#include <lustre/lustre_idl.h>
#include "ofd_internal.h"
#include <obd_cksum.h>
+#include <lustre_ioctl.h>
#include <lustre_quota.h>
#include <lustre_lfsck.h>
stats = exp->exp_nid_stats;
LASSERT(stats != NULL);
- num_stats = NUM_OBD_STATS + LPROC_OFD_LAST;
+ num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
+
stats->nid_stats = lprocfs_alloc_stats(num_stats,
LPROCFS_STATS_FLAG_NOPERCPU);
if (stats->nid_stats == NULL)
return -ENOMEM;
- lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
- lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
- LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
- lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
- LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
-
+ lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
+ ofd_stats_counter_init(stats->nid_stats);
rc = lprocfs_register_stats(stats->nid_proc, "stats",
stats->nid_stats);
if (rc)
CWARN("!!! This export (nid %s) used object group %d "
"earlier; now it's trying to use group %d! This could "
"be a bug in the MDS. Please report to "
- "http://bugs.whamcloud.com/\n",
+ "https://jira.hpdd.intel.com/\n",
obd_export_nid2str(exp), fed->fed_group,
data->ocd_group);
RETURN(-EPROTO);
fed->fed_group = data->ocd_group;
data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
- 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_connect_flags(exp) & OBD_CONNECT_GRANT)
+ if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
new_connection);
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);
- RETURN(rc);
- }
-
- ofd_info_init(env, exp);
rc = ofd_parse_connect_data(env, exp, data, false);
if (rc == 0)
ofd_export_stats_init(ofd, exp, localdata);
exp = class_conn2export(&conn);
LASSERT(exp != NULL);
- rc = lu_env_refill((struct lu_env *)env);
- if (rc != 0) {
- CERROR("Failure to refill session: '%d'\n", rc);
- GOTO(out, rc);
- }
-
- ofd_info_init(env, exp);
-
rc = ofd_parse_connect_data(env, exp, data, true);
if (rc)
GOTO(out, rc);
}
CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
- data->ocd_group);
+ data ? data->ocd_group : -1);
out:
if (rc != 0) {
RETURN(rc);
}
-static int ofd_obd_disconnect(struct obd_export *exp)
+int ofd_obd_disconnect(struct obd_export *exp)
{
- struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
+ struct ofd_device *ofd = ofd_exp(exp);
struct lu_env env;
int rc;
ofd_grant_discard(exp);
- rc = lu_env_init(&env, LCT_DT_THREAD);
- if (rc)
- RETURN(rc);
-
/* Do not erase record for recoverable client. */
if (exp->exp_obd->obd_replayable &&
- (!exp->exp_obd->obd_fail || exp->exp_failed))
- tgt_client_del(&env, exp);
- lu_env_fini(&env);
+ (!exp->exp_obd->obd_fail || exp->exp_failed)) {
+ rc = lu_env_init(&env, LCT_DT_THREAD);
+ if (rc)
+ GOTO(out, rc);
+ tgt_client_del(&env, exp);
+ lu_env_fini(&env);
+ }
+out:
class_export_put(exp);
RETURN(rc);
}
spin_lock_init(&exp->exp_filter_data.fed_lock);
CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
+ atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
spin_lock(&exp->exp_lock);
exp->exp_connecting = 1;
spin_unlock(&exp->exp_lock);
static int ofd_destroy_export(struct obd_export *exp)
{
- struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
+ struct ofd_device *ofd = ofd_exp(exp);
if (exp->exp_filter_data.fed_pending)
CERROR("%s: cli %s/%p has %lu pending on destroyed export"
int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
{
struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
+ struct lfsck_start_param lsp;
+ int rc;
+
+ CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
+
+ lsp.lsp_start = NULL;
+ lsp.lsp_index_valid = 0;
+ rc = lfsck_start(env, ofd->ofd_osd, &lsp);
+ if (rc != 0 && rc != -EALREADY)
+ CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
+ ofd_name(ofd), rc);
- CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
return ldev->ld_ops->ldo_recovery_complete(env, ldev);
}
RETURN(rc);
}
-static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
- struct obd_device *obd, int initial)
-{
- struct filter_obd *fo = &obd->u.filter;
- struct sptlrpc_rule_set tmp_rset;
- int rc;
-
- sptlrpc_rule_set_init(&tmp_rset);
- rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
- if (rc) {
- CERROR("%s: failed get sptlrpc rules: rc = %d\n",
- obd->obd_name, rc);
- return rc;
- }
-
- sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
-
- write_lock(&fo->fo_sptlrpc_lock);
- sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
- fo->fo_sptlrpc_rset = tmp_rset;
- write_unlock(&fo->fo_sptlrpc_lock);
-
- return 0;
-}
-
-static int ofd_set_mds_conn(struct obd_export *exp, void *val)
-{
- int rc = 0;
-
- ENTRY;
-
- LCONSOLE_WARN("%s: received MDS connection from %s\n",
- exp->exp_obd->obd_name, obd_export_nid2str(exp));
- RETURN(rc);
-}
-
+/* This is not called from request handler, check ofd_set_info_hdl() instead
+ * this OBD functions is only used by class_notify_sptlrpc_conf() locally
+ * by direct obd_set_info_async() call */
static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
__u32 keylen, void *key, __u32 vallen, void *val,
struct ptlrpc_request_set *set)
{
- struct ofd_device *ofd;
- int rc = 0;
+ 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)
- CERROR("%s: update capability key failed: rc = %d\n",
- exp->exp_obd->obd_name, rc);
- } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
- ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
- } else if (KEY_IS(KEY_MDS_CONN)) {
- rc = ofd_set_mds_conn(exp, val);
- } 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);
+ if (KEY_IS(KEY_SPTLRPC_CONF)) {
+ rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
} else {
CERROR("%s: Unsupported key %s\n",
exp->exp_obd->obd_name, (char*)key);
rc = -EOPNOTSUPP;
}
-
RETURN(rc);
}
+/* used by nrs_orr_range_fill_physical() in ptlrpc, see LU-3239 */
static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
__u32 keylen, void *key, __u32 *vallen, void *val,
struct lov_stripe_md *lsm)
{
- struct ofd_device *ofd;
- int rc = 0;
+ struct ofd_thread_info *info;
+ struct ofd_device *ofd;
+ struct ll_fiemap_info_key *fm_key = key;
+ struct ll_user_fiemap *fiemap = val;
+ 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))
- GOTO(out, rc = -EOVERFLOW);
- *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
- }
- *vallen = sizeof(*blocksize);
- } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
- __u32 *blocksize_bits = val;
- if (blocksize_bits) {
- if (*vallen < sizeof(*blocksize_bits))
- 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)) {
- 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_device *ofd = ofd_exp(exp);
- struct ofd_object *fo;
- struct ll_fiemap_info_key *fm_key = key;
- struct lu_fid fid;
-
- if (val == NULL) {
- *vallen = fiemap_count_to_size(
- fm_key->fiemap.fm_extent_count);
- GOTO(out, rc = 0);
- }
+ if (KEY_IS(KEY_FIEMAP)) {
+ info = ofd_info_init(env, exp);
- rc = ostid_to_fid(&fid, &fm_key->oa.o_oi, 0);
+ rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi,
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
- GOTO(out, rc);
- CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
- PFID(&fid));
-
- fo = ofd_object_find(env, ofd, &fid);
- if (IS_ERR(fo)) {
- CERROR("%s: error finding object "DFID"\n",
- exp->exp_obd->obd_name, PFID(&fid));
- rc = PTR_ERR(fo);
- } else {
- struct ll_user_fiemap *fiemap = val;
-
- ofd_read_lock(env, fo);
- if (ofd_object_exists(fo)) {
- *fiemap = fm_key->fiemap;
- rc = dt_fiemap_get(env,
- ofd_object_child(fo),
- fiemap);
- } else {
- rc = -ENOENT;
- }
- ofd_read_unlock(env, fo);
- ofd_object_put(env, fo);
- }
- } 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);
+ RETURN(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);
+ rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
} else {
- CERROR("Not supported key %s\n", (char*)key);
+ CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
rc = -EOPNOTSUPP;
}
-out:
+
RETURN(rc);
}
int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
struct obd_statfs *osfs, __u64 max_age, int *from_cache)
{
- int rc;
+ int rc = 0;
+ ENTRY;
+
+ down_read(&ofd->ofd_lastid_rwsem);
+ /* Currently, for safe, we do not distinguish which LAST_ID is broken,
+ * we may do that in the future.
+ * Return -ENOSPC until the LAST_ID rebuilt. */
+ if (unlikely(ofd->ofd_lastid_rebuilding))
+ GOTO(out, rc = -ENOSPC);
spin_lock(&ofd->ofd_osfs_lock);
if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
* call it fairly often as space fills up */
rc = dt_statfs(env, ofd->ofd_osd, osfs);
if (unlikely(rc))
- return rc;
+ GOTO(out, rc);
spin_lock(&ofd->ofd_grant_lock);
spin_lock(&ofd->ofd_osfs_lock);
if (from_cache)
*from_cache = 1;
}
- return 0;
+
+ GOTO(out, rc);
+
+out:
+ up_read(&ofd->ofd_lastid_rwsem);
+
+ return rc;
}
-static int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
- struct obd_statfs *osfs, __u64 max_age, __u32 flags)
+int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
+ struct obd_statfs *osfs, __u64 max_age, __u32 flags)
{
struct obd_device *obd = class_exp2obd(exp);
- struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
+ struct ofd_device *ofd = ofd_exp(exp);
int rc;
ENTRY;
return rc;
}
-int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo, struct obd_trans_info *oti)
+/* needed by echo client only for now, RPC handler uses ofd_setattr_hdl() */
+int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
+ struct obd_info *oinfo, struct obd_trans_info *oti)
{
struct ofd_thread_info *info;
struct ofd_device *ofd = ofd_exp(exp);
struct ldlm_resource *res;
struct ofd_object *fo;
struct obdo *oa = oinfo->oi_oa;
+ struct lu_fid *fid = &oa->o_oi.oi_fid;
struct filter_fid *ff = NULL;
int rc = 0;
ENTRY;
info = ofd_info_init(env, exp);
- ofd_oti2info(info, oti);
- 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);
- 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);
+ ost_fid_build_resid(fid, &info->fti_resid);
/* This would be very bad - accidentally truncating a file when
* changing the time or similar - bug 12203. */
- if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
+ if (oa->o_valid & OBD_MD_FLSIZE &&
oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
static char mdsinum[48];
- if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
+ if (oa->o_valid & OBD_MD_FLFID)
snprintf(mdsinum, sizeof(mdsinum) - 1,
- "of parent "DFID, oinfo->oi_oa->o_parent_seq,
- oinfo->oi_oa->o_parent_oid, 0);
+ "of parent "DFID, oa->o_parent_seq,
+ oa->o_parent_oid, 0);
else
mdsinum[0] = '\0';
CERROR("%s: setattr from %s trying to truncate object "DFID
- " %s\n", exp->exp_obd->obd_name,
- obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
+ " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
+ PFID(fid), mdsinum);
GOTO(out, rc = -EPERM);
}
- fo = ofd_object_find(env, ofd, &info->fti_fid);
+ fo = ofd_object_find_exists(env, ofd, fid);
if (IS_ERR(fo)) {
CERROR("%s: can't find object "DFID"\n",
- exp->exp_obd->obd_name, PFID(&info->fti_fid));
+ ofd_name(ofd), PFID(fid));
GOTO(out, rc = PTR_ERR(fo));
}
- la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
+ la_from_obdo(&info->fti_attr, oa, oa->o_valid);
info->fti_attr.la_valid &= ~LA_TYPE;
if (oa->o_valid & OBD_MD_FLFID) {
if (rc)
GOTO(out_unlock, rc);
- res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
- if (res != NULL) {
- ldlm_res_lvbo_update(res, NULL, 0);
- ldlm_resource_putref(res);
- }
-
- oinfo->oi_oa->o_valid = OBD_MD_FLID;
-
- /* Quota release needs uid/gid info */
- rc = ofd_attr_get(env, fo, &info->fti_attr);
- obdo_from_la(oinfo->oi_oa, &info->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
- ofd_info2oti(info, oti);
-
- ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
+ ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
EXIT;
out_unlock:
ofd_object_put(env, fo);
out:
- return rc;
-}
-
-static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo, struct obd_trans_info *oti,
- struct ptlrpc_request_set *rqset)
-{
- struct ofd_thread_info *info;
- struct ofd_device *ofd = ofd_exp(exp);
- struct ldlm_namespace *ns = ofd->ofd_namespace;
- struct ldlm_resource *res;
- struct ofd_object *fo;
- struct filter_fid *ff = NULL;
- int rc = 0;
-
- ENTRY;
-
- info = ofd_info_init(env, exp);
- ofd_oti2info(info, oti);
-
- 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, ostid_seq(&oinfo->oi_oa->o_oi),
- oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
- if (rc)
- GOTO(out_env, rc);
-
- fo = ofd_object_find(env, ofd, &info->fti_fid);
- if (IS_ERR(fo)) {
- CERROR("%s: error finding object "DFID": rc = %ld\n",
- exp->exp_obd->obd_name, PFID(&info->fti_fid),
- PTR_ERR(fo));
- GOTO(out_env, rc = PTR_ERR(fo));
- }
-
- LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
- if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
- /* Truncate case */
- oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
- } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
- oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
- }
-
- la_from_obdo(&info->fti_attr, oinfo->oi_oa,
- OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
- info->fti_attr.la_valid &= ~LA_TYPE;
- info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
- info->fti_attr.la_valid |= LA_SIZE;
-
- if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
- ff = &info->fti_mds_fid;
- ofd_prepare_fidea(ff, oinfo->oi_oa);
- }
-
- rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
- oinfo->oi_policy.l_extent.end, &info->fti_attr,
- ff);
- if (rc)
- GOTO(out, rc);
-
- res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
- if (res != NULL) {
- ldlm_res_lvbo_update(res, NULL, 0);
- ldlm_resource_putref(res);
+ if (rc == 0) {
+ /* we do not call this before to avoid lu_object_find() in
+ * ->lvbo_update() holding another reference on the object.
+ * otherwise concurrent destroy can make the object unavailable
+ * for 2nd lu_object_find() waiting for the first reference
+ * to go... deadlock! */
+ res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
+ if (!IS_ERR(res)) {
+ ldlm_res_lvbo_update(res, NULL, 0);
+ ldlm_resource_putref(res);
+ }
}
- oinfo->oi_oa->o_valid = OBD_MD_FLID;
- /* Quota release needs uid/gid info */
- rc = ofd_attr_get(env, fo, &info->fti_attr);
- obdo_from_la(oinfo->oi_oa, &info->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
- ofd_info2oti(info, oti);
-
- ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
- EXIT;
-out:
- ofd_object_put(env, fo);
-out_env:
return rc;
}
-static int ofd_destroy_by_fid(const struct lu_env *env,
- struct ofd_device *ofd,
- const struct lu_fid *fid, int orphan)
+int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
+ const struct lu_fid *fid, int orphan)
{
struct ofd_thread_info *info = ofd_info(env);
struct lustre_handle lockh;
ENTRY;
- fo = ofd_object_find(env, ofd, fid);
+ fo = ofd_object_find_exists(env, ofd, fid);
if (IS_ERR(fo))
RETURN(PTR_ERR(fo));
- if (!ofd_object_exists(fo))
- GOTO(out, rc = -ENOENT);
/* Tell the clients that the object is gone now and that they should
* throw away any cached pages. */
rc = ofd_object_destroy(env, fo, orphan);
EXIT;
-out:
+
ofd_object_put(env, fo);
RETURN(rc);
}
-int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
- struct obdo *oa, struct lov_stripe_md *md,
- struct obd_trans_info *oti, struct obd_export *md_exp,
- void *capa)
+/* needed by echo client only for now, RPC handler uses ofd_destroy_hdl() */
+int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa, struct lov_stripe_md *md,
+ struct obd_trans_info *oti, struct obd_export *md_exp,
+ void *capa)
{
struct ofd_device *ofd = ofd_exp(exp);
- struct ofd_thread_info *info;
- obd_count count;
+ struct lu_fid *fid = &oa->o_oi.oi_fid;
int rc = 0;
ENTRY;
- info = ofd_info_init(env, exp);
- ofd_oti2info(info, oti);
-
- if (!(oa->o_valid & OBD_MD_FLGROUP))
- ostid_set_seq_mdt0(&oa->o_oi);
-
- /* check that o_misc makes sense */
- if (oa->o_valid & OBD_MD_FLOBJCOUNT)
- count = oa->o_misc;
- else
- count = 1; /* default case - single destroy */
-
- /**
- * There can be sequence of objects to destroy. Therefore this request
- * may have multiple transaction involved in. It is OK, we need only
- * the highest used transno to be reported back in reply but not for
- * replays, they must report their transno
- */
- 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;
-
- 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,
- "%s: destroying non-existent object "DFID"\n",
- ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
- /* rewrite rc with -ENOENT only if it is 0 */
- if (rc == 0)
- rc = lrc;
- } else if (lrc != 0) {
- CERROR("%s: error destroying object "DFID": %d\n",
- ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
- rc);
- rc = lrc;
- }
- count--;
- ostid_inc_id(&oa->o_oi);
- }
+ ofd_info_init(env, exp);
- /* if we have transaction then there were some deletions, we don't
- * need to return ENOENT in that case because it will not wait
- * for commit of these deletions. The ENOENT must be returned only
- * if there were no transations.
- */
+ CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
+
+ rc = ofd_destroy_by_fid(env, ofd, fid, 0);
if (rc == -ENOENT) {
- if (info->fti_transno != 0)
- rc = 0;
+ CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
+ ofd_name(ofd), PFID(fid));
+ GOTO(out, rc);
} else if (rc != 0) {
- /*
- * If we have at least one transaction then llog record
- * on server will be removed upon commit, so for rc != 0
- * we return no transno and llog record will be reprocessed.
- */
- info->fti_transno = 0;
+ CERROR("%s: error destroying object "DFID": %d\n",
+ ofd_name(ofd), PFID(fid), rc);
+ GOTO(out, rc);
}
- ofd_info2oti(info, oti);
+ EXIT;
out:
- RETURN(rc);
-}
-
-static int ofd_orphans_destroy(const struct lu_env *env,
- struct obd_export *exp, struct ofd_device *ofd,
- struct obdo *oa)
-{
- struct ofd_thread_info *info = ofd_info(env);
- obd_id last;
- 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_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
-
- last = ofd_seq_last_oid(oseq);
- LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
- " to "DOSTID"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
- end_id + 1, ostid_seq(&oa->o_oi), last);
-
- 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("%s: error destroying precreated id "DOSTID
- ": rc = %d\n", ofd_name(ofd), POSTID(&oi), rc);
- if (!skip_orphan) {
- 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 ((ostid_id(&oi) & 511) == 0)
- ofd_seq_last_oid_write(env, ofd, oseq);
- }
- }
- CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
- ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
- if (!skip_orphan) {
- rc = ofd_seq_last_oid_write(env, ofd, oseq);
- } else {
- /* don't reuse orphan object, return last used objid */
- ostid_set_id(&oa->o_oi, last);
- rc = 0;
- }
-out_put:
- ofd_seq_put(env, oseq);
- RETURN(rc);
+ return rc;
}
-int ofd_create(const struct lu_env *env, struct obd_export *exp,
- struct obdo *oa, struct lov_stripe_md **ea,
- struct obd_trans_info *oti)
+/* needed by echo client only for now, RPC handler uses ofd_create_hdl()
+ * It is much simpler and just create objects */
+int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa, struct lov_stripe_md **ea,
+ struct obd_trans_info *oti)
{
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
- obd_seq seq = ostid_seq(&oa->o_oi);
+ obd_seq seq = ostid_seq(&oa->o_oi);
struct ofd_seq *oseq;
- int rc = 0, diff;
- int sync_trans = 0;
+ int rc = 0, diff = 1;
+ obd_id next_id;
+ int count;
ENTRY;
info = ofd_info_init(env, exp);
- ofd_oti2info(info, oti);
- LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
+ LASSERT(seq == FID_SEQ_ECHO);
LASSERT(oa->o_valid & OBD_MD_FLGROUP);
CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
+ down_read(&ofd->ofd_lastid_rwsem);
+ /* Currently, for safe, we do not distinguish which LAST_ID is broken,
+ * we may do that in the future.
+ * Return -ENOSPC until the LAST_ID rebuilt. */
+ if (unlikely(ofd->ofd_lastid_rebuilding))
+ GOTO(out_sem, rc = -ENOSPC);
+
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);
+ GOTO(out_sem, rc = -EINVAL);
}
- if ((oa->o_valid & OBD_MD_FLFLAGS) &&
- (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
- if (!ofd_obd(ofd)->obd_recovering ||
- 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 */
- GOTO(out_nolock, rc = 0);
- }
- /* former ofd_handle_precreate */
- if ((oa->o_valid & OBD_MD_FLFLAGS) &&
- (oa->o_flags & OBD_FL_DELORPHAN)) {
- /* destroy orphans */
- if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
- CERROR("%s: dropping old orphan cleanup request\n",
- ofd_name(ofd));
- GOTO(out_nolock, rc = 0);
- }
- /* This causes inflight precreates to abort and drop lock */
- 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 = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
- CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
- 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);
- oseq->os_destroys_in_progress = 0;
- } else {
- /* XXX: Used by MDS for the first time! */
- oseq->os_destroys_in_progress = 0;
- }
- } else {
- 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);
- GOTO(out, rc = 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 = 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);
- }
- }
+ mutex_lock(&oseq->os_create_lock);
+ rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
+ if (rc < 0) {
+ CDEBUG(D_HA, "%s: failed to acquire grant space for "
+ "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
+ diff = 0;
+ GOTO(out, rc);
}
- if (diff > 0) {
- cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
- obd_id next_id;
- int created = 0;
- int count;
-
- if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
- !(oa->o_flags & OBD_FL_DELORPHAN)) {
- /* don't enforce grant during orphan recovery */
- rc = ofd_grant_create(env,
- ofd_obd(ofd)->obd_self_export,
- &diff);
- if (rc) {
- 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_seq_last_oid(oseq) + 1;
- count = ofd_precreate_batch(ofd, diff);
-
- CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
- " at "LPU64"\n", ofd_obd(ofd)->obd_name,
- count, ostid_seq(&oa->o_oi), next_id);
-
- if (cfs_time_after(jiffies, enough_time)) {
- LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
- " created at a rate of %d/s\n",
- ofd_obd(ofd)->obd_name,
- created, diff + created,
- created / DISK_TIMEOUT);
- break;
- }
-
- rc = ofd_precreate_objects(env, ofd, next_id,
- oseq, count, sync_trans);
- if (rc > 0) {
- created += rc;
- diff -= rc;
- } else if (rc < 0) {
- break;
- }
- }
- if (created > 0)
- /* some objects got created, we can return
- * them, even if last creation failed */
- rc = 0;
- else
- CERROR("%s: unable to precreate: rc = %d\n",
- ofd_name(ofd), rc);
+ next_id = ofd_seq_last_oid(oseq) + 1;
+ count = ofd_precreate_batch(ofd, diff);
+ rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
+ if (rc < 0) {
+ CERROR("%s: unable to precreate: rc = %d\n",
+ ofd_name(ofd), rc);
+ } else {
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) ||
- !(oa->o_flags & OBD_FL_DELORPHAN))
- ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
- rc);
+ rc = 0;
}
- ofd_info2oti(info, oti);
+ ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
out:
mutex_unlock(&oseq->os_create_lock);
-out_nolock:
if (rc == 0 && ea != NULL) {
struct lov_stripe_md *lsm = *ea;
lsm->lsm_oi = oa->o_oi;
}
ofd_seq_put(env, oseq);
+
+out_sem:
+ up_read(&ofd->ofd_lastid_rwsem);
RETURN(rc);
}
-int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo)
+/* needed by echo client only for now, RPC handler uses ofd_getattr_hdl() */
+int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
+ struct obd_info *oinfo)
{
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
+ struct lu_fid *fid = &oinfo->oi_oa->o_oi.oi_fid;
struct ofd_object *fo;
int rc = 0;
info = ofd_info_init(env, exp);
- 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);
-
- fo = ofd_object_find(env, ofd, &info->fti_fid);
+ fo = ofd_object_find_exists(env, ofd, fid);
if (IS_ERR(fo))
GOTO(out, rc = PTR_ERR(fo));
- if (!ofd_object_exists(fo))
- GOTO(out_put, rc = -ENOENT);
LASSERT(fo != NULL);
rc = ofd_attr_get(env, fo, &info->fti_attr);
}
}
-out_put:
ofd_object_put(env, fo);
out:
RETURN(rc);
}
-static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo, obd_size start, obd_size end,
- struct ptlrpc_request_set *set)
-{
- struct ofd_device *ofd = ofd_exp(exp);
- struct ofd_thread_info *info;
- struct ofd_object *fo;
- int rc = 0;
-
- ENTRY;
-
- /* if no objid is specified, it means "sync whole filesystem" */
- if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
- rc = dt_sync(env, ofd->ofd_osd);
- GOTO(out, rc);
- }
-
- info = ofd_info_init(env, exp);
- 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_OSS_TRUNC);
- if (rc)
- GOTO(out, rc);
-
- fo = ofd_object_find(env, ofd, &info->fti_fid);
- if (IS_ERR(fo)) {
- CERROR("%s: error finding object "DFID": rc = %ld\n",
- exp->exp_obd->obd_name, PFID(&info->fti_fid),
- PTR_ERR(fo));
- GOTO(out, rc = PTR_ERR(fo));
- }
-
- if (!ofd_object_exists(fo))
- GOTO(put, rc = -ENOENT);
-
- if (dt_version_get(env, ofd_object_child(fo)) >
- ofd_obd(ofd)->obd_last_committed) {
- rc = dt_object_sync(env, ofd_object_child(fo));
- if (rc)
- GOTO(put, rc);
- }
-
- oinfo->oi_oa->o_valid = OBD_MD_FLID;
- rc = ofd_attr_get(env, fo, &info->fti_attr);
- obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
-
- ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
- EXIT;
-put:
- ofd_object_put(env, fo);
-out:
- return rc;
-}
-
static int ofd_ioc_get_obj_version(const struct lu_env *env,
struct ofd_device *ofd, void *karg)
{
ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
- rc = ostid_to_fid(&fid, &ostid, 0);
+ rc = ostid_to_fid(&fid, &ostid,
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
GOTO(out, rc);
} else {
}
lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
- lsp.lsp_namespace = ofd->ofd_namespace;
+ lsp.lsp_index_valid = 0;
rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
break;
}
case OBD_IOC_STOP_LFSCK: {
- rc = lfsck_stop(&env, ofd->ofd_osd, false);
+ struct obd_ioctl_data *data = karg;
+ struct lfsck_stop stop;
+
+ stop.ls_status = LS_STOPPED;
+ /* Old lfsck utils may pass NULL @stop. */
+ if (data->ioc_inlbuf1 == NULL)
+ stop.ls_flags = 0;
+ else
+ stop.ls_flags =
+ ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
+
+ rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
break;
}
case OBD_IOC_GET_OBJ_VERSION:
GOTO(out, rc = PTR_ERR(th));
rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
- info->fti_buf.lb_len, info->fti_off, th);
+ &info->fti_buf, info->fti_off, th);
if (rc == 0) {
th->th_sync = 1; /* sync IO is needed */
rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
.o_connect = ofd_obd_connect,
.o_reconnect = ofd_obd_reconnect,
.o_disconnect = ofd_obd_disconnect,
- .o_set_info_async = ofd_set_info_async,
- .o_get_info = ofd_get_info,
- .o_create = ofd_create,
+ .o_create = ofd_echo_create,
.o_statfs = ofd_statfs,
- .o_setattr = ofd_setattr,
+ .o_setattr = ofd_echo_setattr,
.o_preprw = ofd_preprw,
.o_commitrw = ofd_commitrw,
- .o_destroy = ofd_destroy,
+ .o_destroy = ofd_echo_destroy,
.o_init_export = ofd_init_export,
.o_destroy_export = ofd_destroy_export,
.o_postrecov = ofd_obd_postrecov,
- .o_punch = ofd_punch,
- .o_getattr = ofd_getattr,
- .o_sync = ofd_sync,
+ .o_getattr = ofd_echo_getattr,
.o_iocontrol = ofd_iocontrol,
.o_precleanup = ofd_precleanup,
.o_ping = ofd_ping,
.o_health_check = ofd_health_check,
.o_quotactl = ofd_quotactl,
+ .o_set_info_async = ofd_set_info_async,
+ .o_get_info = ofd_get_info,
};