* 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/
#define DEBUG_SUBSYSTEM S_FILTER
+#include <lustre/lustre_idl.h>
#include "ofd_internal.h"
#include <obd_cksum.h>
#include <lustre_quota.h>
{
struct obd_device *obd = ofd_obd(ofd);
struct nid_stat *stats;
- int num_stats, i;
+ int num_stats;
int rc, newnid = 0;
ENTRY;
stats = exp->exp_nid_stats;
LASSERT(stats != NULL);
- OBD_ALLOC(stats->nid_brw_stats, sizeof(struct brw_stats));
- if (stats->nid_brw_stats == NULL)
- GOTO(clean, rc = -ENOMEM);
-
- for (i = 0; i < BRW_LAST; i++)
- spin_lock_init(&stats->nid_brw_stats->hist[i].oh_lock);
-
- rc = lprocfs_seq_create(stats->nid_proc, "brw_stats", 0644,
- &ofd_per_nid_stats_fops, stats);
- if (rc)
- CWARN("Error adding the brw_stats file\n");
-
num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
LPROC_OFD_LAST - 1;
static int ofd_parse_connect_data(const struct lu_env *env,
struct obd_export *exp,
- struct obd_connect_data *data)
+ struct obd_connect_data *data,
+ bool new_connection)
{
struct ofd_device *ofd = ofd_exp(exp);
struct filter_export_data *fed = &exp->exp_filter_data;
RETURN(0);
CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
- " ocd_version: %x ocd_grant: %d ocd_index: %u\n",
+ " ocd_version: %x ocd_grant: %d ocd_index: %u"
+ " ocd_group %u\n",
exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
data->ocd_connect_flags, data->ocd_version,
- data->ocd_grant, data->ocd_index);
+ data->ocd_grant, data->ocd_index, data->ocd_group);
if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
CWARN("!!! This export (nid %s) used object group %d "
fed->fed_group = data->ocd_group;
data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
- exp->exp_connect_flags = data->ocd_connect_flags;
+ exp->exp_connect_data = *data;
data->ocd_version = LUSTRE_VERSION_CODE;
/* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
}
- if (exp->exp_connect_flags & OBD_CONNECT_GRANT)
- data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant);
+ if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
+ data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
+ new_connection);
if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
- int index = lsd->lsd_ost_index;
+ int index = lsd->lsd_osd_index;
- if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
- /* this will only happen on the first connect */
- lsd->lsd_ost_index = data->ocd_index;
- lsd->lsd_feature_compat |= OBD_COMPAT_OST;
- /* sync is not needed here as lut_client_add will
- * set exp_need_sync flag */
- tgt_server_data_update(env, &ofd->ofd_lut, 0);
- } else if (index != data->ocd_index) {
+ if (index != data->ocd_index) {
LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
" %u doesn't match actual OST index"
" %u in last_rcvd file, bad "
data->ocd_index);
RETURN(-EBADF);
}
+ if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
+ /* this will only happen on the first connect */
+ lsd->lsd_feature_compat |= OBD_COMPAT_OST;
+ /* sync is not needed here as lut_client_add will
+ * set exp_need_sync flag */
+ tgt_server_data_update(env, &ofd->ofd_lut, 0);
+ }
}
-
if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
data->ocd_brw_size = 65536;
} else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
data->ocd_brw_size = min(data->ocd_brw_size,
- (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
+ (__u32)DT_MAX_BRW_SIZE);
if (data->ocd_brw_size == 0) {
CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
" ocd_version: %x ocd_grant: %d ocd_index: %u "
if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
- RETURN(0);
+ if (OCD_HAS_FLAG(data, PINGLESS)) {
+ if (ptlrpc_pinger_suppress_pings()) {
+ spin_lock(&exp->exp_obd->obd_dev_lock);
+ list_del_init(&exp->exp_obd_chain_timed);
+ spin_unlock(&exp->exp_obd->obd_dev_lock);
+ } else {
+ data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
+ }
+ }
+
+ RETURN(0);
}
static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
struct obd_device *obd, struct obd_uuid *cluuid,
struct obd_connect_data *data, void *localdata)
{
- struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
+ struct ofd_device *ofd;
int rc;
ENTRY;
if (exp == NULL || obd == NULL || cluuid == NULL)
RETURN(-EINVAL);
+ ofd = ofd_dev(obd->obd_lu_dev);
+
rc = lu_env_refill((struct lu_env *)env);
if (rc != 0) {
CERROR("Failure to refill session: '%d'\n", rc);
}
ofd_info_init(env, exp);
- rc = ofd_parse_connect_data(env, exp, data);
+ rc = ofd_parse_connect_data(env, exp, data, false);
if (rc == 0)
ofd_export_stats_init(ofd, exp, localdata);
struct obd_export *exp;
struct ofd_device *ofd;
struct lustre_handle conn = { 0 };
- int rc, group;
-
+ int rc;
ENTRY;
if (_exp == NULL || obd == NULL || cluuid == NULL)
ofd_info_init(env, exp);
- rc = ofd_parse_connect_data(env, exp, data);
+ rc = ofd_parse_connect_data(env, exp, data, true);
if (rc)
GOTO(out, rc);
- group = data->ocd_group;
if (obd->obd_replayable) {
struct tg_export_data *ted = &exp->exp_target_data;
GOTO(out, rc);
ofd_export_stats_init(ofd, exp, localdata);
}
- if (group == 0)
- GOTO(out, rc = 0);
- /* init new group */
- if (group > ofd->ofd_max_group) {
- ofd->ofd_max_group = group;
- rc = ofd_group_load(env, ofd, group);
- }
+ CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
+ data->ocd_group);
+
out:
if (rc != 0) {
class_disconnect(exp);
ofd_grant_discard(exp);
ofd_fmd_cleanup(exp);
- if (exp->exp_connect_flags & OBD_CONNECT_GRANT_SHRINK) {
+ if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
if (ofd->ofd_tot_granted_clients > 0)
ofd->ofd_tot_granted_clients --;
}
__u32 keylen, void *key, __u32 vallen, void *val,
struct ptlrpc_request_set *set)
{
- struct ofd_device *ofd = ofd_exp(exp);
+ struct ofd_device *ofd;
int rc = 0;
ENTRY;
RETURN(-EINVAL);
}
+ ofd = ofd_exp(exp);
+
if (KEY_IS(KEY_CAPA_KEY)) {
rc = ofd_update_capa_key(ofd, val);
if (rc)
} else if (KEY_IS(KEY_GRANT_SHRINK)) {
struct ost_body *body = val;
+ ofd_info_init(env, exp);
/** handle grant shrink, similar to a read request */
ofd_grant_prepare_read(env, exp, &body->oa);
} else {
__u32 keylen, void *key, __u32 *vallen, void *val,
struct lov_stripe_md *lsm)
{
- struct ofd_device *ofd = ofd_exp(exp);
- int rc = 0;
+ struct ofd_device *ofd;
+ int rc = 0;
ENTRY;
RETURN(-EINVAL);
}
+ /* Because ofd_get_info might be called from
+ * handle_request_in as well(see LU-3239), where env might
+ * not be initilaized correctly, and le_ses might be in
+ * an un-initialized state, so only refill le_ctx here */
+ rc = lu_env_refill((struct lu_env *)env);
+ if (rc != 0)
+ RETURN(rc);
+
+ ofd = ofd_exp(exp);
+
if (KEY_IS(KEY_BLOCKSIZE)) {
__u32 *blocksize = val;
if (blocksize) {
if (*vallen < sizeof(*blocksize))
- RETURN(-EOVERFLOW);
+ GOTO(out, rc = -EOVERFLOW);
*blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
}
*vallen = sizeof(*blocksize);
__u32 *blocksize_bits = val;
if (blocksize_bits) {
if (*vallen < sizeof(*blocksize_bits))
- RETURN(-EOVERFLOW);
+ GOTO(out, rc = -EOVERFLOW);
*blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
}
*vallen = sizeof(*blocksize_bits);
} else if (KEY_IS(KEY_LAST_ID)) {
obd_id *last_id = val;
+ struct ofd_seq *oseq;
+
+ if (val == NULL) {
+ *vallen = sizeof(obd_id);
+ GOTO(out, rc = 0);
+ }
+ ofd_info_init(env, exp);
+ oseq = ofd_seq_load(env, ofd,
+ (obd_seq)exp->exp_filter_data.fed_group);
+ LASSERT(!IS_ERR(oseq));
if (last_id) {
- if (*vallen < sizeof(*last_id))
- RETURN(-EOVERFLOW);
- *last_id = ofd_last_id(ofd,
- exp->exp_filter_data.fed_group);
+ if (*vallen < sizeof(*last_id)) {
+ ofd_seq_put(env, oseq);
+ GOTO(out, rc = -EOVERFLOW);
+ }
+ *last_id = ofd_seq_last_oid(oseq);
}
+ ofd_seq_put(env, oseq);
*vallen = sizeof(*last_id);
} else if (KEY_IS(KEY_FIEMAP)) {
struct ofd_thread_info *info;
if (val == NULL) {
*vallen = fiemap_count_to_size(
fm_key->fiemap.fm_extent_count);
- RETURN(0);
+ GOTO(out, rc = 0);
}
info = ofd_info_init(env, exp);
-
- fid_ostid_unpack(&info->fti_fid, &fm_key->oa.o_oi, 0);
-
+ rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0);
+ if (rc != 0)
+ GOTO(out, rc);
CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
PFID(&info->fti_fid));
} else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
*((__u32 *) val) = ofd->ofd_sync_lock_cancel;
*vallen = sizeof(__u32);
+ } else if (KEY_IS(KEY_LAST_FID)) {
+ struct ofd_device *ofd = ofd_exp(exp);
+ struct ofd_seq *oseq;
+ struct lu_fid *fid = val;
+ int rc;
+
+ if (fid == NULL) {
+ *vallen = sizeof(struct lu_fid);
+ GOTO(out, rc = 0);
+ }
+
+ if (*vallen < sizeof(*fid))
+ GOTO(out, rc = -EOVERFLOW);
+
+ ofd_info_init(env, exp);
+
+ fid_le_to_cpu(fid, fid);
+
+ oseq = ofd_seq_load(env, ofd,
+ ostid_seq((struct ost_id *)fid));
+ if (IS_ERR(oseq))
+ GOTO(out, rc = PTR_ERR(oseq));
+
+ rc = ostid_to_fid(fid, &oseq->os_oi,
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
+ PFID(fid));
+ *vallen = sizeof(*fid);
+out_put:
+ ofd_seq_put(env, oseq);
} else {
CERROR("Not supported key %s\n", (char*)key);
rc = -EOPNOTSUPP;
}
-
+out:
RETURN(rc);
}
/* The QoS code on the MDS does not care about space reserved for
* precreate, so take it out. */
- if (exp->exp_connect_flags & OBD_CONNECT_MDS) {
+ if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
struct filter_export_data *fed;
fed = &obd->obd_self_export->exp_filter_data;
osfs->os_files, osfs->os_ffree, osfs->os_state);
if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
- ofd->ofd_lut.lut_lsd.lsd_ost_index))
+ ofd->ofd_lut.lut_lsd.lsd_osd_index))
osfs->os_ffree = 0;
/* OS_STATE_READONLY can be set by OSD already */
}
if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
- ofd->ofd_lut.lut_lsd.lsd_ost_index))
+ ofd->ofd_lut.lut_lsd.lsd_osd_index))
osfs->os_bfree = osfs->os_bavail = 2;
EXIT;
info = ofd_info_init(env, exp);
ofd_oti2info(info, oti);
- fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- ofd_build_resid(&info->fti_fid, &info->fti_resid);
+ rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
+ if (rc != 0)
+ RETURN(rc);
- rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq,
+ ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
+ rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
if (rc)
GOTO(out, rc);
info = ofd_info_init(env, exp);
ofd_oti2info(info, oti);
- fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- ofd_build_resid(&info->fti_fid, &info->fti_resid);
+ rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
+ if (rc != 0)
+ RETURN(rc);
+ ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
oinfo->oi_policy.l_extent.end);
- rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
+ rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
if (rc)
GOTO(out_env, rc);
/* Tell the clients that the object is gone now and that they should
* throw away any cached pages. */
- ofd_build_resid(fid, &info->fti_resid);
+ ost_fid_build_resid(fid, &info->fti_resid);
rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
LDLM_EXTENT, &policy, LCK_PW, &flags,
ldlm_blocking_ast, ldlm_completion_ast,
ofd_oti2info(info, oti);
if (!(oa->o_valid & OBD_MD_FLGROUP))
- oa->o_seq = 0;
+ ostid_set_seq_mdt0(&oa->o_oi);
/* check that o_misc makes sense */
if (oa->o_valid & OBD_MD_FLOBJCOUNT)
*/
if (info->fti_transno == 0) /* not replay */
info->fti_mult_trans = 1;
+
+ CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
+ POSTID(&oa->o_oi), count);
while (count > 0) {
int lrc;
- fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
+ lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
+ if (lrc != 0) {
+ if (rc == 0)
+ rc = lrc;
+ GOTO(out, rc);
+ }
lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
if (lrc == -ENOENT) {
CDEBUG(D_INODE,
rc = lrc;
}
count--;
- oa->o_id++;
+ ostid_inc_id(&oa->o_oi);
}
/* if we have transaction then there were some deletions, we don't
info->fti_transno = 0;
}
ofd_info2oti(info, oti);
+out:
RETURN(rc);
}
int skip_orphan;
int rc = 0;
struct ost_id oi = oa->o_oi;
+ __u64 end_id = ostid_id(&oa->o_oi);
+ struct ofd_seq *oseq;
ENTRY;
+ oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
+ if (oseq == NULL) {
+ CERROR("%s: Can not find seq for "DOSTID"\n",
+ ofd_name(ofd), POSTID(&oa->o_oi));
+ RETURN(-EINVAL);
+ }
+
LASSERT(exp != NULL);
- skip_orphan = !!(exp->exp_connect_flags & OBD_CONNECT_SKIP_ORPHAN);
+ skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
- last = ofd_last_id(ofd, oa->o_seq);
- CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n",
- ofd_obd(ofd)->obd_name, oa->o_id + 1, last);
+ last = ofd_seq_last_oid(oseq);
+ LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
+ " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
+ end_id + 1, last);
- for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
- fid_ostid_unpack(&info->fti_fid, &oi, 0);
+ for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
+ ostid_dec_id(&oi)) {
+ rc = ostid_to_fid(&info->fti_fid, &oi, 0);
+ if (rc != 0)
+ GOTO(out_put, rc);
rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
if (rc && rc != -ENOENT) /* this is pretty fatal... */
- CEMERG("error destroying precreated id "LPU64": %d\n",
- oi.oi_id, rc);
+ CEMERG("error destroying precreated id "DOSTID": %d\n",
+ POSTID(&oi), rc);
if (!skip_orphan) {
- ofd_last_id_set(ofd, oi.oi_id - 1, oa->o_seq);
+ ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
/* update last_id on disk periodically so that if we
* restart * we don't need to re-scan all of the just
* deleted objects. */
- if ((oi.oi_id & 511) == 0)
- ofd_last_id_write(env, ofd, oa->o_seq);
+ if ((ostid_id(&oi) & 511) == 0)
+ ofd_seq_last_oid_write(env, ofd, oseq);
}
}
- CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
- ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
+ CDEBUG(D_HA, "%s: after destroy: set last_objids"DOSTID"\n",
+ ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
if (!skip_orphan) {
- rc = ofd_last_id_write(env, ofd, oa->o_seq);
+ rc = ofd_seq_last_oid_write(env, ofd, oseq);
} else {
/* don't reuse orphan object, return last used objid */
- oa->o_id = last;
+ ostid_set_id(&oa->o_oi, last);
rc = 0;
}
+out_put:
+ ofd_seq_put(env, oseq);
RETURN(rc);
}
{
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
- int rc = 0, diff;
+ obd_seq seq = ostid_seq(&oa->o_oi);
+ struct ofd_seq *oseq;
+ int rc = 0, diff;
+ int sync_trans = 0;
ENTRY;
info = ofd_info_init(env, exp);
ofd_oti2info(info, oti);
- LASSERT(oa->o_seq >= FID_SEQ_OST_MDT0);
+ LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
LASSERT(oa->o_valid & OBD_MD_FLGROUP);
- CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
- oa->o_seq, oa->o_id);
+ CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
+
+ oseq = ofd_seq_load(env, ofd, seq);
+ if (IS_ERR(oseq)) {
+ CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
+ ofd_name(ofd), seq, PTR_ERR(oseq));
+ RETURN(-EINVAL);
+ }
if ((oa->o_valid & OBD_MD_FLFLAGS) &&
(oa->o_flags & OBD_FL_RECREATE_OBJS)) {
if (!ofd_obd(ofd)->obd_recovering ||
- oa->o_id > ofd_last_id(ofd, oa->o_seq)) {
- CERROR("recreate objid "LPU64" > last id "LPU64"\n",
- oa->o_id, ofd_last_id(ofd, oa->o_seq));
+ ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
+ CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
+ POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
GOTO(out_nolock, rc = -EINVAL);
}
/* do nothing because we create objects during first write */
/* destroy orphans */
if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
CERROR("%s: dropping old orphan cleanup request\n",
- ofd_obd(ofd)->obd_name);
+ ofd_name(ofd));
GOTO(out_nolock, rc = 0);
}
/* This causes inflight precreates to abort and drop lock */
- set_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
- mutex_lock(&ofd->ofd_create_locks[oa->o_seq]);
- if (!test_bit(oa->o_seq, &ofd->ofd_destroys_in_progress)) {
- CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
- exp->exp_obd->obd_name, oa->o_seq);
+ oseq->os_destroys_in_progress = 1;
+ mutex_lock(&oseq->os_create_lock);
+ if (!oseq->os_destroys_in_progress) {
+ CERROR("%s:["LPU64"] destroys_in_progress already"
+ " cleared\n", exp->exp_obd->obd_name,
+ ostid_seq(&oa->o_oi));
+ ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
GOTO(out, rc = 0);
}
- diff = oa->o_id - ofd_last_id(ofd, oa->o_seq);
+ diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
- ofd_last_id(ofd, oa->o_seq), diff);
+ ofd_seq_last_oid(oseq), diff);
if (-diff > OST_MAX_PRECREATE) {
/* FIXME: should reset precreate_next_id on MDS */
rc = 0;
} else if (diff < 0) {
rc = ofd_orphans_destroy(env, exp, ofd, oa);
- clear_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
+ oseq->os_destroys_in_progress = 0;
} else {
/* XXX: Used by MDS for the first time! */
- clear_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
+ oseq->os_destroys_in_progress = 0;
}
} else {
- mutex_lock(&ofd->ofd_create_locks[oa->o_seq]);
+ mutex_lock(&oseq->os_create_lock);
if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
CERROR("%s: dropping old precreate request\n",
- ofd_obd(ofd)->obd_name);
+ ofd_obd(ofd)->obd_name);
GOTO(out, rc = 0);
}
- /* only precreate if group == 0 and o_id is specfied */
- if (!fid_seq_is_mdt(oa->o_seq) || oa->o_id == 0) {
+ /* only precreate if seq is 0, IDIF or normal and also o_id
+ * must be specfied */
+ if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
+ !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
+ !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
+ ostid_id(&oa->o_oi) == 0) {
diff = 1; /* shouldn't we create this right now? */
} else {
- diff = oa->o_id - ofd_last_id(ofd, oa->o_seq);
+ diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
+ /* Do sync create if the seq is about to used up */
+ if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
+ fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
+ if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
+ sync_trans = 1;
+ } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
+ if (unlikely(ostid_id(&oa->o_oi) >=
+ LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
+ sync_trans = 1;
+ } else {
+ CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
+ ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
+ GOTO(out, rc = -EINVAL);
+ }
}
}
if (diff > 0) {
ofd_obd(ofd)->obd_self_export,
&diff);
if (rc) {
- CDEBUG(D_HA, "%s: failed to acquire grant space"
- "for precreate (%d)\n",
- ofd_obd(ofd)->obd_name, diff);
+ CDEBUG(D_HA, "%s: failed to acquire grant "
+ "space for precreate (%d): rc = %d\n",
+ ofd_name(ofd), diff, rc);
diff = 0;
}
}
while (diff > 0) {
- next_id = ofd_last_id(ofd, oa->o_seq) + 1;
+ next_id = ofd_seq_last_oid(oseq) + 1;
count = ofd_precreate_batch(ofd, diff);
- CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
+ CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
" at "LPU64"\n", ofd_obd(ofd)->obd_name,
- count, oa->o_seq, next_id);
+ count, ostid_seq(&oa->o_oi), next_id);
if (cfs_time_after(jiffies, enough_time)) {
LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
}
rc = ofd_precreate_objects(env, ofd, next_id,
- oa->o_seq, count);
+ oseq, count, sync_trans);
if (rc > 0) {
created += rc;
diff -= rc;
break;
}
}
- if (created > 0) {
+ if (created > 0)
/* some objects got created, we can return
* them, even if last creation failed */
- oa->o_id = ofd_last_id(ofd, oa->o_seq);
rc = 0;
- } else {
- CERROR("unable to precreate: %d\n", rc);
- oa->o_id = ofd_last_id(ofd, oa->o_seq);
- }
+ else
+ CERROR("%s: unable to precreate: rc = %d\n",
+ ofd_name(ofd), rc);
+ ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
ofd_info2oti(info, oti);
out:
- mutex_unlock(&ofd->ofd_create_locks[oa->o_seq]);
+ mutex_unlock(&oseq->os_create_lock);
out_nolock:
if (rc == 0 && ea != NULL) {
struct lov_stripe_md *lsm = *ea;
- lsm->lsm_object_id = oa->o_id;
+ lsm->lsm_oi = oa->o_oi;
}
- return rc;
+ ofd_seq_put(env, oseq);
+ RETURN(rc);
}
int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
info = ofd_info_init(env, exp);
- fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
+ rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
+ if (rc != 0)
+ GOTO(out, rc);
+ rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_META_READ);
if (rc)
GOTO(out, rc);
}
info = ofd_info_init(env, exp);
- fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
+ rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
+ if (rc != 0)
+ GOTO(out, rc);
- rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
+ rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
if (rc)
GOTO(out, rc);
return rc;
}
+static int ofd_ioc_get_obj_version(const struct lu_env *env,
+ struct ofd_device *ofd, void *karg)
+{
+ struct obd_ioctl_data *data = karg;
+ struct lu_fid fid;
+ struct ofd_object *fo;
+ dt_obj_version_t version;
+ int rc = 0;
+
+ ENTRY;
+
+ if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
+ GOTO(out, rc = -EINVAL);
+
+ if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
+ fid = *(struct lu_fid *)data->ioc_inlbuf1;
+ } else if (data->ioc_inlbuf3 != NULL &&
+ data->ioc_inllen3 == sizeof(__u64) &&
+ data->ioc_inlbuf4 != NULL &&
+ data->ioc_inllen4 == sizeof(__u64)) {
+ struct ost_id ostid;
+
+ ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
+ ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
+ rc = ostid_to_fid(&fid, &ostid, 0);
+ if (rc != 0)
+ GOTO(out, rc);
+ } else {
+ GOTO(out, rc = -EINVAL);
+ }
+
+ if (!fid_is_sane(&fid))
+ GOTO(out, rc = -EINVAL);
+
+ fo = ofd_object_find(env, ofd, &fid);
+ if (IS_ERR(fo))
+ GOTO(out, rc = PTR_ERR(fo));
+
+ if (!ofd_object_exists(fo))
+ GOTO(out_fo, rc = -ENOENT);
+
+ if (lu_object_remote(&fo->ofo_obj.do_lu))
+ GOTO(out_fo, rc = -EREMOTE);
+
+ version = dt_version_get(env, ofd_object_child(fo));
+ if (version == 0)
+ GOTO(out_fo, rc = -EIO);
+
+ *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
+
+ EXIT;
+out_fo:
+ ofd_object_put(env, fo);
+out:
+ return rc;
+}
+
int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
void *karg, void *uarg)
{
ENTRY;
CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
- rc = lu_env_init(&env, LCT_LOCAL);
+ rc = lu_env_init(&env, LCT_DT_THREAD);
if (rc)
RETURN(rc);
if (rc == 0)
rc = dt_ro(&env, ofd->ofd_osd);
break;
+ case OBD_IOC_GET_OBJ_VERSION:
+ rc = ofd_ioc_get_obj_version(&env, ofd, karg);
+ break;
default:
CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
rc = -ENOTTY;
return !!rc;
}
-static int ofd_obd_notify(struct obd_device *obd, struct obd_device *unused,
- enum obd_notify_event ev, void *data)
-{
- switch (ev) {
- case OBD_NOTIFY_CONFIG:
- LASSERT(obd->obd_no_conn);
- spin_lock(&obd->obd_dev_lock);
- obd->obd_no_conn = 0;
- spin_unlock(&obd->obd_dev_lock);
- break;
- default:
- CDEBUG(D_INFO, "%s: Unhandled notification %#x\n",
- obd->obd_name, ev);
- }
- return 0;
-}
-
/*
* Handle quota control requests to consult current usage/limit.
*
.o_precleanup = ofd_precleanup,
.o_ping = ofd_ping,
.o_health_check = ofd_health_check,
- .o_notify = ofd_obd_notify,
.o_quotactl = ofd_quotactl,
};