ GOTO(out, rc = PTR_ERR(mroot));
+
+ lu_object_put(env, &mroot->mo_lu);
+out:
+ if (rc == 0)
+ mdd->mdd_root_fid = *fid;
+
+ RETURN(rc);
+}
+
static int mdd_prepare(const struct lu_env *env,
struct lu_device *pdev,
struct lu_device *cdev)
{
- 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;
+ 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 = &mdd_env_info(env)->mti_fid;
+ int rc;
ENTRY;
- rc = next->ld_ops->ldo_prepare(env, cdev, next);
- if (rc)
- GOTO(out, rc);
- dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
- root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
- &mdd->mdd_root_fid);
- if (!IS_ERR(root)) {
- LASSERT(root != NULL);
- lu_object_put(env, &root->do_lu);
- rc = orph_index_init(env, mdd);
- } else {
- rc = PTR_ERR(root);
- }
- if (rc)
- GOTO(out, rc);
+ rc = next->ld_ops->ldo_prepare(env, cdev, next);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ if (mdd_seq_site(mdd)->ss_node_id == 0) {
+ rc = mdd_find_or_create_root(env, mdd);
+ if (rc != 0) {
+ CERROR("%s: create root fid failed: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
+ GOTO(out, rc);
+ }
+
+ rc = mdd_dot_lustre_setup(env, mdd);
+ if (rc != 0) {
+ CERROR("%s: initializing .lustre failed: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
+ GOTO(out, rc);
+ }
+ }
- rc = mdd_dot_lustre_setup(env, mdd);
- if (rc) {
- CERROR("Error(%d) initializing .lustre objects\n", rc);
+ rc = orph_index_init(env, mdd);
+ if (rc != 0)
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);
+ 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);
+ rc = mdd_changelog_init(env, mdd);
+ if (rc != 0)
+ GOTO(out, rc);
+ rc = mdd_lfsck_setup(env, mdd);
+ if (rc != 0)
+ CERROR("%s: failed to initialize lfsck: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
out:
return rc;
}
/*
* No permission check is needed.
*/
-static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
- int *md_size, int *cookie_size)
-{
- struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
- ENTRY;
-
- *md_size = mdd_lov_mdsize(env, mdd);
- *cookie_size = mdd_lov_cookiesize(env, mdd);
-
- RETURN(0);
-}
-
static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
int mode, unsigned long timeout, __u32 alg,
struct lustre_capa_key *keys)
{
struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
- struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
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);
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_lov_exp;
- int rc;
- ENTRY;
-
- rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
- KEY_CAPA_KEY, sizeof(info), &info, NULL);
- RETURN(rc);
+ /* we do not support capabilities ... */
+ return -EINVAL;
}
static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
return (*h == NULL ? -ENOENT : 0);
}
+static struct lu_device *mdd_device_free(const struct lu_env *env,
+ struct lu_device *lu)
+{
+ struct mdd_device *m = lu2mdd_dev(lu);
+ ENTRY;
+
+ LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
+ md_device_fini(&m->mdd_md_dev);
+ OBD_FREE_PTR(m);
+ RETURN(NULL);
+}
+
static struct lu_device *mdd_device_alloc(const struct lu_env *env,
struct lu_device_type *t,
struct lustre_cfg *lcfg)
if (m == NULL) {
l = ERR_PTR(-ENOMEM);
} else {
- md_device_init(&m->mdd_md_dev, t);
+ int rc;
+
l = mdd2lu_dev(m);
- l->ld_ops = &mdd_lu_ops;
- m->mdd_md_dev.md_ops = &mdd_ops;
- md_upcall_init(&m->mdd_md_dev, NULL);
+ md_device_init(&m->mdd_md_dev, t);
+ rc = mdd_init0(env, m, t, lcfg);
+ if (rc != 0) {
+ mdd_device_free(env, l);
+ l = ERR_PTR(rc);
+ }
}
return l;
}
-static struct lu_device *mdd_device_free(const struct lu_env *env,
- struct lu_device *lu)
+/*
+ * we use exports to track all mdd users
+ */
+static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
+ struct obd_device *obd, struct obd_uuid *cluuid,
+ struct obd_connect_data *data, void *localdata)
{
- struct mdd_device *m = lu2mdd_dev(lu);
- struct lu_device *next = &m->mdd_child->dd_lu_dev;
- ENTRY;
+ struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
+ struct lustre_handle conn;
+ int rc;
+ ENTRY;
+
+ CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
- LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
- md_device_fini(&m->mdd_md_dev);
- OBD_FREE_PTR(m);
- RETURN(next);
+ rc = class_connect(&conn, obd, cluuid);
+ if (rc)
+ RETURN(rc);
+
+ *exp = class_conn2export(&conn);
+
+ /* Why should there ever be more than 1 connect? */
+ LASSERT(mdd->mdd_connects == 0);
+ mdd->mdd_connects++;
+
+ RETURN(0);
}
-static struct obd_ops mdd_obd_device_ops = {
- .o_owner = THIS_MODULE
-};
+/*
+ * once last export (we don't count self-export) disappeared
+ * mdd can be released
+ */
+static int mdd_obd_disconnect(struct obd_export *exp)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
+ int rc, release = 0;
+ ENTRY;
-/* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
-LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
+ mdd->mdd_connects--;
+ if (mdd->mdd_connects == 0)
+ release = 1;
-static struct lu_context_key mdd_ucred_key = {
- .lct_tags = LCT_SESSION,
- .lct_init = mdd_ucred_key_init,
- .lct_fini = mdd_ucred_key_fini
-};
+ rc = class_disconnect(exp);
+
+ if (rc == 0 && release)
+ class_manual_cleanup(obd);
+ RETURN(rc);
+}
-struct md_ucred *md_ucred(const struct lu_env *env)
+static int mdd_obd_health_check(const struct lu_env *env,
+ struct obd_device *obd)
{
- LASSERT(env->le_ses != NULL);
- return lu_context_key_get(env->le_ses, &mdd_ucred_key);
+ struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
+ int rc;
+ ENTRY;
+
+ LASSERT(mdd);
+ rc = obd_health_check(env, mdd->mdd_child_exp->exp_obd);
+ RETURN(rc);
}
-EXPORT_SYMBOL(md_ucred);
+
+static struct obd_ops mdd_obd_device_ops = {
+ .o_owner = THIS_MODULE,
+ .o_connect = mdd_obd_connect,
+ .o_disconnect = mdd_obd_disconnect,
+ .o_health_check = mdd_obd_health_check
+};
/*
* context key constructor/destructor:
}
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)
+static int mdd_changelog_user_register(const struct lu_env *env,
+ struct mdd_device *mdd, int *id)
{
struct llog_ctxt *ctxt;
struct llog_changelog_user_rec *rec;
int rc;
ENTRY;
- ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
+ ctxt = llog_get_context(mdd2obd_dev(mdd),
+ LLOG_CHANGELOG_USER_ORIG_CTXT);
if (ctxt == NULL)
RETURN(-ENXIO);
}
/* Assume we want it on since somebody registered */
- rc = mdd_changelog_on(mdd, 1);
+ rc = mdd_changelog_on(env, mdd, 1);
if (rc)
GOTO(out, rc);
rec->cur_hdr.lrh_len = sizeof(*rec);
rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
- cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
- if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
- 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;
- cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ 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);
+ 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);
- rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
+ rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
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;
- const struct lu_env *mcud_env;
};
#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;
- /* XXX This is a workaround for the deadlock of changelog
- * adding vs. changelog cancelling. LU-81. */
- th = mdd_trans_create(mcud->mcud_env, mdd);
- if (IS_ERR(th)) {
- CERROR("Cannot get thandle\n");
- RETURN(-ENOMEM);
- }
-
- rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
- if (rc)
- GOTO(stop, rc);
-
- rc = mdd_trans_start(mcud->mcud_env, mdd, th);
- if (rc)
- GOTO(stop, rc);
-
- rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
- 1, &cookie);
+ rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
+ 1, &cookie);
if (rc == 0)
mcud->mcud_usercount--;
-stop:
- mdd_trans_stop(mcud->mcud_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(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
+ hdr->lrh_index);
RETURN(rc);
}
data.mcud_minrec = 0;
data.mcud_usercount = 0;
data.mcud_endrec = endrec;
- data.mcud_mdd = mdd;
- data.mcud_env = env;
- cfs_spin_lock(&mdd->mdd_cl.mc_lock);
- endrec = mdd->mdd_cl.mc_index;
- cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ endrec = mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
if ((data.mcud_endrec == 0) ||
((data.mcud_endrec > endrec) &&
(data.mcud_endrec != MCUD_UNREGISTER)))
data.mcud_endrec = endrec;
- ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
+ ctxt = llog_get_context(mdd2obd_dev(mdd),
+ LLOG_CHANGELOG_USER_ORIG_CTXT);
if (ctxt == NULL)
return -ENXIO;
+
LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
rc = llog_cat_process(env, ctxt->loc_handle,
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);
if (!rc && data.mcud_usercount == 0)
/* No more users; turn changelogs off */
- rc = mdd_changelog_on(mdd, 0);
+ rc = mdd_changelog_on(env, mdd, 0);
RETURN (rc);
}
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);
+ rc = mdd_lfsck_start(env, &mdd->mdd_lfsck,
+ (struct lfsck_start *)karg);
RETURN(rc);
}
case OBD_IOC_STOP_LFSCK: {
- rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
+ rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck, false);
+ RETURN(rc);
+ }
+ case OBD_IOC_PAUSE_LFSCK: {
+ rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck, true);
RETURN(rc);
}
}
switch (cmd) {
case OBD_IOC_CHANGELOG_REG:
- rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
+ rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
break;
case OBD_IOC_CHANGELOG_DEREG:
rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
}
/* type constructor/destructor: mdd_type_init, mdd_type_fini */
-LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
- &mdd_quota_key);
+LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_capainfo_key);
const struct md_device_operations mdd_ops = {
.mdo_statfs = mdd_statfs,
.mdo_root_get = mdd_root_get,
- .mdo_maxsize_get = mdd_maxsize_get,
.mdo_init_capa_ctxt = mdd_init_capa_ctxt,
.mdo_update_capa_key= mdd_update_capa_key,
.mdo_llog_ctxt_get = mdd_llog_ctxt_get,
.mdo_iocontrol = mdd_iocontrol,
-#ifdef HAVE_QUOTA_SUPPORT
- .mdo_quota = {
- .mqo_notify = mdd_quota_notify,
- .mqo_setup = mdd_quota_setup,
- .mqo_cleanup = mdd_quota_cleanup,
- .mqo_recovery = mdd_quota_recovery,
- .mqo_check = mdd_quota_check,
- .mqo_on = mdd_quota_on,
- .mqo_off = mdd_quota_off,
- .mqo_setinfo = mdd_quota_setinfo,
- .mqo_getinfo = mdd_quota_getinfo,
- .mqo_setquota = mdd_quota_setquota,
- .mqo_getquota = mdd_quota_getquota,
- .mqo_getoinfo = mdd_quota_getoinfo,
- .mqo_getoquota = mdd_quota_getoquota,
- .mqo_invalidate = mdd_quota_invalidate,
- .mqo_finvalidate = mdd_quota_finvalidate
- }
-#endif
};
static struct lu_device_type_operations mdd_device_type_ops = {
.ldto_device_alloc = mdd_device_alloc,
.ldto_device_free = mdd_device_free,
- .ldto_device_init = mdd_device_init,
.ldto_device_fini = mdd_device_fini
};
.llod_feat = &dt_directory_features,
};
-static struct lu_local_obj_desc llod_mdd_root = {
- .llod_name = mdd_root_dir_name,
- .llod_oid = MDD_ROOT_INDEX_OID,
- .llod_is_index = 1,
- .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 struct lu_local_obj_desc llod_lfsck_namespace = {
+ .llod_name = lfsck_namespace_name,
+ .llod_oid = LFSCK_NAMESPACE_OID,
+ .llod_is_index = 1,
+ .llod_feat = &dt_lfsck_features,
+};
+
static int __init mdd_mod_init(void)
{
struct lprocfs_static_vars lvars;
+ int rc;
+
lprocfs_mdd_init_vars(&lvars);
+ rc = lu_kmem_init(mdd_caches);
+ if (rc)
+ return rc;
+
+ changelog_orig_logops = llog_osd_ops;
+ changelog_orig_logops.lop_cancel = llog_changelog_cancel;
+ changelog_orig_logops.lop_add = llog_cat_add_rec;
+ changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
+
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);
+ llo_local_obj_register(&llod_lfsck_namespace);
- return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
- LUSTRE_MDD_NAME, &mdd_device_type);
+ rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
+ LUSTRE_MDD_NAME, &mdd_device_type);
+ if (rc)
+ lu_kmem_fini(mdd_caches);
+ return rc;
}
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_lfsck_bookmark);
+ llo_local_obj_unregister(&llod_lfsck_namespace);
class_unregister_type(LUSTRE_MDD_NAME);
+ lu_kmem_fini(mdd_caches);
}
MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
git://git.whamcloud.com / fs / lustre-release.git / blobdiff