4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2014 Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ofd/ofd_obd.c
34 * This file contains OBD API methods for OBD Filter Device (OFD) which are
35 * used for export handling, configuration purposes and recovery.
36 * Several methods are used by ECHO client only since it still uses OBD API.
37 * Such methods have _echo_ prefix in name.
39 * Author: Andreas Dilger <andreas.dilger@intel.com>
40 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
41 * Author: Mikhail Pershin <mike.pershin@intel.com>
44 #define DEBUG_SUBSYSTEM S_FILTER
46 #include <lustre/lustre_idl.h>
47 #include "ofd_internal.h"
48 #include <obd_cksum.h>
49 #include <lustre_ioctl.h>
50 #include <lustre_quota.h>
51 #include <lustre_lfsck.h>
52 #include <lustre_nodemap.h>
55 * Initialize OFD per-export statistics.
57 * This function sets up procfs entries for various OFD export counters. These
58 * counters are for per-client statistics tracked on the server.
60 * \param[in] ofd OFD device
61 * \param[in] exp OBD export
62 * \param[in] client_nid NID of client
64 * \retval 0 if successful
65 * \retval negative value on error
67 static int ofd_export_stats_init(struct ofd_device *ofd,
68 struct obd_export *exp,
69 lnet_nid_t *client_nid)
71 struct obd_device *obd = ofd_obd(ofd);
72 struct nid_stat *stats;
78 LASSERT(obd->obd_uses_nid_stats);
80 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
81 /* Self-export gets no proc entry */
84 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
86 /* Mask error for already created
96 stats = exp->exp_nid_stats;
97 LASSERT(stats != NULL);
99 num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
101 stats->nid_stats = lprocfs_alloc_stats(num_stats,
102 LPROCFS_STATS_FLAG_NOPERCPU);
103 if (stats->nid_stats == NULL)
106 lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
107 ofd_stats_counter_init(stats->nid_stats);
108 rc = lprocfs_register_stats(stats->nid_proc, "stats",
113 rc = lprocfs_nid_ldlm_stats_init(stats);
115 lprocfs_free_stats(&stats->nid_stats);
125 * Match client and OST server connection feature flags.
127 * Compute the compatibility flags for a connection request based on
128 * features mutually supported by client and server.
130 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
131 * must not be updated here, otherwise a partially initialized value may
132 * be exposed. After the connection request is successfully processed,
133 * the top-level tgt_connect() request handler atomically updates the export
134 * connect flags from the obd_connect_data::ocd_connect_flags field of the
135 * reply. \see tgt_connect().
137 * \param[in] env execution environment
138 * \param[in] exp the obd_export associated with this
140 * \param[in] data stores data for this connect request
141 * \param[in] new_connection is this connection new or not
143 * \retval 0 if success
144 * \retval -EPROTO client and server feature requirements are
146 * \retval -EBADF OST index in connect request doesn't match
149 static int ofd_parse_connect_data(const struct lu_env *env,
150 struct obd_export *exp,
151 struct obd_connect_data *data,
154 struct ofd_device *ofd = ofd_exp(exp);
155 struct filter_export_data *fed = &exp->exp_filter_data;
160 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
161 " ocd_version: %x ocd_grant: %d ocd_index: %u"
163 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
164 data->ocd_connect_flags, data->ocd_version,
165 data->ocd_grant, data->ocd_index, data->ocd_group);
167 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
168 CWARN("!!! This export (nid %s) used object group %d "
169 "earlier; now it's trying to use group %d! This could "
170 "be a bug in the MDS. Please report to "
171 "https://jira.hpdd.intel.com/\n",
172 obd_export_nid2str(exp), fed->fed_group,
176 fed->fed_group = data->ocd_group;
178 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
179 data->ocd_version = LUSTRE_VERSION_CODE;
181 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
182 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
183 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
184 exp->exp_obd->obd_name, data->ocd_group);
185 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
188 if (ofd_grant_param_supp(exp)) {
189 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
190 /* ocd_{blocksize,inodespace} are log2 values */
191 data->ocd_blocksize = ofd->ofd_blockbits;
192 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
193 /* ocd_grant_extent is in 1K blocks */
194 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
197 if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
198 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
201 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
202 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
203 int index = lsd->lsd_osd_index;
205 if (index != data->ocd_index) {
206 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
207 " %u doesn't match actual OST index"
208 " %u in last_rcvd file, bad "
210 obd_export_nid2str(exp), index,
214 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
215 /* this will only happen on the first connect */
216 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
217 /* sync is not needed here as lut_client_add will
218 * set exp_need_sync flag */
219 tgt_server_data_update(env, &ofd->ofd_lut, 0);
222 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
223 data->ocd_brw_size = 65536;
224 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
225 data->ocd_brw_size = min(data->ocd_brw_size,
226 (__u32)DT_MAX_BRW_SIZE);
227 if (data->ocd_brw_size == 0) {
228 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
229 " ocd_version: %x ocd_grant: %d ocd_index: %u "
230 "ocd_brw_size is unexpectedly zero, "
231 "network data corruption?"
232 "Refusing connection of this client\n",
233 exp->exp_obd->obd_name,
234 exp->exp_client_uuid.uuid,
235 exp, data->ocd_connect_flags, data->ocd_version,
236 data->ocd_grant, data->ocd_index);
241 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
242 __u32 cksum_types = data->ocd_cksum_types;
244 /* The client set in ocd_cksum_types the checksum types it
245 * supports. We have to mask off the algorithms that we don't
247 data->ocd_cksum_types &= cksum_types_supported_server();
249 if (unlikely(data->ocd_cksum_types == 0)) {
250 CERROR("%s: Connect with checksum support but no "
251 "ocd_cksum_types is set\n",
252 exp->exp_obd->obd_name);
256 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
257 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
258 cksum_types, data->ocd_cksum_types);
260 /* This client does not support OBD_CONNECT_CKSUM
261 * fall back to CRC32 */
262 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
263 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
264 exp->exp_obd->obd_name, obd_export_nid2str(exp));
267 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
268 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
270 if (OCD_HAS_FLAG(data, PINGLESS)) {
271 if (ptlrpc_pinger_suppress_pings()) {
272 spin_lock(&exp->exp_obd->obd_dev_lock);
273 list_del_init(&exp->exp_obd_chain_timed);
274 spin_unlock(&exp->exp_obd->obd_dev_lock);
276 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
284 * Re-initialize export upon client reconnection.
286 * This function parses connection data from reconnect and resets
289 * \param[in] env execution environment
290 * \param[in] exp OBD export
291 * \param[in] obd OFD device
292 * \param[in] cluuid NID of client
293 * \param[in] data connection data from request
294 * \param[in] localdata client NID
296 * \retval 0 if successful
297 * \retval negative value on error
299 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
300 struct obd_device *obd, struct obd_uuid *cluuid,
301 struct obd_connect_data *data,
304 struct ofd_device *ofd;
309 if (exp == NULL || obd == NULL || cluuid == NULL)
312 ofd = ofd_dev(obd->obd_lu_dev);
314 rc = ofd_parse_connect_data(env, exp, data, false);
316 ofd_export_stats_init(ofd, exp, client_nid);
318 nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
324 * Initialize new client connection.
326 * This function handles new connection to the OFD. The new export is
327 * created (in context of class_connect()) and persistent client data is
328 * initialized on storage.
330 * \param[in] env execution environment
331 * \param[out] _exp stores pointer to new export
332 * \param[in] obd OFD device
333 * \param[in] cluuid client UUID
334 * \param[in] data connection data from request
335 * \param[in] localdata client NID
337 * \retval 0 if successful
338 * \retval negative value on error
340 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
341 struct obd_device *obd, struct obd_uuid *cluuid,
342 struct obd_connect_data *data, void *localdata)
344 struct obd_export *exp;
345 struct ofd_device *ofd;
346 struct lustre_handle conn = { 0 };
348 lnet_nid_t *client_nid;
351 if (_exp == NULL || obd == NULL || cluuid == NULL)
354 ofd = ofd_dev(obd->obd_lu_dev);
356 rc = class_connect(&conn, obd, cluuid);
360 exp = class_conn2export(&conn);
361 LASSERT(exp != NULL);
363 rc = ofd_parse_connect_data(env, exp, data, true);
367 if (localdata != NULL) {
368 client_nid = localdata;
369 nodemap_add_member(*client_nid, exp);
372 if (obd->obd_replayable) {
373 struct tg_export_data *ted = &exp->exp_target_data;
375 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
376 sizeof(ted->ted_lcd->lcd_uuid));
377 rc = tgt_client_new(env, exp);
380 ofd_export_stats_init(ofd, exp, localdata);
383 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
384 data ? data->ocd_group : -1);
388 nodemap_del_member(exp);
389 class_disconnect(exp);
398 * Disconnect a connected client.
400 * This function terminates the client connection. The client export is
401 * disconnected (cleaned up) and client data on persistent storage is removed.
403 * \param[in] exp OBD export
405 * \retval 0 if successful
406 * \retval negative value on error
408 int ofd_obd_disconnect(struct obd_export *exp)
410 struct ofd_device *ofd = ofd_exp(exp);
417 class_export_get(exp);
419 if (!(exp->exp_flags & OBD_OPT_FORCE))
420 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
422 nodemap_del_member(exp);
423 rc = server_disconnect_export(exp);
425 ofd_grant_discard(exp);
427 /* Do not erase record for recoverable client. */
428 if (exp->exp_obd->obd_replayable &&
429 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
430 rc = lu_env_init(&env, LCT_DT_THREAD);
434 tgt_client_del(&env, exp);
438 class_export_put(exp);
443 * Implementation of obd_ops::o_init_export.
445 * This function is called from class_new_export() and initializes
446 * the OFD-specific data for new export.
448 * \param[in] exp OBD export
450 * \retval 0 if successful
451 * \retval negative value on error
453 static int ofd_init_export(struct obd_export *exp)
457 spin_lock_init(&exp->exp_filter_data.fed_lock);
458 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
459 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
460 spin_lock(&exp->exp_lock);
461 exp->exp_connecting = 1;
462 spin_unlock(&exp->exp_lock);
464 /* self-export doesn't need client data and ldlm initialization */
465 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
466 &exp->exp_client_uuid)))
469 rc = tgt_client_alloc(exp);
471 ldlm_init_export(exp);
473 CERROR("%s: Can't initialize export: rc %d\n",
474 exp->exp_obd->obd_name, rc);
479 * Implementation of obd_ops::o_destroy_export.
481 * This function is called from class_export_destroy() to cleanup
482 * the OFD-specific data for export being destroyed.
484 * \param[in] exp OBD export
486 * \retval 0 if successful
487 * \retval negative value on error
489 static int ofd_destroy_export(struct obd_export *exp)
491 struct ofd_device *ofd = ofd_exp(exp);
493 if (exp->exp_filter_data.fed_pending)
494 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
495 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
496 exp, exp->exp_filter_data.fed_pending);
498 target_destroy_export(exp);
500 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
501 &exp->exp_client_uuid)))
504 ldlm_destroy_export(exp);
505 tgt_client_free(exp);
507 ofd_fmd_cleanup(exp);
510 * discard grants once we're sure no more
511 * interaction with the client is possible
513 ofd_grant_discard(exp);
514 ofd_fmd_cleanup(exp);
516 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
517 if (ofd->ofd_tot_granted_clients > 0)
518 ofd->ofd_tot_granted_clients --;
521 if (!(exp->exp_flags & OBD_OPT_FORCE))
522 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
524 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
529 * Notify all devices in server stack about recovery completion.
531 * This function calls ldo_recovery_complete() for all lower devices in the
532 * server stack so they will be prepared for normal operations.
534 * \param[in] env execution environment
535 * \param[in] ofd OFD device
537 * \retval 0 if successful
538 * \retval negative value on error
540 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
542 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
543 struct lfsck_start_param lsp;
546 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
548 lsp.lsp_start = NULL;
549 lsp.lsp_index_valid = 0;
550 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
551 if (rc != 0 && rc != -EALREADY)
552 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
555 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
559 * Implementation of obd_ops::o_postrecov.
561 * This function is called from target_finish_recovery() upon recovery
564 * \param[in] obd OBD device of OFD
566 * \retval 0 if successful
567 * \retval negative value on error
569 int ofd_obd_postrecov(struct obd_device *obd)
572 struct lu_device *ldev = obd->obd_lu_dev;
577 rc = lu_env_init(&env, LCT_DT_THREAD);
580 ofd_info_init(&env, obd->obd_self_export);
582 rc = ofd_postrecov(&env, ofd_dev(ldev));
589 * Implementation of obd_ops::o_set_info_async.
591 * This function is not called from request handler, it is only used by
592 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
593 * \see ofd_set_info_hdl() for request handler function.
595 * \param[in] env execution environment
596 * \param[in] exp OBD export of OFD device
597 * \param[in] keylen length of \a key
598 * \param[in] key key name
599 * \param[in] vallen length of \a val
600 * \param[in] val the \a key value
601 * \param[in] set not used in OFD
603 * \retval 0 if successful
604 * \retval negative value on error
606 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
607 __u32 keylen, void *key, __u32 vallen, void *val,
608 struct ptlrpc_request_set *set)
614 if (exp->exp_obd == NULL) {
615 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
619 if (KEY_IS(KEY_SPTLRPC_CONF)) {
620 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
622 CERROR("%s: Unsupported key %s\n",
623 exp->exp_obd->obd_name, (char*)key);
630 * Implementation of obd_ops::o_get_info.
632 * This function is not called from request handler, it is only used by
633 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
635 * \see ofd_get_info_hdl() for request handler function.
637 * \param[in] env execution environment
638 * \param[in] exp OBD export of OFD device
639 * \param[in] keylen length of \a key
640 * \param[in] key key name
641 * \param[out] vallen length of key value
642 * \param[out] val the key value to return
643 * \param[in] lsm not used in OFD
645 * \retval 0 if successful
646 * \retval negative value on error
648 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
649 __u32 keylen, void *key, __u32 *vallen, void *val,
650 struct lov_stripe_md *lsm)
652 struct ofd_thread_info *info;
653 struct ofd_device *ofd;
654 struct ll_fiemap_info_key *fm_key = key;
655 struct ll_user_fiemap *fiemap = val;
660 if (exp->exp_obd == NULL) {
661 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
667 if (KEY_IS(KEY_FIEMAP)) {
668 info = ofd_info_init(env, exp);
670 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi,
671 ofd->ofd_lut.lut_lsd.lsd_osd_index);
675 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
677 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
685 * Get file system statistics of OST server.
687 * Helper function for ofd_statfs(), also used by grant code.
688 * Implements caching for statistics to avoid calling OSD device each time.
690 * \param[in] env execution environment
691 * \param[in] ofd OFD device
692 * \param[out] osfs statistic data to return
693 * \param[in] max_age maximum age for cached data
694 * \param[in] from_cache show that data was get from cache or not
696 * \retval 0 if successful
697 * \retval negative value on error
699 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
700 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
705 down_read(&ofd->ofd_lastid_rwsem);
706 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
707 * we may do that in the future.
708 * Return -ENOSPC until the LAST_ID rebuilt. */
709 if (unlikely(ofd->ofd_lastid_rebuilding))
710 GOTO(out, rc = -ENOSPC);
712 spin_lock(&ofd->ofd_osfs_lock);
713 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
716 /* statfs data are too old, get up-to-date one.
717 * we must be cautious here since multiple threads might be
718 * willing to update statfs data concurrently and we must
719 * grant that cached statfs data are always consistent */
721 if (ofd->ofd_statfs_inflight == 0)
722 /* clear inflight counter if no users, although it would
723 * take a while to overflow this 64-bit counter ... */
724 ofd->ofd_osfs_inflight = 0;
725 /* notify ofd_grant_commit() that we want to track writes
726 * completed as of now */
727 ofd->ofd_statfs_inflight++;
728 /* record value of inflight counter before running statfs to
729 * compute the diff once statfs is completed */
730 unstable = ofd->ofd_osfs_inflight;
731 spin_unlock(&ofd->ofd_osfs_lock);
733 /* statfs can sleep ... hopefully not for too long since we can
734 * call it fairly often as space fills up */
735 rc = dt_statfs(env, ofd->ofd_osd, osfs);
739 spin_lock(&ofd->ofd_grant_lock);
740 spin_lock(&ofd->ofd_osfs_lock);
741 /* calculate how much space was written while we released the
743 unstable = ofd->ofd_osfs_inflight - unstable;
744 ofd->ofd_osfs_unstable = 0;
746 /* some writes completed while we were running statfs
747 * w/o the ofd_osfs_lock. Those ones got added to
748 * the cached statfs data that we are about to crunch.
749 * Take them into account in the new statfs data */
750 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
751 unstable >> ofd->ofd_blockbits);
752 /* However, we don't really know if those writes got
753 * accounted in the statfs call, so tell
754 * ofd_grant_space_left() there is some uncertainty
755 * on the accounting of those writes.
756 * The purpose is to prevent spurious error messages in
757 * ofd_grant_space_left() since those writes might be
758 * accounted twice. */
759 ofd->ofd_osfs_unstable += unstable;
761 /* similarly, there is some uncertainty on write requests
762 * between prepare & commit */
763 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
764 spin_unlock(&ofd->ofd_grant_lock);
766 /* finally udpate cached statfs data */
767 ofd->ofd_osfs = *osfs;
768 ofd->ofd_osfs_age = cfs_time_current_64();
770 ofd->ofd_statfs_inflight--; /* stop tracking */
771 if (ofd->ofd_statfs_inflight == 0)
772 ofd->ofd_osfs_inflight = 0;
773 spin_unlock(&ofd->ofd_osfs_lock);
778 /* use cached statfs data */
779 *osfs = ofd->ofd_osfs;
780 spin_unlock(&ofd->ofd_osfs_lock);
788 up_read(&ofd->ofd_lastid_rwsem);
794 * Implementation of obd_ops::o_statfs.
796 * This function returns information about a storage file system.
797 * It is called from several places by using the OBD API as well as
798 * by direct call, e.g. from request handler.
800 * \see ofd_statfs_hdl() for request handler function.
802 * Report also the state of the OST to the caller in osfs->os_state
803 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
805 * \param[in] env execution environment
806 * \param[in] exp OBD export of OFD device
807 * \param[out] osfs statistic data to return
808 * \param[in] max_age maximum age for cached data
809 * \param[in] flags not used in OFD
811 * \retval 0 if successful
812 * \retval negative value on error
814 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
815 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
817 struct obd_device *obd = class_exp2obd(exp);
818 struct ofd_device *ofd = ofd_exp(exp);
823 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
827 /* at least try to account for cached pages. its still racy and
828 * might be under-reporting if clients haven't announced their
829 * caches with brw recently */
831 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
832 " pending "LPU64" free "LPU64" avail "LPU64"\n",
833 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
834 osfs->os_bfree << ofd->ofd_blockbits,
835 osfs->os_bavail << ofd->ofd_blockbits);
837 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
838 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
839 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
841 /* The QoS code on the MDS does not care about space reserved for
842 * precreate, so take it out. */
843 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
844 struct filter_export_data *fed;
846 fed = &obd->obd_self_export->exp_filter_data;
847 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
848 fed->fed_grant >> ofd->ofd_blockbits);
851 ofd_grant_sanity_check(obd, __FUNCTION__);
852 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
853 LPU64" objects: "LPU64" free; state %x\n",
854 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
855 osfs->os_files, osfs->os_ffree, osfs->os_state);
857 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
858 ofd->ofd_lut.lut_lsd.lsd_osd_index))
861 /* OS_STATE_READONLY can be set by OSD already */
862 if (ofd->ofd_raid_degraded)
863 osfs->os_state |= OS_STATE_DEGRADED;
865 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
866 /* clients which don't support OBD_CONNECT_GRANT_PARAM
867 * should not see a block size > page size, otherwise
868 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
869 * block size which is the biggest block size known to work
870 * with all client's page size. */
871 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
872 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
873 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
874 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
877 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
878 ofd->ofd_lut.lut_lsd.lsd_osd_index))
879 osfs->os_bfree = osfs->os_bavail = 2;
887 * Implementation of obd_ops::o_setattr.
889 * This function is only used by ECHO client when it is run on top of OFD,
890 * \see ofd_setattr_hdl() for request handler function.
892 * \param[in] env execution environment
893 * \param[in] exp OBD export of OFD device
894 * \param[in] oinfo obd_info with setattr parameters
895 * \param[in] oti not used in OFD
897 * \retval 0 if successful
898 * \retval negative value on error
900 int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
901 struct obd_info *oinfo, struct obd_trans_info *oti)
903 struct ofd_thread_info *info;
904 struct ofd_device *ofd = ofd_exp(exp);
905 struct ldlm_namespace *ns = ofd->ofd_namespace;
906 struct ldlm_resource *res;
907 struct ofd_object *fo;
908 struct obdo *oa = oinfo->oi_oa;
909 struct lu_fid *fid = &oa->o_oi.oi_fid;
910 struct filter_fid *ff = NULL;
915 info = ofd_info_init(env, exp);
917 ost_fid_build_resid(fid, &info->fti_resid);
919 /* This would be very bad - accidentally truncating a file when
920 * changing the time or similar - bug 12203. */
921 if (oa->o_valid & OBD_MD_FLSIZE &&
922 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
923 static char mdsinum[48];
925 if (oa->o_valid & OBD_MD_FLFID)
926 snprintf(mdsinum, sizeof(mdsinum) - 1,
927 "of parent "DFID, oa->o_parent_seq,
928 oa->o_parent_oid, 0);
932 CERROR("%s: setattr from %s trying to truncate object "DFID
933 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
935 GOTO(out, rc = -EPERM);
938 fo = ofd_object_find_exists(env, ofd, fid);
940 CERROR("%s: can't find object "DFID"\n",
941 ofd_name(ofd), PFID(fid));
942 GOTO(out, rc = PTR_ERR(fo));
945 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
946 info->fti_attr.la_valid &= ~LA_TYPE;
948 if (oa->o_valid & OBD_MD_FLFID) {
949 ff = &info->fti_mds_fid;
950 ofd_prepare_fidea(ff, oa);
953 /* setting objects attributes (including owner/group) */
954 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
956 GOTO(out_unlock, rc);
958 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
961 ofd_object_put(env, fo);
964 /* we do not call this before to avoid lu_object_find() in
965 * ->lvbo_update() holding another reference on the object.
966 * otherwise concurrent destroy can make the object unavailable
967 * for 2nd lu_object_find() waiting for the first reference
968 * to go... deadlock! */
969 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
971 ldlm_res_lvbo_update(res, NULL, 0);
972 ldlm_resource_putref(res);
980 * Destroy OFD object by its FID.
982 * Supplemental function to destroy object by FID, it is used by request
983 * handler and by ofd_echo_destroy() below to find object by FID, lock it
984 * and call ofd_object_destroy() finally.
986 * \param[in] env execution environment
987 * \param[in] ofd OFD device
988 * \param[in] fid FID of object
989 * \param[in] orphan set if object being destroyed is an orphan
991 * \retval 0 if successful
992 * \retval negative value on error
994 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
995 const struct lu_fid *fid, int orphan)
997 struct ofd_thread_info *info = ofd_info(env);
998 struct lustre_handle lockh;
999 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
1001 ldlm_policy_data_t policy = {
1002 .l_extent = { 0, OBD_OBJECT_EOF }
1004 struct ofd_object *fo;
1008 fo = ofd_object_find_exists(env, ofd, fid);
1010 RETURN(PTR_ERR(fo));
1012 /* Tell the clients that the object is gone now and that they should
1013 * throw away any cached pages. */
1014 ost_fid_build_resid(fid, &info->fti_resid);
1015 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1016 LDLM_EXTENT, &policy, LCK_PW, &flags,
1017 ldlm_blocking_ast, ldlm_completion_ast,
1018 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1020 /* We only care about the side-effects, just drop the lock. */
1022 ldlm_lock_decref(&lockh, LCK_PW);
1024 LASSERT(fo != NULL);
1026 rc = ofd_object_destroy(env, fo, orphan);
1029 ofd_object_put(env, fo);
1034 * Implementation of obd_ops::o_destroy.
1036 * This function is only used by ECHO client when it is run on top of OFD,
1037 * \see ofd_destroy_hdl() for request handler function.
1039 * \param[in] env execution environment
1040 * \param[in] exp OBD export of OFD device
1041 * \param[in] oa obdo structure with FID
1042 * \param[in] md not used in OFD
1043 * \param[in] oti not used in OFD
1044 * \param[in] md_exp not used in OFD
1045 * \param[in] capa not used in OFD
1047 * Note: this is OBD API method which is common API for server OBDs and
1048 * client OBDs. Thus some parameters used in client OBDs may not be used
1049 * on server OBDs and vice versa.
1051 * \retval 0 if successful
1052 * \retval negative value on error
1054 int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1055 struct obdo *oa, struct lov_stripe_md *md,
1056 struct obd_trans_info *oti, struct obd_export *md_exp,
1059 struct ofd_device *ofd = ofd_exp(exp);
1060 struct lu_fid *fid = &oa->o_oi.oi_fid;
1065 ofd_info_init(env, exp);
1067 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1069 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1070 if (rc == -ENOENT) {
1071 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1072 ofd_name(ofd), PFID(fid));
1074 } else if (rc != 0) {
1075 CERROR("%s: error destroying object "DFID": %d\n",
1076 ofd_name(ofd), PFID(fid), rc);
1085 * Implementation of obd_ops::o_create.
1087 * This function is only used by ECHO client when it is run on top of OFD
1088 * and just creates an object.
1089 * \see ofd_create_hdl() for request handler function.
1091 * \param[in] env execution environment
1092 * \param[in] exp OBD export of OFD device
1093 * \param[in] oa obdo structure with FID sequence to use
1094 * \param[out] ea contains object ID/SEQ to return
1095 * \param[in] oti not used in OFD
1097 * Note: this is OBD API method which is common API for server OBDs and
1098 * client OBDs. Thus some parameters used in client OBDs may not be used
1099 * on server OBDs and vice versa.
1101 * \retval 0 if successful
1102 * \retval negative value on error
1104 int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1105 struct obdo *oa, struct lov_stripe_md **ea,
1106 struct obd_trans_info *oti)
1108 struct ofd_device *ofd = ofd_exp(exp);
1109 struct ofd_thread_info *info;
1110 obd_seq seq = ostid_seq(&oa->o_oi);
1111 struct ofd_seq *oseq;
1112 int rc = 0, diff = 1;
1118 info = ofd_info_init(env, exp);
1120 LASSERT(seq == FID_SEQ_ECHO);
1121 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1123 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1125 down_read(&ofd->ofd_lastid_rwsem);
1126 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1127 * we may do that in the future.
1128 * Return -ENOSPC until the LAST_ID rebuilt. */
1129 if (unlikely(ofd->ofd_lastid_rebuilding))
1130 GOTO(out_sem, rc = -ENOSPC);
1132 oseq = ofd_seq_load(env, ofd, seq);
1134 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1135 ofd_name(ofd), seq, PTR_ERR(oseq));
1136 GOTO(out_sem, rc = -EINVAL);
1139 mutex_lock(&oseq->os_create_lock);
1140 rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1142 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1143 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
1148 next_id = ofd_seq_last_oid(oseq) + 1;
1149 count = ofd_precreate_batch(ofd, diff);
1151 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1153 CERROR("%s: unable to precreate: rc = %d\n",
1156 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1157 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1161 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
1163 mutex_unlock(&oseq->os_create_lock);
1164 if (rc == 0 && ea != NULL) {
1165 struct lov_stripe_md *lsm = *ea;
1167 lsm->lsm_oi = oa->o_oi;
1169 ofd_seq_put(env, oseq);
1172 up_read(&ofd->ofd_lastid_rwsem);
1177 * Implementation of obd_ops::o_getattr.
1179 * This function is only used by ECHO client when it is run on top of OFD
1180 * and returns attributes of object.
1181 * \see ofd_getattr_hdl() for request handler function.
1183 * \param[in] env execution environment
1184 * \param[in] exp OBD export of OFD device
1185 * \param[in,out] oinfo contains FID of object to get attributes from and
1186 * is used to return attributes back
1188 * \retval 0 if successful
1189 * \retval negative value on error
1191 int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1192 struct obd_info *oinfo)
1194 struct ofd_device *ofd = ofd_exp(exp);
1195 struct ofd_thread_info *info;
1196 struct lu_fid *fid = &oinfo->oi_oa->o_oi.oi_fid;
1197 struct ofd_object *fo;
1202 info = ofd_info_init(env, exp);
1204 fo = ofd_object_find_exists(env, ofd, fid);
1206 GOTO(out, rc = PTR_ERR(fo));
1208 LASSERT(fo != NULL);
1209 rc = ofd_attr_get(env, fo, &info->fti_attr);
1210 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1214 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1215 OFD_VALID_FLAGS | LA_UID | LA_GID);
1217 /* Store object version in reply */
1218 curr_version = dt_version_get(env, ofd_object_child(fo));
1219 if ((__s64)curr_version != -EOPNOTSUPP) {
1220 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1221 oinfo->oi_oa->o_data_version = curr_version;
1225 ofd_object_put(env, fo);
1231 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1233 * This is supplemental function for ofd_iocontrol() to return object
1234 * version for lctl tool.
1236 * \param[in] env execution environment
1237 * \param[in] ofd OFD device
1238 * \param[out] karg ioctl data
1240 * \retval 0 if successful
1241 * \retval negative value on error
1243 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1244 struct ofd_device *ofd, void *karg)
1246 struct obd_ioctl_data *data = karg;
1248 struct ofd_object *fo;
1249 dt_obj_version_t version;
1254 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1255 GOTO(out, rc = -EINVAL);
1257 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1258 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1259 } else if (data->ioc_inlbuf3 != NULL &&
1260 data->ioc_inllen3 == sizeof(__u64) &&
1261 data->ioc_inlbuf4 != NULL &&
1262 data->ioc_inllen4 == sizeof(__u64)) {
1263 struct ost_id ostid;
1265 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1266 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1267 rc = ostid_to_fid(&fid, &ostid,
1268 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1272 GOTO(out, rc = -EINVAL);
1275 if (!fid_is_sane(&fid))
1276 GOTO(out, rc = -EINVAL);
1278 fo = ofd_object_find(env, ofd, &fid);
1280 GOTO(out, rc = PTR_ERR(fo));
1282 if (!ofd_object_exists(fo))
1283 GOTO(out_fo, rc = -ENOENT);
1285 if (lu_object_remote(&fo->ofo_obj.do_lu))
1286 GOTO(out_fo, rc = -EREMOTE);
1288 version = dt_version_get(env, ofd_object_child(fo));
1290 GOTO(out_fo, rc = -EIO);
1292 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1296 ofd_object_put(env, fo);
1302 * Implementation of obd_ops::o_iocontrol.
1304 * This is OFD ioctl handling function which is primary interface for
1305 * Lustre tools like lfs, lctl and lfsck.
1307 * \param[in] cmd ioctl command
1308 * \param[in] exp OBD export of OFD
1309 * \param[in] len not used
1310 * \param[in,out] karg buffer with data
1311 * \param[in] uarg not used
1313 * \retval 0 if successful
1314 * \retval negative value on error
1316 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1317 void *karg, void *uarg)
1320 struct ofd_device *ofd = ofd_exp(exp);
1321 struct obd_device *obd = ofd_obd(ofd);
1326 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1327 rc = lu_env_init(&env, LCT_DT_THREAD);
1332 case OBD_IOC_ABORT_RECOVERY:
1333 CERROR("%s: aborting recovery\n", obd->obd_name);
1334 target_stop_recovery_thread(obd);
1337 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1338 rc = dt_sync(&env, ofd->ofd_osd);
1340 case OBD_IOC_SET_READONLY:
1341 rc = dt_sync(&env, ofd->ofd_osd);
1343 rc = dt_ro(&env, ofd->ofd_osd);
1345 case OBD_IOC_START_LFSCK: {
1346 struct obd_ioctl_data *data = karg;
1347 struct lfsck_start_param lsp;
1349 if (unlikely(data == NULL)) {
1354 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1355 lsp.lsp_index_valid = 0;
1356 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1359 case OBD_IOC_STOP_LFSCK: {
1360 struct obd_ioctl_data *data = karg;
1361 struct lfsck_stop stop;
1363 stop.ls_status = LS_STOPPED;
1364 /* Old lfsck utils may pass NULL @stop. */
1365 if (data->ioc_inlbuf1 == NULL)
1369 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1371 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1374 case OBD_IOC_GET_OBJ_VERSION:
1375 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1378 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1387 * Implementation of obd_ops::o_precleanup.
1389 * This function stops device activity before shutting it down. It is called
1390 * from a cleanup function upon forceful device cleanup. For OFD there are no
1391 * special actions, it just invokes target_recovery_cleanup().
1393 * \param[in] obd OBD device of OFD
1394 * \param[in] stage cleanup stage
1398 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1405 case OBD_CLEANUP_EARLY:
1407 case OBD_CLEANUP_EXPORTS:
1408 target_cleanup_recovery(obd);
1415 * Implementation of obd_ops::o_ping.
1417 * This is OFD-specific part of OBD_PING request handling.
1418 * It controls Filter Modification Data (FMD) expiration each time PING is
1421 * \see ofd_fmd_expire() and ofd_fmd.c for details
1423 * \param[in] env execution environment
1424 * \param[in] exp OBD export of client
1428 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1430 ofd_fmd_expire(exp);
1435 * Implementation of obd_ops::o_health_check.
1437 * This function checks the OFD device health - ability to respond on
1438 * incoming requests. There are two health_check methods:
1439 * - get statfs from the OSD. It checks just responsiveness of
1441 * - do write attempt on bottom device to check it is fully operational and
1442 * is not stuck. This is expensive method and requires special configuration
1443 * option --enable-health-write while building Lustre, it is turned off
1446 * \param[in] nul not used
1447 * \param[in] obd OBD device of OFD
1449 * \retval 0 if successful
1450 * \retval negative value in case of error
1452 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1454 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1455 struct ofd_thread_info *info;
1457 #ifdef USE_HEALTH_CHECK_WRITE
1462 /* obd_proc_read_health pass NULL env, we need real one */
1463 rc = lu_env_init(&env, LCT_DT_THREAD);
1467 info = ofd_info_init(&env, NULL);
1468 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1472 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1473 GOTO(out, rc = -EROFS);
1475 #ifdef USE_HEALTH_CHECK_WRITE
1476 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1477 if (info->fti_buf.lb_buf == NULL)
1478 GOTO(out, rc = -ENOMEM);
1480 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1483 th = dt_trans_create(&env, ofd->ofd_osd);
1485 GOTO(out, rc = PTR_ERR(th));
1487 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1488 &info->fti_buf, info->fti_off, th);
1490 th->th_sync = 1; /* sync IO is needed */
1491 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1493 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1494 &info->fti_buf, &info->fti_off,
1497 dt_trans_stop(&env, ofd->ofd_osd, th);
1499 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1501 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1508 struct obd_ops ofd_obd_ops = {
1509 .o_owner = THIS_MODULE,
1510 .o_connect = ofd_obd_connect,
1511 .o_reconnect = ofd_obd_reconnect,
1512 .o_disconnect = ofd_obd_disconnect,
1513 .o_create = ofd_echo_create,
1514 .o_statfs = ofd_statfs,
1515 .o_setattr = ofd_echo_setattr,
1516 .o_preprw = ofd_preprw,
1517 .o_commitrw = ofd_commitrw,
1518 .o_destroy = ofd_echo_destroy,
1519 .o_init_export = ofd_init_export,
1520 .o_destroy_export = ofd_destroy_export,
1521 .o_postrecov = ofd_obd_postrecov,
1522 .o_getattr = ofd_echo_getattr,
1523 .o_iocontrol = ofd_iocontrol,
1524 .o_precleanup = ofd_precleanup,
1526 .o_health_check = ofd_health_check,
1527 .o_set_info_async = ofd_set_info_async,
1528 .o_get_info = ofd_get_info,