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, 2015, 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;
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);
84 /* Mask error for already created /proc entries */
85 RETURN(rc == -EALREADY ? 0 : rc);
87 stats = exp->exp_nid_stats;
88 stats->nid_stats = lprocfs_alloc_stats(NUM_OBD_STATS +
90 LPROCFS_STATS_FLAG_NOPERCPU);
91 if (stats->nid_stats == NULL)
94 lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
96 ofd_stats_counter_init(stats->nid_stats);
98 rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
100 lprocfs_free_stats(&stats->nid_stats);
104 rc = lprocfs_nid_ldlm_stats_init(stats);
113 * Match client and OST server connection feature flags.
115 * Compute the compatibility flags for a connection request based on
116 * features mutually supported by client and server.
118 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
119 * must not be updated here, otherwise a partially initialized value may
120 * be exposed. After the connection request is successfully processed,
121 * the top-level tgt_connect() request handler atomically updates the export
122 * connect flags from the obd_connect_data::ocd_connect_flags field of the
123 * reply. \see tgt_connect().
125 * \param[in] env execution environment
126 * \param[in] exp the obd_export associated with this
128 * \param[in] data stores data for this connect request
129 * \param[in] new_connection is this connection new or not
131 * \retval 0 if success
132 * \retval -EPROTO client and server feature requirements are
134 * \retval -EBADF OST index in connect request doesn't match
137 static int ofd_parse_connect_data(const struct lu_env *env,
138 struct obd_export *exp,
139 struct obd_connect_data *data,
142 struct ofd_device *ofd = ofd_exp(exp);
143 struct filter_export_data *fed = &exp->exp_filter_data;
148 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
149 " ocd_version: %x ocd_grant: %d ocd_index: %u"
151 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
152 data->ocd_connect_flags, data->ocd_version,
153 data->ocd_grant, data->ocd_index, data->ocd_group);
155 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
156 CWARN("!!! This export (nid %s) used object group %d "
157 "earlier; now it's trying to use group %d! This could "
158 "be a bug in the MDS. Please report to "
159 "https://jira.hpdd.intel.com/\n",
160 obd_export_nid2str(exp), fed->fed_group,
164 fed->fed_group = data->ocd_group;
166 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
167 data->ocd_version = LUSTRE_VERSION_CODE;
169 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
170 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
171 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
172 exp->exp_obd->obd_name, data->ocd_group);
173 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
176 /* Determine optimal brw size before calculating grant */
177 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
178 data->ocd_brw_size = 65536;
179 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
180 data->ocd_brw_size = min(data->ocd_brw_size, ofd->ofd_brw_size);
181 if (data->ocd_brw_size == 0) {
182 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
183 " ocd_version: %x ocd_grant: %d ocd_index: %u "
184 "ocd_brw_size is unexpectedly zero, "
185 "network data corruption?"
186 "Refusing connection of this client\n",
187 exp->exp_obd->obd_name,
188 exp->exp_client_uuid.uuid,
189 exp, data->ocd_connect_flags, data->ocd_version,
190 data->ocd_grant, data->ocd_index);
195 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
196 /* client is reporting its page size, for future use */
197 exp->exp_filter_data.fed_pagebits = data->ocd_grant_blkbits;
198 data->ocd_grant_blkbits = ofd->ofd_blockbits;
199 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
200 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
201 data->ocd_grant_inobits =
202 fls(ofd->ofd_dt_conf.ddp_inodespace - 1);
203 /* ocd_grant_tax_kb is in 1K byte blocks */
204 data->ocd_grant_tax_kb = ofd->ofd_dt_conf.ddp_extent_tax >> 10;
205 data->ocd_grant_max_blks = ofd->ofd_dt_conf.ddp_max_extent_blks;
208 if (OCD_HAS_FLAG(data, GRANT)) {
209 /* Save connect_data we have so far because ofd_grant_connect()
210 * uses it to calculate grant. */
211 exp->exp_connect_data = *data;
212 ofd_grant_connect(env, exp, data, new_connection);
215 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
216 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
217 int index = lsd->lsd_osd_index;
219 if (index != data->ocd_index) {
220 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
221 " %u doesn't match actual OST index"
222 " %u in last_rcvd file, bad "
224 obd_export_nid2str(exp), index,
228 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
229 /* this will only happen on the first connect */
230 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
231 /* sync is not needed here as lut_client_add will
232 * set exp_need_sync flag */
233 tgt_server_data_update(env, &ofd->ofd_lut, 0);
237 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
238 __u32 cksum_types = data->ocd_cksum_types;
240 /* The client set in ocd_cksum_types the checksum types it
241 * supports. We have to mask off the algorithms that we don't
243 data->ocd_cksum_types &= cksum_types_supported_server();
245 if (unlikely(data->ocd_cksum_types == 0)) {
246 CERROR("%s: Connect with checksum support but no "
247 "ocd_cksum_types is set\n",
248 exp->exp_obd->obd_name);
252 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
253 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
254 cksum_types, data->ocd_cksum_types);
256 /* This client does not support OBD_CONNECT_CKSUM
257 * fall back to CRC32 */
258 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
259 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
260 exp->exp_obd->obd_name, obd_export_nid2str(exp));
263 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
264 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
266 if (OCD_HAS_FLAG(data, PINGLESS)) {
267 if (ptlrpc_pinger_suppress_pings()) {
268 spin_lock(&exp->exp_obd->obd_dev_lock);
269 list_del_init(&exp->exp_obd_chain_timed);
270 spin_unlock(&exp->exp_obd->obd_dev_lock);
272 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
280 * Re-initialize export upon client reconnection.
282 * This function parses connection data from reconnect and resets
285 * \param[in] env execution environment
286 * \param[in] exp OBD export
287 * \param[in] obd OFD device
288 * \param[in] cluuid NID of client
289 * \param[in] data connection data from request
290 * \param[in] localdata client NID
292 * \retval 0 if successful
293 * \retval negative value on error
295 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
296 struct obd_device *obd, struct obd_uuid *cluuid,
297 struct obd_connect_data *data,
300 struct ofd_device *ofd;
305 if (exp == NULL || obd == NULL || cluuid == NULL)
308 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
309 if (rc != 0 && rc != -EEXIST)
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_del_member(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 };
350 if (_exp == NULL || obd == NULL || cluuid == NULL)
353 ofd = ofd_dev(obd->obd_lu_dev);
355 rc = class_connect(&conn, obd, cluuid);
359 exp = class_conn2export(&conn);
360 LASSERT(exp != NULL);
362 if (localdata != NULL) {
363 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
364 if (rc != 0 && rc != -EEXIST)
367 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
368 "nid is null\n", obd->obd_name, cluuid->uuid);
371 rc = ofd_parse_connect_data(env, exp, data, true);
375 if (obd->obd_replayable) {
376 struct tg_export_data *ted = &exp->exp_target_data;
378 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
379 sizeof(ted->ted_lcd->lcd_uuid));
380 rc = tgt_client_new(env, exp);
383 ofd_export_stats_init(ofd, exp, localdata);
386 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
387 data ? data->ocd_group : -1);
391 class_disconnect(exp);
392 nodemap_del_member(exp);
401 * Disconnect a connected client.
403 * This function terminates the client connection. The client export is
404 * disconnected (cleaned up) and client data on persistent storage is removed.
406 * \param[in] exp OBD export
408 * \retval 0 if successful
409 * \retval negative value on error
411 int ofd_obd_disconnect(struct obd_export *exp)
413 struct ofd_device *ofd = ofd_exp(exp);
420 class_export_get(exp);
422 if (!(exp->exp_flags & OBD_OPT_FORCE))
423 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
425 rc = server_disconnect_export(exp);
427 ofd_grant_discard(exp);
429 /* Do not erase record for recoverable client. */
430 if (exp->exp_obd->obd_replayable &&
431 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
432 rc = lu_env_init(&env, LCT_DT_THREAD);
436 tgt_client_del(&env, exp);
440 nodemap_del_member(exp);
441 class_export_put(exp);
446 * Implementation of obd_ops::o_init_export.
448 * This function is called from class_new_export() and initializes
449 * the OFD-specific data for new export.
451 * \param[in] exp OBD export
453 * \retval 0 if successful
454 * \retval negative value on error
456 static int ofd_init_export(struct obd_export *exp)
460 spin_lock_init(&exp->exp_filter_data.fed_lock);
461 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
462 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
463 spin_lock(&exp->exp_lock);
464 exp->exp_connecting = 1;
465 spin_unlock(&exp->exp_lock);
467 /* self-export doesn't need client data and ldlm initialization */
468 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
469 &exp->exp_client_uuid)))
472 rc = tgt_client_alloc(exp);
474 ldlm_init_export(exp);
476 CERROR("%s: Can't initialize export: rc %d\n",
477 exp->exp_obd->obd_name, rc);
482 * Implementation of obd_ops::o_destroy_export.
484 * This function is called from class_export_destroy() to cleanup
485 * the OFD-specific data for export being destroyed.
487 * \param[in] exp OBD export
489 * \retval 0 if successful
490 * \retval negative value on error
492 static int ofd_destroy_export(struct obd_export *exp)
494 struct ofd_device *ofd = ofd_exp(exp);
496 if (exp->exp_filter_data.fed_pending)
497 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
498 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
499 exp, exp->exp_filter_data.fed_pending);
501 target_destroy_export(exp);
503 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
504 &exp->exp_client_uuid)))
507 ldlm_destroy_export(exp);
508 tgt_client_free(exp);
510 ofd_fmd_cleanup(exp);
513 * discard grants once we're sure no more
514 * interaction with the client is possible
516 ofd_grant_discard(exp);
517 ofd_fmd_cleanup(exp);
519 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
520 ofd->ofd_tot_granted_clients--;
522 if (!(exp->exp_flags & OBD_OPT_FORCE))
523 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
525 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
530 * Notify all devices in server stack about recovery completion.
532 * This function calls ldo_recovery_complete() for all lower devices in the
533 * server stack so they will be prepared for normal operations.
535 * \param[in] env execution environment
536 * \param[in] ofd OFD device
538 * \retval 0 if successful
539 * \retval negative value on error
541 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
543 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
546 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
548 if (!ofd->ofd_skip_lfsck) {
549 struct lfsck_start_param lsp;
551 lsp.lsp_start = NULL;
552 lsp.lsp_index_valid = 0;
553 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
554 if (rc != 0 && rc != -EALREADY)
555 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
559 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
563 * Implementation of obd_ops::o_postrecov.
565 * This function is called from target_finish_recovery() upon recovery
568 * \param[in] obd OBD device of OFD
570 * \retval 0 if successful
571 * \retval negative value on error
573 static int ofd_obd_postrecov(struct obd_device *obd)
576 struct lu_device *ldev = obd->obd_lu_dev;
581 rc = lu_env_init(&env, LCT_DT_THREAD);
584 ofd_info_init(&env, obd->obd_self_export);
586 rc = ofd_postrecov(&env, ofd_dev(ldev));
593 * Implementation of obd_ops::o_set_info_async.
595 * This function is not called from request handler, it is only used by
596 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
597 * \see ofd_set_info_hdl() for request handler function.
599 * \param[in] env execution environment
600 * \param[in] exp OBD export of OFD device
601 * \param[in] keylen length of \a key
602 * \param[in] key key name
603 * \param[in] vallen length of \a val
604 * \param[in] val the \a key value
605 * \param[in] set not used in OFD
607 * \retval 0 if successful
608 * \retval negative value on error
610 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
611 __u32 keylen, void *key, __u32 vallen, void *val,
612 struct ptlrpc_request_set *set)
618 if (exp->exp_obd == NULL) {
619 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
623 if (KEY_IS(KEY_SPTLRPC_CONF)) {
624 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
626 CERROR("%s: Unsupported key %s\n",
627 exp->exp_obd->obd_name, (char*)key);
634 * Implementation of obd_ops::o_get_info.
636 * This function is not called from request handler, it is only used by
637 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
639 * \see ofd_get_info_hdl() for request handler function.
641 * \param[in] env execution environment
642 * \param[in] exp OBD export of OFD device
643 * \param[in] keylen length of \a key
644 * \param[in] key key name
645 * \param[out] vallen length of key value
646 * \param[out] val the key value to return
648 * \retval 0 if successful
649 * \retval negative value on error
651 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
652 __u32 keylen, void *key, __u32 *vallen, void *val)
654 struct ofd_thread_info *info;
655 struct ofd_device *ofd;
656 struct ll_fiemap_info_key *fm_key = key;
657 struct fiemap *fiemap = val;
662 if (exp->exp_obd == NULL) {
663 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
669 if (KEY_IS(KEY_FIEMAP)) {
670 info = ofd_info_init(env, exp);
672 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
673 ofd->ofd_lut.lut_lsd.lsd_osd_index);
677 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
679 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
687 * Get file system statistics of OST server.
689 * Helper function for ofd_statfs(), also used by grant code.
690 * Implements caching for statistics to avoid calling OSD device each time.
692 * \param[in] env execution environment
693 * \param[in] ofd OFD device
694 * \param[out] osfs statistic data to return
695 * \param[in] max_age maximum age for cached data
696 * \param[in] from_cache show that data was get from cache or not
698 * \retval 0 if successful
699 * \retval negative value on error
701 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
702 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
707 spin_lock(&ofd->ofd_osfs_lock);
708 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
711 /* statfs data are too old, get up-to-date one.
712 * we must be cautious here since multiple threads might be
713 * willing to update statfs data concurrently and we must
714 * grant that cached statfs data are always consistent */
716 if (ofd->ofd_statfs_inflight == 0)
717 /* clear inflight counter if no users, although it would
718 * take a while to overflow this 64-bit counter ... */
719 ofd->ofd_osfs_inflight = 0;
720 /* notify ofd_grant_commit() that we want to track writes
721 * completed as of now */
722 ofd->ofd_statfs_inflight++;
723 /* record value of inflight counter before running statfs to
724 * compute the diff once statfs is completed */
725 unstable = ofd->ofd_osfs_inflight;
726 spin_unlock(&ofd->ofd_osfs_lock);
728 /* statfs can sleep ... hopefully not for too long since we can
729 * call it fairly often as space fills up */
730 rc = dt_statfs(env, ofd->ofd_osd, osfs);
734 spin_lock(&ofd->ofd_grant_lock);
735 spin_lock(&ofd->ofd_osfs_lock);
736 /* calculate how much space was written while we released the
738 unstable = ofd->ofd_osfs_inflight - unstable;
739 ofd->ofd_osfs_unstable = 0;
741 /* some writes committed while we were running statfs
742 * w/o the ofd_osfs_lock. Those ones got added to
743 * the cached statfs data that we are about to crunch.
744 * Take them into account in the new statfs data */
745 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
746 unstable >> ofd->ofd_blockbits);
747 /* However, we don't really know if those writes got
748 * accounted in the statfs call, so tell
749 * ofd_grant_space_left() there is some uncertainty
750 * on the accounting of those writes.
751 * The purpose is to prevent spurious error messages in
752 * ofd_grant_space_left() since those writes might be
753 * accounted twice. */
754 ofd->ofd_osfs_unstable += unstable;
756 /* similarly, there is some uncertainty on write requests
757 * between prepare & commit */
758 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
759 spin_unlock(&ofd->ofd_grant_lock);
761 /* finally udpate cached statfs data */
762 ofd->ofd_osfs = *osfs;
763 ofd->ofd_osfs_age = cfs_time_current_64();
765 ofd->ofd_statfs_inflight--; /* stop tracking */
766 if (ofd->ofd_statfs_inflight == 0)
767 ofd->ofd_osfs_inflight = 0;
768 spin_unlock(&ofd->ofd_osfs_lock);
773 /* use cached statfs data */
774 *osfs = ofd->ofd_osfs;
775 spin_unlock(&ofd->ofd_osfs_lock);
786 * Implementation of obd_ops::o_statfs.
788 * This function returns information about a storage file system.
789 * It is called from several places by using the OBD API as well as
790 * by direct call, e.g. from request handler.
792 * \see ofd_statfs_hdl() for request handler function.
794 * Report also the state of the OST to the caller in osfs->os_state
795 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
797 * \param[in] env execution environment
798 * \param[in] exp OBD export of OFD device
799 * \param[out] osfs statistic data to return
800 * \param[in] max_age maximum age for cached data
801 * \param[in] flags not used in OFD
803 * \retval 0 if successful
804 * \retval negative value on error
806 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
807 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
809 struct obd_device *obd = class_exp2obd(exp);
810 struct ofd_device *ofd = ofd_exp(exp);
815 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
819 /* at least try to account for cached pages. its still racy and
820 * might be under-reporting if clients haven't announced their
821 * caches with brw recently */
823 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
824 " pending "LPU64" free "LPU64" avail "LPU64"\n",
825 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
826 osfs->os_bfree << ofd->ofd_blockbits,
827 osfs->os_bavail << ofd->ofd_blockbits);
829 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
830 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
831 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
833 /* The QoS code on the MDS does not care about space reserved for
834 * precreate, so take it out. */
835 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
836 struct filter_export_data *fed;
838 fed = &obd->obd_self_export->exp_filter_data;
839 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
840 fed->fed_grant >> ofd->ofd_blockbits);
843 ofd_grant_sanity_check(obd, __FUNCTION__);
844 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
845 LPU64" objects: "LPU64" free; state %x\n",
846 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
847 osfs->os_files, osfs->os_ffree, osfs->os_state);
849 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
850 ofd->ofd_lut.lut_lsd.lsd_osd_index))
853 /* OS_STATE_READONLY can be set by OSD already */
854 if (ofd->ofd_raid_degraded)
855 osfs->os_state |= OS_STATE_DEGRADED;
857 if (obd->obd_self_export != exp && !ofd_grant_param_supp(exp) &&
858 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT) {
859 /* clients which don't support OBD_CONNECT_GRANT_PARAM
860 * should not see a block size > page size, otherwise
861 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
862 * block size which is the biggest block size known to work
863 * with all client's page size. */
864 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
865 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
866 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
867 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
870 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
871 ofd->ofd_lut.lut_lsd.lsd_osd_index))
872 osfs->os_bfree = osfs->os_bavail = 2;
880 * Implementation of obd_ops::o_setattr.
882 * This function is only used by ECHO client when it is run on top of OFD,
883 * \see ofd_setattr_hdl() for request handler function.
885 * \param[in] env execution environment
886 * \param[in] exp OBD export of OFD device
887 * \param[in] oa setattr parameters
889 * \retval 0 if successful
890 * \retval negative value on error
892 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
895 struct ofd_thread_info *info;
896 struct ofd_device *ofd = ofd_exp(exp);
897 struct ldlm_namespace *ns = ofd->ofd_namespace;
898 struct ldlm_resource *res;
899 struct ofd_object *fo;
900 struct lu_fid *fid = &oa->o_oi.oi_fid;
901 struct filter_fid *ff = NULL;
906 info = ofd_info_init(env, exp);
908 ost_fid_build_resid(fid, &info->fti_resid);
910 /* This would be very bad - accidentally truncating a file when
911 * changing the time or similar - bug 12203. */
912 if (oa->o_valid & OBD_MD_FLSIZE) {
913 static char mdsinum[48];
915 if (oa->o_valid & OBD_MD_FLFID)
916 snprintf(mdsinum, sizeof(mdsinum) - 1,
917 "of parent "DFID, oa->o_parent_seq,
918 oa->o_parent_oid, 0);
922 CERROR("%s: setattr from %s trying to truncate object "DFID
923 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
925 GOTO(out, rc = -EPERM);
928 fo = ofd_object_find_exists(env, ofd, fid);
930 CERROR("%s: can't find object "DFID"\n",
931 ofd_name(ofd), PFID(fid));
932 GOTO(out, rc = PTR_ERR(fo));
935 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
936 info->fti_attr.la_valid &= ~LA_TYPE;
938 if (oa->o_valid & OBD_MD_FLFID) {
939 ff = &info->fti_mds_fid;
940 ofd_prepare_fidea(ff, oa);
943 /* setting objects attributes (including owner/group) */
944 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
946 GOTO(out_unlock, rc);
948 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
951 ofd_object_put(env, fo);
954 /* we do not call this before to avoid lu_object_find() in
955 * ->lvbo_update() holding another reference on the object.
956 * otherwise concurrent destroy can make the object unavailable
957 * for 2nd lu_object_find() waiting for the first reference
958 * to go... deadlock! */
959 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
961 ldlm_res_lvbo_update(res, NULL, 0);
962 ldlm_resource_putref(res);
970 * Destroy OFD object by its FID.
972 * Supplemental function to destroy object by FID, it is used by request
973 * handler and by ofd_echo_destroy() below to find object by FID, lock it
974 * and call ofd_object_destroy() finally.
976 * \param[in] env execution environment
977 * \param[in] ofd OFD device
978 * \param[in] fid FID of object
979 * \param[in] orphan set if object being destroyed is an orphan
981 * \retval 0 if successful
982 * \retval negative value on error
984 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
985 const struct lu_fid *fid, int orphan)
987 struct ofd_thread_info *info = ofd_info(env);
988 struct lustre_handle lockh;
989 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
990 struct ofd_object *fo;
991 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
996 fo = ofd_object_find_exists(env, ofd, fid);
1000 /* Tell the clients that the object is gone now and that they should
1001 * throw away any cached pages. */
1002 ost_fid_build_resid(fid, &info->fti_resid);
1003 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1004 LDLM_EXTENT, &policy, LCK_PW, &flags,
1005 ldlm_blocking_ast, ldlm_completion_ast,
1006 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1008 /* We only care about the side-effects, just drop the lock. */
1010 ldlm_lock_decref(&lockh, LCK_PW);
1012 LASSERT(fo != NULL);
1014 rc = ofd_object_destroy(env, fo, orphan);
1017 ofd_object_put(env, fo);
1022 * Implementation of obd_ops::o_destroy.
1024 * This function is only used by ECHO client when it is run on top of OFD,
1025 * \see ofd_destroy_hdl() for request handler function.
1027 * \param[in] env execution environment
1028 * \param[in] exp OBD export of OFD device
1029 * \param[in] oa obdo structure with FID
1031 * Note: this is OBD API method which is common API for server OBDs and
1032 * client OBDs. Thus some parameters used in client OBDs may not be used
1033 * on server OBDs and vice versa.
1035 * \retval 0 if successful
1036 * \retval negative value on error
1038 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1041 struct ofd_device *ofd = ofd_exp(exp);
1042 struct lu_fid *fid = &oa->o_oi.oi_fid;
1047 ofd_info_init(env, exp);
1049 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1051 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1052 if (rc == -ENOENT) {
1053 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1054 ofd_name(ofd), PFID(fid));
1056 } else if (rc != 0) {
1057 CERROR("%s: error destroying object "DFID": %d\n",
1058 ofd_name(ofd), PFID(fid), rc);
1067 * Implementation of obd_ops::o_create.
1069 * This function is only used by ECHO client when it is run on top of OFD
1070 * and just creates an object.
1071 * \see ofd_create_hdl() for request handler function.
1073 * \param[in] env execution environment
1074 * \param[in] exp OBD export of OFD device
1075 * \param[in] oa obdo structure with FID sequence to use
1077 * Note: this is OBD API method which is common API for server OBDs and
1078 * client OBDs. Thus some parameters used in client OBDs may not be used
1079 * on server OBDs and vice versa.
1081 * \retval 0 if successful
1082 * \retval negative value on error
1084 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1087 struct ofd_device *ofd = ofd_exp(exp);
1088 struct ofd_thread_info *info;
1089 u64 seq = ostid_seq(&oa->o_oi);
1090 struct ofd_seq *oseq;
1091 int rc = 0, diff = 1;
1098 info = ofd_info_init(env, exp);
1100 LASSERT(seq == FID_SEQ_ECHO);
1101 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1103 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1105 down_read(&ofd->ofd_lastid_rwsem);
1106 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1107 * we may do that in the future.
1108 * Return -ENOSPC until the LAST_ID rebuilt. */
1109 if (unlikely(ofd->ofd_lastid_rebuilding))
1110 GOTO(out_sem, rc = -ENOSPC);
1112 oseq = ofd_seq_load(env, ofd, seq);
1114 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1115 ofd_name(ofd), seq, PTR_ERR(oseq));
1116 GOTO(out_sem, rc = -EINVAL);
1119 mutex_lock(&oseq->os_create_lock);
1120 granted = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1124 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1125 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
1130 next_id = ofd_seq_last_oid(oseq) + 1;
1131 count = ofd_precreate_batch(ofd, diff);
1133 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1135 CERROR("%s: unable to precreate: rc = %d\n",
1138 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1139 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1143 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1145 mutex_unlock(&oseq->os_create_lock);
1146 ofd_seq_put(env, oseq);
1149 up_read(&ofd->ofd_lastid_rwsem);
1154 * Implementation of obd_ops::o_getattr.
1156 * This function is only used by ECHO client when it is run on top of OFD
1157 * and returns attributes of object.
1158 * \see ofd_getattr_hdl() for request handler function.
1160 * \param[in] env execution environment
1161 * \param[in] exp OBD export of OFD device
1162 * \param[in,out] oa contains FID of object to get attributes from and
1163 * is used to return attributes back
1165 * \retval 0 if successful
1166 * \retval negative value on error
1168 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1171 struct ofd_device *ofd = ofd_exp(exp);
1172 struct ofd_thread_info *info;
1173 struct lu_fid *fid = &oa->o_oi.oi_fid;
1174 struct ofd_object *fo;
1179 info = ofd_info_init(env, exp);
1181 fo = ofd_object_find_exists(env, ofd, fid);
1183 GOTO(out, rc = PTR_ERR(fo));
1185 LASSERT(fo != NULL);
1186 rc = ofd_attr_get(env, fo, &info->fti_attr);
1187 oa->o_valid = OBD_MD_FLID;
1191 obdo_from_la(oa, &info->fti_attr,
1192 OFD_VALID_FLAGS | LA_UID | LA_GID);
1194 /* Store object version in reply */
1195 curr_version = dt_version_get(env, ofd_object_child(fo));
1196 if ((__s64)curr_version != -EOPNOTSUPP) {
1197 oa->o_valid |= OBD_MD_FLDATAVERSION;
1198 oa->o_data_version = curr_version;
1202 ofd_object_put(env, fo);
1208 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1210 * This is supplemental function for ofd_iocontrol() to return object
1211 * version for lctl tool.
1213 * \param[in] env execution environment
1214 * \param[in] ofd OFD device
1215 * \param[out] karg ioctl data
1217 * \retval 0 if successful
1218 * \retval negative value on error
1220 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1221 struct ofd_device *ofd, void *karg)
1223 struct obd_ioctl_data *data = karg;
1225 struct ofd_object *fo;
1226 dt_obj_version_t version;
1231 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1232 GOTO(out, rc = -EINVAL);
1234 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1235 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1236 } else if (data->ioc_inlbuf3 != NULL &&
1237 data->ioc_inllen3 == sizeof(__u64) &&
1238 data->ioc_inlbuf4 != NULL &&
1239 data->ioc_inllen4 == sizeof(__u64)) {
1240 struct ost_id ostid;
1242 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1243 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1244 rc = ostid_to_fid(&fid, &ostid,
1245 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1249 GOTO(out, rc = -EINVAL);
1252 if (!fid_is_sane(&fid))
1253 GOTO(out, rc = -EINVAL);
1255 fo = ofd_object_find(env, ofd, &fid);
1257 GOTO(out, rc = PTR_ERR(fo));
1259 if (!ofd_object_exists(fo))
1260 GOTO(out_fo, rc = -ENOENT);
1262 if (lu_object_remote(&fo->ofo_obj.do_lu))
1263 GOTO(out_fo, rc = -EREMOTE);
1265 version = dt_version_get(env, ofd_object_child(fo));
1267 GOTO(out_fo, rc = -EIO);
1269 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1273 ofd_object_put(env, fo);
1279 * Implementation of obd_ops::o_iocontrol.
1281 * This is OFD ioctl handling function which is primary interface for
1282 * Lustre tools like lfs, lctl and lfsck.
1284 * \param[in] cmd ioctl command
1285 * \param[in] exp OBD export of OFD
1286 * \param[in] len not used
1287 * \param[in,out] karg buffer with data
1288 * \param[in] uarg not used
1290 * \retval 0 if successful
1291 * \retval negative value on error
1293 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1294 void *karg, void __user *uarg)
1297 struct ofd_device *ofd = ofd_exp(exp);
1298 struct obd_device *obd = ofd_obd(ofd);
1303 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1304 rc = lu_env_init(&env, LCT_DT_THREAD);
1309 case OBD_IOC_ABORT_RECOVERY:
1310 CERROR("%s: aborting recovery\n", obd->obd_name);
1311 obd->obd_abort_recovery = 1;
1312 target_stop_recovery_thread(obd);
1315 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1316 rc = dt_sync(&env, ofd->ofd_osd);
1318 case OBD_IOC_SET_READONLY:
1319 rc = dt_sync(&env, ofd->ofd_osd);
1321 rc = dt_ro(&env, ofd->ofd_osd);
1323 case OBD_IOC_START_LFSCK: {
1324 struct obd_ioctl_data *data = karg;
1325 struct lfsck_start_param lsp;
1327 if (unlikely(data == NULL)) {
1332 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1333 lsp.lsp_index_valid = 0;
1334 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1337 case OBD_IOC_STOP_LFSCK: {
1338 struct obd_ioctl_data *data = karg;
1339 struct lfsck_stop stop;
1341 stop.ls_status = LS_STOPPED;
1342 /* Old lfsck utils may pass NULL @stop. */
1343 if (data->ioc_inlbuf1 == NULL)
1347 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1349 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1352 case OBD_IOC_GET_OBJ_VERSION:
1353 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1356 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1365 * Implementation of obd_ops::o_precleanup.
1367 * This function stops device activity before shutting it down. It is called
1368 * from a cleanup function upon forceful device cleanup. For OFD there are no
1369 * special actions, it just invokes target_recovery_cleanup().
1371 * \param[in] obd OBD device of OFD
1375 static int ofd_precleanup(struct obd_device *obd)
1378 target_cleanup_recovery(obd);
1383 * Implementation of obd_ops::o_ping.
1385 * This is OFD-specific part of OBD_PING request handling.
1386 * It controls Filter Modification Data (FMD) expiration each time PING is
1389 * \see ofd_fmd_expire() and ofd_fmd.c for details
1391 * \param[in] env execution environment
1392 * \param[in] exp OBD export of client
1396 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1398 ofd_fmd_expire(exp);
1403 * Implementation of obd_ops::o_health_check.
1405 * This function checks the OFD device health - ability to respond on
1406 * incoming requests. There are two health_check methods:
1407 * - get statfs from the OSD. It checks just responsiveness of
1409 * - do write attempt on bottom device to check it is fully operational and
1410 * is not stuck. This is expensive method and requires special configuration
1411 * option --enable-health-write while building Lustre, it is turned off
1414 * \param[in] nul not used
1415 * \param[in] obd OBD device of OFD
1417 * \retval 0 if successful
1418 * \retval negative value in case of error
1420 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1422 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1423 struct ofd_thread_info *info;
1425 #ifdef USE_HEALTH_CHECK_WRITE
1430 /* obd_proc_read_health pass NULL env, we need real one */
1431 rc = lu_env_init(&env, LCT_DT_THREAD);
1435 info = ofd_info_init(&env, NULL);
1436 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1440 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1441 GOTO(out, rc = -EROFS);
1443 #ifdef USE_HEALTH_CHECK_WRITE
1444 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1445 if (info->fti_buf.lb_buf == NULL)
1446 GOTO(out, rc = -ENOMEM);
1448 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1451 th = dt_trans_create(&env, ofd->ofd_osd);
1453 GOTO(out, rc = PTR_ERR(th));
1455 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1456 &info->fti_buf, info->fti_off, th);
1458 th->th_sync = 1; /* sync IO is needed */
1459 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1461 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1462 &info->fti_buf, &info->fti_off,
1465 dt_trans_stop(&env, ofd->ofd_osd, th);
1467 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1469 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1476 struct obd_ops ofd_obd_ops = {
1477 .o_owner = THIS_MODULE,
1478 .o_connect = ofd_obd_connect,
1479 .o_reconnect = ofd_obd_reconnect,
1480 .o_disconnect = ofd_obd_disconnect,
1481 .o_create = ofd_echo_create,
1482 .o_statfs = ofd_statfs,
1483 .o_setattr = ofd_echo_setattr,
1484 .o_preprw = ofd_preprw,
1485 .o_commitrw = ofd_commitrw,
1486 .o_destroy = ofd_echo_destroy,
1487 .o_init_export = ofd_init_export,
1488 .o_destroy_export = ofd_destroy_export,
1489 .o_postrecov = ofd_obd_postrecov,
1490 .o_getattr = ofd_echo_getattr,
1491 .o_iocontrol = ofd_iocontrol,
1492 .o_precleanup = ofd_precleanup,
1494 .o_health_check = ofd_health_check,
1495 .o_set_info_async = ofd_set_info_async,
1496 .o_get_info = ofd_get_info,