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_connect_flags2 &= OST_CONNECT_SUPPORTED2;
168 data->ocd_version = LUSTRE_VERSION_CODE;
170 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
171 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
172 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
173 exp->exp_obd->obd_name, data->ocd_group);
174 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
177 /* Determine optimal brw size before calculating grant */
178 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
179 data->ocd_brw_size = 65536;
180 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
181 data->ocd_brw_size = min(data->ocd_brw_size, ofd->ofd_brw_size);
182 if (data->ocd_brw_size == 0) {
183 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
184 " ocd_version: %x ocd_grant: %d ocd_index: %u "
185 "ocd_brw_size is unexpectedly zero, "
186 "network data corruption?"
187 "Refusing connection of this client\n",
188 exp->exp_obd->obd_name,
189 exp->exp_client_uuid.uuid,
190 exp, data->ocd_connect_flags, data->ocd_version,
191 data->ocd_grant, data->ocd_index);
196 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
197 /* client is reporting its page size, for future use */
198 exp->exp_filter_data.fed_pagebits = data->ocd_grant_blkbits;
199 data->ocd_grant_blkbits = ofd->ofd_blockbits;
200 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
201 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
202 data->ocd_grant_inobits =
203 fls(ofd->ofd_dt_conf.ddp_inodespace - 1);
204 /* ocd_grant_tax_kb is in 1K byte blocks */
205 data->ocd_grant_tax_kb = ofd->ofd_dt_conf.ddp_extent_tax >> 10;
206 data->ocd_grant_max_blks = ofd->ofd_dt_conf.ddp_max_extent_blks;
209 if (OCD_HAS_FLAG(data, GRANT)) {
210 /* Save connect_data we have so far because ofd_grant_connect()
211 * uses it to calculate grant. */
212 exp->exp_connect_data = *data;
213 ofd_grant_connect(env, exp, data, new_connection);
216 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
217 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
218 int index = lsd->lsd_osd_index;
220 if (index != data->ocd_index) {
221 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
222 " %u doesn't match actual OST index"
223 " %u in last_rcvd file, bad "
225 obd_export_nid2str(exp), index,
229 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
230 /* this will only happen on the first connect */
231 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
232 /* sync is not needed here as lut_client_add will
233 * set exp_need_sync flag */
234 tgt_server_data_update(env, &ofd->ofd_lut, 0);
238 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
239 __u32 cksum_types = data->ocd_cksum_types;
241 /* The client set in ocd_cksum_types the checksum types it
242 * supports. We have to mask off the algorithms that we don't
244 data->ocd_cksum_types &= cksum_types_supported_server();
246 if (unlikely(data->ocd_cksum_types == 0)) {
247 CERROR("%s: Connect with checksum support but no "
248 "ocd_cksum_types is set\n",
249 exp->exp_obd->obd_name);
253 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
254 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
255 cksum_types, data->ocd_cksum_types);
257 /* This client does not support OBD_CONNECT_CKSUM
258 * fall back to CRC32 */
259 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
260 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
261 exp->exp_obd->obd_name, obd_export_nid2str(exp));
264 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
265 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
267 if (OCD_HAS_FLAG(data, PINGLESS)) {
268 if (ptlrpc_pinger_suppress_pings()) {
269 spin_lock(&exp->exp_obd->obd_dev_lock);
270 list_del_init(&exp->exp_obd_chain_timed);
271 spin_unlock(&exp->exp_obd->obd_dev_lock);
273 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
281 * Re-initialize export upon client reconnection.
283 * This function parses connection data from reconnect and resets
286 * \param[in] env execution environment
287 * \param[in] exp OBD export
288 * \param[in] obd OFD device
289 * \param[in] cluuid NID of client
290 * \param[in] data connection data from request
291 * \param[in] localdata client NID
293 * \retval 0 if successful
294 * \retval negative value on error
296 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
297 struct obd_device *obd, struct obd_uuid *cluuid,
298 struct obd_connect_data *data,
301 struct ofd_device *ofd;
306 if (exp == NULL || obd == NULL || cluuid == NULL)
309 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
310 if (rc != 0 && rc != -EEXIST)
313 ofd = ofd_dev(obd->obd_lu_dev);
315 rc = ofd_parse_connect_data(env, exp, data, false);
317 ofd_export_stats_init(ofd, exp, client_nid);
319 nodemap_del_member(exp);
325 * Initialize new client connection.
327 * This function handles new connection to the OFD. The new export is
328 * created (in context of class_connect()) and persistent client data is
329 * initialized on storage.
331 * \param[in] env execution environment
332 * \param[out] _exp stores pointer to new export
333 * \param[in] obd OFD device
334 * \param[in] cluuid client UUID
335 * \param[in] data connection data from request
336 * \param[in] localdata client NID
338 * \retval 0 if successful
339 * \retval negative value on error
341 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
342 struct obd_device *obd, struct obd_uuid *cluuid,
343 struct obd_connect_data *data, void *localdata)
345 struct obd_export *exp;
346 struct ofd_device *ofd;
347 struct lustre_handle conn = { 0 };
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 if (localdata != NULL) {
364 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
365 if (rc != 0 && rc != -EEXIST)
368 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
369 "nid is null\n", obd->obd_name, cluuid->uuid);
372 rc = ofd_parse_connect_data(env, exp, data, true);
376 if (obd->obd_replayable) {
377 struct tg_export_data *ted = &exp->exp_target_data;
379 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
380 sizeof(ted->ted_lcd->lcd_uuid));
381 rc = tgt_client_new(env, exp);
384 ofd_export_stats_init(ofd, exp, localdata);
387 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
388 data ? data->ocd_group : -1);
392 class_disconnect(exp);
393 nodemap_del_member(exp);
402 * Disconnect a connected client.
404 * This function terminates the client connection. The client export is
405 * disconnected (cleaned up) and client data on persistent storage is removed.
407 * \param[in] exp OBD export
409 * \retval 0 if successful
410 * \retval negative value on error
412 int ofd_obd_disconnect(struct obd_export *exp)
414 struct ofd_device *ofd = ofd_exp(exp);
421 class_export_get(exp);
423 if (!(exp->exp_flags & OBD_OPT_FORCE))
424 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
426 rc = server_disconnect_export(exp);
428 ofd_grant_discard(exp);
430 /* Do not erase record for recoverable client. */
431 if (exp->exp_obd->obd_replayable &&
432 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
433 rc = lu_env_init(&env, LCT_DT_THREAD);
437 tgt_client_del(&env, exp);
441 nodemap_del_member(exp);
442 class_export_put(exp);
447 * Implementation of obd_ops::o_init_export.
449 * This function is called from class_new_export() and initializes
450 * the OFD-specific data for new export.
452 * \param[in] exp OBD export
454 * \retval 0 if successful
455 * \retval negative value on error
457 static int ofd_init_export(struct obd_export *exp)
461 spin_lock_init(&exp->exp_filter_data.fed_lock);
462 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
463 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
464 spin_lock(&exp->exp_lock);
465 exp->exp_connecting = 1;
466 spin_unlock(&exp->exp_lock);
468 /* self-export doesn't need client data and ldlm initialization */
469 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
470 &exp->exp_client_uuid)))
473 rc = tgt_client_alloc(exp);
475 ldlm_init_export(exp);
477 CERROR("%s: Can't initialize export: rc %d\n",
478 exp->exp_obd->obd_name, rc);
483 * Implementation of obd_ops::o_destroy_export.
485 * This function is called from class_export_destroy() to cleanup
486 * the OFD-specific data for export being destroyed.
488 * \param[in] exp OBD export
490 * \retval 0 if successful
491 * \retval negative value on error
493 static int ofd_destroy_export(struct obd_export *exp)
495 struct ofd_device *ofd = ofd_exp(exp);
497 if (exp->exp_filter_data.fed_pending)
498 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
499 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
500 exp, exp->exp_filter_data.fed_pending);
502 target_destroy_export(exp);
504 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
505 &exp->exp_client_uuid)))
508 ldlm_destroy_export(exp);
509 tgt_client_free(exp);
511 ofd_fmd_cleanup(exp);
514 * discard grants once we're sure no more
515 * interaction with the client is possible
517 ofd_grant_discard(exp);
518 ofd_fmd_cleanup(exp);
520 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
521 ofd->ofd_tot_granted_clients--;
523 if (!(exp->exp_flags & OBD_OPT_FORCE))
524 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
526 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
531 * Notify all devices in server stack about recovery completion.
533 * This function calls ldo_recovery_complete() for all lower devices in the
534 * server stack so they will be prepared for normal operations.
536 * \param[in] env execution environment
537 * \param[in] ofd OFD device
539 * \retval 0 if successful
540 * \retval negative value on error
542 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
544 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
547 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
549 if (!ofd->ofd_skip_lfsck) {
550 struct lfsck_start_param lsp;
552 lsp.lsp_start = NULL;
553 lsp.lsp_index_valid = 0;
554 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
555 if (rc != 0 && rc != -EALREADY)
556 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
560 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
564 * Implementation of obd_ops::o_postrecov.
566 * This function is called from target_finish_recovery() upon recovery
569 * \param[in] obd OBD device of OFD
571 * \retval 0 if successful
572 * \retval negative value on error
574 static int ofd_obd_postrecov(struct obd_device *obd)
577 struct lu_device *ldev = obd->obd_lu_dev;
582 rc = lu_env_init(&env, LCT_DT_THREAD);
585 ofd_info_init(&env, obd->obd_self_export);
587 rc = ofd_postrecov(&env, ofd_dev(ldev));
594 * Implementation of obd_ops::o_set_info_async.
596 * This function is not called from request handler, it is only used by
597 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
598 * \see ofd_set_info_hdl() for request handler function.
600 * \param[in] env execution environment
601 * \param[in] exp OBD export of OFD device
602 * \param[in] keylen length of \a key
603 * \param[in] key key name
604 * \param[in] vallen length of \a val
605 * \param[in] val the \a key value
606 * \param[in] set not used in OFD
608 * \retval 0 if successful
609 * \retval negative value on error
611 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
612 __u32 keylen, void *key, __u32 vallen, void *val,
613 struct ptlrpc_request_set *set)
619 if (exp->exp_obd == NULL) {
620 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
624 if (KEY_IS(KEY_SPTLRPC_CONF)) {
625 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
627 CERROR("%s: Unsupported key %s\n",
628 exp->exp_obd->obd_name, (char*)key);
635 * Implementation of obd_ops::o_get_info.
637 * This function is not called from request handler, it is only used by
638 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
640 * \see ofd_get_info_hdl() for request handler function.
642 * \param[in] env execution environment
643 * \param[in] exp OBD export of OFD device
644 * \param[in] keylen length of \a key
645 * \param[in] key key name
646 * \param[out] vallen length of key value
647 * \param[out] val the key value to return
649 * \retval 0 if successful
650 * \retval negative value on error
652 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
653 __u32 keylen, void *key, __u32 *vallen, void *val)
655 struct ofd_thread_info *info;
656 struct ofd_device *ofd;
657 struct ll_fiemap_info_key *fm_key = key;
658 struct fiemap *fiemap = val;
663 if (exp->exp_obd == NULL) {
664 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
670 if (KEY_IS(KEY_FIEMAP)) {
671 info = ofd_info_init(env, exp);
673 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
674 ofd->ofd_lut.lut_lsd.lsd_osd_index);
678 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
680 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
688 * Get file system statistics of OST server.
690 * Helper function for ofd_statfs(), also used by grant code.
691 * Implements caching for statistics to avoid calling OSD device each time.
693 * \param[in] env execution environment
694 * \param[in] ofd OFD device
695 * \param[out] osfs statistic data to return
696 * \param[in] max_age maximum age for cached data
697 * \param[in] from_cache show that data was get from cache or not
699 * \retval 0 if successful
700 * \retval negative value on error
702 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
703 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
708 spin_lock(&ofd->ofd_osfs_lock);
709 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
712 /* statfs data are too old, get up-to-date one.
713 * we must be cautious here since multiple threads might be
714 * willing to update statfs data concurrently and we must
715 * grant that cached statfs data are always consistent */
717 if (ofd->ofd_statfs_inflight == 0)
718 /* clear inflight counter if no users, although it would
719 * take a while to overflow this 64-bit counter ... */
720 ofd->ofd_osfs_inflight = 0;
721 /* notify ofd_grant_commit() that we want to track writes
722 * completed as of now */
723 ofd->ofd_statfs_inflight++;
724 /* record value of inflight counter before running statfs to
725 * compute the diff once statfs is completed */
726 unstable = ofd->ofd_osfs_inflight;
727 spin_unlock(&ofd->ofd_osfs_lock);
729 /* statfs can sleep ... hopefully not for too long since we can
730 * call it fairly often as space fills up */
731 rc = dt_statfs(env, ofd->ofd_osd, osfs);
735 spin_lock(&ofd->ofd_grant_lock);
736 spin_lock(&ofd->ofd_osfs_lock);
737 /* calculate how much space was written while we released the
739 unstable = ofd->ofd_osfs_inflight - unstable;
740 ofd->ofd_osfs_unstable = 0;
742 /* some writes committed while we were running statfs
743 * w/o the ofd_osfs_lock. Those ones got added to
744 * the cached statfs data that we are about to crunch.
745 * Take them into account in the new statfs data */
746 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
747 unstable >> ofd->ofd_blockbits);
748 /* However, we don't really know if those writes got
749 * accounted in the statfs call, so tell
750 * ofd_grant_space_left() there is some uncertainty
751 * on the accounting of those writes.
752 * The purpose is to prevent spurious error messages in
753 * ofd_grant_space_left() since those writes might be
754 * accounted twice. */
755 ofd->ofd_osfs_unstable += unstable;
757 /* similarly, there is some uncertainty on write requests
758 * between prepare & commit */
759 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
760 spin_unlock(&ofd->ofd_grant_lock);
762 /* finally udpate cached statfs data */
763 ofd->ofd_osfs = *osfs;
764 ofd->ofd_osfs_age = cfs_time_current_64();
766 ofd->ofd_statfs_inflight--; /* stop tracking */
767 if (ofd->ofd_statfs_inflight == 0)
768 ofd->ofd_osfs_inflight = 0;
769 spin_unlock(&ofd->ofd_osfs_lock);
774 /* use cached statfs data */
775 *osfs = ofd->ofd_osfs;
776 spin_unlock(&ofd->ofd_osfs_lock);
787 * Implementation of obd_ops::o_statfs.
789 * This function returns information about a storage file system.
790 * It is called from several places by using the OBD API as well as
791 * by direct call, e.g. from request handler.
793 * \see ofd_statfs_hdl() for request handler function.
795 * Report also the state of the OST to the caller in osfs->os_state
796 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
798 * \param[in] env execution environment
799 * \param[in] exp OBD export of OFD device
800 * \param[out] osfs statistic data to return
801 * \param[in] max_age maximum age for cached data
802 * \param[in] flags not used in OFD
804 * \retval 0 if successful
805 * \retval negative value on error
807 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
808 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
810 struct obd_device *obd = class_exp2obd(exp);
811 struct ofd_device *ofd = ofd_exp(exp);
816 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
820 /* at least try to account for cached pages. its still racy and
821 * might be under-reporting if clients haven't announced their
822 * caches with brw recently */
824 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
825 " pending "LPU64" free "LPU64" avail "LPU64"\n",
826 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
827 osfs->os_bfree << ofd->ofd_blockbits,
828 osfs->os_bavail << ofd->ofd_blockbits);
830 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
831 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
832 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
834 /* The QoS code on the MDS does not care about space reserved for
835 * precreate, so take it out. */
836 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
837 struct filter_export_data *fed;
839 fed = &obd->obd_self_export->exp_filter_data;
840 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
841 fed->fed_grant >> ofd->ofd_blockbits);
844 ofd_grant_sanity_check(obd, __FUNCTION__);
845 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
846 LPU64" objects: "LPU64" free; state %x\n",
847 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
848 osfs->os_files, osfs->os_ffree, osfs->os_state);
850 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
851 ofd->ofd_lut.lut_lsd.lsd_osd_index))
854 /* OS_STATE_READONLY can be set by OSD already */
855 if (ofd->ofd_raid_degraded)
856 osfs->os_state |= OS_STATE_DEGRADED;
858 if (obd->obd_self_export != exp && !ofd_grant_param_supp(exp) &&
859 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT) {
860 /* clients which don't support OBD_CONNECT_GRANT_PARAM
861 * should not see a block size > page size, otherwise
862 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
863 * block size which is the biggest block size known to work
864 * with all client's page size. */
865 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
866 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
867 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
868 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
871 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
872 ofd->ofd_lut.lut_lsd.lsd_osd_index))
873 osfs->os_bfree = osfs->os_bavail = 2;
881 * Implementation of obd_ops::o_setattr.
883 * This function is only used by ECHO client when it is run on top of OFD,
884 * \see ofd_setattr_hdl() for request handler function.
886 * \param[in] env execution environment
887 * \param[in] exp OBD export of OFD device
888 * \param[in] oa setattr parameters
890 * \retval 0 if successful
891 * \retval negative value on error
893 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
896 struct ofd_thread_info *info;
897 struct ofd_device *ofd = ofd_exp(exp);
898 struct ldlm_namespace *ns = ofd->ofd_namespace;
899 struct ldlm_resource *res;
900 struct ofd_object *fo;
901 struct lu_fid *fid = &oa->o_oi.oi_fid;
902 struct filter_fid *ff = NULL;
907 info = ofd_info_init(env, exp);
909 ost_fid_build_resid(fid, &info->fti_resid);
911 /* This would be very bad - accidentally truncating a file when
912 * changing the time or similar - bug 12203. */
913 if (oa->o_valid & OBD_MD_FLSIZE) {
914 static char mdsinum[48];
916 if (oa->o_valid & OBD_MD_FLFID)
917 snprintf(mdsinum, sizeof(mdsinum) - 1,
918 "of parent "DFID, oa->o_parent_seq,
919 oa->o_parent_oid, 0);
923 CERROR("%s: setattr from %s trying to truncate object "DFID
924 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
926 GOTO(out, rc = -EPERM);
929 fo = ofd_object_find_exists(env, ofd, fid);
931 CERROR("%s: can't find object "DFID"\n",
932 ofd_name(ofd), PFID(fid));
933 GOTO(out, rc = PTR_ERR(fo));
936 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
937 info->fti_attr.la_valid &= ~LA_TYPE;
939 if (oa->o_valid & OBD_MD_FLFID) {
940 ff = &info->fti_mds_fid;
941 ofd_prepare_fidea(ff, oa);
944 /* setting objects attributes (including owner/group) */
945 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
947 GOTO(out_unlock, rc);
949 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
952 ofd_object_put(env, fo);
955 /* we do not call this before to avoid lu_object_find() in
956 * ->lvbo_update() holding another reference on the object.
957 * otherwise concurrent destroy can make the object unavailable
958 * for 2nd lu_object_find() waiting for the first reference
959 * to go... deadlock! */
960 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
962 ldlm_res_lvbo_update(res, NULL, 0);
963 ldlm_resource_putref(res);
971 * Destroy OFD object by its FID.
973 * Supplemental function to destroy object by FID, it is used by request
974 * handler and by ofd_echo_destroy() below to find object by FID, lock it
975 * and call ofd_object_destroy() finally.
977 * \param[in] env execution environment
978 * \param[in] ofd OFD device
979 * \param[in] fid FID of object
980 * \param[in] orphan set if object being destroyed is an orphan
982 * \retval 0 if successful
983 * \retval negative value on error
985 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
986 const struct lu_fid *fid, int orphan)
988 struct ofd_thread_info *info = ofd_info(env);
989 struct lustre_handle lockh;
990 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
991 struct ofd_object *fo;
992 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
997 fo = ofd_object_find_exists(env, ofd, fid);
1001 /* Tell the clients that the object is gone now and that they should
1002 * throw away any cached pages. */
1003 ost_fid_build_resid(fid, &info->fti_resid);
1004 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1005 LDLM_EXTENT, &policy, LCK_PW, &flags,
1006 ldlm_blocking_ast, ldlm_completion_ast,
1007 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1009 /* We only care about the side-effects, just drop the lock. */
1011 ldlm_lock_decref(&lockh, LCK_PW);
1013 LASSERT(fo != NULL);
1015 rc = ofd_object_destroy(env, fo, orphan);
1018 ofd_object_put(env, fo);
1023 * Implementation of obd_ops::o_destroy.
1025 * This function is only used by ECHO client when it is run on top of OFD,
1026 * \see ofd_destroy_hdl() for request handler function.
1028 * \param[in] env execution environment
1029 * \param[in] exp OBD export of OFD device
1030 * \param[in] oa obdo structure with FID
1032 * Note: this is OBD API method which is common API for server OBDs and
1033 * client OBDs. Thus some parameters used in client OBDs may not be used
1034 * on server OBDs and vice versa.
1036 * \retval 0 if successful
1037 * \retval negative value on error
1039 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1042 struct ofd_device *ofd = ofd_exp(exp);
1043 struct lu_fid *fid = &oa->o_oi.oi_fid;
1048 ofd_info_init(env, exp);
1050 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1052 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1053 if (rc == -ENOENT) {
1054 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1055 ofd_name(ofd), PFID(fid));
1057 } else if (rc != 0) {
1058 CERROR("%s: error destroying object "DFID": %d\n",
1059 ofd_name(ofd), PFID(fid), rc);
1068 * Implementation of obd_ops::o_create.
1070 * This function is only used by ECHO client when it is run on top of OFD
1071 * and just creates an object.
1072 * \see ofd_create_hdl() for request handler function.
1074 * \param[in] env execution environment
1075 * \param[in] exp OBD export of OFD device
1076 * \param[in] oa obdo structure with FID sequence to use
1078 * Note: this is OBD API method which is common API for server OBDs and
1079 * client OBDs. Thus some parameters used in client OBDs may not be used
1080 * on server OBDs and vice versa.
1082 * \retval 0 if successful
1083 * \retval negative value on error
1085 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1088 struct ofd_device *ofd = ofd_exp(exp);
1089 struct ofd_thread_info *info;
1090 u64 seq = ostid_seq(&oa->o_oi);
1091 struct ofd_seq *oseq;
1092 int rc = 0, diff = 1;
1099 info = ofd_info_init(env, exp);
1101 LASSERT(seq == FID_SEQ_ECHO);
1102 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1104 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1106 down_read(&ofd->ofd_lastid_rwsem);
1107 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1108 * we may do that in the future.
1109 * Return -ENOSPC until the LAST_ID rebuilt. */
1110 if (unlikely(ofd->ofd_lastid_rebuilding))
1111 GOTO(out_sem, rc = -ENOSPC);
1113 oseq = ofd_seq_load(env, ofd, seq);
1115 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1116 ofd_name(ofd), seq, PTR_ERR(oseq));
1117 GOTO(out_sem, rc = -EINVAL);
1120 mutex_lock(&oseq->os_create_lock);
1121 granted = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1125 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1126 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
1131 next_id = ofd_seq_last_oid(oseq) + 1;
1132 count = ofd_precreate_batch(ofd, diff);
1134 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1136 CERROR("%s: unable to precreate: rc = %d\n",
1139 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1140 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1144 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1146 mutex_unlock(&oseq->os_create_lock);
1147 ofd_seq_put(env, oseq);
1150 up_read(&ofd->ofd_lastid_rwsem);
1155 * Implementation of obd_ops::o_getattr.
1157 * This function is only used by ECHO client when it is run on top of OFD
1158 * and returns attributes of object.
1159 * \see ofd_getattr_hdl() for request handler function.
1161 * \param[in] env execution environment
1162 * \param[in] exp OBD export of OFD device
1163 * \param[in,out] oa contains FID of object to get attributes from and
1164 * is used to return attributes back
1166 * \retval 0 if successful
1167 * \retval negative value on error
1169 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1172 struct ofd_device *ofd = ofd_exp(exp);
1173 struct ofd_thread_info *info;
1174 struct lu_fid *fid = &oa->o_oi.oi_fid;
1175 struct ofd_object *fo;
1180 info = ofd_info_init(env, exp);
1182 fo = ofd_object_find_exists(env, ofd, fid);
1184 GOTO(out, rc = PTR_ERR(fo));
1186 LASSERT(fo != NULL);
1187 rc = ofd_attr_get(env, fo, &info->fti_attr);
1188 oa->o_valid = OBD_MD_FLID;
1192 obdo_from_la(oa, &info->fti_attr,
1193 OFD_VALID_FLAGS | LA_UID | LA_GID);
1195 /* Store object version in reply */
1196 curr_version = dt_version_get(env, ofd_object_child(fo));
1197 if ((__s64)curr_version != -EOPNOTSUPP) {
1198 oa->o_valid |= OBD_MD_FLDATAVERSION;
1199 oa->o_data_version = curr_version;
1203 ofd_object_put(env, fo);
1209 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1211 * This is supplemental function for ofd_iocontrol() to return object
1212 * version for lctl tool.
1214 * \param[in] env execution environment
1215 * \param[in] ofd OFD device
1216 * \param[out] karg ioctl data
1218 * \retval 0 if successful
1219 * \retval negative value on error
1221 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1222 struct ofd_device *ofd, void *karg)
1224 struct obd_ioctl_data *data = karg;
1226 struct ofd_object *fo;
1227 dt_obj_version_t version;
1232 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1233 GOTO(out, rc = -EINVAL);
1235 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1236 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1237 } else if (data->ioc_inlbuf3 != NULL &&
1238 data->ioc_inllen3 == sizeof(__u64) &&
1239 data->ioc_inlbuf4 != NULL &&
1240 data->ioc_inllen4 == sizeof(__u64)) {
1241 struct ost_id ostid;
1243 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1244 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1245 rc = ostid_to_fid(&fid, &ostid,
1246 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1250 GOTO(out, rc = -EINVAL);
1253 if (!fid_is_sane(&fid))
1254 GOTO(out, rc = -EINVAL);
1256 fo = ofd_object_find(env, ofd, &fid);
1258 GOTO(out, rc = PTR_ERR(fo));
1260 if (!ofd_object_exists(fo))
1261 GOTO(out_fo, rc = -ENOENT);
1263 if (lu_object_remote(&fo->ofo_obj.do_lu))
1264 GOTO(out_fo, rc = -EREMOTE);
1266 version = dt_version_get(env, ofd_object_child(fo));
1268 GOTO(out_fo, rc = -EIO);
1270 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1274 ofd_object_put(env, fo);
1280 * Implementation of obd_ops::o_iocontrol.
1282 * This is OFD ioctl handling function which is primary interface for
1283 * Lustre tools like lfs, lctl and lfsck.
1285 * \param[in] cmd ioctl command
1286 * \param[in] exp OBD export of OFD
1287 * \param[in] len not used
1288 * \param[in,out] karg buffer with data
1289 * \param[in] uarg not used
1291 * \retval 0 if successful
1292 * \retval negative value on error
1294 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1295 void *karg, void __user *uarg)
1298 struct ofd_device *ofd = ofd_exp(exp);
1299 struct obd_device *obd = ofd_obd(ofd);
1304 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1305 rc = lu_env_init(&env, LCT_DT_THREAD);
1310 case OBD_IOC_ABORT_RECOVERY:
1311 CERROR("%s: aborting recovery\n", obd->obd_name);
1312 obd->obd_abort_recovery = 1;
1313 target_stop_recovery_thread(obd);
1316 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1317 rc = dt_sync(&env, ofd->ofd_osd);
1319 case OBD_IOC_SET_READONLY:
1320 rc = dt_sync(&env, ofd->ofd_osd);
1322 rc = dt_ro(&env, ofd->ofd_osd);
1324 case OBD_IOC_START_LFSCK: {
1325 struct obd_ioctl_data *data = karg;
1326 struct lfsck_start_param lsp;
1328 if (unlikely(data == NULL)) {
1333 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1334 lsp.lsp_index_valid = 0;
1335 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1338 case OBD_IOC_STOP_LFSCK: {
1339 struct obd_ioctl_data *data = karg;
1340 struct lfsck_stop stop;
1342 stop.ls_status = LS_STOPPED;
1343 /* Old lfsck utils may pass NULL @stop. */
1344 if (data->ioc_inlbuf1 == NULL)
1348 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1350 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1353 case OBD_IOC_GET_OBJ_VERSION:
1354 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1357 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1366 * Implementation of obd_ops::o_precleanup.
1368 * This function stops device activity before shutting it down. It is called
1369 * from a cleanup function upon forceful device cleanup. For OFD there are no
1370 * special actions, it just invokes target_recovery_cleanup().
1372 * \param[in] obd OBD device of OFD
1376 static int ofd_precleanup(struct obd_device *obd)
1379 target_cleanup_recovery(obd);
1384 * Implementation of obd_ops::o_ping.
1386 * This is OFD-specific part of OBD_PING request handling.
1387 * It controls Filter Modification Data (FMD) expiration each time PING is
1390 * \see ofd_fmd_expire() and ofd_fmd.c for details
1392 * \param[in] env execution environment
1393 * \param[in] exp OBD export of client
1397 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1399 ofd_fmd_expire(exp);
1404 * Implementation of obd_ops::o_health_check.
1406 * This function checks the OFD device health - ability to respond on
1407 * incoming requests. There are two health_check methods:
1408 * - get statfs from the OSD. It checks just responsiveness of
1410 * - do write attempt on bottom device to check it is fully operational and
1411 * is not stuck. This is expensive method and requires special configuration
1412 * option --enable-health-write while building Lustre, it is turned off
1415 * \param[in] nul not used
1416 * \param[in] obd OBD device of OFD
1418 * \retval 0 if successful
1419 * \retval negative value in case of error
1421 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1423 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1424 struct ofd_thread_info *info;
1426 #ifdef USE_HEALTH_CHECK_WRITE
1431 /* obd_proc_read_health pass NULL env, we need real one */
1432 rc = lu_env_init(&env, LCT_DT_THREAD);
1436 info = ofd_info_init(&env, NULL);
1437 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1441 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1442 GOTO(out, rc = -EROFS);
1444 #ifdef USE_HEALTH_CHECK_WRITE
1445 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1446 if (info->fti_buf.lb_buf == NULL)
1447 GOTO(out, rc = -ENOMEM);
1449 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1452 th = dt_trans_create(&env, ofd->ofd_osd);
1454 GOTO(out, rc = PTR_ERR(th));
1456 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1457 &info->fti_buf, info->fti_off, th);
1459 th->th_sync = 1; /* sync IO is needed */
1460 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1462 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1463 &info->fti_buf, &info->fti_off,
1466 dt_trans_stop(&env, ofd->ofd_osd, th);
1468 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1470 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1477 struct obd_ops ofd_obd_ops = {
1478 .o_owner = THIS_MODULE,
1479 .o_connect = ofd_obd_connect,
1480 .o_reconnect = ofd_obd_reconnect,
1481 .o_disconnect = ofd_obd_disconnect,
1482 .o_create = ofd_echo_create,
1483 .o_statfs = ofd_statfs,
1484 .o_setattr = ofd_echo_setattr,
1485 .o_preprw = ofd_preprw,
1486 .o_commitrw = ofd_commitrw,
1487 .o_destroy = ofd_echo_destroy,
1488 .o_init_export = ofd_init_export,
1489 .o_destroy_export = ofd_destroy_export,
1490 .o_postrecov = ofd_obd_postrecov,
1491 .o_getattr = ofd_echo_getattr,
1492 .o_iocontrol = ofd_iocontrol,
1493 .o_precleanup = ofd_precleanup,
1495 .o_health_check = ofd_health_check,
1496 .o_set_info_async = ofd_set_info_async,
1497 .o_get_info = ofd_get_info,