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 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
177 /* client is reporting its page size, for future use */
178 exp->exp_filter_data.fed_pagebits = data->ocd_grant_blkbits;
179 data->ocd_grant_blkbits = ofd->ofd_blockbits;
180 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
181 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
182 data->ocd_grant_inobits =
183 fls(ofd->ofd_dt_conf.ddp_inodespace - 1);
184 /* ocd_grant_tax_kb is in 1K byte blocks */
185 data->ocd_grant_tax_kb = ofd->ofd_dt_conf.ddp_extent_tax >> 10;
186 data->ocd_grant_max_blks = ofd->ofd_dt_conf.ddp_max_extent_blks;
189 if (OCD_HAS_FLAG(data, GRANT))
190 ofd_grant_connect(env, exp, data, new_connection);
192 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
193 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
194 int index = lsd->lsd_osd_index;
196 if (index != data->ocd_index) {
197 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
198 " %u doesn't match actual OST index"
199 " %u in last_rcvd file, bad "
201 obd_export_nid2str(exp), index,
205 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
206 /* this will only happen on the first connect */
207 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
208 /* sync is not needed here as lut_client_add will
209 * set exp_need_sync flag */
210 tgt_server_data_update(env, &ofd->ofd_lut, 0);
213 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
214 data->ocd_brw_size = 65536;
215 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
216 data->ocd_brw_size = min(data->ocd_brw_size,
217 (__u32)DT_MAX_BRW_SIZE);
218 if (data->ocd_brw_size == 0) {
219 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
220 " ocd_version: %x ocd_grant: %d ocd_index: %u "
221 "ocd_brw_size is unexpectedly zero, "
222 "network data corruption?"
223 "Refusing connection of this client\n",
224 exp->exp_obd->obd_name,
225 exp->exp_client_uuid.uuid,
226 exp, data->ocd_connect_flags, data->ocd_version,
227 data->ocd_grant, data->ocd_index);
232 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
233 __u32 cksum_types = data->ocd_cksum_types;
235 /* The client set in ocd_cksum_types the checksum types it
236 * supports. We have to mask off the algorithms that we don't
238 data->ocd_cksum_types &= cksum_types_supported_server();
240 if (unlikely(data->ocd_cksum_types == 0)) {
241 CERROR("%s: Connect with checksum support but no "
242 "ocd_cksum_types is set\n",
243 exp->exp_obd->obd_name);
247 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
248 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
249 cksum_types, data->ocd_cksum_types);
251 /* This client does not support OBD_CONNECT_CKSUM
252 * fall back to CRC32 */
253 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
254 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
255 exp->exp_obd->obd_name, obd_export_nid2str(exp));
258 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
259 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
261 if (OCD_HAS_FLAG(data, PINGLESS)) {
262 if (ptlrpc_pinger_suppress_pings()) {
263 spin_lock(&exp->exp_obd->obd_dev_lock);
264 list_del_init(&exp->exp_obd_chain_timed);
265 spin_unlock(&exp->exp_obd->obd_dev_lock);
267 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
275 * Re-initialize export upon client reconnection.
277 * This function parses connection data from reconnect and resets
280 * \param[in] env execution environment
281 * \param[in] exp OBD export
282 * \param[in] obd OFD device
283 * \param[in] cluuid NID of client
284 * \param[in] data connection data from request
285 * \param[in] localdata client NID
287 * \retval 0 if successful
288 * \retval negative value on error
290 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
291 struct obd_device *obd, struct obd_uuid *cluuid,
292 struct obd_connect_data *data,
295 struct ofd_device *ofd;
300 if (exp == NULL || obd == NULL || cluuid == NULL)
303 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
304 if (rc != 0 && rc != -EEXIST)
307 ofd = ofd_dev(obd->obd_lu_dev);
309 rc = ofd_parse_connect_data(env, exp, data, false);
311 ofd_export_stats_init(ofd, exp, client_nid);
313 nodemap_del_member(exp);
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 if (localdata != NULL) {
358 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
359 if (rc != 0 && rc != -EEXIST)
362 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
363 "nid is null\n", obd->obd_name, cluuid->uuid);
366 rc = ofd_parse_connect_data(env, exp, data, true);
370 if (obd->obd_replayable) {
371 struct tg_export_data *ted = &exp->exp_target_data;
373 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
374 sizeof(ted->ted_lcd->lcd_uuid));
375 rc = tgt_client_new(env, exp);
378 ofd_export_stats_init(ofd, exp, localdata);
381 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
382 data ? data->ocd_group : -1);
386 class_disconnect(exp);
387 nodemap_del_member(exp);
396 * Disconnect a connected client.
398 * This function terminates the client connection. The client export is
399 * disconnected (cleaned up) and client data on persistent storage is removed.
401 * \param[in] exp OBD export
403 * \retval 0 if successful
404 * \retval negative value on error
406 int ofd_obd_disconnect(struct obd_export *exp)
408 struct ofd_device *ofd = ofd_exp(exp);
415 class_export_get(exp);
417 if (!(exp->exp_flags & OBD_OPT_FORCE))
418 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
420 rc = server_disconnect_export(exp);
422 ofd_grant_discard(exp);
424 /* Do not erase record for recoverable client. */
425 if (exp->exp_obd->obd_replayable &&
426 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
427 rc = lu_env_init(&env, LCT_DT_THREAD);
431 tgt_client_del(&env, exp);
435 nodemap_del_member(exp);
436 class_export_put(exp);
441 * Implementation of obd_ops::o_init_export.
443 * This function is called from class_new_export() and initializes
444 * the OFD-specific data for new export.
446 * \param[in] exp OBD export
448 * \retval 0 if successful
449 * \retval negative value on error
451 static int ofd_init_export(struct obd_export *exp)
455 spin_lock_init(&exp->exp_filter_data.fed_lock);
456 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
457 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
458 spin_lock(&exp->exp_lock);
459 exp->exp_connecting = 1;
460 spin_unlock(&exp->exp_lock);
462 /* self-export doesn't need client data and ldlm initialization */
463 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
464 &exp->exp_client_uuid)))
467 rc = tgt_client_alloc(exp);
469 ldlm_init_export(exp);
471 CERROR("%s: Can't initialize export: rc %d\n",
472 exp->exp_obd->obd_name, rc);
477 * Implementation of obd_ops::o_destroy_export.
479 * This function is called from class_export_destroy() to cleanup
480 * the OFD-specific data for export being destroyed.
482 * \param[in] exp OBD export
484 * \retval 0 if successful
485 * \retval negative value on error
487 static int ofd_destroy_export(struct obd_export *exp)
489 struct ofd_device *ofd = ofd_exp(exp);
491 if (exp->exp_filter_data.fed_pending)
492 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
493 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
494 exp, exp->exp_filter_data.fed_pending);
496 target_destroy_export(exp);
498 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
499 &exp->exp_client_uuid)))
502 ldlm_destroy_export(exp);
503 tgt_client_free(exp);
505 ofd_fmd_cleanup(exp);
508 * discard grants once we're sure no more
509 * interaction with the client is possible
511 ofd_grant_discard(exp);
512 ofd_fmd_cleanup(exp);
514 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
515 ofd->ofd_tot_granted_clients--;
517 if (!(exp->exp_flags & OBD_OPT_FORCE))
518 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
520 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
525 * Notify all devices in server stack about recovery completion.
527 * This function calls ldo_recovery_complete() for all lower devices in the
528 * server stack so they will be prepared for normal operations.
530 * \param[in] env execution environment
531 * \param[in] ofd OFD device
533 * \retval 0 if successful
534 * \retval negative value on error
536 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
538 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
541 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
543 if (!ofd->ofd_skip_lfsck) {
544 struct lfsck_start_param lsp;
546 lsp.lsp_start = NULL;
547 lsp.lsp_index_valid = 0;
548 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
549 if (rc != 0 && rc != -EALREADY)
550 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
554 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
558 * Implementation of obd_ops::o_postrecov.
560 * This function is called from target_finish_recovery() upon recovery
563 * \param[in] obd OBD device of OFD
565 * \retval 0 if successful
566 * \retval negative value on error
568 static int ofd_obd_postrecov(struct obd_device *obd)
571 struct lu_device *ldev = obd->obd_lu_dev;
576 rc = lu_env_init(&env, LCT_DT_THREAD);
579 ofd_info_init(&env, obd->obd_self_export);
581 rc = ofd_postrecov(&env, ofd_dev(ldev));
588 * Implementation of obd_ops::o_set_info_async.
590 * This function is not called from request handler, it is only used by
591 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
592 * \see ofd_set_info_hdl() for request handler function.
594 * \param[in] env execution environment
595 * \param[in] exp OBD export of OFD device
596 * \param[in] keylen length of \a key
597 * \param[in] key key name
598 * \param[in] vallen length of \a val
599 * \param[in] val the \a key value
600 * \param[in] set not used in OFD
602 * \retval 0 if successful
603 * \retval negative value on error
605 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
606 __u32 keylen, void *key, __u32 vallen, void *val,
607 struct ptlrpc_request_set *set)
613 if (exp->exp_obd == NULL) {
614 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
618 if (KEY_IS(KEY_SPTLRPC_CONF)) {
619 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
621 CERROR("%s: Unsupported key %s\n",
622 exp->exp_obd->obd_name, (char*)key);
629 * Implementation of obd_ops::o_get_info.
631 * This function is not called from request handler, it is only used by
632 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
634 * \see ofd_get_info_hdl() for request handler function.
636 * \param[in] env execution environment
637 * \param[in] exp OBD export of OFD device
638 * \param[in] keylen length of \a key
639 * \param[in] key key name
640 * \param[out] vallen length of key value
641 * \param[out] val the key value to return
643 * \retval 0 if successful
644 * \retval negative value on error
646 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
647 __u32 keylen, void *key, __u32 *vallen, void *val)
649 struct ofd_thread_info *info;
650 struct ofd_device *ofd;
651 struct ll_fiemap_info_key *fm_key = key;
652 struct fiemap *fiemap = val;
657 if (exp->exp_obd == NULL) {
658 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
664 if (KEY_IS(KEY_FIEMAP)) {
665 info = ofd_info_init(env, exp);
667 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
668 ofd->ofd_lut.lut_lsd.lsd_osd_index);
672 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
674 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
682 * Get file system statistics of OST server.
684 * Helper function for ofd_statfs(), also used by grant code.
685 * Implements caching for statistics to avoid calling OSD device each time.
687 * \param[in] env execution environment
688 * \param[in] ofd OFD device
689 * \param[out] osfs statistic data to return
690 * \param[in] max_age maximum age for cached data
691 * \param[in] from_cache show that data was get from cache or not
693 * \retval 0 if successful
694 * \retval negative value on error
696 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
697 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
702 spin_lock(&ofd->ofd_osfs_lock);
703 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
706 /* statfs data are too old, get up-to-date one.
707 * we must be cautious here since multiple threads might be
708 * willing to update statfs data concurrently and we must
709 * grant that cached statfs data are always consistent */
711 if (ofd->ofd_statfs_inflight == 0)
712 /* clear inflight counter if no users, although it would
713 * take a while to overflow this 64-bit counter ... */
714 ofd->ofd_osfs_inflight = 0;
715 /* notify ofd_grant_commit() that we want to track writes
716 * completed as of now */
717 ofd->ofd_statfs_inflight++;
718 /* record value of inflight counter before running statfs to
719 * compute the diff once statfs is completed */
720 unstable = ofd->ofd_osfs_inflight;
721 spin_unlock(&ofd->ofd_osfs_lock);
723 /* statfs can sleep ... hopefully not for too long since we can
724 * call it fairly often as space fills up */
725 rc = dt_statfs(env, ofd->ofd_osd, osfs);
729 spin_lock(&ofd->ofd_grant_lock);
730 spin_lock(&ofd->ofd_osfs_lock);
731 /* calculate how much space was written while we released the
733 unstable = ofd->ofd_osfs_inflight - unstable;
734 ofd->ofd_osfs_unstable = 0;
736 /* some writes committed while we were running statfs
737 * w/o the ofd_osfs_lock. Those ones got added to
738 * the cached statfs data that we are about to crunch.
739 * Take them into account in the new statfs data */
740 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
741 unstable >> ofd->ofd_blockbits);
742 /* However, we don't really know if those writes got
743 * accounted in the statfs call, so tell
744 * ofd_grant_space_left() there is some uncertainty
745 * on the accounting of those writes.
746 * The purpose is to prevent spurious error messages in
747 * ofd_grant_space_left() since those writes might be
748 * accounted twice. */
749 ofd->ofd_osfs_unstable += unstable;
751 /* similarly, there is some uncertainty on write requests
752 * between prepare & commit */
753 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
754 spin_unlock(&ofd->ofd_grant_lock);
756 /* finally udpate cached statfs data */
757 ofd->ofd_osfs = *osfs;
758 ofd->ofd_osfs_age = cfs_time_current_64();
760 ofd->ofd_statfs_inflight--; /* stop tracking */
761 if (ofd->ofd_statfs_inflight == 0)
762 ofd->ofd_osfs_inflight = 0;
763 spin_unlock(&ofd->ofd_osfs_lock);
768 /* use cached statfs data */
769 *osfs = ofd->ofd_osfs;
770 spin_unlock(&ofd->ofd_osfs_lock);
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(u64, 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(u64, 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_param_supp(exp) &&
853 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT) {
854 /* clients which don't support OBD_CONNECT_GRANT_PARAM
855 * should not see a block size > page size, otherwise
856 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
857 * block size which is the biggest block size known to work
858 * with all client's page size. */
859 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
860 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
861 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
862 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
865 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
866 ofd->ofd_lut.lut_lsd.lsd_osd_index))
867 osfs->os_bfree = osfs->os_bavail = 2;
875 * Implementation of obd_ops::o_setattr.
877 * This function is only used by ECHO client when it is run on top of OFD,
878 * \see ofd_setattr_hdl() for request handler function.
880 * \param[in] env execution environment
881 * \param[in] exp OBD export of OFD device
882 * \param[in] oa setattr parameters
884 * \retval 0 if successful
885 * \retval negative value on error
887 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
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 lu_fid *fid = &oa->o_oi.oi_fid;
896 struct filter_fid *ff = NULL;
901 info = ofd_info_init(env, exp);
903 ost_fid_build_resid(fid, &info->fti_resid);
905 /* This would be very bad - accidentally truncating a file when
906 * changing the time or similar - bug 12203. */
907 if (oa->o_valid & OBD_MD_FLSIZE) {
908 static char mdsinum[48];
910 if (oa->o_valid & OBD_MD_FLFID)
911 snprintf(mdsinum, sizeof(mdsinum) - 1,
912 "of parent "DFID, oa->o_parent_seq,
913 oa->o_parent_oid, 0);
917 CERROR("%s: setattr from %s trying to truncate object "DFID
918 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
920 GOTO(out, rc = -EPERM);
923 fo = ofd_object_find_exists(env, ofd, fid);
925 CERROR("%s: can't find object "DFID"\n",
926 ofd_name(ofd), PFID(fid));
927 GOTO(out, rc = PTR_ERR(fo));
930 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
931 info->fti_attr.la_valid &= ~LA_TYPE;
933 if (oa->o_valid & OBD_MD_FLFID) {
934 ff = &info->fti_mds_fid;
935 ofd_prepare_fidea(ff, oa);
938 /* setting objects attributes (including owner/group) */
939 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
941 GOTO(out_unlock, rc);
943 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
946 ofd_object_put(env, fo);
949 /* we do not call this before to avoid lu_object_find() in
950 * ->lvbo_update() holding another reference on the object.
951 * otherwise concurrent destroy can make the object unavailable
952 * for 2nd lu_object_find() waiting for the first reference
953 * to go... deadlock! */
954 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
956 ldlm_res_lvbo_update(res, NULL, 0);
957 ldlm_resource_putref(res);
965 * Destroy OFD object by its FID.
967 * Supplemental function to destroy object by FID, it is used by request
968 * handler and by ofd_echo_destroy() below to find object by FID, lock it
969 * and call ofd_object_destroy() finally.
971 * \param[in] env execution environment
972 * \param[in] ofd OFD device
973 * \param[in] fid FID of object
974 * \param[in] orphan set if object being destroyed is an orphan
976 * \retval 0 if successful
977 * \retval negative value on error
979 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
980 const struct lu_fid *fid, int orphan)
982 struct ofd_thread_info *info = ofd_info(env);
983 struct lustre_handle lockh;
984 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
985 struct ofd_object *fo;
986 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
991 fo = ofd_object_find_exists(env, ofd, fid);
995 /* Tell the clients that the object is gone now and that they should
996 * throw away any cached pages. */
997 ost_fid_build_resid(fid, &info->fti_resid);
998 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
999 LDLM_EXTENT, &policy, LCK_PW, &flags,
1000 ldlm_blocking_ast, ldlm_completion_ast,
1001 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1003 /* We only care about the side-effects, just drop the lock. */
1005 ldlm_lock_decref(&lockh, LCK_PW);
1007 LASSERT(fo != NULL);
1009 rc = ofd_object_destroy(env, fo, orphan);
1012 ofd_object_put(env, fo);
1017 * Implementation of obd_ops::o_destroy.
1019 * This function is only used by ECHO client when it is run on top of OFD,
1020 * \see ofd_destroy_hdl() for request handler function.
1022 * \param[in] env execution environment
1023 * \param[in] exp OBD export of OFD device
1024 * \param[in] oa obdo structure with FID
1026 * Note: this is OBD API method which is common API for server OBDs and
1027 * client OBDs. Thus some parameters used in client OBDs may not be used
1028 * on server OBDs and vice versa.
1030 * \retval 0 if successful
1031 * \retval negative value on error
1033 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1036 struct ofd_device *ofd = ofd_exp(exp);
1037 struct lu_fid *fid = &oa->o_oi.oi_fid;
1042 ofd_info_init(env, exp);
1044 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1046 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1047 if (rc == -ENOENT) {
1048 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1049 ofd_name(ofd), PFID(fid));
1051 } else if (rc != 0) {
1052 CERROR("%s: error destroying object "DFID": %d\n",
1053 ofd_name(ofd), PFID(fid), rc);
1062 * Implementation of obd_ops::o_create.
1064 * This function is only used by ECHO client when it is run on top of OFD
1065 * and just creates an object.
1066 * \see ofd_create_hdl() for request handler function.
1068 * \param[in] env execution environment
1069 * \param[in] exp OBD export of OFD device
1070 * \param[in] oa obdo structure with FID sequence to use
1072 * Note: this is OBD API method which is common API for server OBDs and
1073 * client OBDs. Thus some parameters used in client OBDs may not be used
1074 * on server OBDs and vice versa.
1076 * \retval 0 if successful
1077 * \retval negative value on error
1079 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1082 struct ofd_device *ofd = ofd_exp(exp);
1083 struct ofd_thread_info *info;
1084 u64 seq = ostid_seq(&oa->o_oi);
1085 struct ofd_seq *oseq;
1086 int rc = 0, diff = 1;
1093 info = ofd_info_init(env, exp);
1095 LASSERT(seq == FID_SEQ_ECHO);
1096 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1098 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1100 down_read(&ofd->ofd_lastid_rwsem);
1101 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1102 * we may do that in the future.
1103 * Return -ENOSPC until the LAST_ID rebuilt. */
1104 if (unlikely(ofd->ofd_lastid_rebuilding))
1105 GOTO(out_sem, rc = -ENOSPC);
1107 oseq = ofd_seq_load(env, ofd, seq);
1109 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1110 ofd_name(ofd), seq, PTR_ERR(oseq));
1111 GOTO(out_sem, rc = -EINVAL);
1114 mutex_lock(&oseq->os_create_lock);
1115 granted = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1119 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1120 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
1125 next_id = ofd_seq_last_oid(oseq) + 1;
1126 count = ofd_precreate_batch(ofd, diff);
1128 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1130 CERROR("%s: unable to precreate: rc = %d\n",
1133 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1134 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1138 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1140 mutex_unlock(&oseq->os_create_lock);
1141 ofd_seq_put(env, oseq);
1144 up_read(&ofd->ofd_lastid_rwsem);
1149 * Implementation of obd_ops::o_getattr.
1151 * This function is only used by ECHO client when it is run on top of OFD
1152 * and returns attributes of object.
1153 * \see ofd_getattr_hdl() for request handler function.
1155 * \param[in] env execution environment
1156 * \param[in] exp OBD export of OFD device
1157 * \param[in,out] oa contains FID of object to get attributes from and
1158 * is used to return attributes back
1160 * \retval 0 if successful
1161 * \retval negative value on error
1163 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1166 struct ofd_device *ofd = ofd_exp(exp);
1167 struct ofd_thread_info *info;
1168 struct lu_fid *fid = &oa->o_oi.oi_fid;
1169 struct ofd_object *fo;
1174 info = ofd_info_init(env, exp);
1176 fo = ofd_object_find_exists(env, ofd, fid);
1178 GOTO(out, rc = PTR_ERR(fo));
1180 LASSERT(fo != NULL);
1181 rc = ofd_attr_get(env, fo, &info->fti_attr);
1182 oa->o_valid = OBD_MD_FLID;
1186 obdo_from_la(oa, &info->fti_attr,
1187 OFD_VALID_FLAGS | LA_UID | LA_GID);
1189 /* Store object version in reply */
1190 curr_version = dt_version_get(env, ofd_object_child(fo));
1191 if ((__s64)curr_version != -EOPNOTSUPP) {
1192 oa->o_valid |= OBD_MD_FLDATAVERSION;
1193 oa->o_data_version = curr_version;
1197 ofd_object_put(env, fo);
1203 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1205 * This is supplemental function for ofd_iocontrol() to return object
1206 * version for lctl tool.
1208 * \param[in] env execution environment
1209 * \param[in] ofd OFD device
1210 * \param[out] karg ioctl data
1212 * \retval 0 if successful
1213 * \retval negative value on error
1215 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1216 struct ofd_device *ofd, void *karg)
1218 struct obd_ioctl_data *data = karg;
1220 struct ofd_object *fo;
1221 dt_obj_version_t version;
1226 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1227 GOTO(out, rc = -EINVAL);
1229 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1230 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1231 } else if (data->ioc_inlbuf3 != NULL &&
1232 data->ioc_inllen3 == sizeof(__u64) &&
1233 data->ioc_inlbuf4 != NULL &&
1234 data->ioc_inllen4 == sizeof(__u64)) {
1235 struct ost_id ostid;
1237 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1238 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1239 rc = ostid_to_fid(&fid, &ostid,
1240 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1244 GOTO(out, rc = -EINVAL);
1247 if (!fid_is_sane(&fid))
1248 GOTO(out, rc = -EINVAL);
1250 fo = ofd_object_find(env, ofd, &fid);
1252 GOTO(out, rc = PTR_ERR(fo));
1254 if (!ofd_object_exists(fo))
1255 GOTO(out_fo, rc = -ENOENT);
1257 if (lu_object_remote(&fo->ofo_obj.do_lu))
1258 GOTO(out_fo, rc = -EREMOTE);
1260 version = dt_version_get(env, ofd_object_child(fo));
1262 GOTO(out_fo, rc = -EIO);
1264 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1268 ofd_object_put(env, fo);
1274 * Implementation of obd_ops::o_iocontrol.
1276 * This is OFD ioctl handling function which is primary interface for
1277 * Lustre tools like lfs, lctl and lfsck.
1279 * \param[in] cmd ioctl command
1280 * \param[in] exp OBD export of OFD
1281 * \param[in] len not used
1282 * \param[in,out] karg buffer with data
1283 * \param[in] uarg not used
1285 * \retval 0 if successful
1286 * \retval negative value on error
1288 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1289 void *karg, void __user *uarg)
1292 struct ofd_device *ofd = ofd_exp(exp);
1293 struct obd_device *obd = ofd_obd(ofd);
1298 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1299 rc = lu_env_init(&env, LCT_DT_THREAD);
1304 case OBD_IOC_ABORT_RECOVERY:
1305 CERROR("%s: aborting recovery\n", obd->obd_name);
1306 obd->obd_abort_recovery = 1;
1307 target_stop_recovery_thread(obd);
1310 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1311 rc = dt_sync(&env, ofd->ofd_osd);
1313 case OBD_IOC_SET_READONLY:
1314 rc = dt_sync(&env, ofd->ofd_osd);
1316 rc = dt_ro(&env, ofd->ofd_osd);
1318 case OBD_IOC_START_LFSCK: {
1319 struct obd_ioctl_data *data = karg;
1320 struct lfsck_start_param lsp;
1322 if (unlikely(data == NULL)) {
1327 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1328 lsp.lsp_index_valid = 0;
1329 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1332 case OBD_IOC_STOP_LFSCK: {
1333 struct obd_ioctl_data *data = karg;
1334 struct lfsck_stop stop;
1336 stop.ls_status = LS_STOPPED;
1337 /* Old lfsck utils may pass NULL @stop. */
1338 if (data->ioc_inlbuf1 == NULL)
1342 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1344 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1347 case OBD_IOC_GET_OBJ_VERSION:
1348 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1351 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1360 * Implementation of obd_ops::o_precleanup.
1362 * This function stops device activity before shutting it down. It is called
1363 * from a cleanup function upon forceful device cleanup. For OFD there are no
1364 * special actions, it just invokes target_recovery_cleanup().
1366 * \param[in] obd OBD device of OFD
1370 static int ofd_precleanup(struct obd_device *obd)
1373 target_cleanup_recovery(obd);
1378 * Implementation of obd_ops::o_ping.
1380 * This is OFD-specific part of OBD_PING request handling.
1381 * It controls Filter Modification Data (FMD) expiration each time PING is
1384 * \see ofd_fmd_expire() and ofd_fmd.c for details
1386 * \param[in] env execution environment
1387 * \param[in] exp OBD export of client
1391 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1393 ofd_fmd_expire(exp);
1398 * Implementation of obd_ops::o_health_check.
1400 * This function checks the OFD device health - ability to respond on
1401 * incoming requests. There are two health_check methods:
1402 * - get statfs from the OSD. It checks just responsiveness of
1404 * - do write attempt on bottom device to check it is fully operational and
1405 * is not stuck. This is expensive method and requires special configuration
1406 * option --enable-health-write while building Lustre, it is turned off
1409 * \param[in] nul not used
1410 * \param[in] obd OBD device of OFD
1412 * \retval 0 if successful
1413 * \retval negative value in case of error
1415 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1417 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1418 struct ofd_thread_info *info;
1420 #ifdef USE_HEALTH_CHECK_WRITE
1425 /* obd_proc_read_health pass NULL env, we need real one */
1426 rc = lu_env_init(&env, LCT_DT_THREAD);
1430 info = ofd_info_init(&env, NULL);
1431 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1435 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1436 GOTO(out, rc = -EROFS);
1438 #ifdef USE_HEALTH_CHECK_WRITE
1439 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1440 if (info->fti_buf.lb_buf == NULL)
1441 GOTO(out, rc = -ENOMEM);
1443 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1446 th = dt_trans_create(&env, ofd->ofd_osd);
1448 GOTO(out, rc = PTR_ERR(th));
1450 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1451 &info->fti_buf, info->fti_off, th);
1453 th->th_sync = 1; /* sync IO is needed */
1454 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1456 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1457 &info->fti_buf, &info->fti_off,
1460 dt_trans_stop(&env, ofd->ofd_osd, th);
1462 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1464 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1471 struct obd_ops ofd_obd_ops = {
1472 .o_owner = THIS_MODULE,
1473 .o_connect = ofd_obd_connect,
1474 .o_reconnect = ofd_obd_reconnect,
1475 .o_disconnect = ofd_obd_disconnect,
1476 .o_create = ofd_echo_create,
1477 .o_statfs = ofd_statfs,
1478 .o_setattr = ofd_echo_setattr,
1479 .o_preprw = ofd_preprw,
1480 .o_commitrw = ofd_commitrw,
1481 .o_destroy = ofd_echo_destroy,
1482 .o_init_export = ofd_init_export,
1483 .o_destroy_export = ofd_destroy_export,
1484 .o_postrecov = ofd_obd_postrecov,
1485 .o_getattr = ofd_echo_getattr,
1486 .o_iocontrol = ofd_iocontrol,
1487 .o_precleanup = ofd_precleanup,
1489 .o_health_check = ofd_health_check,
1490 .o_set_info_async = ofd_set_info_async,
1491 .o_get_info = ofd_get_info,