X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_io.c;h=b90bc8a40110bab92022d503de4c10ccfb4acac9;hb=16184a05de0db78bf4a8056525db09fc96c31256;hp=a75e5cd75cfeaeb0681e6d915b574b3f05180a31;hpb=a67ea1c51daad1a2c774bdceeabbfac927d348e6;p=fs%2Flustre-release.git diff --git a/lustre/ofd/ofd_io.c b/lustre/ofd/ofd_io.c index a75e5cd..b90bc8a 100644 --- a/lustre/ofd/ofd_io.c +++ b/lustre/ofd/ofd_io.c @@ -27,7 +27,7 @@ * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2012, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -42,10 +42,12 @@ #include "ofd_internal.h" -static int ofd_preprw_read(const struct lu_env *env, struct ofd_device *ofd, - struct lu_fid *fid, struct lu_attr *la, int niocount, +static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp, + struct ofd_device *ofd, struct lu_fid *fid, + struct lu_attr *la, int niocount, struct niobuf_remote *rnb, int *nr_local, - struct niobuf_local *lnb) + struct niobuf_local *lnb, + struct obd_trans_info *oti) { struct ofd_object *fo; int i, j, rc, tot_bytes = 0; @@ -63,18 +65,20 @@ static int ofd_preprw_read(const struct lu_env *env, struct ofd_device *ofd, GOTO(unlock, rc = -ENOENT); /* parse remote buffers to local buffers and prepare the latter */ + *nr_local = 0; for (i = 0, j = 0; i < niocount; i++) { rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i, lnb + j, 0, ofd_object_capa(env, fo)); - LASSERT(rc > 0); + if (unlikely(rc < 0)) + GOTO(buf_put, rc); LASSERT(rc <= PTLRPC_MAX_BRW_PAGES); /* correct index for local buffers to continue with */ j += rc; + *nr_local += rc; LASSERT(j <= PTLRPC_MAX_BRW_PAGES); tot_bytes += rnb[i].rnb_len; } - *nr_local = j; LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES); rc = dt_attr_get(env, ofd_object_child(fo), la, ofd_object_capa(env, fo)); @@ -86,6 +90,8 @@ static int ofd_preprw_read(const struct lu_env *env, struct ofd_device *ofd, GOTO(buf_put, rc); lprocfs_counter_add(ofd_obd(ofd)->obd_stats, LPROC_OFD_READ_BYTES, tot_bytes); + ofd_counter_incr(exp, LPROC_OFD_STATS_READ, + oti->oti_jobid, tot_bytes); RETURN(0); buf_put: @@ -139,8 +145,8 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, ofd_read_lock(env, fo); if (!ofd_object_exists(fo)) { - CERROR("%s: BRW to missing obj "LPU64"/"LPU64"\n", - exp->exp_obd->obd_name, obj->ioo_id, obj->ioo_seq); + CERROR("%s: BRW to missing obj "DOSTID"\n", + exp->exp_obd->obd_name, POSTID(&obj->ioo_oid)); ofd_read_unlock(env, fo); ofd_object_put(env, fo); GOTO(out, rc = -ENOENT); @@ -154,11 +160,13 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt); /* parse remote buffers to local buffers and prepare the latter */ + *nr_local = 0; for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) { rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i, lnb + j, 1, ofd_object_capa(env, fo)); - LASSERT(rc > 0); + if (unlikely(rc < 0)) + GOTO(err, rc); LASSERT(rc <= PTLRPC_MAX_BRW_PAGES); /* correct index for local buffers to continue with */ for (k = 0; k < rc; k++) { @@ -167,31 +175,37 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, lnb[j+k].lnb_rc = -ENOSPC; if (!(rnb[i].rnb_flags & OBD_BRW_ASYNC)) oti->oti_sync_write = 1; + /* remote client can't break through quota */ + if (exp_connect_rmtclient(exp)) + lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA; } j += rc; + *nr_local += rc; LASSERT(j <= PTLRPC_MAX_BRW_PAGES); tot_bytes += rnb[i].rnb_len; } - *nr_local = j; LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES); - lprocfs_counter_add(ofd_obd(ofd)->obd_stats, - LPROC_OFD_WRITE_BYTES, tot_bytes); rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local); - if (unlikely(rc != 0)) { - dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local); - ofd_read_unlock(env, fo); - /* ofd_grant_prepare_write() was called, so we must commit */ - ofd_grant_commit(env, exp, rc); - } + if (unlikely(rc != 0)) + GOTO(err, rc); - RETURN(rc); + lprocfs_counter_add(ofd_obd(ofd)->obd_stats, + LPROC_OFD_WRITE_BYTES, tot_bytes); + ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, + oti->oti_jobid, tot_bytes); + RETURN(0); +err: + dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local); + ofd_read_unlock(env, fo); + /* ofd_grant_prepare_write() was called, so we must commit */ + ofd_grant_commit(env, exp, rc); out: /* let's still process incoming grant information packed in the oa, * but without enforcing grant since we won't proceed with the write. * Just like a read request actually. */ ofd_grant_prepare_read(env, exp, oa); - RETURN(rc); + return rc; } int ofd_preprw(const struct lu_env* env, int cmd, struct obd_export *exp, @@ -204,38 +218,63 @@ int ofd_preprw(const struct lu_env* env, int cmd, struct obd_export *exp, struct ofd_thread_info *info; int rc = 0; - if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT) && - ofd->ofd_destroys_in_progress == 0) { - /* don't fail lookups for orphan recovery, it causes - * later LBUGs when objects still exist during precreate */ - CDEBUG(D_INFO, "*** obd_fail_loc=%x ***\n",OBD_FAIL_OST_ENOENT); - RETURN(-ENOENT); + if (*nr_local > PTLRPC_MAX_BRW_PAGES) { + CERROR("%s: bulk has too many pages %d, which exceeds the" + "maximum pages per RPC of %d\n", + exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES); + RETURN(-EPROTO); } + rc = lu_env_refill((struct lu_env *)env); + LASSERT(rc == 0); info = ofd_info_init(env, exp); + LASSERT(oa != NULL); + + if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) { + struct ofd_seq *oseq; + oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi)); + if (IS_ERR(oseq)) { + CERROR("%s: Can not find seq for "DOSTID + ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi), + PTR_ERR(oseq)); + RETURN(-EINVAL); + } + + if (oseq->os_destroys_in_progress == 0) { + /* don't fail lookups for orphan recovery, it causes + * later LBUGs when objects still exist during + * precreate */ + ofd_seq_put(env, oseq); + RETURN(-ENOENT); + } + ofd_seq_put(env, oseq); + } + LASSERT(objcount == 1); LASSERT(obj->ioo_bufcnt > 0); - fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0); + rc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0); + if (unlikely(rc != 0)) + RETURN(rc); + if (cmd == OBD_BRW_WRITE) { - rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq, + rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi), capa, CAPA_OPC_OSS_WRITE); if (rc == 0) { - LASSERT(oa != NULL); la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR); rc = ofd_preprw_write(env, exp, ofd, &info->fti_fid, &info->fti_attr, oa, objcount, obj, rnb, nr_local, lnb, oti); } } else if (cmd == OBD_BRW_READ) { - rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq, + rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi), capa, CAPA_OPC_OSS_READ); if (rc == 0) { ofd_grant_prepare_read(env, exp, oa); - rc = ofd_preprw_read(env, ofd, &info->fti_fid, + rc = ofd_preprw_read(env, exp, ofd, &info->fti_fid, &info->fti_attr, obj->ioo_bufcnt, - rnb, nr_local, lnb); + rnb, nr_local, lnb, oti); obdo_from_la(oa, &info->fti_attr, LA_ATIME); } } else { @@ -405,6 +444,9 @@ retry: th->th_sync |= oti->oti_sync_write; + if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET)) + GOTO(out_stop, rc = -EINPROGRESS); + rc = dt_declare_write_commit(env, o, lnb, niocount, th); if (rc) GOTO(out_stop, rc); @@ -431,7 +473,7 @@ retry: } /* get attr to return */ - dt_attr_get(env, o, la, ofd_object_capa(env, fo)); + rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo)); out_stop: /* Force commit to make the just-deleted blocks @@ -474,7 +516,9 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp, LASSERT(npages > 0); - fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0); + rc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0); + if (unlikely(rc != 0)) + RETURN(rc); if (cmd == OBD_BRW_WRITE) { /* Don't update timestamps if this write is older than a * setattr which modifies the timestamps. b=10150 */ @@ -505,12 +549,27 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp, else obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID); - if (ofd_grant_prohibit(exp, ofd)) - /* Trick to prevent clients from waiting for bulk write - * in flight since they won't get any grant in the reply - * anyway so they had better firing the sync write RPC - * straight away */ + /* don't report overquota flag if we failed before reaching + * commit */ + if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) { + /* return the overquota flags to client */ + if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) { + if (oa->o_valid & OBD_MD_FLFLAGS) + oa->o_flags |= OBD_FL_NO_USRQUOTA; + else + oa->o_flags = OBD_FL_NO_USRQUOTA; + } + + if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) { + if (oa->o_valid & OBD_MD_FLFLAGS) + oa->o_flags |= OBD_FL_NO_GRPQUOTA; + else + oa->o_flags = OBD_FL_NO_GRPQUOTA; + } + + oa->o_valid |= OBD_MD_FLFLAGS; oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA; + } } else if (cmd == OBD_BRW_READ) { struct ldlm_namespace *ns = ofd->ofd_namespace; @@ -520,7 +579,7 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp, if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) { struct ldlm_resource *rs = NULL; - ofd_build_resid(&info->fti_fid, &info->fti_resid); + ost_fid_build_resid(&info->fti_fid, &info->fti_resid); rs = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0); if (rs != NULL) {