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, 2016, 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 * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
126 * full struct obd_connect_data. So care must be taken when accessing fields
127 * that are not present in struct obd_connect_data_v1. See LU-16.
129 * \param[in] env execution environment
130 * \param[in] exp the obd_export associated with this
132 * \param[in] data stores data for this connect request
133 * \param[in] new_connection is this connection new or not
135 * \retval 0 if success
136 * \retval -EPROTO client and server feature requirements are
138 * \retval -EBADF OST index in connect request doesn't match
141 static int ofd_parse_connect_data(const struct lu_env *env,
142 struct obd_export *exp,
143 struct obd_connect_data *data,
146 struct ofd_device *ofd = ofd_exp(exp);
147 struct filter_export_data *fed = &exp->exp_filter_data;
152 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: %#llx"
153 " ocd_version: %x ocd_grant: %d ocd_index: %u"
155 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
156 data->ocd_connect_flags, data->ocd_version,
157 data->ocd_grant, data->ocd_index, data->ocd_group);
159 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
160 CWARN("!!! This export (nid %s) used object group %d "
161 "earlier; now it's trying to use group %d! This could "
162 "be a bug in the MDS. Please report to "
163 "https://jira.hpdd.intel.com/\n",
164 obd_export_nid2str(exp), fed->fed_group,
168 fed->fed_group = data->ocd_group;
170 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
172 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
173 data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2;
175 data->ocd_version = LUSTRE_VERSION_CODE;
177 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
178 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
179 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
180 exp->exp_obd->obd_name, data->ocd_group);
181 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
184 /* Determine optimal brw size before calculating grant */
185 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
186 data->ocd_brw_size = 65536;
187 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
188 if (data->ocd_brw_size > ofd->ofd_brw_size)
189 data->ocd_brw_size = ofd->ofd_brw_size;
190 if (data->ocd_brw_size == 0) {
191 CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
192 " ocd_version: %x ocd_grant: %d ocd_index: %u "
193 "ocd_brw_size is unexpectedly zero, "
194 "network data corruption?"
195 "Refusing connection of this client\n",
196 exp->exp_obd->obd_name,
197 exp->exp_client_uuid.uuid,
198 exp, data->ocd_connect_flags, data->ocd_version,
199 data->ocd_grant, data->ocd_index);
204 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
205 /* client is reporting its page size, for future use */
206 exp->exp_filter_data.fed_pagebits = data->ocd_grant_blkbits;
207 data->ocd_grant_blkbits = ofd->ofd_blockbits;
208 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
209 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
210 data->ocd_grant_inobits =
211 fls(ofd->ofd_dt_conf.ddp_inodespace - 1);
212 /* ocd_grant_tax_kb is in 1K byte blocks */
213 data->ocd_grant_tax_kb = ofd->ofd_dt_conf.ddp_extent_tax >> 10;
214 data->ocd_grant_max_blks = ofd->ofd_dt_conf.ddp_max_extent_blks;
217 if (OCD_HAS_FLAG(data, GRANT)) {
218 /* Save connect_data we have so far because ofd_grant_connect()
219 * uses it to calculate grant. */
220 exp->exp_connect_data = *data;
221 ofd_grant_connect(env, exp, data, new_connection);
224 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
225 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
226 int index = lsd->lsd_osd_index;
228 if (index != data->ocd_index) {
229 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
230 " %u doesn't match actual OST index"
231 " %u in last_rcvd file, bad "
233 obd_export_nid2str(exp), index,
237 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
238 /* this will only happen on the first connect */
239 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
240 /* sync is not needed here as tgt_client_new will
241 * set exp_need_sync flag */
242 tgt_server_data_update(env, &ofd->ofd_lut, 0);
246 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
247 __u32 cksum_types = data->ocd_cksum_types;
249 /* The client set in ocd_cksum_types the checksum types it
250 * supports. We have to mask off the algorithms that we don't
252 data->ocd_cksum_types &= ofd->ofd_cksum_types_supported;
254 if (unlikely(data->ocd_cksum_types == 0)) {
255 CERROR("%s: Connect with checksum support but no "
256 "ocd_cksum_types is set\n",
257 exp->exp_obd->obd_name);
261 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
262 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
263 cksum_types, data->ocd_cksum_types);
265 /* This client does not support OBD_CONNECT_CKSUM.
266 * Report failure to negotiate checksum at connect */
267 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
268 "OBD_CONNECT_CKSUM\n",
269 exp->exp_obd->obd_name, obd_export_nid2str(exp));
272 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
273 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
275 if (OCD_HAS_FLAG(data, PINGLESS)) {
276 if (ptlrpc_pinger_suppress_pings()) {
277 spin_lock(&exp->exp_obd->obd_dev_lock);
278 list_del_init(&exp->exp_obd_chain_timed);
279 spin_unlock(&exp->exp_obd->obd_dev_lock);
281 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
289 * Re-initialize export upon client reconnection.
291 * This function parses connection data from reconnect and resets
294 * \param[in] env execution environment
295 * \param[in] exp OBD export
296 * \param[in] obd OFD device
297 * \param[in] cluuid NID of client
298 * \param[in] data connection data from request
299 * \param[in] localdata client NID
301 * \retval 0 if successful
302 * \retval negative value on error
304 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
305 struct obd_device *obd, struct obd_uuid *cluuid,
306 struct obd_connect_data *data,
309 struct ofd_device *ofd;
314 if (exp == NULL || obd == NULL || cluuid == NULL)
317 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
318 if (rc != 0 && rc != -EEXIST)
321 ofd = ofd_dev(obd->obd_lu_dev);
323 rc = ofd_parse_connect_data(env, exp, data, false);
325 ofd_export_stats_init(ofd, exp, client_nid);
327 nodemap_del_member(exp);
333 * Initialize new client connection.
335 * This function handles new connection to the OFD. The new export is
336 * created (in context of class_connect()) and persistent client data is
337 * initialized on storage.
339 * \param[in] env execution environment
340 * \param[out] _exp stores pointer to new export
341 * \param[in] obd OFD device
342 * \param[in] cluuid client UUID
343 * \param[in] data connection data from request
344 * \param[in] localdata client NID
346 * \retval 0 if successful
347 * \retval negative value on error
349 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
350 struct obd_device *obd, struct obd_uuid *cluuid,
351 struct obd_connect_data *data, void *localdata)
353 struct obd_export *exp;
354 struct ofd_device *ofd;
355 struct lustre_handle conn = { 0 };
359 if (_exp == NULL || obd == NULL || cluuid == NULL)
362 ofd = ofd_dev(obd->obd_lu_dev);
364 rc = class_connect(&conn, obd, cluuid);
368 exp = class_conn2export(&conn);
369 LASSERT(exp != NULL);
371 if (localdata != NULL) {
372 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
373 if (rc != 0 && rc != -EEXIST)
376 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
377 "nid is null\n", obd->obd_name, cluuid->uuid);
380 rc = ofd_parse_connect_data(env, exp, data, true);
384 if (obd->obd_replayable) {
385 struct tg_export_data *ted = &exp->exp_target_data;
387 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
388 sizeof(ted->ted_lcd->lcd_uuid));
389 rc = tgt_client_new(env, exp);
392 ofd_export_stats_init(ofd, exp, localdata);
395 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
396 data ? data->ocd_group : -1);
400 class_disconnect(exp);
401 nodemap_del_member(exp);
410 * Disconnect a connected client.
412 * This function terminates the client connection. The client export is
413 * disconnected (cleaned up) and client data on persistent storage is removed.
415 * \param[in] exp OBD export
417 * \retval 0 if successful
418 * \retval negative value on error
420 int ofd_obd_disconnect(struct obd_export *exp)
422 struct ofd_device *ofd = ofd_exp(exp);
429 class_export_get(exp);
431 if (!(exp->exp_flags & OBD_OPT_FORCE))
432 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
434 rc = server_disconnect_export(exp);
436 ofd_grant_discard(exp);
438 /* Do not erase record for recoverable client. */
439 if (exp->exp_obd->obd_replayable &&
440 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
441 rc = lu_env_init(&env, LCT_DT_THREAD);
445 tgt_client_del(&env, exp);
449 nodemap_del_member(exp);
450 class_export_put(exp);
455 * Implementation of obd_ops::o_init_export.
457 * This function is called from class_new_export() and initializes
458 * the OFD-specific data for new export.
460 * \param[in] exp OBD export
462 * \retval 0 if successful
463 * \retval negative value on error
465 static int ofd_init_export(struct obd_export *exp)
469 spin_lock_init(&exp->exp_filter_data.fed_lock);
470 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
471 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
472 spin_lock(&exp->exp_lock);
473 exp->exp_connecting = 1;
474 spin_unlock(&exp->exp_lock);
476 /* self-export doesn't need client data and ldlm initialization */
477 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
478 &exp->exp_client_uuid)))
481 rc = tgt_client_alloc(exp);
483 ldlm_init_export(exp);
485 CERROR("%s: Can't initialize export: rc %d\n",
486 exp->exp_obd->obd_name, rc);
491 * Implementation of obd_ops::o_destroy_export.
493 * This function is called from class_export_destroy() to cleanup
494 * the OFD-specific data for export being destroyed.
496 * \param[in] exp OBD export
498 * \retval 0 if successful
499 * \retval negative value on error
501 static int ofd_destroy_export(struct obd_export *exp)
503 struct ofd_device *ofd = ofd_exp(exp);
505 if (exp->exp_filter_data.fed_pending)
506 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
507 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
508 exp, exp->exp_filter_data.fed_pending);
510 target_destroy_export(exp);
512 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
513 &exp->exp_client_uuid)))
516 ldlm_destroy_export(exp);
517 tgt_client_free(exp);
519 ofd_fmd_cleanup(exp);
522 * discard grants once we're sure no more
523 * interaction with the client is possible
525 ofd_grant_discard(exp);
526 ofd_fmd_cleanup(exp);
528 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
529 ofd->ofd_tot_granted_clients--;
531 if (!(exp->exp_flags & OBD_OPT_FORCE))
532 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
534 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
539 * Notify all devices in server stack about recovery completion.
541 * This function calls ldo_recovery_complete() for all lower devices in the
542 * server stack so they will be prepared for normal operations.
544 * \param[in] env execution environment
545 * \param[in] ofd OFD device
547 * \retval 0 if successful
548 * \retval negative value on error
550 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
552 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
555 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
557 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
558 struct lfsck_start_param lsp;
560 lsp.lsp_start = NULL;
561 lsp.lsp_index_valid = 0;
562 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
563 if (rc != 0 && rc != -EALREADY)
564 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
568 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
572 * Implementation of obd_ops::o_postrecov.
574 * This function is called from target_finish_recovery() upon recovery
577 * \param[in] obd OBD device of OFD
579 * \retval 0 if successful
580 * \retval negative value on error
582 static int ofd_obd_postrecov(struct obd_device *obd)
585 struct lu_device *ldev = obd->obd_lu_dev;
590 rc = lu_env_init(&env, LCT_DT_THREAD);
593 ofd_info_init(&env, obd->obd_self_export);
595 rc = ofd_postrecov(&env, ofd_dev(ldev));
602 * Implementation of obd_ops::o_set_info_async.
604 * This function is not called from request handler, it is only used by
605 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
606 * \see ofd_set_info_hdl() for request handler function.
608 * \param[in] env execution environment
609 * \param[in] exp OBD export of OFD device
610 * \param[in] keylen length of \a key
611 * \param[in] key key name
612 * \param[in] vallen length of \a val
613 * \param[in] val the \a key value
614 * \param[in] set not used in OFD
616 * \retval 0 if successful
617 * \retval negative value on error
619 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
620 __u32 keylen, void *key, __u32 vallen, void *val,
621 struct ptlrpc_request_set *set)
627 if (exp->exp_obd == NULL) {
628 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
632 if (KEY_IS(KEY_SPTLRPC_CONF)) {
633 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
635 CERROR("%s: Unsupported key %s\n",
636 exp->exp_obd->obd_name, (char*)key);
643 * Implementation of obd_ops::o_get_info.
645 * This function is not called from request handler, it is only used by
646 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
648 * \see ofd_get_info_hdl() for request handler function.
650 * \param[in] env execution environment
651 * \param[in] exp OBD export of OFD device
652 * \param[in] keylen length of \a key
653 * \param[in] key key name
654 * \param[out] vallen length of key value
655 * \param[out] val the key value to return
657 * \retval 0 if successful
658 * \retval negative value on error
660 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
661 __u32 keylen, void *key, __u32 *vallen, void *val)
663 struct ofd_thread_info *info;
664 struct ofd_device *ofd;
665 struct ll_fiemap_info_key *fm_key = key;
666 struct fiemap *fiemap = val;
671 if (exp->exp_obd == NULL) {
672 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
678 if (KEY_IS(KEY_FIEMAP)) {
679 info = ofd_info_init(env, exp);
681 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
682 ofd->ofd_lut.lut_lsd.lsd_osd_index);
686 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
688 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
696 * Get file system statistics of OST server.
698 * Helper function for ofd_statfs(), also used by grant code.
699 * Implements caching for statistics to avoid calling OSD device each time.
701 * \param[in] env execution environment
702 * \param[in] ofd OFD device
703 * \param[out] osfs statistic data to return
704 * \param[in] max_age maximum age for cached data
705 * \param[in] from_cache show that data was get from cache or not
707 * \retval 0 if successful
708 * \retval negative value on error
710 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
711 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
716 spin_lock(&ofd->ofd_osfs_lock);
717 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
720 /* statfs data are too old, get up-to-date one.
721 * we must be cautious here since multiple threads might be
722 * willing to update statfs data concurrently and we must
723 * grant that cached statfs data are always consistent */
725 if (ofd->ofd_statfs_inflight == 0)
726 /* clear inflight counter if no users, although it would
727 * take a while to overflow this 64-bit counter ... */
728 ofd->ofd_osfs_inflight = 0;
729 /* notify ofd_grant_commit() that we want to track writes
730 * completed as of now */
731 ofd->ofd_statfs_inflight++;
732 /* record value of inflight counter before running statfs to
733 * compute the diff once statfs is completed */
734 unstable = ofd->ofd_osfs_inflight;
735 spin_unlock(&ofd->ofd_osfs_lock);
737 /* statfs can sleep ... hopefully not for too long since we can
738 * call it fairly often as space fills up */
739 rc = dt_statfs(env, ofd->ofd_osd, osfs);
743 spin_lock(&ofd->ofd_grant_lock);
744 spin_lock(&ofd->ofd_osfs_lock);
745 /* calculate how much space was written while we released the
747 unstable = ofd->ofd_osfs_inflight - unstable;
748 ofd->ofd_osfs_unstable = 0;
750 /* some writes committed while we were running statfs
751 * w/o the ofd_osfs_lock. Those ones got added to
752 * the cached statfs data that we are about to crunch.
753 * Take them into account in the new statfs data */
754 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
755 unstable >> ofd->ofd_blockbits);
756 /* However, we don't really know if those writes got
757 * accounted in the statfs call, so tell
758 * ofd_grant_space_left() there is some uncertainty
759 * on the accounting of those writes.
760 * The purpose is to prevent spurious error messages in
761 * ofd_grant_space_left() since those writes might be
762 * accounted twice. */
763 ofd->ofd_osfs_unstable += unstable;
765 /* similarly, there is some uncertainty on write requests
766 * between prepare & commit */
767 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
768 spin_unlock(&ofd->ofd_grant_lock);
770 /* finally udpate cached statfs data */
771 ofd->ofd_osfs = *osfs;
772 ofd->ofd_osfs_age = cfs_time_current_64();
774 ofd->ofd_statfs_inflight--; /* stop tracking */
775 if (ofd->ofd_statfs_inflight == 0)
776 ofd->ofd_osfs_inflight = 0;
777 spin_unlock(&ofd->ofd_osfs_lock);
782 /* use cached statfs data */
783 *osfs = ofd->ofd_osfs;
784 spin_unlock(&ofd->ofd_osfs_lock);
795 * Implementation of obd_ops::o_statfs.
797 * This function returns information about a storage file system.
798 * It is called from several places by using the OBD API as well as
799 * by direct call, e.g. from request handler.
801 * \see ofd_statfs_hdl() for request handler function.
803 * Report also the state of the OST to the caller in osfs->os_state
804 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
806 * \param[in] env execution environment
807 * \param[in] exp OBD export of OFD device
808 * \param[out] osfs statistic data to return
809 * \param[in] max_age maximum age for cached data
810 * \param[in] flags not used in OFD
812 * \retval 0 if successful
813 * \retval negative value on error
815 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
816 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
818 struct obd_device *obd = class_exp2obd(exp);
819 struct ofd_device *ofd = ofd_exp(exp);
824 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
828 /* at least try to account for cached pages. its still racy and
829 * might be under-reporting if clients haven't announced their
830 * caches with brw recently */
832 CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
833 " pending %llu free %llu avail %llu\n",
834 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
835 osfs->os_bfree << ofd->ofd_blockbits,
836 osfs->os_bavail << ofd->ofd_blockbits);
838 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
839 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
840 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
842 /* The QoS code on the MDS does not care about space reserved for
843 * precreate, so take it out. */
844 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
845 struct filter_export_data *fed;
847 fed = &obd->obd_self_export->exp_filter_data;
848 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
849 fed->fed_grant >> ofd->ofd_blockbits);
852 ofd_grant_sanity_check(obd, __FUNCTION__);
853 CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
854 "%llu objects: %llu free; state %x\n",
855 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
856 osfs->os_files, osfs->os_ffree, osfs->os_state);
858 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
859 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
860 /* Reduce free inode count to zero, but keep "used" intact */
861 osfs->os_files -= osfs->os_ffree;
862 osfs->os_ffree -= osfs->os_ffree;
865 /* OS_STATE_READONLY can be set by OSD already */
866 if (ofd->ofd_raid_degraded)
867 osfs->os_state |= OS_STATE_DEGRADED;
869 if (obd->obd_self_export != exp && !ofd_grant_param_supp(exp) &&
870 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT) {
871 /* clients which don't support OBD_CONNECT_GRANT_PARAM
872 * should not see a block size > page size, otherwise
873 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
874 * block size which is the biggest block size known to work
875 * with all client's page size. */
876 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
877 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
878 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
879 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
882 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
883 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
884 /* Reduce free blocks count near zero, but keep "used" intact */
885 osfs->os_bavail -= osfs->os_bavail - 2;
886 osfs->os_blocks -= osfs->os_bfree - 2;
887 osfs->os_bfree -= osfs->os_bfree - 2;
896 * Implementation of obd_ops::o_setattr.
898 * This function is only used by ECHO client when it is run on top of OFD,
899 * \see ofd_setattr_hdl() for request handler function.
901 * \param[in] env execution environment
902 * \param[in] exp OBD export of OFD device
903 * \param[in] oa setattr parameters
905 * \retval 0 if successful
906 * \retval negative value on error
908 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
911 struct ofd_thread_info *info;
912 struct ofd_device *ofd = ofd_exp(exp);
913 struct ldlm_namespace *ns = ofd->ofd_namespace;
914 struct ldlm_resource *res;
915 struct ofd_object *fo;
916 struct lu_fid *fid = &oa->o_oi.oi_fid;
917 struct filter_fid *ff = NULL;
922 info = ofd_info_init(env, exp);
924 ost_fid_build_resid(fid, &info->fti_resid);
926 /* This would be very bad - accidentally truncating a file when
927 * changing the time or similar - bug 12203. */
928 if (oa->o_valid & OBD_MD_FLSIZE) {
929 static char mdsinum[48];
931 if (oa->o_valid & OBD_MD_FLFID)
932 snprintf(mdsinum, sizeof(mdsinum) - 1,
933 "of parent "DFID, oa->o_parent_seq,
934 oa->o_parent_oid, 0);
938 CERROR("%s: setattr from %s trying to truncate object "DFID
939 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
941 GOTO(out, rc = -EPERM);
944 fo = ofd_object_find_exists(env, ofd, fid);
946 CERROR("%s: can't find object "DFID"\n",
947 ofd_name(ofd), PFID(fid));
948 GOTO(out, rc = PTR_ERR(fo));
951 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
952 info->fti_attr.la_valid &= ~LA_TYPE;
954 if (oa->o_valid & OBD_MD_FLFID) {
955 ff = &info->fti_mds_fid;
956 ofd_prepare_fidea(ff, oa);
959 /* setting objects attributes (including owner/group) */
960 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
962 GOTO(out_unlock, rc);
964 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
967 ofd_object_put(env, fo);
970 /* we do not call this before to avoid lu_object_find() in
971 * ->lvbo_update() holding another reference on the object.
972 * otherwise concurrent destroy can make the object unavailable
973 * for 2nd lu_object_find() waiting for the first reference
974 * to go... deadlock! */
975 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
977 ldlm_res_lvbo_update(res, NULL, 0);
978 ldlm_resource_putref(res);
986 * Destroy OFD object by its FID.
988 * Supplemental function to destroy object by FID, it is used by request
989 * handler and by ofd_echo_destroy() below to find object by FID, lock it
990 * and call ofd_object_destroy() finally.
992 * \param[in] env execution environment
993 * \param[in] ofd OFD device
994 * \param[in] fid FID of object
995 * \param[in] orphan set if object being destroyed is an orphan
997 * \retval 0 if successful
998 * \retval negative value on error
1000 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
1001 const struct lu_fid *fid, int orphan)
1003 struct ofd_thread_info *info = ofd_info(env);
1004 struct lustre_handle lockh;
1005 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1006 struct ofd_object *fo;
1007 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
1012 fo = ofd_object_find_exists(env, ofd, fid);
1014 RETURN(PTR_ERR(fo));
1016 /* Tell the clients that the object is gone now and that they should
1017 * throw away any cached pages. */
1018 ost_fid_build_resid(fid, &info->fti_resid);
1019 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1020 LDLM_EXTENT, &policy, LCK_PW, &flags,
1021 ldlm_blocking_ast, ldlm_completion_ast,
1022 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1024 /* We only care about the side-effects, just drop the lock. */
1026 ldlm_lock_decref(&lockh, LCK_PW);
1028 LASSERT(fo != NULL);
1030 rc = ofd_object_destroy(env, fo, orphan);
1033 ofd_object_put(env, fo);
1038 * Implementation of obd_ops::o_destroy.
1040 * This function is only used by ECHO client when it is run on top of OFD,
1041 * \see ofd_destroy_hdl() for request handler function.
1043 * \param[in] env execution environment
1044 * \param[in] exp OBD export of OFD device
1045 * \param[in] oa obdo structure with FID
1047 * Note: this is OBD API method which is common API for server OBDs and
1048 * client OBDs. Thus some parameters used in client OBDs may not be used
1049 * on server OBDs and vice versa.
1051 * \retval 0 if successful
1052 * \retval negative value on error
1054 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1057 struct ofd_device *ofd = ofd_exp(exp);
1058 struct lu_fid *fid = &oa->o_oi.oi_fid;
1063 ofd_info_init(env, exp);
1065 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
1069 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1071 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1072 if (rc == -ENOENT) {
1073 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1074 ofd_name(ofd), PFID(fid));
1076 } else if (rc != 0) {
1077 CERROR("%s: error destroying object "DFID": %d\n",
1078 ofd_name(ofd), PFID(fid), rc);
1087 * Implementation of obd_ops::o_create.
1089 * This function is only used by ECHO client when it is run on top of OFD
1090 * and just creates an object.
1091 * \see ofd_create_hdl() for request handler function.
1093 * \param[in] env execution environment
1094 * \param[in] exp OBD export of OFD device
1095 * \param[in] oa obdo structure with FID sequence to use
1097 * Note: this is OBD API method which is common API for server OBDs and
1098 * client OBDs. Thus some parameters used in client OBDs may not be used
1099 * on server OBDs and vice versa.
1101 * \retval 0 if successful
1102 * \retval negative value on error
1104 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1107 struct ofd_device *ofd = ofd_exp(exp);
1108 u64 seq = ostid_seq(&oa->o_oi);
1109 struct ofd_seq *oseq;
1110 int rc = 0, diff = 1;
1117 ofd_info_init(env, exp);
1119 LASSERT(seq == FID_SEQ_ECHO);
1120 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1122 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1124 down_read(&ofd->ofd_lastid_rwsem);
1125 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1126 * we may do that in the future.
1127 * Return -ENOSPC until the LAST_ID rebuilt. */
1128 if (unlikely(ofd->ofd_lastid_rebuilding))
1129 GOTO(out_sem, rc = -ENOSPC);
1131 rc = ofd_validate_seq(exp, seq);
1135 oseq = ofd_seq_load(env, ofd, seq);
1137 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1138 ofd_name(ofd), seq, PTR_ERR(oseq));
1139 GOTO(out_sem, rc = -EINVAL);
1142 mutex_lock(&oseq->os_create_lock);
1143 granted = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1147 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1148 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
1153 next_id = ofd_seq_last_oid(oseq) + 1;
1154 count = ofd_precreate_batch(ofd, diff);
1156 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1158 CERROR("%s: unable to precreate: rc = %d\n",
1161 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1162 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1166 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1168 mutex_unlock(&oseq->os_create_lock);
1169 ofd_seq_put(env, oseq);
1172 up_read(&ofd->ofd_lastid_rwsem);
1177 * Implementation of obd_ops::o_getattr.
1179 * This function is only used by ECHO client when it is run on top of OFD
1180 * and returns attributes of object.
1181 * \see ofd_getattr_hdl() for request handler function.
1183 * \param[in] env execution environment
1184 * \param[in] exp OBD export of OFD device
1185 * \param[in,out] oa contains FID of object to get attributes from and
1186 * is used to return attributes back
1188 * \retval 0 if successful
1189 * \retval negative value on error
1191 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1194 struct ofd_device *ofd = ofd_exp(exp);
1195 struct ofd_thread_info *info;
1196 struct lu_fid *fid = &oa->o_oi.oi_fid;
1197 struct ofd_object *fo;
1202 info = ofd_info_init(env, exp);
1204 fo = ofd_object_find_exists(env, ofd, fid);
1206 GOTO(out, rc = PTR_ERR(fo));
1208 LASSERT(fo != NULL);
1209 rc = ofd_attr_get(env, fo, &info->fti_attr);
1210 oa->o_valid = OBD_MD_FLID;
1214 obdo_from_la(oa, &info->fti_attr,
1215 OFD_VALID_FLAGS | LA_UID | LA_GID);
1217 /* Store object version in reply */
1218 curr_version = dt_version_get(env, ofd_object_child(fo));
1219 if ((__s64)curr_version != -EOPNOTSUPP) {
1220 oa->o_valid |= OBD_MD_FLDATAVERSION;
1221 oa->o_data_version = curr_version;
1225 ofd_object_put(env, fo);
1231 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1233 * This is supplemental function for ofd_iocontrol() to return object
1234 * version for lctl tool.
1236 * \param[in] env execution environment
1237 * \param[in] ofd OFD device
1238 * \param[out] karg ioctl data
1240 * \retval 0 if successful
1241 * \retval negative value on error
1243 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1244 struct ofd_device *ofd, void *karg)
1246 struct obd_ioctl_data *data = karg;
1248 struct ofd_object *fo;
1249 dt_obj_version_t version;
1254 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1255 GOTO(out, rc = -EINVAL);
1257 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1258 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1259 } else if (data->ioc_inlbuf3 != NULL &&
1260 data->ioc_inllen3 == sizeof(__u64) &&
1261 data->ioc_inlbuf4 != NULL &&
1262 data->ioc_inllen4 == sizeof(__u64)) {
1263 struct ost_id ostid;
1265 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1266 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1267 rc = ostid_to_fid(&fid, &ostid,
1268 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1272 GOTO(out, rc = -EINVAL);
1275 if (!fid_is_sane(&fid))
1276 GOTO(out, rc = -EINVAL);
1278 fo = ofd_object_find(env, ofd, &fid);
1280 GOTO(out, rc = PTR_ERR(fo));
1282 if (!ofd_object_exists(fo))
1283 GOTO(out_fo, rc = -ENOENT);
1285 if (lu_object_remote(&fo->ofo_obj.do_lu))
1286 GOTO(out_fo, rc = -EREMOTE);
1288 version = dt_version_get(env, ofd_object_child(fo));
1290 GOTO(out_fo, rc = -EIO);
1292 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1296 ofd_object_put(env, fo);
1302 * Implementation of obd_ops::o_iocontrol.
1304 * This is OFD ioctl handling function which is primary interface for
1305 * Lustre tools like lfs, lctl and lfsck.
1307 * \param[in] cmd ioctl command
1308 * \param[in] exp OBD export of OFD
1309 * \param[in] len not used
1310 * \param[in,out] karg buffer with data
1311 * \param[in] uarg not used
1313 * \retval 0 if successful
1314 * \retval negative value on error
1316 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1317 void *karg, void __user *uarg)
1320 struct ofd_device *ofd = ofd_exp(exp);
1321 struct obd_device *obd = ofd_obd(ofd);
1326 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1327 rc = lu_env_init(&env, LCT_DT_THREAD);
1332 case OBD_IOC_ABORT_RECOVERY:
1333 CERROR("%s: aborting recovery\n", obd->obd_name);
1334 obd->obd_abort_recovery = 1;
1335 target_stop_recovery_thread(obd);
1338 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1339 rc = dt_sync(&env, ofd->ofd_osd);
1341 case OBD_IOC_SET_READONLY:
1342 rc = dt_sync(&env, ofd->ofd_osd);
1344 rc = dt_ro(&env, ofd->ofd_osd);
1346 case OBD_IOC_START_LFSCK: {
1347 struct obd_ioctl_data *data = karg;
1348 struct lfsck_start_param lsp;
1350 if (unlikely(data == NULL)) {
1355 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1356 lsp.lsp_index_valid = 0;
1357 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1360 case OBD_IOC_STOP_LFSCK: {
1361 struct obd_ioctl_data *data = karg;
1362 struct lfsck_stop stop;
1364 stop.ls_status = LS_STOPPED;
1365 /* Old lfsck utils may pass NULL @stop. */
1366 if (data->ioc_inlbuf1 == NULL)
1370 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1372 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1375 case OBD_IOC_GET_OBJ_VERSION:
1376 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1379 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1388 * Implementation of obd_ops::o_precleanup.
1390 * This function stops device activity before shutting it down. It is called
1391 * from a cleanup function upon forceful device cleanup. For OFD there are no
1392 * special actions, it just invokes target_recovery_cleanup().
1394 * \param[in] obd OBD device of OFD
1398 static int ofd_precleanup(struct obd_device *obd)
1401 target_cleanup_recovery(obd);
1406 * Implementation of obd_ops::o_ping.
1408 * This is OFD-specific part of OBD_PING request handling.
1409 * It controls Filter Modification Data (FMD) expiration each time PING is
1412 * \see ofd_fmd_expire() and ofd_fmd.c for details
1414 * \param[in] env execution environment
1415 * \param[in] exp OBD export of client
1419 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1421 ofd_fmd_expire(exp);
1426 * Implementation of obd_ops::o_health_check.
1428 * This function checks the OFD device health - ability to respond on
1429 * incoming requests. There are two health_check methods:
1430 * - get statfs from the OSD. It checks just responsiveness of
1432 * - do write attempt on bottom device to check it is fully operational and
1433 * is not stuck. This is expensive method and requires special configuration
1434 * option --enable-health-write while building Lustre, it is turned off
1437 * \param[in] nul not used
1438 * \param[in] obd OBD device of OFD
1440 * \retval 0 if successful
1441 * \retval negative value in case of error
1443 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1445 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1446 struct ofd_thread_info *info;
1448 #ifdef USE_HEALTH_CHECK_WRITE
1453 /* obd_proc_read_health pass NULL env, we need real one */
1454 rc = lu_env_init(&env, LCT_DT_THREAD);
1458 info = ofd_info_init(&env, NULL);
1459 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1463 if (info->fti_u.osfs.os_state & OS_STATE_READONLY)
1464 GOTO(out, rc = -EROFS);
1466 #ifdef USE_HEALTH_CHECK_WRITE
1467 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1468 if (info->fti_buf.lb_buf == NULL)
1469 GOTO(out, rc = -ENOMEM);
1471 info->fti_buf.lb_len = PAGE_SIZE;
1474 th = dt_trans_create(&env, ofd->ofd_osd);
1476 GOTO(out, rc = PTR_ERR(th));
1478 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1479 &info->fti_buf, info->fti_off, th);
1481 th->th_sync = 1; /* sync IO is needed */
1482 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1484 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1485 &info->fti_buf, &info->fti_off,
1488 dt_trans_stop(&env, ofd->ofd_osd, th);
1490 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1492 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1499 struct obd_ops ofd_obd_ops = {
1500 .o_owner = THIS_MODULE,
1501 .o_connect = ofd_obd_connect,
1502 .o_reconnect = ofd_obd_reconnect,
1503 .o_disconnect = ofd_obd_disconnect,
1504 .o_create = ofd_echo_create,
1505 .o_statfs = ofd_statfs,
1506 .o_setattr = ofd_echo_setattr,
1507 .o_preprw = ofd_preprw,
1508 .o_commitrw = ofd_commitrw,
1509 .o_destroy = ofd_echo_destroy,
1510 .o_init_export = ofd_init_export,
1511 .o_destroy_export = ofd_destroy_export,
1512 .o_postrecov = ofd_obd_postrecov,
1513 .o_getattr = ofd_echo_getattr,
1514 .o_iocontrol = ofd_iocontrol,
1515 .o_precleanup = ofd_precleanup,
1517 .o_health_check = ofd_health_check,
1518 .o_set_info_async = ofd_set_info_async,
1519 .o_get_info = ofd_get_info,