X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_dev.c;h=0c43df0b0ad9eca777a3e84d18cd5189cb68f605;hb=a21ce928aa641cbfacb9dde74e4be1cd1b658fb0;hp=8d585528f7ed371a946ad3d84904c721ae4601ce;hpb=aa82cc83612dbd4c7d05fc101b98e8660f1373db;p=fs%2Flustre-release.git diff --git a/lustre/ofd/ofd_dev.c b/lustre/ofd/ofd_dev.c index 8d58552..0c43df0 100644 --- a/lustre/ofd/ofd_dev.c +++ b/lustre/ofd/ofd_dev.c @@ -27,7 +27,6 @@ */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/ofd/ofd_dev.c * @@ -77,6 +76,7 @@ #include #include #include +#include #include "ofd_internal.h" @@ -161,7 +161,8 @@ out: * \retval negative value on error */ static int ofd_stack_init(const struct lu_env *env, - struct ofd_device *m, struct lustre_cfg *cfg) + struct ofd_device *m, struct lustre_cfg *cfg, + u32 *lmd_flags) { const char *dev = lustre_cfg_string(cfg, 0); struct lu_device *d; @@ -180,8 +181,13 @@ static int ofd_stack_init(const struct lu_env *env, } lmd = s2lsi(lmi->lmi_sb)->lsi_lmd; - if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) - m->ofd_skip_lfsck = 1; + if (lmd) { + if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) + m->ofd_skip_lfsck = 1; + if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE) + m->ofd_no_precreate = 1; + *lmd_flags = lmd->lmd_flags; + } /* find bottom osd */ OBD_ALLOC(osdname, MTI_NAME_MAXLEN); @@ -250,11 +256,8 @@ static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m, } lu_site_purge(env, top->ld_site, ~0); - if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) { - LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL); - lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer); - } - + lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems, + D_OTHER, lu_cdebug_printer); LASSERT(m->ofd_osd_exp); obd_disconnect(m->ofd_osd_exp); @@ -449,6 +452,14 @@ static int ofd_object_init(const struct lu_env *env, struct lu_object *o, RETURN(rc); } +static void ofd_object_free_rcu(struct rcu_head *head) +{ + struct ofd_object *of = container_of(head, struct ofd_object, + ofo_header.loh_rcu); + + kmem_cache_free(ofd_object_kmem, of); +} + /** * Implementation of lu_object_operations::loo_object_free. * @@ -470,7 +481,8 @@ static void ofd_object_free(const struct lu_env *env, struct lu_object *o) lu_object_fini(o); lu_object_header_fini(h); - OBD_SLAB_FREE_PTR(of, ofd_object_kmem); + OBD_FREE_PRE(of, sizeof(*of), "slab-freed"); + call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu); EXIT; } @@ -494,7 +506,7 @@ static int ofd_object_print(const struct lu_env *env, void *cookie, return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o); } -static struct lu_object_operations ofd_obj_ops = { +static const struct lu_object_operations ofd_obj_ops = { .loo_object_init = ofd_object_init, .loo_object_free = ofd_object_free, .loo_object_print = ofd_object_print @@ -689,7 +701,7 @@ static int ofd_recovery_complete(const struct lu_env *env, /** * lu_device_operations matrix for OFD device. */ -static struct lu_device_operations ofd_lu_ops = { +static const struct lu_device_operations ofd_lu_ops = { .ldo_object_alloc = ofd_object_alloc, .ldo_process_config = ofd_process_config, .ldo_recovery_complete = ofd_recovery_complete, @@ -827,17 +839,12 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) GOTO(out_server, rc = -ENOMEM); snprintf(name, len, "%s-super", obd_name); - rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, - name, NULL); - if (rc) { - CERROR("%s: seq client init error: rc = %d\n", obd_name, rc); - GOTO(out_client, rc); - } + seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, + name, NULL); rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq); if (rc) { -out_client: seq_client_fini(ss->ss_client_seq); OBD_FREE_PTR(ss->ss_client_seq); ss->ss_client_seq = NULL; @@ -869,6 +876,7 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) void *key, *val = NULL; int keylen, vallen, rc = 0; bool is_grant_shrink; + ktime_t kstart = ktime_get(); ENTRY; @@ -901,6 +909,19 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) if (is_grant_shrink) { body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY); + /* + * Because we already sync grant info with client when + * reconnect, grant info will be cleared for resent + * req, otherwise, outdated grant count in the rpc + * would de-sync grant counters + */ + if (lustre_msg_get_flags(req->rq_reqmsg) & + (MSG_RESENT | MSG_REPLAY)) { + DEBUG_REQ(D_CACHE, req, + "clear resent/replay req grant info"); + body->oa.o_valid &= ~OBD_MD_FLGRANT; + } + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); *repbody = *body; @@ -917,7 +938,7 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) rc = -EOPNOTSUPP; } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -977,7 +998,7 @@ static int ofd_lock_unlock_region(const struct lu_env *env, return rc; CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh); - tgt_extent_unlock(&lh, LCK_PR); + tgt_data_unlock(&lh, LCK_PR); return 0; } @@ -1063,6 +1084,7 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) void *key; int keylen; int replylen, rc = 0; + ktime_t kstart = ktime_get(); ENTRY; @@ -1147,6 +1169,9 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) if (rc) RETURN(err_serious(rc)); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_GET_LAST_FID)) + RETURN(-EAGAIN); + fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID); if (fid == NULL) RETURN(err_serious(-EPROTO)); @@ -1160,12 +1185,12 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) oseq = ofd_seq_load(tsi->tsi_env, ofd, ostid_seq(&fti->fti_ostid)); if (IS_ERR(oseq)) - RETURN(PTR_ERR(oseq)); + RETURN(-EFAULT); rc = ostid_to_fid(fid, &oseq->os_oi, ofd->ofd_lut.lut_lsd.lsd_osd_index); if (rc != 0) - GOTO(out_put, rc); + GOTO(out_put, rc = -EFAULT); CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd), PFID(fid)); @@ -1177,7 +1202,7 @@ out_put: rc = -EOPNOTSUPP; } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1202,6 +1227,7 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) struct ofd_object *fo; __u64 flags = 0; enum ldlm_mode lock_mode = LCK_PR; + ktime_t kstart = ktime_get(); bool srvlock; int rc; ENTRY; @@ -1263,10 +1289,10 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) ofd_object_put(tsi->tsi_env, fo); out: if (srvlock) - tgt_extent_unlock(&lh, lock_mode); + tgt_data_unlock(&lh, lock_mode); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); repbody->oa.o_valid |= OBD_MD_FLFLAGS; repbody->oa.o_flags = OBD_FL_FLUSH; @@ -1293,6 +1319,7 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi) struct ost_body *repbody; struct ldlm_resource *res; struct ofd_object *fo; + ktime_t kstart = ktime_get(); int rc = 0; ENTRY; @@ -1341,7 +1368,7 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi) OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); EXIT; out_put: ofd_object_put(tsi->tsi_env, fo); @@ -1478,16 +1505,20 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) u64 seq = ostid_seq(&oa->o_oi); u64 oid = ostid_id(&oa->o_oi); struct ofd_seq *oseq; - s64 diff; - int rc = 0; int sync_trans = 0; long granted = 0; + ktime_t kstart = ktime_get(); + s64 diff; + int rc = 0; ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) RETURN(-EROFS); + if (ofd->ofd_no_precreate) + return -EPERM; + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); if (repbody == NULL) RETURN(-ENOMEM); @@ -1607,7 +1638,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) GOTO(out, rc = -EINVAL); } - if (diff < 0) { + if (diff <= -OST_MAX_PRECREATE) { /* LU-5648 */ CERROR("%s: invalid precreate request for " DOSTID", last_id %llu. " @@ -1615,6 +1646,15 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) ofd_name(ofd), POSTID(&oa->o_oi), ofd_seq_last_oid(oseq)); GOTO(out, rc = -EINVAL); + } else if (diff < 0) { + LCONSOLE(D_INFO, + "%s: MDS LAST_ID "DFID" (%llu) is %lld behind OST LAST_ID "DFID" (%llu), trust the OST\n", + ofd_name(ofd), PFID(&oa->o_oi.oi_fid), + oid, -diff, PFID(&oseq->os_oi.oi_fid), + ofd_seq_last_oid(oseq)); + /* Let MDS know that we are so far ahead. */ + rc = ostid_set_id(&rep_oa->o_oi, + ofd_seq_last_oid(oseq) + 1); } } } @@ -1645,24 +1685,14 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) * in place of an old one at the same index. Instead of * precreating potentially millions of deleted old objects * (possibly filling the OST), only precreate the last batch. - * LFSCK will eventually clean up any orphans. LU-14 - * 500000 is a theoretical maximum number of objects that could - * be created(but not stored on disk in case of failover) - * during default journal commit interval - 5 seconds. */ - if (diff > 500000) { + * LFSCK will eventually clean up any orphans. LU-14 */ + if (diff > 5 * OST_MAX_PRECREATE) { + /* Message below is checked in conf-sanity test_122b */ + LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %lld higher than LAST_ID "DOSTID", only precreating the last %u objects. OST replaced or reformatted?\n", + ofd_name(ofd), POSTID(&oa->o_oi), diff, + POSTID(&oseq->os_oi), + OST_MAX_PRECREATE / 2); diff = OST_MAX_PRECREATE / 2; - LCONSOLE_WARN("%s: Too many FIDs to precreate " - "precreate FID"DOSTID" LAST_ID"DOSTID - "OST replaced or reformatted: " - "LFSCK will clean up\n", - ofd_name(ofd), POSTID(&oa->o_oi), - POSTID(&oseq->os_oi)); - - CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over 500000" - "larger than the LAST_ID "DOSTID", only " - "precreating the last %lld objects.\n", - ofd_name(ofd), POSTID(&oa->o_oi), - POSTID(&oseq->os_oi), diff); ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff); } @@ -1720,24 +1750,14 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) } EXIT; ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (unlikely(!oseq->os_last_id_synced)) oseq->os_last_id_synced = 1; out: mutex_unlock(&oseq->os_create_lock); out_nolock: - if (rc == 0) { -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0) - struct ofd_thread_info *info = ofd_info(tsi->tsi_env); - struct lu_fid *fid = &info->fti_fid; - - /* For compatible purpose, it needs to convert back to - * OST ID before put it on wire. */ - *fid = rep_oa->o_oi.oi_fid; - fid_to_ostid(fid, &rep_oa->o_oi); -#endif + if (rc == 0) rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP; - } ofd_seq_put(tsi->tsi_env, oseq); out_sem: @@ -1763,6 +1783,7 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi) struct ofd_device *ofd = ofd_exp(tsi->tsi_exp); struct ofd_thread_info *fti = tsi2ofd_info(tsi); struct lu_fid *fid = &fti->fti_fid; + ktime_t kstart = ktime_get(); u64 oid; u32 count; int rc = 0; @@ -1824,7 +1845,7 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi) } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); GOTO(out, rc); @@ -1846,6 +1867,7 @@ out: */ static int ofd_statfs_hdl(struct tgt_session_info *tsi) { + ktime_t kstart = ktime_get(); struct obd_statfs *osfs; int rc; @@ -1865,7 +1887,7 @@ static int ofd_statfs_hdl(struct tgt_session_info *tsi) rc = -EINPROGRESS; ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1888,6 +1910,7 @@ static int ofd_sync_hdl(struct tgt_session_info *tsi) struct ofd_thread_info *fti = tsi2ofd_info(tsi); struct ofd_device *ofd = ofd_exp(tsi->tsi_exp); struct ofd_object *fo = NULL; + ktime_t kstart = ktime_get(); int rc = 0; ENTRY; @@ -1908,7 +1931,7 @@ static int ofd_sync_hdl(struct tgt_session_info *tsi) GOTO(put, rc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (fo == NULL) RETURN(0); @@ -1930,6 +1953,144 @@ put: } /** + * OFD request handler for OST_FALLOCATE RPC. + * + * This is part of request processing. Validate request fields, + * preallocate the given OFD object and pack reply. + * + * \param[in] tsi target session environment for this request + * + * \retval 0 if successful + * \retval negative value on error + */ +static int ofd_fallocate_hdl(struct tgt_session_info *tsi) +{ + struct obdo *oa = &tsi->tsi_ost_body->oa; + struct ost_body *repbody; + struct ofd_thread_info *info = tsi2ofd_info(tsi); + struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace; + struct ldlm_resource *res; + struct ofd_object *fo; + __u64 flags = 0; + __u64 valid; + struct lustre_handle lh = { 0, }; + int rc, mode; + __u64 start, end; + bool srvlock; + ktime_t kstart = ktime_get(); + + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); + if (repbody == NULL) + RETURN(err_serious(-ENOMEM)); + + /* + * fallocate() start and end are passed in o_size and o_blocks + * on the wire. Clients 2.15.0 and newer should always set + * the OBD_MD_FLSIZE and OBD_MD_FLBLOCKS valid flags, but some + * older client versions did not. We permit older clients to + * not set these flags, checking their version by proxy using + * the lack of OBD_CONNECT_TRUNCLOCK to imply 2.14.0 and older. + * + * Return -EOPNOTSUPP to also work with older clients not + * supporting newer server modes. + */ + if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != + (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS) +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 21, 53, 0) + && (tgt_conn_flags(tsi) & OBD_CONNECT_OLD_FALLOC) +#endif + ) + RETURN(-EOPNOTSUPP); + + start = oa->o_size; + end = oa->o_blocks; + /* client should already limit len >= 0 */ + if (start >= end) + RETURN(-EINVAL); + + mode = oa->o_falloc_mode; + /* + * mode == 0 (which is standard prealloc) and PUNCH is supported + * Rest of mode options are not supported yet. + */ + if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) + RETURN(-EOPNOTSUPP); + + /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag + * Check that and add the missing flag for such invalid call with + * warning. + */ + if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) { + CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n", + tsi->tsi_tgt->lut_obd->obd_name); + mode |= FALLOC_FL_KEEP_SIZE; + } + + repbody->oa.o_oi = oa->o_oi; + repbody->oa.o_valid = OBD_MD_FLID; + + srvlock = oa->o_valid & OBD_MD_FLFLAGS && + oa->o_flags & OBD_FL_SRVLOCK; + + if (srvlock) { + rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, + start, end, &lh, LCK_PW, &flags); + if (rc != 0) + RETURN(rc); + } + + fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp), + &tsi->tsi_fid); + if (IS_ERR(fo)) + GOTO(out, rc = PTR_ERR(fo)); + + valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID | + OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME; + la_from_obdo(&info->fti_attr, oa, valid); + + rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode, + &info->fti_attr, oa); + if (rc) + GOTO(out_put, rc); + + rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr); + if (rc == 0) + obdo_from_la(&repbody->oa, &info->fti_attr, OFD_VALID_FLAGS); + else + rc = 0; + + ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC, + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); + + EXIT; +out_put: + ofd_object_put(tsi->tsi_env, fo); +out: + if (srvlock) + tgt_data_unlock(&lh, LCK_PW); + if (rc == 0) { + res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid, + LDLM_EXTENT, 0); + if (!IS_ERR(res)) { + struct ost_lvb *res_lvb; + + ldlm_res_lvbo_update(res, NULL, 0); + res_lvb = res->lr_lvb_data; + /* Blocks */ + repbody->oa.o_valid |= OBD_MD_FLBLOCKS; + repbody->oa.o_blocks = res_lvb->lvb_blocks; + /* Size */ + repbody->oa.o_valid |= OBD_MD_FLSIZE; + repbody->oa.o_size = res_lvb->lvb_size; + + ldlm_resource_putref(res); + } + } + + RETURN(rc); +} + +/** * OFD request handler for OST_PUNCH RPC. * * This is part of request processing. Validate request fields, @@ -1950,17 +2111,15 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) struct ofd_object *fo; __u64 flags = 0; struct lustre_handle lh = { 0, }; - int rc; __u64 start, end; bool srvlock; + ktime_t kstart = ktime_get(); + int rc; ENTRY; OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val); - /* check that we do support OBD_CONNECT_TRUNCLOCK. */ - CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK); - if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) RETURN(err_serious(-EPROTO)); @@ -2014,13 +2173,13 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) GOTO(out_put, rc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); EXIT; out_put: ofd_object_put(tsi->tsi_env, fo); out: if (srvlock) - tgt_extent_unlock(&lh, LCK_PW); + tgt_data_unlock(&lh, LCK_PW); if (rc == 0) { /* we do not call this before to avoid lu_object_find() in * ->lvbo_update() holding another reference on the object. @@ -2079,7 +2238,8 @@ static int ofd_ladvise_prefetch(const struct lu_env *env, PTLRPC_MAX_BRW_PAGES; rnb.rnb_offset = start_index << PAGE_SHIFT; rnb.rnb_len = nr_local << PAGE_SHIFT; - rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt); + rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, + PTLRPC_MAX_BRW_PAGES, dbt); if (unlikely(rc < 0)) break; nr_local = rc; @@ -2208,7 +2368,7 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) req->rq_status = ofd_ladvise_prefetch(env, fo, tbc->local, start, end, dbt); - tgt_extent_unlock(&lockh, LCK_PR); + tgt_data_unlock(&lockh, LCK_PR); break; case LU_LADVISE_DONTNEED: rc = dt_ladvise(env, dob, ladvise->lla_start, @@ -2239,6 +2399,7 @@ static int ofd_quotactl(struct tgt_session_info *tsi) { struct obd_quotactl *oqctl, *repoqc; struct lu_nodemap *nodemap; + ktime_t kstart = ktime_get(); int id; int rc; @@ -2267,6 +2428,10 @@ static int ofd_quotactl(struct tgt_session_info *tsi) id = nodemap_map_id(nodemap, NODEMAP_GID, NODEMAP_CLIENT_TO_FS, repoqc->qc_id); + else if (oqctl->qc_type == PRJQUOTA) + id = nodemap_map_id(nodemap, NODEMAP_PROJID, + NODEMAP_CLIENT_TO_FS, + repoqc->qc_id); nodemap_putref(nodemap); @@ -2276,7 +2441,7 @@ static int ofd_quotactl(struct tgt_session_info *tsi) rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (repoqc->qc_id != id) swap(repoqc->qc_id, id); @@ -2285,29 +2450,6 @@ static int ofd_quotactl(struct tgt_session_info *tsi) } /** - * Calculate the amount of time for lock prolongation. - * - * This is helper for ofd_prolong_extent_locks() function to get - * the timeout extra time. - * - * \param[in] req current request - * - * \retval amount of time to extend the timeout with - */ -static inline time64_t prolong_timeout(struct ptlrpc_request *req) -{ - struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt; - time64_t req_timeout; - - if (AT_OFF) - return obd_timeout / 2; - - req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec; - return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)), - req_timeout); -} - -/** * Prolong lock timeout for the given extent. * * This function finds all locks related with incoming request and @@ -2483,14 +2625,14 @@ static int ofd_rw_hpreq_check(struct ptlrpc_request *req) rnb += ioo->ioo_bufcnt - 1; pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1; - DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID - " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt), - current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start, - pa.lpa_extent.end); + DEBUG_REQ(D_RPCTRACE, req, + "%s %s: refresh rw locks for "DFID" (%llu->%llu)", + tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid), + pa.lpa_extent.start, pa.lpa_extent.end); ofd_prolong_extent_locks(tsi, &pa); - CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n", + CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n", tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req); if (pa.lpa_blocks_cnt > 0) @@ -2734,6 +2876,8 @@ TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE, TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl), TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl), TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek), }; static struct tgt_opc_slice ofd_common_slice[] = { @@ -2838,6 +2982,7 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, struct lu_fid fid; struct nm_config_file *nodemap_config; struct obd_device_target *obt; + u32 lmd_flags = 0; int rc; ENTRY; @@ -2857,16 +3002,16 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, spin_lock_init(&m->ofd_flags_lock); m->ofd_raid_degraded = 0; - m->ofd_checksum_t10pi_enforce = 0; m->ofd_sync_journal = 0; ofd_slc_set(m); m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT; m->ofd_seq_count = 0; - init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq); INIT_LIST_HEAD(&m->ofd_inconsistency_list); spin_lock_init(&m->ofd_inconsistency_lock); + m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */ + spin_lock_init(&m->ofd_batch_lock); init_rwsem(&m->ofd_lastid_rwsem); @@ -2893,7 +3038,7 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, if (info == NULL) RETURN(-EFAULT); - rc = ofd_stack_init(env, m, cfg); + rc = ofd_stack_init(env, m, cfg, &lmd_flags); if (rc) { CERROR("%s: can't init device stack, rc %d\n", obd->obd_name, rc); @@ -2910,8 +3055,13 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, LDLM_NAMESPACE_SERVER, LDLM_NAMESPACE_GREEDY, LDLM_NS_TYPE_OST); - if (m->ofd_namespace == NULL) - GOTO(err_fini_stack, rc = -ENOMEM); + if (IS_ERR(m->ofd_namespace)) { + rc = PTR_ERR(m->ofd_namespace); + CERROR("%s: unable to create server namespace: rc = %d\n", + obd->obd_name, rc); + m->ofd_namespace = NULL; + GOTO(err_fini_stack, rc); + } /* set obd_namespace for compatibility with old code */ obd->obd_namespace = m->ofd_namespace; ldlm_register_intent(m->ofd_namespace, ofd_intent_policy); @@ -2927,6 +3077,11 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, if (rc) GOTO(err_free_ns, rc); + if (lmd_flags & LMD_FLG_SKIP_LFSCK) + m->ofd_skip_lfsck = 1; + if (lmd_flags & LMD_FLG_LOCAL_RECOV) + m->ofd_lut.lut_local_recovery = 1; + rc = ofd_tunables_init(m); if (rc) GOTO(err_fini_lut, rc); @@ -2934,12 +3089,11 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, tgd->tgd_reserved_pcnt = 0; m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size; - m->ofd_cksum_types_supported = - obd_cksum_types_supported_server(obd->obd_name); m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT; if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks < OFD_PRECREATE_SMALL_FS) m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL; + m->ofd_atime_diff = OFD_DEF_ATIME_DIFF; rc = ofd_fs_setup(env, m, obd); if (rc) @@ -3031,6 +3185,9 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) d->ld_obd->obd_namespace = m->ofd_namespace = NULL; } + ofd_access_log_delete(m->ofd_access_log); + m->ofd_access_log = NULL; + ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev); LASSERT(atomic_read(&d->ld_ref) == 0); @@ -3116,7 +3273,7 @@ static struct lu_device *ofd_device_alloc(const struct lu_env *env, /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */ LU_TYPE_INIT_FINI(ofd, &ofd_thread_key); -static struct lu_device_type_operations ofd_device_type_ops = { +static const struct lu_device_type_operations ofd_device_type_ops = { .ldto_init = ofd_type_init, .ldto_fini = ofd_type_fini, @@ -3146,13 +3303,28 @@ static struct lu_device_type ofd_device_type = { */ static int __init ofd_init(void) { - int rc; + int rc; rc = lu_kmem_init(ofd_caches); if (rc) return rc; - rc = class_register_type(&ofd_obd_ops, NULL, true, NULL, + + rc = ofd_access_log_module_init(); + if (rc) + goto out_caches; + + rc = class_register_type(&ofd_obd_ops, NULL, true, LUSTRE_OST_NAME, &ofd_device_type); + if (rc) + goto out_ofd_access_log; + + return 0; + +out_ofd_access_log: + ofd_access_log_module_exit(); +out_caches: + lu_kmem_fini(ofd_caches); + return rc; } @@ -3164,8 +3336,9 @@ static int __init ofd_init(void) */ static void __exit ofd_exit(void) { - lu_kmem_fini(ofd_caches); class_unregister_type(LUSTRE_OST_NAME); + ofd_access_log_module_exit(); + lu_kmem_fini(ofd_caches); } MODULE_AUTHOR("OpenSFS, Inc. ");