*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2014 Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* lustre/ofd/ofd_obd.c
*
- * Author: Andreas Dilger <adilger@whamcloud.com>
- * Author: Alex Zhuravlev <bzzz@whamcloud.com>
- * Author: Mike Pershin <tappro@whamcloud.com>
+ * This file contains OBD API methods for OBD Filter Device (OFD) which are
+ * used for export handling, configuration purposes and recovery.
+ * Several methods are used by ECHO client only since it still uses OBD API.
+ * Such methods have _echo_ prefix in name.
+ *
+ * Author: Andreas Dilger <andreas.dilger@intel.com>
+ * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
+ * Author: Mikhail Pershin <mike.pershin@intel.com>
*/
#define DEBUG_SUBSYSTEM S_FILTER
#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>
+#include <lustre_nodemap.h>
+/**
+ * Initialize OFD per-export statistics.
+ *
+ * This function sets up procfs entries for various OFD export counters. These
+ * counters are for per-client statistics tracked on the server.
+ *
+ * \param[in] ofd OFD device
+ * \param[in] exp OBD export
+ * \param[in] client_nid NID of client
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_export_stats_init(struct ofd_device *ofd,
- struct obd_export *exp, void *client_nid)
+ struct obd_export *exp,
+ lnet_nid_t *client_nid)
{
struct obd_device *obd = ofd_obd(ofd);
struct nid_stat *stats;
- int num_stats;
- int rc, newnid = 0;
-
+ int rc;
ENTRY;
+ LASSERT(obd->obd_uses_nid_stats);
+
if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
/* Self-export gets no proc entry */
RETURN(0);
- rc = lprocfs_exp_setup(exp, client_nid, &newnid);
- if (rc) {
- /* Mask error for already created
- * /proc entries */
- if (rc == -EALREADY)
- rc = 0;
- RETURN(rc);
- }
-
- if (newnid == 0)
- RETURN(0);
+ rc = lprocfs_exp_setup(exp, client_nid);
+ if (rc != 0)
+ /* Mask error for already created /proc entries */
+ RETURN(rc == -EALREADY ? 0 : rc);
stats = exp->exp_nid_stats;
- LASSERT(stats != NULL);
-
- num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
- LPROC_OFD_LAST - 1;
-
- stats->nid_stats = lprocfs_alloc_stats(num_stats,
- LPROCFS_STATS_FLAG_NOPERCPU);
+ stats->nid_stats = lprocfs_alloc_stats(NUM_OBD_STATS +
+ LPROC_OFD_STATS_LAST,
+ LPROCFS_STATS_FLAG_NOPERCPU);
if (stats->nid_stats == NULL)
- return -ENOMEM;
+ 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);
- rc = lprocfs_register_stats(stats->nid_proc, "stats",
- stats->nid_stats);
- if (rc)
- GOTO(clean, rc);
+ ofd_stats_counter_init(stats->nid_stats);
- rc = lprocfs_nid_ldlm_stats_init(stats);
- if (rc) {
+ rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
+ if (rc != 0) {
lprocfs_free_stats(&stats->nid_stats);
- GOTO(clean, rc);
+ GOTO(out, rc);
}
- RETURN(0);
-clean:
- return rc;
+ rc = lprocfs_nid_ldlm_stats_init(stats);
+ if (rc != 0)
+ GOTO(out, rc);
+
+out:
+ RETURN(rc);
}
+/**
+ * Match client and OST server connection feature flags.
+ *
+ * Compute the compatibility flags for a connection request based on
+ * features mutually supported by client and server.
+ *
+ * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
+ * must not be updated here, otherwise a partially initialized value may
+ * be exposed. After the connection request is successfully processed,
+ * the top-level tgt_connect() request handler atomically updates the export
+ * connect flags from the obd_connect_data::ocd_connect_flags field of the
+ * reply. \see tgt_connect().
+ *
+ * \param[in] env execution environment
+ * \param[in] exp the obd_export associated with this
+ * client/target pair
+ * \param[in] data stores data for this connect request
+ * \param[in] new_connection is this connection new or not
+ *
+ * \retval 0 if success
+ * \retval -EPROTO client and server feature requirements are
+ * incompatible
+ * \retval -EBADF OST index in connect request doesn't match
+ * real OST index
+ */
static int ofd_parse_connect_data(const struct lu_env *env,
struct obd_export *exp,
struct obd_connect_data *data,
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);
RETURN(0);
}
+/**
+ * Re-initialize export upon client reconnection.
+ *
+ * This function parses connection data from reconnect and resets
+ * export statistics.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export
+ * \param[in] obd OFD device
+ * \param[in] cluuid NID of client
+ * \param[in] data connection data from request
+ * \param[in] localdata client NID
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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 obd_connect_data *data,
+ void *client_nid)
{
struct ofd_device *ofd;
int rc;
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);
+ ofd_export_stats_init(ofd, exp, client_nid);
+
+ nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
RETURN(rc);
}
+/**
+ * Initialize new client connection.
+ *
+ * This function handles new connection to the OFD. The new export is
+ * created (in context of class_connect()) and persistent client data is
+ * initialized on storage.
+ *
+ * \param[in] env execution environment
+ * \param[out] _exp stores pointer to new export
+ * \param[in] obd OFD device
+ * \param[in] cluuid client UUID
+ * \param[in] data connection data from request
+ * \param[in] localdata client NID
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_obd_connect(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;
struct lustre_handle conn = { 0 };
int rc;
+ lnet_nid_t *client_nid;
ENTRY;
if (_exp == NULL || obd == NULL || cluuid == NULL)
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);
+ if (localdata != NULL) {
+ client_nid = localdata;
+ nodemap_add_member(*client_nid, exp);
+ }
+
if (obd->obd_replayable) {
struct tg_export_data *ted = &exp->exp_target_data;
}
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) {
+ nodemap_del_member(exp);
class_disconnect(exp);
*_exp = NULL;
} else {
RETURN(rc);
}
-static int ofd_obd_disconnect(struct obd_export *exp)
+/**
+ * Disconnect a connected client.
+ *
+ * This function terminates the client connection. The client export is
+ * disconnected (cleaned up) and client data on persistent storage is removed.
+ *
+ * \param[in] exp OBD export
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+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;
if (!(exp->exp_flags & OBD_OPT_FORCE))
ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
+ nodemap_del_member(exp);
rc = server_disconnect_export(exp);
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);
}
+/**
+ * Implementation of obd_ops::o_init_export.
+ *
+ * This function is called from class_new_export() and initializes
+ * the OFD-specific data for new export.
+ *
+ * \param[in] exp OBD export
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_init_export(struct obd_export *exp)
{
int rc;
spin_lock_init(&exp->exp_filter_data.fed_lock);
- CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
+ 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);
return rc;
}
+/**
+ * Implementation of obd_ops::o_destroy_export.
+ *
+ * This function is called from class_export_destroy() to cleanup
+ * the OFD-specific data for export being destroyed.
+ *
+ * \param[in] exp OBD export
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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"
if (!(exp->exp_flags & OBD_OPT_FORCE))
ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
- LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
+ LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
return 0;
}
+/**
+ * Notify all devices in server stack about recovery completion.
+ *
+ * This function calls ldo_recovery_complete() for all lower devices in the
+ * server stack so they will be prepared for normal operations.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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);
}
-int ofd_obd_postrecov(struct obd_device *obd)
+/**
+ * Implementation of obd_ops::o_postrecov.
+ *
+ * This function is called from target_finish_recovery() upon recovery
+ * completion.
+ *
+ * \param[in] obd OBD device of OFD
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_obd_postrecov(struct obd_device *obd)
{
struct lu_env env;
struct lu_device *ldev = obd->obd_lu_dev;
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);
-}
-
+/**
+ * Implementation of obd_ops::o_set_info_async.
+ *
+ * This function is not called from request handler, it is only used by
+ * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
+ * \see ofd_set_info_hdl() for request handler function.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[in] keylen length of \a key
+ * \param[in] key key name
+ * \param[in] vallen length of \a val
+ * \param[in] val the \a key value
+ * \param[in] set not used in OFD
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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);
}
+/**
+ * Implementation of obd_ops::o_get_info.
+ *
+ * This function is not called from request handler, it is only used by
+ * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
+ *
+ * \see ofd_get_info_hdl() for request handler function.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[in] keylen length of \a key
+ * \param[in] key key name
+ * \param[out] vallen length of key value
+ * \param[out] val the key value to return
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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)
+ __u32 keylen, void *key, __u32 *vallen, void *val)
{
- 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 fiemap *fiemap = val;
+ int rc = 0;
ENTRY;
ofd = ofd_exp(exp);
- if (KEY_IS(KEY_BLOCKSIZE)) {
- __u32 *blocksize = val;
- if (blocksize) {
- if (*vallen < sizeof(*blocksize))
- RETURN(-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))
- RETURN(-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);
- RETURN(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);
- RETURN(-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;
- struct ofd_device *ofd = ofd_exp(exp);
- struct ofd_object *fo;
- struct ll_fiemap_info_key *fm_key = key;
-
- if (val == NULL) {
- *vallen = fiemap_count_to_size(
- fm_key->fiemap.fm_extent_count);
- RETURN(0);
- }
-
+ if (KEY_IS(KEY_FIEMAP)) {
info = ofd_info_init(env, exp);
- rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0);
- if (rc != 0)
- RETURN(rc);
- CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
- PFID(&info->fti_fid));
-
- fo = ofd_object_find(env, ofd, &info->fti_fid);
- if (IS_ERR(fo)) {
- CERROR("%s: error finding object "DFID"\n",
- exp->exp_obd->obd_name, PFID(&info->fti_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 lu_env env;
- 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);
- RETURN(0);
- }
-
- if (*vallen < sizeof(*fid))
- RETURN(-EOVERFLOW);
- rc = lu_env_init(&env, LCT_DT_THREAD);
+ rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
RETURN(rc);
- 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_fini, 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);
-out_fini:
- lu_env_fini(&env);
+ 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;
}
RETURN(rc);
}
-/** helper function for statfs, also used by grant code */
+/**
+ * Get file system statistics of OST server.
+ *
+ * Helper function for ofd_statfs(), also used by grant code.
+ * Implements caching for statistics to avoid calling OSD device each time.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[out] osfs statistic data to return
+ * \param[in] max_age maximum age for cached data
+ * \param[in] from_cache show that data was get from cache or not
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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) {
- obd_size unstable;
+ u64 unstable;
/* statfs data are too old, get up-to-date one.
* we must be cautious here since multiple threads might be
* 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);
* w/o the ofd_osfs_lock. Those ones got added to
* the cached statfs data that we are about to crunch.
* Take them into account in the new statfs data */
- osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
+ osfs->os_bavail -= min_t(u64, osfs->os_bavail,
unstable >> ofd->ofd_blockbits);
/* However, we don't really know if those writes got
* accounted in the statfs call, so tell
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)
+/**
+ * Implementation of obd_ops::o_statfs.
+ *
+ * This function returns information about a storage file system.
+ * It is called from several places by using the OBD API as well as
+ * by direct call, e.g. from request handler.
+ *
+ * \see ofd_statfs_hdl() for request handler function.
+ *
+ * Report also the state of the OST to the caller in osfs->os_state
+ * (OS_STATE_READONLY, OS_STATE_DEGRADED).
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[out] osfs statistic data to return
+ * \param[in] max_age maximum age for cached data
+ * \param[in] flags not used in OFD
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+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;
osfs->os_bfree << ofd->ofd_blockbits,
osfs->os_bavail << ofd->ofd_blockbits);
- osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
+ osfs->os_bavail -= min_t(u64, osfs->os_bavail,
((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
osfs->os_bsize - 1) >> ofd->ofd_blockbits));
struct filter_export_data *fed;
fed = &obd->obd_self_export->exp_filter_data;
- osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
+ osfs->os_bavail -= min_t(u64, osfs->os_bavail,
fed->fed_grant >> ofd->ofd_blockbits);
}
return rc;
}
-int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo, struct obd_trans_info *oti)
+/**
+ * Implementation of obd_ops::o_setattr.
+ *
+ * This function is only used by ECHO client when it is run on top of OFD,
+ * \see ofd_setattr_hdl() for request handler function.
+
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[in] oa setattr parameters
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa)
{
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 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 &&
- oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
+ if (oa->o_valid & OBD_MD_FLSIZE) {
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)
+/**
+ * Destroy OFD object by its FID.
+ *
+ * Supplemental function to destroy object by FID, it is used by request
+ * handler and by ofd_echo_destroy() below to find object by FID, lock it
+ * and call ofd_object_destroy() finally.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] fid FID of object
+ * \param[in] orphan set if object being destroyed is an orphan
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+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;
- __u64 flags = LDLM_AST_DISCARD_DATA, rc = 0;
+ __u64 flags = LDLM_FL_AST_DISCARD_DATA;
+ __u64 rc = 0;
ldlm_policy_data_t policy = {
.l_extent = { 0, OBD_OBJECT_EOF }
};
ENTRY;
- fo = ofd_object_find(env, ofd, fid);
+ fo = ofd_object_find_exists(env, ofd, fid);
if (IS_ERR(fo))
RETURN(PTR_ERR(fo));
LASSERT(fo != NULL);
rc = ofd_object_destroy(env, fo, orphan);
+ EXIT;
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)
+/**
+ * Implementation of obd_ops::o_destroy.
+ *
+ * This function is only used by ECHO client when it is run on top of OFD,
+ * \see ofd_destroy_hdl() for request handler function.
+
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[in] oa obdo structure with FID
+ *
+ * Note: this is OBD API method which is common API for server OBDs and
+ * client OBDs. Thus some parameters used in client OBDs may not be used
+ * on server OBDs and vice versa.
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa)
{
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_INFO("%s: deleting orphan objects from "LPX64":"LPU64
- " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
- end_id + 1, 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("error destroying precreated id "DOSTID": %d\n",
- 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_objids"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)
+/**
+ * Implementation of obd_ops::o_create.
+ *
+ * This function is only used by ECHO client when it is run on top of OFD
+ * and just creates an object.
+ * \see ofd_create_hdl() for request handler function.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[in] oa obdo structure with FID sequence to use
+ *
+ * Note: this is OBD API method which is common API for server OBDs and
+ * client OBDs. Thus some parameters used in client OBDs may not be used
+ * on server OBDs and vice versa.
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa)
{
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
- obd_seq seq = ostid_seq(&oa->o_oi);
+ u64 seq = ostid_seq(&oa->o_oi);
struct ofd_seq *oseq;
- int rc = 0, diff;
- int sync_trans = 0;
+ int rc = 0, diff = 1;
+ u64 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)
+/**
+ * Implementation of obd_ops::o_getattr.
+ *
+ * This function is only used by ECHO client when it is run on top of OFD
+ * and returns attributes of object.
+ * \see ofd_getattr_hdl() for request handler function.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of OFD device
+ * \param[in,out] oa contains FID of object to get attributes from and
+ * is used to return attributes back
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa)
{
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
+ struct lu_fid *fid = &oa->o_oi.oi_fid;
struct ofd_object *fo;
- __u64 curr_version;
int rc = 0;
ENTRY;
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));
+
LASSERT(fo != NULL);
rc = ofd_attr_get(env, fo, &info->fti_attr);
- oinfo->oi_oa->o_valid = OBD_MD_FLID;
- if (rc == 0)
- obdo_from_la(oinfo->oi_oa, &info->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
-
- /* Store object version in reply */
- curr_version = dt_version_get(env, ofd_object_child(fo));
- if ((__s64)curr_version != -EOPNOTSUPP) {
- oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
- oinfo->oi_oa->o_data_version = curr_version;
- }
- 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));
- }
+ oa->o_valid = OBD_MD_FLID;
+ if (rc == 0) {
+ __u64 curr_version;
- if (!ofd_object_exists(fo))
- GOTO(put, rc = -ENOENT);
+ obdo_from_la(oa, &info->fti_attr,
+ OFD_VALID_FLAGS | LA_UID | LA_GID);
- 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);
+ /* Store object version in reply */
+ curr_version = dt_version_get(env, ofd_object_child(fo));
+ if ((__s64)curr_version != -EOPNOTSUPP) {
+ oa->o_valid |= OBD_MD_FLDATAVERSION;
+ oa->o_data_version = curr_version;
+ }
}
- 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;
+ RETURN(rc);
}
+/**
+ * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
+ *
+ * This is supplemental function for ofd_iocontrol() to return object
+ * version for lctl tool.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[out] karg ioctl data
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
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 {
return rc;
}
-int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
- void *karg, void *uarg)
+/**
+ * Implementation of obd_ops::o_iocontrol.
+ *
+ * This is OFD ioctl handling function which is primary interface for
+ * Lustre tools like lfs, lctl and lfsck.
+ *
+ * \param[in] cmd ioctl command
+ * \param[in] exp OBD export of OFD
+ * \param[in] len not used
+ * \param[in,out] karg buffer with data
+ * \param[in] uarg not used
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
+ void *karg, void *uarg)
{
struct lu_env env;
struct ofd_device *ofd = ofd_exp(exp);
switch (cmd) {
case OBD_IOC_ABORT_RECOVERY:
CERROR("%s: aborting recovery\n", obd->obd_name);
+ obd->obd_force_abort_recovery = 1;
target_stop_recovery_thread(obd);
break;
case OBD_IOC_SYNC:
if (rc == 0)
rc = dt_ro(&env, ofd->ofd_osd);
break;
+ case OBD_IOC_START_LFSCK: {
+ struct obd_ioctl_data *data = karg;
+ struct lfsck_start_param lsp;
+
+ if (unlikely(data == NULL)) {
+ rc = -EINVAL;
+ break;
+ }
+
+ lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
+ lsp.lsp_index_valid = 0;
+ rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
+ break;
+ }
+ case OBD_IOC_STOP_LFSCK: {
+ 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:
rc = ofd_ioc_get_obj_version(&env, ofd, karg);
break;
default:
- CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
+ CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
rc = -ENOTTY;
}
RETURN(rc);
}
+/**
+ * Implementation of obd_ops::o_precleanup.
+ *
+ * This function stops device activity before shutting it down. It is called
+ * from a cleanup function upon forceful device cleanup. For OFD there are no
+ * special actions, it just invokes target_recovery_cleanup().
+ *
+ * \param[in] obd OBD device of OFD
+ * \param[in] stage cleanup stage
+ *
+ * \retval 0
+ */
static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
{
int rc = 0;
RETURN(rc);
}
+/**
+ * Implementation of obd_ops::o_ping.
+ *
+ * This is OFD-specific part of OBD_PING request handling.
+ * It controls Filter Modification Data (FMD) expiration each time PING is
+ * received.
+ *
+ * \see ofd_fmd_expire() and ofd_fmd.c for details
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of client
+ *
+ * \retval 0
+ */
static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
{
ofd_fmd_expire(exp);
return 0;
}
+/**
+ * Implementation of obd_ops::o_health_check.
+ *
+ * This function checks the OFD device health - ability to respond on
+ * incoming requests. There are two health_check methods:
+ * - get statfs from the OSD. It checks just responsiveness of
+ * bottom device
+ * - do write attempt on bottom device to check it is fully operational and
+ * is not stuck. This is expensive method and requires special configuration
+ * option --enable-health-write while building Lustre, it is turned off
+ * by default.
+ *
+ * \param[in] nul not used
+ * \param[in] obd OBD device of OFD
+ *
+ * \retval 0 if successful
+ * \retval negative value in case of error
+ */
static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
{
struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
GOTO(out, rc = -EROFS);
#ifdef USE_HEALTH_CHECK_WRITE
- OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
+ OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
if (info->fti_buf.lb_buf == NULL)
GOTO(out, rc = -ENOMEM);
- info->fti_buf.lb_len = CFS_PAGE_SIZE;
+ info->fti_buf.lb_len = PAGE_CACHE_SIZE;
info->fti_off = 0;
th = dt_trans_create(&env, ofd->ofd_osd);
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);
}
dt_trans_stop(&env, ofd->ofd_osd, th);
- OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
+ OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
#endif
return !!rc;
}
-/*
- * Handle quota control requests to consult current usage/limit.
- *
- * \param obd - is the obd device associated with the ofd
- * \param exp - is the client's export
- * \param oqctl - is the obd_quotactl request to be processed
- */
-static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
- struct obd_quotactl *oqctl)
-{
- struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
- struct lu_env env;
- int rc;
- ENTRY;
-
- /* report success for quota on/off for interoperability with current MDT
- * stack */
- if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
- RETURN(0);
-
- rc = lu_env_init(&env, LCT_DT_THREAD);
- if (rc)
- RETURN(rc);
-
- rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
- lu_env_fini(&env);
-
- RETURN(rc);
-}
-
struct obd_ops ofd_obd_ops = {
.o_owner = THIS_MODULE,
.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,
};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff