* 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 Whamcloud, Inc.
+ */
+/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*
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);
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;
" 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);
}
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;
}
struct llog_ctxt *ctxt;
int rc;
- if ((mdd->mdd_cl.mc_mask & (1 << rec->cr.cr_type)) == 0)
- return 0;
-
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();
- spin_lock(&mdd->mdd_cl.mc_lock);
+ 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;
- spin_unlock(&mdd->mdd_cl.mc_lock);
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
if (ctxt == NULL)
return -ENXIO;
rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
llog_ctxt_put(ctxt);
- cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
-
return rc;
}
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();
+ /* XXX: transaction is started by llog itself */
rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
out:
llog_ctxt_put(ctxt);
/* Status and action flags */
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.cr_time;
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
OBD_FREE(rec, reclen);
RETURN(rc);
}
static int dot_lustre_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);
return -ENOSYS;
}
-static dt_obj_version_t dot_lustre_mdd_version_get(const struct lu_env *env,
- struct md_object *obj)
+static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
+ char *path, int pathlen, __u64 *recno, int *linkno)
{
- return 0;
+ return -ENOSYS;
}
-static void dot_lustre_mdd_version_set(const struct lu_env *env,
- struct md_object *obj,
- dt_obj_version_t version)
+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;
+ 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)
+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_permission = dot_lustre_mdd_permission,
.moo_attr_get = dot_lustre_mdd_attr_get,
.moo_close = dot_lustre_mdd_close,
.moo_capa_get = mdd_capa_get,
.moo_object_sync = dot_lustre_mdd_object_sync,
- .moo_version_get = dot_lustre_mdd_version_get,
- .moo_version_set = dot_lustre_mdd_version_set,
.moo_path = dot_lustre_mdd_path,
+ .moo_file_lock = dot_file_lock,
+ .moo_file_unlock = dot_file_unlock,
};
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;
}
}
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);
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;
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))
+ mdd->mdd_capa = root;
+ else
+ rc = PTR_ERR(root);
+
out:
RETURN(rc);
}
* No permission check is needed.
*/
static int mdd_statfs(const struct lu_env *env, struct md_device *m,
- struct kstatfs *sfs)
+ cfs_kstatfs_t *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 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;
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);
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
/* 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);
if (rc == 0)
mcud->mcud_usercount--;
+
+stop:
+ mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
RETURN(rc);
}
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;
+ data.mcud_env = env;
+ 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)))
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;
- 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);
+ }
+ }
/* 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,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff