* GPL HEADER END
*/
/*
- * Copyright (c) 2012 Intel, Inc.
+ * Copyright (c) 2012, Intel Corporation.
* Use is subject to license terms.
*
* Author: Johann Lombardi <johann.lombardi@intel.com>
#include <obd_class.h>
#include "qsd_internal.h"
+cfs_mem_cache_t *upd_kmem;
+
+struct lu_kmem_descr qsd_caches[] = {
+ {
+ .ckd_cache = &upd_kmem,
+ .ckd_name = "upd_kmem",
+ .ckd_size = sizeof(struct qsd_upd_rec)
+ },
+ {
+ .ckd_cache = NULL
+ }
+};
+
/* define qsd thread key */
LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
int count, int *eof, void *data)
{
struct qsd_instance *qsd = (struct qsd_instance *)data;
+ char enabled[5];
+ int rc;
LASSERT(qsd != NULL);
- return snprintf(page, count,
- "target name: %s\n"
- "pool ID: %d\n"
- "type: %s\n"
- "quota enabled: none\n",
- qsd->qsd_svname, qsd->qsd_pool_id,
- qsd->qsd_is_md ? "md" : "dt");
+ memset(enabled, 0, sizeof(enabled));
+ if (qsd_type_enabled(qsd, USRQUOTA))
+ strcat(enabled, "u");
+ if (qsd_type_enabled(qsd, GRPQUOTA))
+ strcat(enabled, "g");
+ if (strlen(enabled) == 0)
+ strcat(enabled, "none");
+
+ rc = snprintf(page, count,
+ "target name: %s\n"
+ "pool ID: %d\n"
+ "type: %s\n"
+ "quota enabled: %s\n"
+ "conn to master: %s\n",
+ qsd->qsd_svname, qsd->qsd_pool_id,
+ qsd->qsd_is_md ? "md" : "dt", enabled,
+ qsd->qsd_exp_valid ? "setup" : "not setup yet");
+
+ if (qsd->qsd_prepared) {
+ memset(enabled, 0, sizeof(enabled));
+ if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
+ strcat(enabled, "u");
+ if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
+ strcat(enabled, "g");
+ if (strlen(enabled) == 0)
+ strcat(enabled, "none");
+ rc += snprintf(page + rc, count - rc,
+ "space acct: %s\n"
+ "user uptodate: glb[%d],slv[%d],reint[%d]\n"
+ "group uptodate: glb[%d],slv[%d],reint[%d]\n",
+ enabled,
+ qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
+ qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
+ qsd->qsd_type_array[USRQUOTA]->qqi_reint,
+ qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
+ qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
+ qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
+ }
+ return rc;
}
static int lprocfs_qsd_rd_enabled(char *page, char **start, off_t off,
{
struct qsd_instance *qsd = (struct qsd_instance *)data;
char enabled[5];
+
LASSERT(qsd != NULL);
memset(enabled, 0, sizeof(enabled));
return snprintf(page, count, "%s\n", enabled);
}
+/* force reintegration procedure to be executed.
+ * Used for test/debugging purpose */
+static int lprocfs_qsd_wr_force_reint(struct file *file, const char *buffer,
+ unsigned long count, void *data)
+{
+ struct qsd_instance *qsd = (struct qsd_instance *)data;
+ int rc = 0, qtype;
+
+ LASSERT(qsd != NULL);
+
+ write_lock(&qsd->qsd_lock);
+ if (qsd->qsd_stopping) {
+ /* don't mess up with shutdown procedure, it is already
+ * complicated enough */
+ rc = -ESHUTDOWN;
+ } else if (!qsd->qsd_prepared) {
+ rc = -EAGAIN;
+ } else {
+ /* mark all indexes as stale */
+ for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
+ qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
+ qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
+ }
+ }
+ write_unlock(&qsd->qsd_lock);
+
+ if (rc)
+ return rc;
+
+ /* kick off reintegration */
+ for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
+ rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
+ if (rc)
+ break;
+ }
+ return rc == 0 ? count : rc;
+}
+
+static int lprocfs_qsd_rd_timeout(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
+{
+ struct qsd_instance *qsd = (struct qsd_instance *)data;
+ LASSERT(qsd != NULL);
+
+ return snprintf(page, count, "%d\n", qsd_wait_timeout(qsd));
+}
+
+static int lprocfs_qsd_wr_timeout(struct file *file, const char *buffer,
+ unsigned long count, void *data)
+{
+ struct qsd_instance *qsd = (struct qsd_instance *)data;
+ int timeout, rc;
+ LASSERT(qsd != NULL);
+
+ rc = lprocfs_write_helper(buffer, count, &timeout);
+ if (rc)
+ return rc;
+ if (timeout < 0)
+ return -EINVAL;
+
+ qsd->qsd_timeout = timeout;
+ return count;
+}
+
static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
{ "info", lprocfs_qsd_rd_state, 0, 0},
{ "enabled", lprocfs_qsd_rd_enabled, 0, 0},
+ { "force_reint", 0, lprocfs_qsd_wr_force_reint, 0},
+ { "timeout", lprocfs_qsd_rd_timeout, lprocfs_qsd_wr_timeout, 0},
{ NULL }
};
/*
+ * Callback function invoked by the OSP layer when the connection to the master
+ * has been set up.
+ *
+ * \param data - is a pointer to the qsd_instance
+ *
+ * \retval - 0 on success, appropriate error on failure
+ */
+static int qsd_conn_callback(void *data)
+{
+ struct qsd_instance *qsd = (struct qsd_instance *)data;
+ int type;
+ ENTRY;
+
+ /* qsd_exp should now be valid */
+ LASSERT(qsd->qsd_exp);
+
+ /* grab reference on namespace */
+ ldlm_namespace_get(class_exp2obd(qsd->qsd_exp)->obd_namespace);
+ qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
+
+ write_lock(&qsd->qsd_lock);
+ /* notify that qsd_exp is now valid */
+ qsd->qsd_exp_valid = true;
+ write_unlock(&qsd->qsd_lock);
+
+ /* Now that the connection to master is setup, we can initiate the
+ * reintegration procedure for quota types which are enabled.
+ * It is worth noting that, if the qsd_instance hasn't been started
+ * already, then we can only complete the first two steps of the
+ * reintegration procedure (i.e. global lock enqueue and slave
+ * index transfer) since the space usage reconciliation (i.e.
+ * step 3) will have to wait for qsd_start() to be called */
+ for (type = USRQUOTA; type < MAXQUOTAS; type++) {
+ struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
+ cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
+ }
+
+ RETURN(0);
+}
+
+/*
* Release qsd_qtype_info structure which contains data associated with a
* given quota type. This releases the accounting objects.
* It's called on OSD cleanup when the qsd instance is released.
qqi = qsd->qsd_type_array[qtype];
qsd->qsd_type_array[qtype] = NULL;
+ /* all deferred work lists should be empty */
+ LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
+ LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
+
+ /* shutdown lquota site */
+ if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
+ lquota_site_free(env, qqi->qqi_site);
+ qqi->qqi_site = NULL;
+ }
+
/* by now, all qqi users should have gone away */
LASSERT(cfs_atomic_read(&qqi->qqi_ref) == 1);
lu_ref_fini(&qqi->qqi_reference);
qqi->qqi_glb_uptodate = false;
qqi->qqi_slv_uptodate = false;
qqi->qqi_reint = false;
- memset(&qqi->qqi_lockh, 0, sizeof(qqi->qqi_lockh));
-
- /* open accounting object */
- LASSERT(qqi->qqi_acct_obj == NULL);
- qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev,
- qtype == USRQUOTA ? ACCT_USER_OID
- : ACCT_GROUP_OID);
- /* don't print any error message on failure in order not to confuse
- * non-OFD user (e.g. 2.3 MDT stack) */
- if (IS_ERR(qqi->qqi_acct_obj))
+ cfs_waitq_init(&qqi->qqi_reint_thread.t_ctl_waitq);
+ thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
+ CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
+ CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
+
+ /* open accounting object */
+ LASSERT(qqi->qqi_acct_obj == NULL);
+ qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
+ if (IS_ERR(qqi->qqi_acct_obj)) {
+ CDEBUG(D_QUOTA, "%s: no %s space accounting support rc:%ld\n",
+ qsd->qsd_svname, QTYPE_NAME(qtype),
+ PTR_ERR(qqi->qqi_acct_obj));
qqi->qqi_acct_obj = NULL;
+ qsd->qsd_acct_failed = true;
+ }
/* open global index copy */
LASSERT(qqi->qqi_glb_obj == NULL);
}
qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
- /* register proc entry for accounting object */
+ /* allocate site */
+ qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
+ if (IS_ERR(qqi->qqi_site)) {
+ CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
+ PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
+ GOTO(out, rc = PTR_ERR(qqi->qqi_site));
+ }
+
+ /* register proc entry for accounting & global index copy objects */
rc = lprocfs_seq_create(qsd->qsd_proc,
qtype == USRQUOTA ? "acct_user" : "acct_group",
0444, &lprocfs_quota_seq_fops,
qqi->qqi_acct_obj);
if (rc) {
- CWARN("%s: can't add procfs entry for accounting file %d\n",
- qsd->qsd_svname, rc);
+ CERROR("%s: can't add procfs entry for accounting file %d\n",
+ qsd->qsd_svname, rc);
GOTO(out, rc);
}
+ rc = lprocfs_seq_create(qsd->qsd_proc,
+ qtype == USRQUOTA ? "limit_user" : "limit_group",
+ 0444, &lprocfs_quota_seq_fops,
+ qqi->qqi_glb_obj);
+ if (rc) {
+ CERROR("%s: can't add procfs entry for global index copy %d\n",
+ qsd->qsd_svname, rc);
+ GOTO(out, rc);
+ }
EXIT;
out:
if (rc)
/*
* Release a qsd_instance. Companion of qsd_init(). This releases all data
- * structures associated with the quota slave.
+ * structures associated with the quota slave (on-disk objects, lquota entry
+ * tables, ...).
* This function should be called when the OSD is shutting down.
*
* \param env - is the environment passed by the caller
int qtype;
ENTRY;
+ if (unlikely(qsd == NULL))
+ RETURN_EXIT;
+
CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
+ write_lock(&qsd->qsd_lock);
qsd->qsd_stopping = true;
-
- /* remove from the list of fsinfo */
- if (!cfs_list_empty(&qsd->qsd_link)) {
- LASSERT(qsd->qsd_fsinfo != NULL);
- cfs_down(&qsd->qsd_fsinfo->qfs_sem);
- cfs_list_del_init(&qsd->qsd_link);
- cfs_up(&qsd->qsd_fsinfo->qfs_sem);
- }
+ write_unlock(&qsd->qsd_lock);
/* remove qsd proc entry */
if (qsd->qsd_proc != NULL) {
qsd->qsd_proc = NULL;
}
+ /* stop the writeback thread */
+ qsd_stop_upd_thread(qsd);
+
+ /* shutdown the reintegration threads */
+ for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
+ if (qsd->qsd_type_array[qtype] == NULL)
+ continue;
+ qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
+ }
+
+ /* release reference on namespace */
+ if (qsd->qsd_ns != NULL) {
+ ldlm_namespace_put(qsd->qsd_ns);
+ qsd->qsd_ns = NULL;
+ }
+
/* free per-quota type data */
for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
qsd_qtype_fini(env, qsd, qtype);
+ /* deregister connection to the quota master */
+ qsd->qsd_exp_valid = false;
+ lustre_deregister_lwp_item(&qsd->qsd_exp);
+
/* release per-filesystem information */
- if (qsd->qsd_fsinfo != NULL)
+ if (qsd->qsd_fsinfo != NULL) {
+ down(&qsd->qsd_fsinfo->qfs_sem);
+ /* remove from the list of fsinfo */
+ cfs_list_del_init(&qsd->qsd_link);
+ up(&qsd->qsd_fsinfo->qfs_sem);
qsd_put_fsinfo(qsd->qsd_fsinfo);
+ qsd->qsd_fsinfo = NULL;
+ }
/* release quota root directory */
if (qsd->qsd_root != NULL) {
qsd->qsd_dev = NULL;
}
+ CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
OBD_FREE_PTR(qsd);
EXIT;
}
{
struct qsd_thread_info *qti = qsd_info(env);
struct qsd_instance *qsd;
- int rc;
+ int rc, type, idx;
ENTRY;
+ /* only configure qsd for MDT & OST */
+ type = server_name2index(svname, &idx, NULL);
+ if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
+ RETURN(NULL);
+
/* allocate qsd instance */
OBD_ALLOC_PTR(qsd);
if (qsd == NULL)
RETURN(ERR_PTR(-ENOMEM));
- cfs_rwlock_init(&qsd->qsd_lock);
+ /* generic initializations */
+ rwlock_init(&qsd->qsd_lock);
CFS_INIT_LIST_HEAD(&qsd->qsd_link);
+ thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
+ cfs_waitq_init(&qsd->qsd_upd_thread.t_ctl_waitq);
+ CFS_INIT_LIST_HEAD(&qsd->qsd_upd_list);
+ spin_lock_init(&qsd->qsd_adjust_lock);
+ CFS_INIT_LIST_HEAD(&qsd->qsd_adjust_list);
+ qsd->qsd_prepared = false;
+ qsd->qsd_started = false;
+
/* copy service name */
- strncpy(qsd->qsd_svname, svname, MAX_OBD_NAME);
+ if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
+ >= sizeof(qsd->qsd_svname))
+ GOTO(out, rc = -E2BIG);
/* grab reference on osd device */
lu_device_get(&dev->dd_lu_dev);
}
/* add in the list of lquota_fsinfo */
- cfs_down(&qsd->qsd_fsinfo->qfs_sem);
+ down(&qsd->qsd_fsinfo->qfs_sem);
list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
- cfs_up(&qsd->qsd_fsinfo->qfs_sem);
+ up(&qsd->qsd_fsinfo->qfs_sem);
/* register procfs directory */
qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
svname, rc);
GOTO(out, rc);
}
+ EXIT;
out:
if (rc) {
qsd_fini(env, qsd);
*/
int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
{
- int rc, qtype;
+ struct qsd_thread_info *qti = qsd_info(env);
+ int qtype, rc = 0;
ENTRY;
- LASSERT(qsd != NULL);
+ if (unlikely(qsd == NULL))
+ RETURN(0);
+
+ read_lock(&qsd->qsd_lock);
+ if (qsd->qsd_prepared) {
+ CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
+ rc = -EALREADY;
+ }
+ read_unlock(&qsd->qsd_lock);
+ if (rc)
+ RETURN(rc);
/* Record whether this qsd instance is managing quota enforcement for a
* MDT (i.e. inode quota) or OST (block quota) */
- if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev))
+ if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
qsd->qsd_is_md = true;
+ qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
+ } else {
+ qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
+ }
/* look-up on-disk directory for the quota slave */
qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
RETURN(rc);
}
+ /* pools successfully setup, mark the qsd as prepared */
+ write_lock(&qsd->qsd_lock);
+ qsd->qsd_prepared = true;
+ write_unlock(&qsd->qsd_lock);
+
+ /* start reintegration thread for each type, if required */
+ for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
+ struct qsd_qtype_info *qqi = qsd->qsd_type_array[qtype];
+
+ if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
+ LCONSOLE_ERROR("%s: can't enable quota enforcement "
+ "since space accounting isn't functional"
+ ". Please run tunefs.lustre --quota on "
+ "an unmounted filesystem if not done "
+ "already\n", qsd->qsd_svname);
+ break;
+ }
+
+ rc = qsd_start_reint_thread(qqi);
+ if (rc) {
+ CERROR("%s: failed to start reint thread for type %s "
+ "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
+ rc);
+ RETURN(rc);
+ }
+ }
+
+ /* start writeback thread */
+ rc = qsd_start_upd_thread(qsd);
+ if (rc) {
+ CERROR("%s: failed to start writeback thread (%d)\n",
+ qsd->qsd_svname, rc);
+ RETURN(rc);
+ }
+
+ /* generate osp name */
+ rc = tgt_name2lwpname((char *)qsd->qsd_svname, qti->qti_buf);
+ if (rc) {
+ CERROR("%s: failed to generate ospname (%d)\n",
+ qsd->qsd_svname, rc);
+ RETURN(rc);
+ }
+
+ /* the connection callback will start the reintegration
+ * procedure if quota is enabled */
+ rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
+ qsd_conn_callback, (void *)qsd);
+ if (rc) {
+ CERROR("%s: fail to get connection to master (%d)\n",
+ qsd->qsd_svname, rc);
+ RETURN(rc);
+ }
+
RETURN(0);
}
EXPORT_SYMBOL(qsd_prepare);
+/*
+ * Start a qsd instance. This will complete the last step of the reintegration
+ * procedure as soon as possible (provided that the master is reachable).
+ * This should be called when recovery has been completed and quota should now
+ * be enforced on every operations.
+ *
+ * \param env - the environment passed by the caller
+ * \param qsd - is the qsd instance associated with the osd device to start
+ */
+int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
+{
+ int type, rc = 0;
+ ENTRY;
+
+ if (unlikely(qsd == NULL))
+ RETURN(0);
+
+ write_lock(&qsd->qsd_lock);
+ if (!qsd->qsd_prepared) {
+ CERROR("%s: can't start qsd instance since it wasn't properly "
+ "initialized\n", qsd->qsd_svname);
+ rc = -EFAULT;
+ } else if (qsd->qsd_started) {
+ CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
+ rc = -EALREADY;
+ } else {
+ /* notify that the qsd_instance is now started */
+ qsd->qsd_started = true;
+ }
+ write_unlock(&qsd->qsd_lock);
+
+ if (rc)
+ RETURN(rc);
+
+ /* Trigger the 3rd step of reintegration: If usage > granted, acquire
+ * up to usage; If usage < granted, release down to usage. */
+ for (type = USRQUOTA; type < MAXQUOTAS; type++) {
+ struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
+ cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
+ }
+
+ RETURN(rc);
+}
+EXPORT_SYMBOL(qsd_start);
+
void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
/*
*/
int qsd_glb_init(void)
{
+ int rc;
+
+ rc = lu_kmem_init(qsd_caches);
+ if (rc)
+ return rc;
+
qsd_key_init_generic(&qsd_thread_key, NULL);
lu_context_key_register(&qsd_thread_key);
lustre_register_quota_process_config(qsd_process_config);
+
return 0;
}
void qsd_glb_fini(void)
{
lustre_register_quota_process_config(NULL);
+ lu_kmem_fini(qsd_caches);
lu_context_key_degister(&qsd_thread_key);
}