* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_FILTER
#include <obd_class.h>
-#include <lustre_param.h>
+#include <obd_cksum.h>
+#include <uapi/linux/lustre_param.h>
#include <lustre_fid.h>
#include <lustre_lfsck.h>
#include <lustre/lustre_idl.h>
#include <lustre_dlm.h>
#include <lustre_quota.h>
#include <lustre_nodemap.h>
+#include <lustre_log.h>
#include "ofd_internal.h"
if (obd->obd_fail)
strcat(flags, "A");
lustre_cfg_bufs_set_string(&bufs, 1, flags);
- lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
- if (lcfg == NULL)
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
+ if (!lcfg)
RETURN_EXIT;
+ lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
LASSERT(top);
top->ld_ops->ldo_process_config(env, top, lcfg);
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
lu_site_purge(env, top->ld_site, ~0);
if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
* Grant space for object precreation on the self export.
* The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
* is enough to create 10k objects. More space is then acquired for
- * precreation in ofd_grant_create().
+ * precreation in tgt_grant_create().
*/
memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
- oti->fti_ocd.ocd_grant *= ofd->ofd_dt_conf.ddp_inodespace;
+ oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
OBD_CONNECT_GRANT_PARAM;
- ofd_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
+ tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
true);
rc = next->ld_ops->ldo_recovery_complete(env, next);
RETURN(rc);
/* lprocfs must be setup before the ofd so state can be safely added
* to /proc incrementally as the ofd is setup */
obd->obd_vars = lprocfs_ofd_obd_vars;
- rc = lprocfs_obd_setup(obd);
+ rc = lprocfs_obd_setup(obd, false);
if (rc) {
CERROR("%s: lprocfs_obd_setup failed: %d.\n",
obd->obd_name, rc);
*repbody = *body;
/** handle grant shrink, similar to a read request */
- ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
+ tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
&repbody->oa);
} else if (KEY_IS(KEY_EVICT_BY_NID)) {
if (vallen > 0)
obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
rc = 0;
- } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
- rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
} else {
CERROR("%s: Unsupported key %s\n",
tgt_name(tsi->tsi_tgt), (char *)key);
__u64 curr_version;
obdo_from_la(&repbody->oa, &fti->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
+ OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
/* Store object version in reply */
curr_version = dt_version_get(tsi->tsi_env,
GOTO(out_put, rc);
obdo_from_la(&repbody->oa, &fti->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
+ OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
tsi->tsi_jobid, 1);
rc = ofd_seq_last_oid_write(env, ofd, oseq);
} else {
/* don't reuse orphan object, return last used objid */
- ostid_set_id(oi, last);
- rc = 0;
+ rc = ostid_set_id(oi, last);
}
GOTO(out_put, rc);
u64 seq = ostid_seq(&oa->o_oi);
u64 oid = ostid_id(&oa->o_oi);
struct ofd_seq *oseq;
- int rc = 0, diff;
+ s64 diff;
+ int rc = 0;
int sync_trans = 0;
long granted = 0;
rep_oa = &repbody->oa;
rep_oa->o_oi = oa->o_oi;
- LASSERT(seq >= FID_SEQ_OST_MDT0);
LASSERT(oa->o_valid & OBD_MD_FLGROUP);
CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
if (!oseq->os_destroys_in_progress) {
CERROR("%s:[%llu] destroys_in_progress already"
" cleared\n", ofd_name(ofd), seq);
- ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
- GOTO(out, rc = 0);
+ rc = ostid_set_id(&rep_oa->o_oi,
+ ofd_seq_last_oid(oseq));
+ GOTO(out, rc);
}
diff = oid - ofd_seq_last_oid(oseq);
- CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %d\n",
- ofd_seq_last_oid(oseq), diff);
+ CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
+ ofd_seq_last_oid(oseq), diff);
if (-diff > OST_MAX_PRECREATE) {
+ LCONSOLE(D_INFO, "%s: too large difference between MDS "
+ "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
+ "(%llu), trust the OST\n",
+ ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
+ PFID(&oseq->os_oi.oi_fid),
+ ofd_seq_last_oid(oseq));
+
/* Let MDS know that we are so far ahead. */
- ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq) + 1);
- rc = 0;
+ rc = ostid_set_id(&rep_oa->o_oi,
+ ofd_seq_last_oid(oseq) + 1);
} else if (diff < 0) {
rc = ofd_orphans_destroy(tsi->tsi_env, exp,
ofd, rep_oa);
if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
!(oa->o_flags & OBD_FL_DELORPHAN)) {
/* don't enforce grant during orphan recovery */
- granted = ofd_grant_create(tsi->tsi_env,
- ofd_obd(ofd)->obd_self_export,
- &diff);
+ granted = tgt_grant_create(tsi->tsi_env,
+ ofd_obd(ofd)->obd_self_export,
+ &diff);
if (granted < 0) {
rc = granted;
granted = 0;
CDEBUG(D_HA, "%s: failed to acquire grant "
- "space for precreate (%d): rc = %d\n",
+ "space for precreate (%lld): rc = %d\n",
ofd_name(ofd), diff, rc);
diff = 0;
}
CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
"%u larger than the LAST_ID "DOSTID", only "
- "precreating the last %u objects.\n",
+ "precreating the last %lld objects.\n",
ofd_name(ofd), POSTID(&oa->o_oi),
5 * OST_MAX_PRECREATE,
POSTID(&oseq->os_oi), diff);
while (diff > 0) {
next_id = ofd_seq_last_oid(oseq) + 1;
- count = ofd_precreate_batch(ofd, diff);
+ count = ofd_precreate_batch(ofd, (int)diff);
CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
" at %llu\n", ofd_name(ofd),
if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
&& cfs_time_after(jiffies, enough_time)) {
- CDEBUG(D_HA, "%s: Slow creates, %d/%d objects"
+ CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
" created at a rate of %d/s\n",
ofd_name(ofd), created, diff + created,
created / DISK_TIMEOUT);
lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
LCONSOLE_WARN("%s: can't create the same count of"
" objects when replaying the request"
- " (diff is %d). see LU-4621\n",
+ " (diff is %lld). see LU-4621\n",
ofd_name(ofd), diff);
if (created > 0)
if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
!(oa->o_flags & OBD_FL_DELORPHAN)) {
- ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted,
- rc);
+ tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
+ granted, rc);
granted = 0;
}
- ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
+ rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
}
EXIT;
ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
static int ofd_ladvise_prefetch(const struct lu_env *env,
struct ofd_object *fo,
- __u64 start, __u64 end)
+ struct niobuf_local *lnb,
+ __u64 start, __u64 end, enum dt_bufs_type dbt)
{
- struct ofd_thread_info *info = ofd_info(env);
- pgoff_t start_index, end_index, pages;
- struct niobuf_remote rnb;
- unsigned long nr_local;
- struct niobuf_local *lnb;
- int rc = 0;
+ struct ofd_thread_info *info = ofd_info(env);
+ pgoff_t start_index, end_index, pages;
+ struct niobuf_remote rnb;
+ unsigned long nr_local;
+ int rc = 0;
if (end <= start)
RETURN(-EINVAL);
- OBD_ALLOC_LARGE(lnb, sizeof(*lnb) * PTLRPC_MAX_BRW_PAGES);
- if (lnb == NULL)
- RETURN(-ENOMEM);
-
ofd_read_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(out_unlock, rc = -ENOENT);
if (end > info->fti_attr.la_size)
end = info->fti_attr.la_size;
- if (end == 0)
+ if (end <= start)
GOTO(out_unlock, rc);
/* We need page aligned offset and length */
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, 0);
+ rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
if (unlikely(rc < 0))
break;
nr_local = rc;
out_unlock:
ofd_read_unlock(env, fo);
- OBD_FREE_LARGE(lnb, sizeof(*lnb) * PTLRPC_MAX_BRW_PAGES);
RETURN(rc);
}
*/
static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
{
- struct ptlrpc_request *req = tgt_ses_req(tsi);
- struct obd_export *exp = tsi->tsi_exp;
- struct ofd_device *ofd = ofd_exp(exp);
- struct ost_body *body, *repbody;
- struct ofd_thread_info *info;
- struct ofd_object *fo;
- const struct lu_env *env = req->rq_svc_thread->t_env;
- int rc = 0;
- struct lu_ladvise *ladvise;
- int num_advise;
- struct ladvise_hdr *ladvise_hdr;
- struct obd_ioobj ioo;
- struct lustre_handle lockh = { 0 };
- __u64 flags = 0;
- int i;
- struct dt_object *dob;
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct obd_export *exp = tsi->tsi_exp;
+ struct ofd_device *ofd = ofd_exp(exp);
+ struct ost_body *body, *repbody;
+ struct ofd_thread_info *info;
+ struct ofd_object *fo;
+ struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
+ const struct lu_env *env = svc_thread->t_env;
+ struct tgt_thread_big_cache *tbc = svc_thread->t_data;
+ enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
+ struct lu_ladvise *ladvise;
+ int num_advise;
+ struct ladvise_hdr *ladvise_hdr;
+ struct obd_ioobj ioo;
+ struct lustre_handle lockh = { 0 };
+ __u64 flags = 0;
+ int i;
+ struct dt_object *dob;
+ __u64 start;
+ __u64 end;
+ int rc = 0;
ENTRY;
CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
LASSERT(fo != NULL);
dob = ofd_object_child(fo);
+ if (ptlrpc_connection_is_local(exp->exp_connection))
+ dbt |= DT_BUFS_TYPE_LOCAL;
+
for (i = 0; i < num_advise; i++, ladvise++) {
- if (ladvise->lla_end <= ladvise->lla_start) {
+ start = ladvise->lla_start;
+ end = ladvise->lla_end;
+ if (end <= start) {
rc = err_serious(-EPROTO);
break;
}
rc = -ENOTSUPP;
break;
case LU_LADVISE_WILLREAD:
+ if (tbc == NULL)
+ RETURN(-ENOMEM);
+
ioo.ioo_oid = body->oa.o_oi;
ioo.ioo_bufcnt = 1;
rc = tgt_extent_lock(exp->exp_obd->obd_namespace,
- &tsi->tsi_resid,
- ladvise->lla_start,
- ladvise->lla_end - 1,
+ &tsi->tsi_resid, start, end - 1,
&lockh, LCK_PR, &flags);
if (rc != 0)
break;
- req->rq_status = ofd_ladvise_prefetch(env,
- fo,
- ladvise->lla_start,
- ladvise->lla_end);
+ req->rq_status = ofd_ladvise_prefetch(env, fo,
+ tbc->local,
+ start, end, dbt);
tgt_extent_unlock(&lockh, LCK_PR);
break;
case LU_LADVISE_DONTNEED:
static inline int prolong_timeout(struct ptlrpc_request *req)
{
struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
+ time_t req_timeout;
if (AT_OFF)
return obd_timeout / 2;
- return at_est2timeout(at_get(&svcpt->scp_at_estimate));
+ req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
+ return max_t(time_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
+ req_timeout);
}
/**
info->fti_xid = 0;
info->fti_pre_version = 0;
- info->fti_used = 0;
memset(&info->fti_attr, 0, sizeof info->fti_attr);
}
const char *dev = lustre_cfg_string(cfg, 0);
struct ofd_thread_info *info = NULL;
struct obd_device *obd;
+ struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
struct obd_statfs *osfs;
struct lu_fid fid;
struct nm_config_file *nodemap_config;
+ struct obd_device_target *obt;
int rc;
ENTRY;
if (rc != 0)
RETURN(rc);
- obd->u.obt.obt_magic = OBT_MAGIC;
+ obt = &obd->u.obt;
+ obt->obt_magic = OBT_MAGIC;
m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
m->ofd_raid_degraded = 0;
m->ofd_syncjournal = 0;
ofd_slc_set(m);
- m->ofd_grant_compat_disable = 0;
+ tgd->tgd_grant_compat_disable = 0;
m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
/* statfs data */
- spin_lock_init(&m->ofd_osfs_lock);
- m->ofd_osfs_age = cfs_time_shift_64(-1000);
- m->ofd_osfs_unstable = 0;
- m->ofd_statfs_inflight = 0;
- m->ofd_osfs_inflight = 0;
+ spin_lock_init(&tgd->tgd_osfs_lock);
+ tgd->tgd_osfs_age = cfs_time_shift_64(-1000);
+ tgd->tgd_osfs_unstable = 0;
+ tgd->tgd_statfs_inflight = 0;
+ tgd->tgd_osfs_inflight = 0;
/* grant data */
- spin_lock_init(&m->ofd_grant_lock);
- m->ofd_tot_dirty = 0;
- m->ofd_tot_granted = 0;
- m->ofd_tot_pending = 0;
+ spin_lock_init(&tgd->tgd_grant_lock);
+ tgd->tgd_tot_dirty = 0;
+ tgd->tgd_tot_granted = 0;
+ tgd->tgd_tot_pending = 0;
+
m->ofd_seq_count = 0;
init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
INIT_LIST_HEAD(&m->ofd_inconsistency_list);
ofd_procfs_add_brw_stats_symlink(m);
- /* populate cached statfs data */
- osfs = &ofd_info(env)->fti_u.osfs;
- rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
- if (rc != 0) {
- CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
- GOTO(err_fini_stack, rc);
- }
- if (!IS_PO2(osfs->os_bsize)) {
- CERROR("%s: blocksize (%d) is not a power of 2\n",
- obd->obd_name, osfs->os_bsize);
- GOTO(err_fini_stack, rc = -EPROTO);
- }
- m->ofd_blockbits = fls(osfs->os_bsize) - 1;
-
- if (ONE_MB_BRW_SIZE < (1U << m->ofd_blockbits))
- m->ofd_brw_size = 1U << m->ofd_blockbits;
- else
- m->ofd_brw_size = ONE_MB_BRW_SIZE;
-
- m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
- if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
- m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
-
snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
"filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
"filter_ldlm_cb_client", &obd->obd_ldlm_client);
- dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
+ dt_conf_get(env, m->ofd_osd, &m->ofd_lut.lut_dt_conf);
rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
OBD_FAIL_OST_ALL_REQUEST_NET,
if (rc)
GOTO(err_free_ns, rc);
+ /* populate cached statfs data */
+ osfs = &ofd_info(env)->fti_u.osfs;
+ rc = tgt_statfs_internal(env, &m->ofd_lut, osfs, 0, NULL);
+ if (rc != 0) {
+ CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
+ GOTO(err_fini_lut, rc);
+ }
+ if (!is_power_of_2(osfs->os_bsize)) {
+ CERROR("%s: blocksize (%d) is not a power of 2\n",
+ obd->obd_name, osfs->os_bsize);
+ GOTO(err_fini_lut, rc = -EPROTO);
+ }
+ tgd->tgd_blockbits = fls(osfs->os_bsize) - 1;
+
+ if (DT_DEF_BRW_SIZE < (1U << tgd->tgd_blockbits))
+ m->ofd_brw_size = 1U << tgd->tgd_blockbits;
+ else
+ m->ofd_brw_size = DT_DEF_BRW_SIZE;
+
+ m->ofd_cksum_types_supported = cksum_types_supported_server();
+ m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
+ if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
+ m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
+
rc = ofd_fs_setup(env, m, obd);
if (rc)
GOTO(err_fini_lut, rc);
nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
m->ofd_los);
- if (IS_ERR(nodemap_config))
- GOTO(err_fini_los, rc = PTR_ERR(nodemap_config));
-
- obd->u.obt.obt_nodemap_config_file = nodemap_config;
+ if (IS_ERR(nodemap_config)) {
+ rc = PTR_ERR(nodemap_config);
+ if (rc != -EROFS)
+ GOTO(err_fini_los, rc);
+ } else {
+ obt->obt_nodemap_config_file = nodemap_config;
+ }
rc = ofd_start_inconsistency_verification_thread(m);
if (rc != 0)
GOTO(err_fini_nm, rc);
- tgt_adapt_sptlrpc_conf(&m->ofd_lut, 1);
+ tgt_adapt_sptlrpc_conf(&m->ofd_lut);
RETURN(0);
err_fini_nm:
- nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
- obd->u.obt.obt_nodemap_config_file = NULL;
+ nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
+ obt->obt_nodemap_config_file = NULL;
err_fini_los:
local_oid_storage_fini(env, m->ofd_los);
m->ofd_los = NULL;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff