X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosp%2Fosp_precreate.c;h=574cd0dc9427b23d542c7a2a17c02430755fc412;hb=de73bdafb0e0ab187865dcdcc76c0f48e1b5239e;hp=1836f18de9bffc64591855e82221dca91587c154;hpb=707642fa4062a273ef5352b7cc44e43e18681dc0;p=fs%2Flustre-release.git diff --git a/lustre/osp/osp_precreate.c b/lustre/osp/osp_precreate.c index 1836f18..574cd0d 100644 --- a/lustre/osp/osp_precreate.c +++ b/lustre/osp/osp_precreate.c @@ -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 */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #include "osp_internal.h" @@ -68,12 +65,29 @@ static inline int osp_statfs_need_update(struct osp_device *d) 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, @@ -111,7 +125,9 @@ 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); @@ -177,27 +193,10 @@ 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); } } - -/* - * 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) { @@ -212,11 +211,11 @@ 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 +380,16 @@ 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; } @@ -462,7 +461,7 @@ 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); } @@ -473,7 +472,7 @@ static int osp_precreate_send(const struct lu_env *env, struct osp_device *d) fid->f_seq = 0; } - fid_ostid_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); @@ -490,14 +489,11 @@ 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); @@ -516,12 +512,13 @@ 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); @@ -530,10 +527,9 @@ out_req: static int osp_get_lastfid_from_ost(const struct lu_env *env, struct osp_device *d) { - struct ost_id *oi = &osp_env_info(env)->osi_oi; 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; @@ -545,26 +541,22 @@ static int osp_get_lastfid_from_ost(const struct lu_env *env, 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(*oi)); - 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; - fid_ostid_pack(last_fid, oi); - ostid_cpu_to_le(oi, oi); - tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL); - memcpy(tmp, oi, sizeof(*oi)); + 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); @@ -578,14 +570,13 @@ static int osp_get_lastfid_from_ost(const struct lu_env *env, GOTO(out, rc); } - oi = req_capsule_server_get(&req->rq_pill, &RMF_OST_ID); - if (oi == NULL) { + last_fid = req_capsule_server_get(&req->rq_pill, &RMF_FID); + if (last_fid == NULL) { CERROR("%s: Got last_fid failed.\n", d->opd_obd->obd_name); GOTO(out, rc = -EPROTO); } - rc = fid_ostid_unpack(last_fid, oi, d->opd_index); - if (rc != 0 || !fid_is_sane(last_fid)) { + 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); @@ -682,7 +673,7 @@ 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; - fid_ostid_pack(&d->opd_last_used_fid, &body->oa.o_oi); + fid_to_ostid(&d->opd_last_used_fid, &body->oa.o_oi); ptlrpc_request_set_replen(req); @@ -702,10 +693,7 @@ 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); @@ -733,7 +721,9 @@ 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. @@ -751,7 +741,7 @@ 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); } } @@ -807,9 +797,11 @@ 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, @@ -819,10 +811,10 @@ 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) +int osp_init_pre_fid(struct osp_device *osp) { struct lu_env env; struct osp_thread_info *osi; @@ -831,6 +823,8 @@ static int osp_init_pre_fid(struct osp_device *osp) 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); @@ -881,15 +875,11 @@ 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", 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, @@ -900,7 +890,7 @@ 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)) { /* @@ -924,19 +914,18 @@ static int osp_precreate_thread(void *_arg) 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); @@ -1002,7 +991,7 @@ 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); } @@ -1018,11 +1007,23 @@ 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; } @@ -1053,7 +1054,6 @@ 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; @@ -1069,18 +1069,7 @@ int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d) * - 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 @@ -1106,7 +1095,7 @@ 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; } @@ -1135,12 +1124,14 @@ 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); @@ -1180,7 +1171,7 @@ 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; } @@ -1229,7 +1220,7 @@ 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; @@ -1238,7 +1229,7 @@ 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); */ @@ -1258,10 +1249,14 @@ out: 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); @@ -1276,9 +1271,9 @@ 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 @@ -1293,10 +1288,11 @@ 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, @@ -1308,16 +1304,24 @@ int osp_init_precreate(struct osp_device *d) 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; }