* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Di Wang <di.wang@intel.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MDS
#include "osp_internal.h"
d->opd_statfs_fresh_till);
}
+/*
+ * OSP tries to maintain pool of available objects so that calls to create
+ * objects don't block most of time
+ *
+ * each time OSP gets connected to OST, we should start from precreation cleanup
+ */
+static inline bool osp_precreate_running(struct osp_device *d)
+{
+ return !!(d->opd_pre_thread.t_flags & SVC_RUNNING);
+}
+
+static inline bool osp_precreate_stopped(struct osp_device *d)
+{
+ return !!(d->opd_pre_thread.t_flags & SVC_STOPPED);
+}
+
static void osp_statfs_timer_cb(unsigned long _d)
{
struct osp_device *d = (struct osp_device *) _d;
LASSERT(d);
- cfs_waitq_signal(&d->opd_pre_waitq);
+ if (d->opd_pre != NULL && osp_precreate_running(d))
+ wake_up(&d->opd_pre_waitq);
}
static int osp_statfs_interpret(const struct lu_env *env,
RETURN(0);
out:
/* couldn't update statfs, try again as soon as possible */
- cfs_waitq_signal(&d->opd_pre_waitq);
+ if (d->opd_pre != NULL && osp_precreate_running(d))
+ wake_up(&d->opd_pre_waitq);
+
if (req->rq_import_generation == imp->imp_generation)
CDEBUG(D_CACHE, "%s: couldn't update statfs: rc = %d\n",
d->opd_obd->obd_name, rc);
*/
d->opd_statfs_fresh_till = cfs_time_shift(-1);
cfs_timer_disarm(&d->opd_statfs_timer);
- cfs_waitq_signal(&d->opd_pre_waitq);
+ wake_up(&d->opd_pre_waitq);
}
}
-
-/*
- * OSP tries to maintain pool of available objects so that calls to create
- * objects don't block most of time
- *
- * each time OSP gets connected to OST, we should start from precreation cleanup
- */
-static inline int osp_precreate_running(struct osp_device *d)
-{
- return !!(d->opd_pre_thread.t_flags & SVC_RUNNING);
-}
-
-static inline int osp_precreate_stopped(struct osp_device *d)
-{
- return !!(d->opd_pre_thread.t_flags & SVC_STOPPED);
-}
-
static inline int osp_objs_precreated(const struct lu_env *env,
struct osp_device *osp)
{
- struct lu_fid *fid1 = &osp->opd_pre_last_created_fid;
- struct lu_fid *fid2 = &osp->opd_pre_used_fid;
-
- LASSERTF(fid_seq(fid1) == fid_seq(fid2),
- "Created fid"DFID" Next fid "DFID"\n", PFID(fid1), PFID(fid2));
-
- if (fid_is_idif(fid1)) {
- struct ost_id *oi1 = &osp_env_info(env)->osi_oi;
- struct ost_id *oi2 = &osp_env_info(env)->osi_oi2;
-
- LASSERT(fid_is_idif(fid1) && fid_is_idif(fid2));
- fid_to_ostid(fid1, oi1);
- fid_to_ostid(fid2, oi2);
- LASSERT(ostid_id(oi1) >= ostid_id(oi2));
-
- return ostid_id(oi1) - ostid_id(oi2);
- }
-
- return fid_oid(fid1) - fid_oid(fid2);
+ return osp_fid_diff(&osp->opd_pre_last_created_fid,
+ &osp->opd_pre_used_fid);
}
static inline int osp_precreate_near_empty_nolock(const struct lu_env *env,
th->th_sync |= sync;
rc = dt_declare_record_write(env, osp->opd_last_used_oid_file,
- lb_oid->lb_len, oid_off, th);
+ lb_oid, oid_off, th);
if (rc != 0)
GOTO(out, rc);
rc = dt_declare_record_write(env, osp->opd_last_used_seq_file,
- lb_oseq->lb_len, oseq_off, th);
+ lb_oseq, oseq_off, th);
if (rc != 0)
GOTO(out, rc);
osp_pre_update_status(d, -ENOSPC);
rc = -ENOSPC;
}
- cfs_waitq_signal(&d->opd_pre_waitq);
+ wake_up(&d->opd_pre_waitq);
GOTO(out_req, rc);
}
GOTO(out_req, rc = -EPROTO);
ostid_to_fid(fid, &body->oa.o_oi, d->opd_index);
- LASSERTF(lu_fid_diff(fid, &d->opd_pre_used_fid) > 0,
+ LASSERTF(osp_fid_diff(fid, &d->opd_pre_used_fid) > 0,
"reply fid "DFID" pre used fid "DFID"\n", PFID(fid),
PFID(&d->opd_pre_used_fid));
- diff = lu_fid_diff(fid, &d->opd_pre_last_created_fid);
+ diff = osp_fid_diff(fid, &d->opd_pre_last_created_fid);
spin_lock(&d->opd_pre_lock);
if (diff < grow) {
out_req:
/* now we can wakeup all users awaiting for objects */
osp_pre_update_status(d, rc);
- cfs_waitq_signal(&d->opd_pre_user_waitq);
+ wake_up(&d->opd_pre_user_waitq);
ptlrpc_req_finished(req);
RETURN(rc);
if (req == NULL)
RETURN(-ENOMEM);
- req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY, RCL_CLIENT,
+ req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY, RCL_CLIENT,
sizeof(KEY_LAST_FID));
- req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL, RCL_CLIENT,
- sizeof(struct lu_fid));
-
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
- tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
+ tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
memcpy(tmp, KEY_LAST_FID, sizeof(KEY_LAST_FID));
req->rq_no_delay = req->rq_no_resend = 1;
- tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
- fid_cpu_to_le((struct lu_fid *)tmp, &d->opd_last_used_fid);
+ last_fid = req_capsule_client_get(&req->rq_pill, &RMF_FID);
+ fid_cpu_to_le(last_fid, &d->opd_last_used_fid);
+
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
ostid_to_fid(last_fid, &body->oa.o_oi, d->opd_index);
spin_lock(&d->opd_pre_lock);
- diff = lu_fid_diff(&d->opd_last_used_fid, last_fid);
+ diff = osp_fid_diff(&d->opd_last_used_fid, last_fid);
if (diff > 0) {
d->opd_pre_grow_count = OST_MIN_PRECREATE + diff;
d->opd_pre_last_created_fid = d->opd_last_used_fid;
if (req)
ptlrpc_req_finished(req);
+ spin_lock(&d->opd_pre_lock);
d->opd_pre_recovering = 0;
+ spin_unlock(&d->opd_pre_lock);
/*
* If rc is zero, the pre-creation window should have been emptied.
* this OSP isn't quite functional yet */
osp_pre_update_status(d, rc);
} else {
- cfs_waitq_signal(&d->opd_pre_user_waitq);
+ wake_up(&d->opd_pre_user_waitq);
}
}
d->opd_syn_changes, d->opd_syn_rpc_in_progress);
} else if (old == -ENOSPC) {
d->opd_pre_status = 0;
+ spin_lock(&d->opd_pre_lock);
d->opd_pre_grow_slow = 0;
d->opd_pre_grow_count = OST_MIN_PRECREATE;
- cfs_waitq_signal(&d->opd_pre_waitq);
+ spin_unlock(&d->opd_pre_lock);
+ wake_up(&d->opd_pre_waitq);
CDEBUG(D_INFO, "%s: no space: "LPU64" blocks, "LPU64
" free, "LPU64" used, "LPU64" avail -> %d: "
"rc = %d\n", d->opd_obd->obd_name,
}
out:
- cfs_waitq_signal(&d->opd_pre_user_waitq);
+ wake_up(&d->opd_pre_user_waitq);
}
-static int osp_init_pre_fid(struct osp_device *osp)
+int osp_init_pre_fid(struct osp_device *osp)
{
struct lu_env env;
struct osp_thread_info *osi;
int rc;
ENTRY;
+ LASSERT(osp->opd_pre != NULL);
+
/* Return if last_used fid has been initialized */
if (!fid_is_zero(&osp->opd_last_used_fid))
RETURN(0);
struct osp_device *d = _arg;
struct ptlrpc_thread *thread = &d->opd_pre_thread;
struct l_wait_info lwi = { 0 };
- char pname[16];
struct lu_env env;
int rc;
ENTRY;
- sprintf(pname, "osp-pre-%u", d->opd_index);
- cfs_daemonize(pname);
-
rc = lu_env_init(&env, d->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
if (rc) {
CERROR("%s: init env error: rc = %d\n", d->opd_obd->obd_name,
spin_lock(&d->opd_pre_lock);
thread->t_flags = SVC_RUNNING;
spin_unlock(&d->opd_pre_lock);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
while (osp_precreate_running(d)) {
/*
break;
LASSERT(d->opd_obd->u.cli.cl_seq != NULL);
- if (d->opd_obd->u.cli.cl_seq->lcs_exp == NULL) {
- /* Get new sequence for client first */
- LASSERT(d->opd_exp != NULL);
- d->opd_obd->u.cli.cl_seq->lcs_exp =
- class_export_get(d->opd_exp);
- rc = osp_init_pre_fid(d);
- if (rc != 0) {
- class_export_put(d->opd_exp);
- d->opd_obd->u.cli.cl_seq->lcs_exp = NULL;
- CERROR("%s: init pre fid error: rc = %d\n",
- d->opd_obd->obd_name, rc);
- continue;
- }
+ /* Sigh, fid client is not ready yet */
+ if (d->opd_obd->u.cli.cl_seq->lcs_exp == NULL)
+ continue;
+
+ /* Init fid for osp_precreate if necessary */
+ rc = osp_init_pre_fid(d);
+ if (rc != 0) {
+ class_export_put(d->opd_exp);
+ d->opd_obd->u.cli.cl_seq->lcs_exp = NULL;
+ CERROR("%s: init pre fid error: rc = %d\n",
+ d->opd_obd->obd_name, rc);
+ continue;
}
osp_statfs_update(d);
thread->t_flags = SVC_STOPPED;
lu_env_fini(&env);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
RETURN(0);
}
if (d->opd_pre_reserved + 1 < osp_objs_precreated(env, d))
return 1;
- /* ready if OST reported no space and no destoys in progress */
+ /* ready if OST reported no space and no destroys in progress */
if (d->opd_syn_changes + d->opd_syn_rpc_in_progress == 0 &&
d->opd_pre_status == -ENOSPC)
return 1;
+ /* Bail out I/O fails to OST */
+ if (d->opd_pre_status != 0 &&
+ d->opd_pre_status != -EAGAIN &&
+ d->opd_pre_status != -ENODEV &&
+ d->opd_pre_status != -ENOSPC) {
+ /* DEBUG LU-3230 */
+ if (d->opd_pre_status != -EIO)
+ CERROR("%s: precreate failed opd_pre_status %d\n",
+ d->opd_obd->obd_name, d->opd_pre_status);
+ return 1;
+ }
+
return 0;
}
struct l_wait_info lwi;
cfs_time_t expire = cfs_time_shift(obd_timeout);
int precreated, rc;
- int count = 0;
ENTRY;
* - OST can allocate fid sequence.
*/
while ((rc = d->opd_pre_status) == 0 || rc == -ENOSPC ||
- rc == -ENODEV || rc == -EAGAIN) {
-
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
- /*
- * to address Andreas's concern on possible busy-loop
- * between this thread and osp_precreate_send()
- */
- if (unlikely(count++ == 1000)) {
- osp_precreate_timeout_condition(d);
- LBUG();
- }
-#endif
+ rc == -ENODEV || rc == -EAGAIN || rc == -ENOTCONN) {
/*
* increase number of precreations
/* XXX: don't wake up if precreation is in progress */
if (osp_precreate_near_empty_nolock(env, d) &&
!osp_precreate_end_seq_nolock(env, d))
- cfs_waitq_signal(&d->opd_pre_waitq);
+ wake_up(&d->opd_pre_waitq);
break;
}
}
/* XXX: don't wake up if precreation is in progress */
- cfs_waitq_signal(&d->opd_pre_waitq);
+ wake_up(&d->opd_pre_waitq);
lwi = LWI_TIMEOUT(expire - cfs_time_current(),
osp_precreate_timeout_condition, d);
/* grab next id from the pool */
spin_lock(&d->opd_pre_lock);
- LASSERTF(lu_fid_diff(&d->opd_pre_used_fid,
+ LASSERTF(osp_fid_diff(&d->opd_pre_used_fid,
&d->opd_pre_last_created_fid) < 0,
"next fid "DFID" last created fid "DFID"\n",
PFID(&d->opd_pre_used_fid),
* osp_precreate_thread() just before orphan cleanup
*/
if (unlikely(d->opd_pre_reserved == 0 && d->opd_pre_status))
- cfs_waitq_signal(&d->opd_pre_waitq);
+ wake_up(&d->opd_pre_waitq);
return 0;
}
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body);
- lustre_set_wire_obdo(&body->oa, oa);
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
/* XXX: capa support? */
/* osc_pack_capa(req, body, capa); */
int osp_init_precreate(struct osp_device *d)
{
struct l_wait_info lwi = { 0 };
- int rc;
+ struct task_struct *task;
ENTRY;
+ OBD_ALLOC_PTR(d->opd_pre);
+ if (d->opd_pre == NULL)
+ RETURN(-ENOMEM);
+
/* initially precreation isn't ready */
d->opd_pre_status = -EAGAIN;
fid_zero(&d->opd_pre_used_fid);
d->opd_pre_max_grow_count = OST_MAX_PRECREATE;
spin_lock_init(&d->opd_pre_lock);
- cfs_waitq_init(&d->opd_pre_waitq);
- cfs_waitq_init(&d->opd_pre_user_waitq);
- cfs_waitq_init(&d->opd_pre_thread.t_ctl_waitq);
+ init_waitqueue_head(&d->opd_pre_waitq);
+ init_waitqueue_head(&d->opd_pre_user_waitq);
+ init_waitqueue_head(&d->opd_pre_thread.t_ctl_waitq);
/*
* Initialize statfs-related things
/*
* start thread handling precreation and statfs updates
*/
- rc = cfs_create_thread(osp_precreate_thread, d, 0);
- if (rc < 0) {
- CERROR("can't start precreate thread %d\n", rc);
- RETURN(rc);
+ task = kthread_run(osp_precreate_thread, d,
+ "osp-pre-%u-%u", d->opd_index, d->opd_group);
+ if (IS_ERR(task)) {
+ CERROR("can't start precreate thread %ld\n", PTR_ERR(task));
+ RETURN(PTR_ERR(task));
}
l_wait_event(d->opd_pre_thread.t_ctl_waitq,
void osp_precreate_fini(struct osp_device *d)
{
- struct ptlrpc_thread *thread = &d->opd_pre_thread;
+ struct ptlrpc_thread *thread;
ENTRY;
cfs_timer_disarm(&d->opd_statfs_timer);
+ if (d->opd_pre == NULL)
+ RETURN_EXIT;
+
+ thread = &d->opd_pre_thread;
+
thread->t_flags = SVC_STOPPING;
- cfs_waitq_signal(&d->opd_pre_waitq);
+ wake_up(&d->opd_pre_waitq);
+
+ wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
- cfs_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
+ OBD_FREE_PTR(d->opd_pre);
+ d->opd_pre = NULL;
EXIT;
}