* 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/
#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, char *jobid)
{
struct ofd_object *fo;
int i, j, rc, tot_bytes = 0;
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));
rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
if (unlikely(rc))
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, jobid, tot_bytes);
RETURN(0);
buf_put:
struct lu_attr *la, struct obdo *oa,
int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
- struct niobuf_local *lnb,
- struct obd_trans_info *oti)
+ struct niobuf_local *lnb, char *jobid)
{
struct ofd_object *fo;
int i, j, k, rc = 0, tot_bytes = 0;
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);
}
- /* Always sync if syncjournal parameter is set */
- oti->oti_sync_write = ofd->ofd_syncjournal;
-
/* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
* space back if possible */
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++) {
lnb[j+k].lnb_flags = rnb[i].rnb_flags;
if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
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);
+ ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, 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,
+int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, struct obd_trans_info *oti,
struct lustre_capa *capa)
{
+ struct tgt_session_info *tsi = tgt_ses_info(env);
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
+ char *jobid;
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);
}
- info = ofd_info_init(env, exp);
+ if (tgt_ses_req(tsi) == NULL) { /* echo client case */
+ LASSERT(oti != NULL);
+ lu_env_refill((struct lu_env *)env);
+ info = ofd_info_init(env, exp);
+ ofd_oti2info(info, oti);
+ jobid = oti->oti_jobid;
+ } else {
+ info = tsi2ofd_info(tsi);
+ jobid = tsi->tsi_jobid;
+ }
+
+ 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);
+ info->fti_fid = oa->o_oi.oi_fid;
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);
+ obj, rnb, nr_local, lnb, jobid);
}
} 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, jobid);
obdo_from_la(oa, &info->fti_attr, LA_ATIME);
}
} else {
return rc;
}
+struct ofd_soft_sync_callback {
+ struct dt_txn_commit_cb ossc_cb;
+ struct obd_export *ossc_exp;
+};
+
+static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
+ struct dt_txn_commit_cb *cb, int err)
+{
+ struct ofd_soft_sync_callback *ossc;
+
+ ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
+
+ CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
+ atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
+
+ class_export_cb_put(ossc->ossc_exp);
+ OBD_FREE_PTR(ossc);
+}
+
+static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
+{
+ struct ofd_soft_sync_callback *ossc;
+ struct dt_txn_commit_cb *dcb;
+ int rc;
+
+ OBD_ALLOC_PTR(ossc);
+ if (ossc == NULL)
+ return -ENOMEM;
+
+ ossc->ossc_exp = class_export_cb_get(exp);
+
+ dcb = &ossc->ossc_cb;
+ dcb->dcb_func = ofd_cb_soft_sync;
+ CFS_INIT_LIST_HEAD(&dcb->dcb_linkage);
+ strncpy(dcb->dcb_name, "ofd_cb_soft_sync", MAX_COMMIT_CB_STR_LEN);
+ dcb->dcb_name[MAX_COMMIT_CB_STR_LEN - 1] = '\0';
+
+ rc = dt_trans_cb_add(th, dcb);
+ if (rc) {
+ class_export_cb_put(exp);
+ OBD_FREE_PTR(ossc);
+ }
+
+ return rc;
+}
+
static int
-ofd_commitrw_write(const struct lu_env *env, struct ofd_device *ofd,
- struct lu_fid *fid, struct lu_attr *la,
- struct filter_fid *ff, int objcount,
- int niocount, struct niobuf_local *lnb,
- struct obd_trans_info *oti, int old_rc)
+ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
+ struct ofd_device *ofd, struct lu_fid *fid,
+ struct lu_attr *la, struct filter_fid *ff, int objcount,
+ int niocount, struct niobuf_local *lnb, int old_rc)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo;
struct thandle *th;
int rc = 0;
int retries = 0;
+ int i;
+ struct filter_export_data *fed = &exp->exp_filter_data;
+ bool soft_sync = false;
+ bool cb_registered = false;
ENTRY;
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
- th->th_sync |= oti->oti_sync_write;
+ th->th_sync |= ofd->ofd_syncjournal;
+ if (th->th_sync == 0) {
+ for (i = 0; i < niocount; i++) {
+ if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
+ th->th_sync = 1;
+ break;
+ }
+ if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
+ soft_sync = true;
+ }
+ }
if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
GOTO(out_stop, rc = -EINPROGRESS);
}
/* 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
if (rc == -ENOSPC)
th->th_sync = 1;
+ /* do this before trans stop in case commit has finished */
+ if (!th->th_sync && soft_sync && !cb_registered) {
+ ofd_soft_sync_cb_add(th, exp);
+ cb_registered = true;
+ }
+
ofd_trans_stop(env, ofd, th, rc);
if (rc == -ENOSPC && retries++ < 3) {
CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
goto retry;
}
+ if (!soft_sync)
+ /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
+ atomic_set(&fed->fed_soft_sync_count, 0);
+ else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
+ ofd->ofd_soft_sync_limit)
+ dt_commit_async(env, ofd->ofd_osd);
+
out:
dt_bufs_put(env, o, lnb, niocount);
ofd_read_unlock(env, fo);
struct niobuf_local *lnb, struct obd_trans_info *oti,
int old_rc)
{
- struct ofd_thread_info *info;
+ struct ofd_thread_info *info = ofd_info(env);
struct ofd_mod_data *fmd;
__u64 valid;
struct ofd_device *ofd = ofd_exp(exp);
struct filter_fid *ff = NULL;
int rc = 0;
- info = ofd_info(env);
- ofd_oti2info(info, oti);
-
LASSERT(npages > 0);
- fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
+ info->fti_fid = oa->o_oi.oi_fid;
if (cmd == OBD_BRW_WRITE) {
/* Don't update timestamps if this write is older than a
* setattr which modifies the timestamps. b=10150 */
ofd_prepare_fidea(ff, oa);
}
- rc = ofd_commitrw_write(env, ofd, &info->fti_fid,
+ rc = ofd_commitrw_write(env, exp, ofd, &info->fti_fid,
&info->fti_attr, ff, objcount, npages,
- lnb, oti, old_rc);
+ lnb, old_rc);
if (rc == 0)
obdo_from_la(oa, &info->fti_attr,
OFD_VALID_FLAGS | LA_GID | LA_UID);
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) {
}
}
rc = ofd_commitrw_read(env, ofd, &info->fti_fid, objcount,
- npages, lnb);
+ npages, lnb);
if (old_rc)
rc = old_rc;
} else {
rc = -EPROTO;
}
- ofd_info2oti(info, oti);
+ if (oti != NULL)
+ ofd_info2oti(info, oti);
RETURN(rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff