Whamcloud - gitweb
LU-3230 osp: unstick precreates on unmount
[fs/lustre-release.git] / lustre / osp / osp_precreate.c
index 0a0f9f0..4bd43d3 100644 (file)
@@ -27,7 +27,7 @@
  * 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/
@@ -42,9 +42,6 @@
  * Author: Di Wang <di.wang@intel.com>
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include "osp_internal.h"
@@ -73,7 +70,7 @@ 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);
+       wake_up(&d->opd_pre_waitq);
 }
 
 static int osp_statfs_interpret(const struct lu_env *env,
@@ -111,7 +108,7 @@ 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);
+       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);
@@ -177,7 +174,7 @@ void osp_statfs_need_now(struct osp_device *d)
                 */
                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);
        }
 }
 
@@ -212,11 +209,11 @@ static inline int osp_objs_precreated(const struct lu_env *env,
                struct ost_id *oi2 = &osp_env_info(env)->osi_oi2;
 
                LASSERT(fid_is_idif(fid1) && fid_is_idif(fid2));
-               ostid_idif_pack(fid1, oi1);
-               ostid_idif_pack(fid2, oi2);
-               LASSERT(oi1->oi_id >= oi2->oi_id);
+               fid_to_ostid(fid1, oi1);
+               fid_to_ostid(fid2, oi2);
+               LASSERT(ostid_id(oi1) >= ostid_id(oi2));
 
-               return oi1->oi_id - oi2->oi_id;
+               return ostid_id(oi1) - ostid_id(oi2);
        }
 
        return fid_oid(fid1) - fid_oid(fid2);
@@ -381,16 +378,16 @@ static int osp_precreate_fids(const struct lu_env *env, struct osp_device *osp,
 
                spin_lock(&osp->opd_pre_lock);
                last_fid = &osp->opd_pre_last_created_fid;
-               ostid_idif_pack(last_fid, oi);
-               end = min(oi->oi_id + *grow, IDIF_MAX_OID);
-               *grow = end - oi->oi_id;
-               oi->oi_id += *grow;
+               fid_to_ostid(last_fid, oi);
+               end = min(ostid_id(oi) + *grow, IDIF_MAX_OID);
+               *grow = end - ostid_id(oi);
+               ostid_set_id(oi, ostid_id(oi) + *grow);
                spin_unlock(&osp->opd_pre_lock);
 
                if (*grow == 0)
                        return 1;
 
-               ostid_idif_unpack(oi, fid, osp->opd_index);
+               ostid_to_fid(fid, oi, osp->opd_index);
                return 0;
        }
 
@@ -454,6 +451,7 @@ static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
        LASSERT(body);
 
+       *fid = d->opd_pre_last_created_fid;
        rc = osp_precreate_fids(env, d, fid, &grow);
        if (rc == 1) {
                /* Current seq has been used up*/
@@ -461,7 +459,7 @@ static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
                        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);
        }
 
@@ -472,7 +470,7 @@ static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
                fid->f_seq = 0;
        }
 
-       ostid_fid_pack(fid, &body->oa.o_oi);
+       fid_to_ostid(fid, &body->oa.o_oi);
        body->oa.o_valid = OBD_MD_FLGROUP;
 
        ptlrpc_request_set_replen(req);
@@ -489,14 +487,11 @@ static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
        if (body == NULL)
                GOTO(out_req, rc = -EPROTO);
 
-       fid_ostid_unpack(fid, &body->oa.o_oi, d->opd_index);
+       ostid_to_fid(fid, &body->oa.o_oi, d->opd_index);
        LASSERTF(lu_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));
 
-       CDEBUG(D_HA, "%s: new last_created "DFID"\n", d->opd_obd->obd_name,
-              PFID(fid));
-
        diff = lu_fid_diff(fid, &d->opd_pre_last_created_fid);
 
        spin_lock(&d->opd_pre_lock);
@@ -515,22 +510,24 @@ static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
        d->opd_pre_last_created_fid = *fid;
        spin_unlock(&d->opd_pre_lock);
 
-       CDEBUG(D_OTHER, "current precreated pool: "DFID"-"DFID"\n",
-              PFID(&d->opd_pre_used_fid), PFID(&d->opd_pre_last_created_fid));
+       CDEBUG(D_HA, "%s: current precreated pool: "DFID"-"DFID"\n",
+              d->opd_obd->obd_name, PFID(&d->opd_pre_used_fid),
+              PFID(&d->opd_pre_last_created_fid));
 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);
 }
 
-static int osp_get_lastfid_from_ost(struct osp_device *d)
+static int osp_get_lastfid_from_ost(const struct lu_env *env,
+                                   struct osp_device *d)
 {
        struct ptlrpc_request   *req = NULL;
        struct obd_import       *imp;
-       struct lu_fid           *last_fid = &d->opd_last_used_fid;
+       struct lu_fid           *last_fid;
        char                    *tmp;
        int                     rc;
        ENTRY;
@@ -542,11 +539,8 @@ static int osp_get_lastfid_from_ost(struct osp_device *d)
        if (req == NULL)
                RETURN(-ENOMEM);
 
-       req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
-                            RCL_CLIENT, sizeof(KEY_LAST_FID));
-
-       req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
-                            RCL_CLIENT, sizeof(*last_fid));
+       req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY, RCL_CLIENT,
+                            sizeof(KEY_LAST_FID));
 
        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
        if (rc) {
@@ -554,13 +548,13 @@ static int osp_get_lastfid_from_ost(struct osp_device *d)
                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;
-       fid_cpu_to_le(last_fid, last_fid);
-       tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
-       memcpy(tmp, last_fid, sizeof(*last_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);
@@ -575,7 +569,12 @@ static int osp_get_lastfid_from_ost(struct osp_device *d)
        }
 
        last_fid = req_capsule_server_get(&req->rq_pill, &RMF_FID);
-       if (last_fid == NULL || !fid_is_sane(last_fid)) {
+       if (last_fid == NULL) {
+               CERROR("%s: Got last_fid failed.\n", d->opd_obd->obd_name);
+               GOTO(out, rc = -EPROTO);
+       }
+
+       if (!fid_is_sane(last_fid)) {
                CERROR("%s: Got insane last_fid "DFID"\n",
                       d->opd_obd->obd_name, PFID(last_fid));
                GOTO(out, rc = -EPROTO);
@@ -586,7 +585,7 @@ static int osp_get_lastfid_from_ost(struct osp_device *d)
        if (fid_oid(last_fid) > 0)
                d->opd_last_used_fid = *last_fid;
 
-       CDEBUG(D_HA, "%s: Got insane last_fid "DFID"\n", d->opd_obd->obd_name,
+       CDEBUG(D_HA, "%s: Got last_fid "DFID"\n", d->opd_obd->obd_name,
               PFID(last_fid));
 
 out:
@@ -646,7 +645,7 @@ static int osp_precreate_cleanup_orphans(struct lu_env *env,
        LASSERT(!fid_is_zero(last_fid));
        if (fid_oid(&d->opd_last_used_fid) < 2) {
                /* lastfid looks strange... ask OST */
-               rc = osp_get_lastfid_from_ost(d);
+               rc = osp_get_lastfid_from_ost(env, d);
                if (rc)
                        GOTO(out, rc);
        }
@@ -671,12 +670,8 @@ static int osp_precreate_cleanup_orphans(struct lu_env *env,
 
        body->oa.o_flags = OBD_FL_DELORPHAN;
        body->oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
-       if (osp_is_fid_client(d))
-               body->oa.o_seq = fid_seq(&d->opd_last_used_fid);
-       else
-               body->oa.o_seq = 0;
-       /* remove from NEXT after used one */
-       body->oa.o_id = fid_oid(&d->opd_last_used_fid);
+
+       fid_to_ostid(&d->opd_last_used_fid, &body->oa.o_oi);
 
        ptlrpc_request_set_replen(req);
 
@@ -696,10 +691,7 @@ static int osp_precreate_cleanup_orphans(struct lu_env *env,
        /*
         * OST provides us with id new pool starts from in body->oa.o_id
         */
-       fid_ostid_unpack(last_fid, &body->oa.o_oi, d->opd_index);
-       CDEBUG(D_INFO, "%s: last_fid "DFID" server last fid "DFID"\n",
-              d->opd_obd->obd_name, PFID(&d->opd_last_used_fid),
-              PFID(last_fid));
+       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);
@@ -727,7 +719,9 @@ out:
        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.
@@ -745,7 +739,7 @@ out:
                         * 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);
                }
        }
 
@@ -801,9 +795,11 @@ void osp_pre_update_status(struct osp_device *d, int rc)
                               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,
@@ -813,7 +809,7 @@ void osp_pre_update_status(struct osp_device *d, int rc)
        }
 
 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)
@@ -840,7 +836,7 @@ static int osp_init_pre_fid(struct osp_device *osp)
        last_fid = &osi->osi_fid;
        fid_zero(last_fid);
        /* For a freshed fs, it will allocate a new sequence first */
-       if (osp_is_fid_client(osp)) {
+       if (osp_is_fid_client(osp) && osp->opd_group != 0) {
                cli_seq = osp->opd_obd->u.cli.cl_seq;
                rc = seq_client_get_seq(&env, cli_seq, &last_fid->f_seq);
                if (rc != 0) {
@@ -849,7 +845,7 @@ static int osp_init_pre_fid(struct osp_device *osp)
                        GOTO(out, rc);
                }
        } else {
-               last_fid->f_seq = fid_idif_seq(1, osp->opd_index);
+               last_fid->f_seq = fid_idif_seq(0, osp->opd_index);
        }
        last_fid->f_oid = 1;
        last_fid->f_ver = 0;
@@ -875,15 +871,11 @@ static int osp_precreate_thread(void *_arg)
        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\n", 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,
@@ -894,7 +886,7 @@ static int osp_precreate_thread(void *_arg)
        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)) {
                /*
@@ -906,9 +898,6 @@ static int osp_precreate_thread(void *_arg)
                                     d->opd_new_connection,
                                     &lwi);
 
-                       if (!osp_precreate_running(d))
-                               break;
-
                        if (!d->opd_new_connection)
                                continue;
 
@@ -917,6 +906,10 @@ static int osp_precreate_thread(void *_arg)
                        break;
                }
 
+               if (!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);
@@ -995,7 +988,7 @@ static int osp_precreate_thread(void *_arg)
 
        thread->t_flags = SVC_STOPPED;
        lu_env_fini(&env);
-       cfs_waitq_signal(&thread->t_ctl_waitq);
+       wake_up(&thread->t_ctl_waitq);
 
        RETURN(0);
 }
@@ -1011,11 +1004,23 @@ static int osp_precreate_ready_condition(const struct lu_env *env,
        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;
 }
 
@@ -1046,7 +1051,6 @@ int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d)
        struct l_wait_info       lwi;
        cfs_time_t               expire = cfs_time_shift(obd_timeout);
        int                      precreated, rc;
-       int                      count = 0;
 
        ENTRY;
 
@@ -1064,17 +1068,6 @@ int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d)
        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
-
                /*
                 * increase number of precreations
                 */
@@ -1099,7 +1092,7 @@ int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d)
                        /* 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;
                }
@@ -1128,12 +1121,14 @@ int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d)
                }
 
                /* 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);
-               if (cfs_time_aftereq(cfs_time_current(), expire))
+               if (cfs_time_aftereq(cfs_time_current(), expire)) {
+                       rc = -ETIMEDOUT;
                        break;
+               }
 
                l_wait_event(d->opd_pre_user_waitq,
                             osp_precreate_ready_condition(env, d), &lwi);
@@ -1173,7 +1168,7 @@ int osp_precreate_get_fid(const struct lu_env *env, struct osp_device *d,
         * 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;
 }
@@ -1222,7 +1217,7 @@ int osp_object_truncate(const struct lu_env *env, struct dt_object *dt,
        if (oa == NULL)
                GOTO(out, rc = -ENOMEM);
 
-       rc = fid_ostid_pack(lu_object_fid(&dt->do_lu), &oa->o_oi);
+       rc = fid_to_ostid(lu_object_fid(&dt->do_lu), &oa->o_oi);
        LASSERT(rc == 0);
        oa->o_size = size;
        oa->o_blocks = OBD_OBJECT_EOF;
@@ -1231,7 +1226,7 @@ int osp_object_truncate(const struct lu_env *env, struct dt_object *dt,
 
        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); */
@@ -1251,7 +1246,7 @@ out:
 int osp_init_precreate(struct osp_device *d)
 {
        struct l_wait_info       lwi = { 0 };
-       int                      rc;
+       struct task_struct              *task;
 
        ENTRY;
 
@@ -1269,9 +1264,9 @@ int osp_init_precreate(struct osp_device *d)
        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
@@ -1286,10 +1281,11 @@ int osp_init_precreate(struct osp_device *d)
        /*
         * 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", d->opd_index);
+       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,
@@ -1308,9 +1304,9 @@ void osp_precreate_fini(struct osp_device *d)
        cfs_timer_disarm(&d->opd_statfs_timer);
 
        thread->t_flags = SVC_STOPPING;
-       cfs_waitq_signal(&d->opd_pre_waitq);
+       wake_up(&d->opd_pre_waitq);
 
-       cfs_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
+       wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
 
        EXIT;
 }