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>
54 * Initialize OFD per-export statistics.
56 * This function sets up procfs entries for various OFD export counters. These
57 * counters are for per-client statistics tracked on the server.
59 * \param[in] ofd OFD device
60 * \param[in] exp OBD export
61 * \param[in] client_nid NID of client
63 * \retval 0 if successful
64 * \retval negative value on error
66 static int ofd_export_stats_init(struct ofd_device *ofd,
67 struct obd_export *exp, void *client_nid)
69 struct obd_device *obd = ofd_obd(ofd);
70 struct nid_stat *stats;
76 LASSERT(obd->obd_uses_nid_stats);
78 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
79 /* Self-export gets no proc entry */
82 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
84 /* Mask error for already created
94 stats = exp->exp_nid_stats;
95 LASSERT(stats != NULL);
97 num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
99 stats->nid_stats = lprocfs_alloc_stats(num_stats,
100 LPROCFS_STATS_FLAG_NOPERCPU);
101 if (stats->nid_stats == NULL)
104 lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
105 ofd_stats_counter_init(stats->nid_stats);
106 rc = lprocfs_register_stats(stats->nid_proc, "stats",
111 rc = lprocfs_nid_ldlm_stats_init(stats);
113 lprocfs_free_stats(&stats->nid_stats);
123 * Match client and OST server connection feature flags.
125 * Compute the compatibility flags for a connection request based on
126 * features mutually supported by client and server.
128 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
129 * must not be updated here, otherwise a partially initialized value may
130 * be exposed. After the connection request is successfully processed,
131 * the top-level tgt_connect() request handler atomically updates the export
132 * connect flags from the obd_connect_data::ocd_connect_flags field of the
133 * reply. \see tgt_connect().
135 * \param[in] env execution environment
136 * \param[in] exp the obd_export associated with this
138 * \param[in] data stores data for this connect request
139 * \param[in] new_connection is this connection new or not
141 * \retval 0 if success
142 * \retval -EPROTO client and server feature requirements are
144 * \retval -EBADF OST index in connect request doesn't match
147 static int ofd_parse_connect_data(const struct lu_env *env,
148 struct obd_export *exp,
149 struct obd_connect_data *data,
152 struct ofd_device *ofd = ofd_exp(exp);
153 struct filter_export_data *fed = &exp->exp_filter_data;
158 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
159 " ocd_version: %x ocd_grant: %d ocd_index: %u"
161 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
162 data->ocd_connect_flags, data->ocd_version,
163 data->ocd_grant, data->ocd_index, data->ocd_group);
165 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
166 CWARN("!!! This export (nid %s) used object group %d "
167 "earlier; now it's trying to use group %d! This could "
168 "be a bug in the MDS. Please report to "
169 "https://jira.hpdd.intel.com/\n",
170 obd_export_nid2str(exp), fed->fed_group,
174 fed->fed_group = data->ocd_group;
176 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
177 data->ocd_version = LUSTRE_VERSION_CODE;
179 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
180 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
181 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
182 exp->exp_obd->obd_name, data->ocd_group);
183 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
186 if (ofd_grant_param_supp(exp)) {
187 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
188 /* ocd_{blocksize,inodespace} are log2 values */
189 data->ocd_blocksize = ofd->ofd_blockbits;
190 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
191 /* ocd_grant_extent is in 1K blocks */
192 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
195 if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
196 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
199 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
200 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
201 int index = lsd->lsd_osd_index;
203 if (index != data->ocd_index) {
204 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
205 " %u doesn't match actual OST index"
206 " %u in last_rcvd file, bad "
208 obd_export_nid2str(exp), index,
212 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
213 /* this will only happen on the first connect */
214 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
215 /* sync is not needed here as lut_client_add will
216 * set exp_need_sync flag */
217 tgt_server_data_update(env, &ofd->ofd_lut, 0);
220 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
221 data->ocd_brw_size = 65536;
222 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
223 data->ocd_brw_size = min(data->ocd_brw_size,
224 (__u32)DT_MAX_BRW_SIZE);
225 if (data->ocd_brw_size == 0) {
226 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
227 " ocd_version: %x ocd_grant: %d ocd_index: %u "
228 "ocd_brw_size is unexpectedly zero, "
229 "network data corruption?"
230 "Refusing connection of this client\n",
231 exp->exp_obd->obd_name,
232 exp->exp_client_uuid.uuid,
233 exp, data->ocd_connect_flags, data->ocd_version,
234 data->ocd_grant, data->ocd_index);
239 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
240 __u32 cksum_types = data->ocd_cksum_types;
242 /* The client set in ocd_cksum_types the checksum types it
243 * supports. We have to mask off the algorithms that we don't
245 data->ocd_cksum_types &= cksum_types_supported_server();
247 if (unlikely(data->ocd_cksum_types == 0)) {
248 CERROR("%s: Connect with checksum support but no "
249 "ocd_cksum_types is set\n",
250 exp->exp_obd->obd_name);
254 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
255 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
256 cksum_types, data->ocd_cksum_types);
258 /* This client does not support OBD_CONNECT_CKSUM
259 * fall back to CRC32 */
260 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
261 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
262 exp->exp_obd->obd_name, obd_export_nid2str(exp));
265 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
266 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
268 if (OCD_HAS_FLAG(data, PINGLESS)) {
269 if (ptlrpc_pinger_suppress_pings()) {
270 spin_lock(&exp->exp_obd->obd_dev_lock);
271 list_del_init(&exp->exp_obd_chain_timed);
272 spin_unlock(&exp->exp_obd->obd_dev_lock);
274 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
282 * Re-initialize export upon client reconnection.
284 * This function parses connection data from reconnect and resets
287 * \param[in] env execution environment
288 * \param[in] exp OBD export
289 * \param[in] obd OFD device
290 * \param[in] cluuid NID of client
291 * \param[in] data connection data from request
292 * \param[in] localdata client NID
294 * \retval 0 if successful
295 * \retval negative value on error
297 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
298 struct obd_device *obd, struct obd_uuid *cluuid,
299 struct obd_connect_data *data, void *localdata)
301 struct ofd_device *ofd;
306 if (exp == NULL || obd == NULL || cluuid == NULL)
309 ofd = ofd_dev(obd->obd_lu_dev);
311 rc = ofd_parse_connect_data(env, exp, data, false);
313 ofd_export_stats_init(ofd, exp, localdata);
319 * Initialize new client connection.
321 * This function handles new connection to the OFD. The new export is
322 * created (in context of class_connect()) and persistent client data is
323 * initialized on storage.
325 * \param[in] env execution environment
326 * \param[out] _exp stores pointer to new export
327 * \param[in] obd OFD device
328 * \param[in] cluuid client UUID
329 * \param[in] data connection data from request
330 * \param[in] localdata client NID
332 * \retval 0 if successful
333 * \retval negative value on error
335 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
336 struct obd_device *obd, struct obd_uuid *cluuid,
337 struct obd_connect_data *data, void *localdata)
339 struct obd_export *exp;
340 struct ofd_device *ofd;
341 struct lustre_handle conn = { 0 };
345 if (_exp == NULL || obd == NULL || cluuid == NULL)
348 ofd = ofd_dev(obd->obd_lu_dev);
350 rc = class_connect(&conn, obd, cluuid);
354 exp = class_conn2export(&conn);
355 LASSERT(exp != NULL);
357 rc = ofd_parse_connect_data(env, exp, data, true);
361 if (obd->obd_replayable) {
362 struct tg_export_data *ted = &exp->exp_target_data;
364 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
365 sizeof(ted->ted_lcd->lcd_uuid));
366 rc = tgt_client_new(env, exp);
369 ofd_export_stats_init(ofd, exp, localdata);
372 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
373 data ? data->ocd_group : -1);
377 class_disconnect(exp);
386 * Disconnect a connected client.
388 * This function terminates the client connection. The client export is
389 * disconnected (cleaned up) and client data on persistent storage is removed.
391 * \param[in] exp OBD export
393 * \retval 0 if successful
394 * \retval negative value on error
396 int ofd_obd_disconnect(struct obd_export *exp)
398 struct ofd_device *ofd = ofd_exp(exp);
405 class_export_get(exp);
407 if (!(exp->exp_flags & OBD_OPT_FORCE))
408 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
410 rc = server_disconnect_export(exp);
412 ofd_grant_discard(exp);
414 /* Do not erase record for recoverable client. */
415 if (exp->exp_obd->obd_replayable &&
416 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
417 rc = lu_env_init(&env, LCT_DT_THREAD);
421 tgt_client_del(&env, exp);
425 class_export_put(exp);
430 * Implementation of obd_ops::o_init_export.
432 * This function is called from class_new_export() and initializes
433 * the OFD-specific data for new export.
435 * \param[in] exp OBD export
437 * \retval 0 if successful
438 * \retval negative value on error
440 static int ofd_init_export(struct obd_export *exp)
444 spin_lock_init(&exp->exp_filter_data.fed_lock);
445 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
446 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
447 spin_lock(&exp->exp_lock);
448 exp->exp_connecting = 1;
449 spin_unlock(&exp->exp_lock);
451 /* self-export doesn't need client data and ldlm initialization */
452 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
453 &exp->exp_client_uuid)))
456 rc = tgt_client_alloc(exp);
458 ldlm_init_export(exp);
460 CERROR("%s: Can't initialize export: rc %d\n",
461 exp->exp_obd->obd_name, rc);
466 * Implementation of obd_ops::o_destroy_export.
468 * This function is called from class_export_destroy() to cleanup
469 * the OFD-specific data for export being destroyed.
471 * \param[in] exp OBD export
473 * \retval 0 if successful
474 * \retval negative value on error
476 static int ofd_destroy_export(struct obd_export *exp)
478 struct ofd_device *ofd = ofd_exp(exp);
480 if (exp->exp_filter_data.fed_pending)
481 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
482 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
483 exp, exp->exp_filter_data.fed_pending);
485 target_destroy_export(exp);
487 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
488 &exp->exp_client_uuid)))
491 ldlm_destroy_export(exp);
492 tgt_client_free(exp);
494 ofd_fmd_cleanup(exp);
497 * discard grants once we're sure no more
498 * interaction with the client is possible
500 ofd_grant_discard(exp);
501 ofd_fmd_cleanup(exp);
503 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
504 if (ofd->ofd_tot_granted_clients > 0)
505 ofd->ofd_tot_granted_clients --;
508 if (!(exp->exp_flags & OBD_OPT_FORCE))
509 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
511 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
516 * Notify all devices in server stack about recovery completion.
518 * This function calls ldo_recovery_complete() for all lower devices in the
519 * server stack so they will be prepared for normal operations.
521 * \param[in] env execution environment
522 * \param[in] ofd OFD device
524 * \retval 0 if successful
525 * \retval negative value on error
527 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
529 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
530 struct lfsck_start_param lsp;
533 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
535 lsp.lsp_start = NULL;
536 lsp.lsp_index_valid = 0;
537 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
538 if (rc != 0 && rc != -EALREADY)
539 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
542 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
546 * Implementation of obd_ops::o_postrecov.
548 * This function is called from target_finish_recovery() upon recovery
551 * \param[in] obd OBD device of OFD
553 * \retval 0 if successful
554 * \retval negative value on error
556 int ofd_obd_postrecov(struct obd_device *obd)
559 struct lu_device *ldev = obd->obd_lu_dev;
564 rc = lu_env_init(&env, LCT_DT_THREAD);
567 ofd_info_init(&env, obd->obd_self_export);
569 rc = ofd_postrecov(&env, ofd_dev(ldev));
576 * Implementation of obd_ops::o_set_info_async.
578 * This function is not called from request handler, it is only used by
579 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
580 * \see ofd_set_info_hdl() for request handler function.
582 * \param[in] env execution environment
583 * \param[in] exp OBD export of OFD device
584 * \param[in] keylen length of \a key
585 * \param[in] key key name
586 * \param[in] vallen length of \a val
587 * \param[in] val the \a key value
588 * \param[in] set not used in OFD
590 * \retval 0 if successful
591 * \retval negative value on error
593 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
594 __u32 keylen, void *key, __u32 vallen, void *val,
595 struct ptlrpc_request_set *set)
601 if (exp->exp_obd == NULL) {
602 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
606 if (KEY_IS(KEY_SPTLRPC_CONF)) {
607 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
609 CERROR("%s: Unsupported key %s\n",
610 exp->exp_obd->obd_name, (char*)key);
617 * Implementation of obd_ops::o_get_info.
619 * This function is not called from request handler, it is only used by
620 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
622 * \see ofd_get_info_hdl() for request handler function.
624 * \param[in] env execution environment
625 * \param[in] exp OBD export of OFD device
626 * \param[in] keylen length of \a key
627 * \param[in] key key name
628 * \param[out] vallen length of key value
629 * \param[out] val the key value to return
630 * \param[in] lsm not used in OFD
632 * \retval 0 if successful
633 * \retval negative value on error
635 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
636 __u32 keylen, void *key, __u32 *vallen, void *val,
637 struct lov_stripe_md *lsm)
639 struct ofd_thread_info *info;
640 struct ofd_device *ofd;
641 struct ll_fiemap_info_key *fm_key = key;
642 struct ll_user_fiemap *fiemap = val;
647 if (exp->exp_obd == NULL) {
648 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
654 if (KEY_IS(KEY_FIEMAP)) {
655 info = ofd_info_init(env, exp);
657 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi,
658 ofd->ofd_lut.lut_lsd.lsd_osd_index);
662 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
664 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
672 * Get file system statistics of OST server.
674 * Helper function for ofd_statfs(), also used by grant code.
675 * Implements caching for statistics to avoid calling OSD device each time.
677 * \param[in] env execution environment
678 * \param[in] ofd OFD device
679 * \param[out] osfs statistic data to return
680 * \param[in] max_age maximum age for cached data
681 * \param[in] from_cache show that data was get from cache or not
683 * \retval 0 if successful
684 * \retval negative value on error
686 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
687 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
692 down_read(&ofd->ofd_lastid_rwsem);
693 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
694 * we may do that in the future.
695 * Return -ENOSPC until the LAST_ID rebuilt. */
696 if (unlikely(ofd->ofd_lastid_rebuilding))
697 GOTO(out, rc = -ENOSPC);
699 spin_lock(&ofd->ofd_osfs_lock);
700 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
703 /* statfs data are too old, get up-to-date one.
704 * we must be cautious here since multiple threads might be
705 * willing to update statfs data concurrently and we must
706 * grant that cached statfs data are always consistent */
708 if (ofd->ofd_statfs_inflight == 0)
709 /* clear inflight counter if no users, although it would
710 * take a while to overflow this 64-bit counter ... */
711 ofd->ofd_osfs_inflight = 0;
712 /* notify ofd_grant_commit() that we want to track writes
713 * completed as of now */
714 ofd->ofd_statfs_inflight++;
715 /* record value of inflight counter before running statfs to
716 * compute the diff once statfs is completed */
717 unstable = ofd->ofd_osfs_inflight;
718 spin_unlock(&ofd->ofd_osfs_lock);
720 /* statfs can sleep ... hopefully not for too long since we can
721 * call it fairly often as space fills up */
722 rc = dt_statfs(env, ofd->ofd_osd, osfs);
726 spin_lock(&ofd->ofd_grant_lock);
727 spin_lock(&ofd->ofd_osfs_lock);
728 /* calculate how much space was written while we released the
730 unstable = ofd->ofd_osfs_inflight - unstable;
731 ofd->ofd_osfs_unstable = 0;
733 /* some writes completed while we were running statfs
734 * w/o the ofd_osfs_lock. Those ones got added to
735 * the cached statfs data that we are about to crunch.
736 * Take them into account in the new statfs data */
737 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
738 unstable >> ofd->ofd_blockbits);
739 /* However, we don't really know if those writes got
740 * accounted in the statfs call, so tell
741 * ofd_grant_space_left() there is some uncertainty
742 * on the accounting of those writes.
743 * The purpose is to prevent spurious error messages in
744 * ofd_grant_space_left() since those writes might be
745 * accounted twice. */
746 ofd->ofd_osfs_unstable += unstable;
748 /* similarly, there is some uncertainty on write requests
749 * between prepare & commit */
750 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
751 spin_unlock(&ofd->ofd_grant_lock);
753 /* finally udpate cached statfs data */
754 ofd->ofd_osfs = *osfs;
755 ofd->ofd_osfs_age = cfs_time_current_64();
757 ofd->ofd_statfs_inflight--; /* stop tracking */
758 if (ofd->ofd_statfs_inflight == 0)
759 ofd->ofd_osfs_inflight = 0;
760 spin_unlock(&ofd->ofd_osfs_lock);
765 /* use cached statfs data */
766 *osfs = ofd->ofd_osfs;
767 spin_unlock(&ofd->ofd_osfs_lock);
775 up_read(&ofd->ofd_lastid_rwsem);
781 * Implementation of obd_ops::o_statfs.
783 * This function returns information about a storage file system.
784 * It is called from several places by using the OBD API as well as
785 * by direct call, e.g. from request handler.
787 * \see ofd_statfs_hdl() for request handler function.
789 * Report also the state of the OST to the caller in osfs->os_state
790 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
792 * \param[in] env execution environment
793 * \param[in] exp OBD export of OFD device
794 * \param[out] osfs statistic data to return
795 * \param[in] max_age maximum age for cached data
796 * \param[in] flags not used in OFD
798 * \retval 0 if successful
799 * \retval negative value on error
801 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
802 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
804 struct obd_device *obd = class_exp2obd(exp);
805 struct ofd_device *ofd = ofd_exp(exp);
810 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
814 /* at least try to account for cached pages. its still racy and
815 * might be under-reporting if clients haven't announced their
816 * caches with brw recently */
818 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
819 " pending "LPU64" free "LPU64" avail "LPU64"\n",
820 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
821 osfs->os_bfree << ofd->ofd_blockbits,
822 osfs->os_bavail << ofd->ofd_blockbits);
824 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
825 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
826 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
828 /* The QoS code on the MDS does not care about space reserved for
829 * precreate, so take it out. */
830 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
831 struct filter_export_data *fed;
833 fed = &obd->obd_self_export->exp_filter_data;
834 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
835 fed->fed_grant >> ofd->ofd_blockbits);
838 ofd_grant_sanity_check(obd, __FUNCTION__);
839 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
840 LPU64" objects: "LPU64" free; state %x\n",
841 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
842 osfs->os_files, osfs->os_ffree, osfs->os_state);
844 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
845 ofd->ofd_lut.lut_lsd.lsd_osd_index))
848 /* OS_STATE_READONLY can be set by OSD already */
849 if (ofd->ofd_raid_degraded)
850 osfs->os_state |= OS_STATE_DEGRADED;
852 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
853 /* clients which don't support OBD_CONNECT_GRANT_PARAM
854 * should not see a block size > page size, otherwise
855 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
856 * block size which is the biggest block size known to work
857 * with all client's page size. */
858 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
859 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
860 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
861 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
864 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
865 ofd->ofd_lut.lut_lsd.lsd_osd_index))
866 osfs->os_bfree = osfs->os_bavail = 2;
874 * Implementation of obd_ops::o_setattr.
876 * This function is only used by ECHO client when it is run on top of OFD,
877 * \see ofd_setattr_hdl() for request handler function.
879 * \param[in] env execution environment
880 * \param[in] exp OBD export of OFD device
881 * \param[in] oinfo obd_info with setattr parameters
882 * \param[in] oti not used in OFD
884 * \retval 0 if successful
885 * \retval negative value on error
887 int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
888 struct obd_info *oinfo, struct obd_trans_info *oti)
890 struct ofd_thread_info *info;
891 struct ofd_device *ofd = ofd_exp(exp);
892 struct ldlm_namespace *ns = ofd->ofd_namespace;
893 struct ldlm_resource *res;
894 struct ofd_object *fo;
895 struct obdo *oa = oinfo->oi_oa;
896 struct lu_fid *fid = &oa->o_oi.oi_fid;
897 struct filter_fid *ff = NULL;
902 info = ofd_info_init(env, exp);
904 ost_fid_build_resid(fid, &info->fti_resid);
906 /* This would be very bad - accidentally truncating a file when
907 * changing the time or similar - bug 12203. */
908 if (oa->o_valid & OBD_MD_FLSIZE &&
909 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
910 static char mdsinum[48];
912 if (oa->o_valid & OBD_MD_FLFID)
913 snprintf(mdsinum, sizeof(mdsinum) - 1,
914 "of parent "DFID, oa->o_parent_seq,
915 oa->o_parent_oid, 0);
919 CERROR("%s: setattr from %s trying to truncate object "DFID
920 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
922 GOTO(out, rc = -EPERM);
925 fo = ofd_object_find_exists(env, ofd, fid);
927 CERROR("%s: can't find object "DFID"\n",
928 ofd_name(ofd), PFID(fid));
929 GOTO(out, rc = PTR_ERR(fo));
932 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
933 info->fti_attr.la_valid &= ~LA_TYPE;
935 if (oa->o_valid & OBD_MD_FLFID) {
936 ff = &info->fti_mds_fid;
937 ofd_prepare_fidea(ff, oa);
940 /* setting objects attributes (including owner/group) */
941 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
943 GOTO(out_unlock, rc);
945 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
948 ofd_object_put(env, fo);
951 /* we do not call this before to avoid lu_object_find() in
952 * ->lvbo_update() holding another reference on the object.
953 * otherwise concurrent destroy can make the object unavailable
954 * for 2nd lu_object_find() waiting for the first reference
955 * to go... deadlock! */
956 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
958 ldlm_res_lvbo_update(res, NULL, 0);
959 ldlm_resource_putref(res);
967 * Destroy OFD object by its FID.
969 * Supplemental function to destroy object by FID, it is used by request
970 * handler and by ofd_echo_destroy() below to find object by FID, lock it
971 * and call ofd_object_destroy() finally.
973 * \param[in] env execution environment
974 * \param[in] ofd OFD device
975 * \param[in] fid FID of object
976 * \param[in] orphan set if object being destroyed is an orphan
978 * \retval 0 if successful
979 * \retval negative value on error
981 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
982 const struct lu_fid *fid, int orphan)
984 struct ofd_thread_info *info = ofd_info(env);
985 struct lustre_handle lockh;
986 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
988 ldlm_policy_data_t policy = {
989 .l_extent = { 0, OBD_OBJECT_EOF }
991 struct ofd_object *fo;
995 fo = ofd_object_find_exists(env, ofd, fid);
999 /* Tell the clients that the object is gone now and that they should
1000 * throw away any cached pages. */
1001 ost_fid_build_resid(fid, &info->fti_resid);
1002 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1003 LDLM_EXTENT, &policy, LCK_PW, &flags,
1004 ldlm_blocking_ast, ldlm_completion_ast,
1005 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1007 /* We only care about the side-effects, just drop the lock. */
1009 ldlm_lock_decref(&lockh, LCK_PW);
1011 LASSERT(fo != NULL);
1013 rc = ofd_object_destroy(env, fo, orphan);
1016 ofd_object_put(env, fo);
1021 * Implementation of obd_ops::o_destroy.
1023 * This function is only used by ECHO client when it is run on top of OFD,
1024 * \see ofd_destroy_hdl() for request handler function.
1026 * \param[in] env execution environment
1027 * \param[in] exp OBD export of OFD device
1028 * \param[in] oa obdo structure with FID
1029 * \param[in] md not used in OFD
1030 * \param[in] oti not used in OFD
1031 * \param[in] md_exp not used in OFD
1032 * \param[in] capa not used in OFD
1034 * Note: this is OBD API method which is common API for server OBDs and
1035 * client OBDs. Thus some parameters used in client OBDs may not be used
1036 * on server OBDs and vice versa.
1038 * \retval 0 if successful
1039 * \retval negative value on error
1041 int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1042 struct obdo *oa, struct lov_stripe_md *md,
1043 struct obd_trans_info *oti, struct obd_export *md_exp,
1046 struct ofd_device *ofd = ofd_exp(exp);
1047 struct lu_fid *fid = &oa->o_oi.oi_fid;
1052 ofd_info_init(env, exp);
1054 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1056 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1057 if (rc == -ENOENT) {
1058 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1059 ofd_name(ofd), PFID(fid));
1061 } else if (rc != 0) {
1062 CERROR("%s: error destroying object "DFID": %d\n",
1063 ofd_name(ofd), PFID(fid), rc);
1072 * Implementation of obd_ops::o_create.
1074 * This function is only used by ECHO client when it is run on top of OFD
1075 * and just creates an object.
1076 * \see ofd_create_hdl() for request handler function.
1078 * \param[in] env execution environment
1079 * \param[in] exp OBD export of OFD device
1080 * \param[in] oa obdo structure with FID sequence to use
1081 * \param[out] ea contains object ID/SEQ to return
1082 * \param[in] oti not used in OFD
1084 * Note: this is OBD API method which is common API for server OBDs and
1085 * client OBDs. Thus some parameters used in client OBDs may not be used
1086 * on server OBDs and vice versa.
1088 * \retval 0 if successful
1089 * \retval negative value on error
1091 int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1092 struct obdo *oa, struct lov_stripe_md **ea,
1093 struct obd_trans_info *oti)
1095 struct ofd_device *ofd = ofd_exp(exp);
1096 struct ofd_thread_info *info;
1097 obd_seq seq = ostid_seq(&oa->o_oi);
1098 struct ofd_seq *oseq;
1099 int rc = 0, diff = 1;
1105 info = ofd_info_init(env, exp);
1107 LASSERT(seq == FID_SEQ_ECHO);
1108 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1110 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1112 down_read(&ofd->ofd_lastid_rwsem);
1113 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1114 * we may do that in the future.
1115 * Return -ENOSPC until the LAST_ID rebuilt. */
1116 if (unlikely(ofd->ofd_lastid_rebuilding))
1117 GOTO(out_sem, rc = -ENOSPC);
1119 oseq = ofd_seq_load(env, ofd, seq);
1121 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1122 ofd_name(ofd), seq, PTR_ERR(oseq));
1123 GOTO(out_sem, rc = -EINVAL);
1126 mutex_lock(&oseq->os_create_lock);
1127 rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1129 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1130 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
1135 next_id = ofd_seq_last_oid(oseq) + 1;
1136 count = ofd_precreate_batch(ofd, diff);
1138 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1140 CERROR("%s: unable to precreate: rc = %d\n",
1143 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1144 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1148 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
1150 mutex_unlock(&oseq->os_create_lock);
1151 if (rc == 0 && ea != NULL) {
1152 struct lov_stripe_md *lsm = *ea;
1154 lsm->lsm_oi = oa->o_oi;
1156 ofd_seq_put(env, oseq);
1159 up_read(&ofd->ofd_lastid_rwsem);
1164 * Implementation of obd_ops::o_getattr.
1166 * This function is only used by ECHO client when it is run on top of OFD
1167 * and returns attributes of object.
1168 * \see ofd_getattr_hdl() for request handler function.
1170 * \param[in] env execution environment
1171 * \param[in] exp OBD export of OFD device
1172 * \param[in,out] oinfo contains FID of object to get attributes from and
1173 * is used to return attributes back
1175 * \retval 0 if successful
1176 * \retval negative value on error
1178 int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1179 struct obd_info *oinfo)
1181 struct ofd_device *ofd = ofd_exp(exp);
1182 struct ofd_thread_info *info;
1183 struct lu_fid *fid = &oinfo->oi_oa->o_oi.oi_fid;
1184 struct ofd_object *fo;
1189 info = ofd_info_init(env, exp);
1191 fo = ofd_object_find_exists(env, ofd, fid);
1193 GOTO(out, rc = PTR_ERR(fo));
1195 LASSERT(fo != NULL);
1196 rc = ofd_attr_get(env, fo, &info->fti_attr);
1197 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1201 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1202 OFD_VALID_FLAGS | LA_UID | LA_GID);
1204 /* Store object version in reply */
1205 curr_version = dt_version_get(env, ofd_object_child(fo));
1206 if ((__s64)curr_version != -EOPNOTSUPP) {
1207 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1208 oinfo->oi_oa->o_data_version = curr_version;
1212 ofd_object_put(env, fo);
1218 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1220 * This is supplemental function for ofd_iocontrol() to return object
1221 * version for lctl tool.
1223 * \param[in] env execution environment
1224 * \param[in] ofd OFD device
1225 * \param[out] karg ioctl data
1227 * \retval 0 if successful
1228 * \retval negative value on error
1230 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1231 struct ofd_device *ofd, void *karg)
1233 struct obd_ioctl_data *data = karg;
1235 struct ofd_object *fo;
1236 dt_obj_version_t version;
1241 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1242 GOTO(out, rc = -EINVAL);
1244 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1245 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1246 } else if (data->ioc_inlbuf3 != NULL &&
1247 data->ioc_inllen3 == sizeof(__u64) &&
1248 data->ioc_inlbuf4 != NULL &&
1249 data->ioc_inllen4 == sizeof(__u64)) {
1250 struct ost_id ostid;
1252 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1253 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1254 rc = ostid_to_fid(&fid, &ostid,
1255 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1259 GOTO(out, rc = -EINVAL);
1262 if (!fid_is_sane(&fid))
1263 GOTO(out, rc = -EINVAL);
1265 fo = ofd_object_find(env, ofd, &fid);
1267 GOTO(out, rc = PTR_ERR(fo));
1269 if (!ofd_object_exists(fo))
1270 GOTO(out_fo, rc = -ENOENT);
1272 if (lu_object_remote(&fo->ofo_obj.do_lu))
1273 GOTO(out_fo, rc = -EREMOTE);
1275 version = dt_version_get(env, ofd_object_child(fo));
1277 GOTO(out_fo, rc = -EIO);
1279 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1283 ofd_object_put(env, fo);
1289 * Implementation of obd_ops::o_iocontrol.
1291 * This is OFD ioctl handling function which is primary interface for
1292 * Lustre tools like lfs, lctl and lfsck.
1294 * \param[in] cmd ioctl command
1295 * \param[in] exp OBD export of OFD
1296 * \param[in] len not used
1297 * \param[in,out] karg buffer with data
1298 * \param[in] uarg not used
1300 * \retval 0 if successful
1301 * \retval negative value on error
1303 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1304 void *karg, void *uarg)
1307 struct ofd_device *ofd = ofd_exp(exp);
1308 struct obd_device *obd = ofd_obd(ofd);
1313 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1314 rc = lu_env_init(&env, LCT_DT_THREAD);
1319 case OBD_IOC_ABORT_RECOVERY:
1320 CERROR("%s: aborting recovery\n", obd->obd_name);
1321 target_stop_recovery_thread(obd);
1324 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1325 rc = dt_sync(&env, ofd->ofd_osd);
1327 case OBD_IOC_SET_READONLY:
1328 rc = dt_sync(&env, ofd->ofd_osd);
1330 rc = dt_ro(&env, ofd->ofd_osd);
1332 case OBD_IOC_START_LFSCK: {
1333 struct obd_ioctl_data *data = karg;
1334 struct lfsck_start_param lsp;
1336 if (unlikely(data == NULL)) {
1341 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1342 lsp.lsp_index_valid = 0;
1343 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1346 case OBD_IOC_STOP_LFSCK: {
1347 struct obd_ioctl_data *data = karg;
1348 struct lfsck_stop stop;
1350 stop.ls_status = LS_STOPPED;
1351 /* Old lfsck utils may pass NULL @stop. */
1352 if (data->ioc_inlbuf1 == NULL)
1356 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1358 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1361 case OBD_IOC_GET_OBJ_VERSION:
1362 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1365 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1374 * Implementation of obd_ops::o_precleanup.
1376 * This function stops device activity before shutting it down. It is called
1377 * from a cleanup function upon forceful device cleanup. For OFD there are no
1378 * special actions, it just invokes target_recovery_cleanup().
1380 * \param[in] obd OBD device of OFD
1381 * \param[in] stage cleanup stage
1385 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1392 case OBD_CLEANUP_EARLY:
1394 case OBD_CLEANUP_EXPORTS:
1395 target_cleanup_recovery(obd);
1402 * Implementation of obd_ops::o_ping.
1404 * This is OFD-specific part of OBD_PING request handling.
1405 * It controls Filter Modification Data (FMD) expiration each time PING is
1408 * \see ofd_fmd_expire() and ofd_fmd.c for details
1410 * \param[in] env execution environment
1411 * \param[in] exp OBD export of client
1415 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1417 ofd_fmd_expire(exp);
1422 * Implementation of obd_ops::o_health_check.
1424 * This function checks the OFD device health - ability to respond on
1425 * incoming requests. There are two health_check methods:
1426 * - get statfs from the OSD. It checks just responsiveness of
1428 * - do write attempt on bottom device to check it is fully operational and
1429 * is not stuck. This is expensive method and requires special configuration
1430 * option --enable-health-write while building Lustre, it is turned off
1433 * \param[in] nul not used
1434 * \param[in] obd OBD device of OFD
1436 * \retval 0 if successful
1437 * \retval negative value in case of error
1439 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1441 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1442 struct ofd_thread_info *info;
1444 #ifdef USE_HEALTH_CHECK_WRITE
1449 /* obd_proc_read_health pass NULL env, we need real one */
1450 rc = lu_env_init(&env, LCT_DT_THREAD);
1454 info = ofd_info_init(&env, NULL);
1455 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1459 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1460 GOTO(out, rc = -EROFS);
1462 #ifdef USE_HEALTH_CHECK_WRITE
1463 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1464 if (info->fti_buf.lb_buf == NULL)
1465 GOTO(out, rc = -ENOMEM);
1467 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1470 th = dt_trans_create(&env, ofd->ofd_osd);
1472 GOTO(out, rc = PTR_ERR(th));
1474 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1475 &info->fti_buf, info->fti_off, th);
1477 th->th_sync = 1; /* sync IO is needed */
1478 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1480 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1481 &info->fti_buf, &info->fti_off,
1484 dt_trans_stop(&env, ofd->ofd_osd, th);
1486 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1488 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1495 struct obd_ops ofd_obd_ops = {
1496 .o_owner = THIS_MODULE,
1497 .o_connect = ofd_obd_connect,
1498 .o_reconnect = ofd_obd_reconnect,
1499 .o_disconnect = ofd_obd_disconnect,
1500 .o_create = ofd_echo_create,
1501 .o_statfs = ofd_statfs,
1502 .o_setattr = ofd_echo_setattr,
1503 .o_preprw = ofd_preprw,
1504 .o_commitrw = ofd_commitrw,
1505 .o_destroy = ofd_echo_destroy,
1506 .o_init_export = ofd_init_export,
1507 .o_destroy_export = ofd_destroy_export,
1508 .o_postrecov = ofd_obd_postrecov,
1509 .o_getattr = ofd_echo_getattr,
1510 .o_iocontrol = ofd_iocontrol,
1511 .o_precleanup = ofd_precleanup,
1513 .o_health_check = ofd_health_check,
1514 .o_set_info_async = ofd_set_info_async,
1515 .o_get_info = ofd_get_info,