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 <uapi/linux/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 struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf;
207 /* client is reporting its page size, for future use */
208 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
209 data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits;
210 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
211 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
212 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
213 /* ocd_grant_tax_kb is in 1K byte blocks */
214 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
215 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
218 if (OCD_HAS_FLAG(data, GRANT)) {
219 /* Save connect_data we have so far because tgt_grant_connect()
220 * uses it to calculate grant. */
221 exp->exp_connect_data = *data;
222 tgt_grant_connect(env, exp, data, new_connection);
225 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
226 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
227 int index = lsd->lsd_osd_index;
229 if (index != data->ocd_index) {
230 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
231 " %u doesn't match actual OST index"
232 " %u in last_rcvd file, bad "
234 obd_export_nid2str(exp), index,
238 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
239 /* this will only happen on the first connect */
240 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
241 /* sync is not needed here as tgt_client_new will
242 * set exp_need_sync flag */
243 tgt_server_data_update(env, &ofd->ofd_lut, 0);
247 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
248 __u32 cksum_types = data->ocd_cksum_types;
250 /* The client set in ocd_cksum_types the checksum types it
251 * supports. We have to mask off the algorithms that we don't
253 data->ocd_cksum_types &= ofd->ofd_cksum_types_supported;
255 if (unlikely(data->ocd_cksum_types == 0)) {
256 CERROR("%s: Connect with checksum support but no "
257 "ocd_cksum_types is set\n",
258 exp->exp_obd->obd_name);
262 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
263 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
264 cksum_types, data->ocd_cksum_types);
266 /* This client does not support OBD_CONNECT_CKSUM.
267 * Report failure to negotiate checksum at connect */
268 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
269 "OBD_CONNECT_CKSUM\n",
270 exp->exp_obd->obd_name, obd_export_nid2str(exp));
273 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
274 data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes;
276 if (OCD_HAS_FLAG(data, PINGLESS)) {
277 if (ptlrpc_pinger_suppress_pings()) {
278 spin_lock(&exp->exp_obd->obd_dev_lock);
279 list_del_init(&exp->exp_obd_chain_timed);
280 spin_unlock(&exp->exp_obd->obd_dev_lock);
282 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
290 * Re-initialize export upon client reconnection.
292 * This function parses connection data from reconnect and resets
295 * \param[in] env execution environment
296 * \param[in] exp OBD export
297 * \param[in] obd OFD device
298 * \param[in] cluuid NID of client
299 * \param[in] data connection data from request
300 * \param[in] localdata client NID
302 * \retval 0 if successful
303 * \retval negative value on error
305 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
306 struct obd_device *obd, struct obd_uuid *cluuid,
307 struct obd_connect_data *data,
310 struct ofd_device *ofd;
315 if (exp == NULL || obd == NULL || cluuid == NULL)
318 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
319 if (rc != 0 && rc != -EEXIST)
322 ofd = ofd_dev(obd->obd_lu_dev);
324 rc = ofd_parse_connect_data(env, exp, data, false);
326 ofd_export_stats_init(ofd, exp, client_nid);
328 nodemap_del_member(exp);
334 * Initialize new client connection.
336 * This function handles new connection to the OFD. The new export is
337 * created (in context of class_connect()) and persistent client data is
338 * initialized on storage.
340 * \param[in] env execution environment
341 * \param[out] _exp stores pointer to new export
342 * \param[in] obd OFD device
343 * \param[in] cluuid client UUID
344 * \param[in] data connection data from request
345 * \param[in] localdata client NID
347 * \retval 0 if successful
348 * \retval negative value on error
350 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
351 struct obd_device *obd, struct obd_uuid *cluuid,
352 struct obd_connect_data *data, void *localdata)
354 struct obd_export *exp;
355 struct ofd_device *ofd;
356 struct lustre_handle conn = { 0 };
360 if (_exp == NULL || obd == NULL || cluuid == NULL)
363 ofd = ofd_dev(obd->obd_lu_dev);
365 rc = class_connect(&conn, obd, cluuid);
369 exp = class_conn2export(&conn);
370 LASSERT(exp != NULL);
372 if (localdata != NULL) {
373 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
374 if (rc != 0 && rc != -EEXIST)
377 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
378 "nid is null\n", obd->obd_name, cluuid->uuid);
381 rc = ofd_parse_connect_data(env, exp, data, true);
385 if (obd->obd_replayable) {
386 struct tg_export_data *ted = &exp->exp_target_data;
388 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
389 sizeof(ted->ted_lcd->lcd_uuid));
390 rc = tgt_client_new(env, exp);
393 ofd_export_stats_init(ofd, exp, localdata);
396 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
397 data ? data->ocd_group : -1);
401 class_disconnect(exp);
402 nodemap_del_member(exp);
411 * Disconnect a connected client.
413 * This function terminates the client connection. The client export is
414 * disconnected (cleaned up) and client data on persistent storage is removed.
416 * \param[in] exp OBD export
418 * \retval 0 if successful
419 * \retval negative value on error
421 int ofd_obd_disconnect(struct obd_export *exp)
423 struct ofd_device *ofd = ofd_exp(exp);
430 class_export_get(exp);
432 if (!(exp->exp_flags & OBD_OPT_FORCE))
433 tgt_grant_sanity_check(ofd_obd(ofd), __func__);
435 rc = server_disconnect_export(exp);
437 tgt_grant_discard(exp);
439 /* Do not erase record for recoverable client. */
440 if (exp->exp_obd->obd_replayable &&
441 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
442 rc = lu_env_init(&env, LCT_DT_THREAD);
446 tgt_client_del(&env, exp);
450 nodemap_del_member(exp);
451 class_export_put(exp);
456 * Implementation of obd_ops::o_init_export.
458 * This function is called from class_new_export() and initializes
459 * the OFD-specific data for new export.
461 * \param[in] exp OBD export
463 * \retval 0 if successful
464 * \retval negative value on error
466 static int ofd_init_export(struct obd_export *exp)
470 spin_lock_init(&exp->exp_filter_data.fed_lock);
471 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
472 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
473 spin_lock(&exp->exp_lock);
474 exp->exp_connecting = 1;
475 spin_unlock(&exp->exp_lock);
477 /* self-export doesn't need client data and ldlm initialization */
478 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
479 &exp->exp_client_uuid)))
482 rc = tgt_client_alloc(exp);
484 ldlm_init_export(exp);
486 CERROR("%s: Can't initialize export: rc %d\n",
487 exp->exp_obd->obd_name, rc);
492 * Implementation of obd_ops::o_destroy_export.
494 * This function is called from class_export_destroy() to cleanup
495 * the OFD-specific data for export being destroyed.
497 * \param[in] exp OBD export
499 * \retval 0 if successful
500 * \retval negative value on error
502 static int ofd_destroy_export(struct obd_export *exp)
504 struct ofd_device *ofd = ofd_exp(exp);
506 if (exp->exp_target_data.ted_pending)
507 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
508 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
509 exp, exp->exp_target_data.ted_pending);
511 target_destroy_export(exp);
513 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
514 &exp->exp_client_uuid)))
517 ldlm_destroy_export(exp);
518 tgt_client_free(exp);
520 ofd_fmd_cleanup(exp);
523 * discard grants once we're sure no more
524 * interaction with the client is possible
526 tgt_grant_discard(exp);
527 ofd_fmd_cleanup(exp);
529 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
530 ofd->ofd_lut.lut_tgd.tgd_tot_granted_clients--;
532 if (!(exp->exp_flags & OBD_OPT_FORCE))
533 tgt_grant_sanity_check(exp->exp_obd, __func__);
535 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
540 * Notify all devices in server stack about recovery completion.
542 * This function calls ldo_recovery_complete() for all lower devices in the
543 * server stack so they will be prepared for normal operations.
545 * \param[in] env execution environment
546 * \param[in] ofd OFD device
548 * \retval 0 if successful
549 * \retval negative value on error
551 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
553 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
556 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
558 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
559 struct lfsck_start_param lsp;
561 lsp.lsp_start = NULL;
562 lsp.lsp_index_valid = 0;
563 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
564 if (rc != 0 && rc != -EALREADY)
565 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
569 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
573 * Implementation of obd_ops::o_postrecov.
575 * This function is called from target_finish_recovery() upon recovery
578 * \param[in] obd OBD device of OFD
580 * \retval 0 if successful
581 * \retval negative value on error
583 static int ofd_obd_postrecov(struct obd_device *obd)
586 struct lu_device *ldev = obd->obd_lu_dev;
591 rc = lu_env_init(&env, LCT_DT_THREAD);
594 ofd_info_init(&env, obd->obd_self_export);
596 rc = ofd_postrecov(&env, ofd_dev(ldev));
603 * Implementation of obd_ops::o_set_info_async.
605 * This function is not called from request handler, it is only used by
606 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
607 * \see ofd_set_info_hdl() for request handler function.
609 * \param[in] env execution environment
610 * \param[in] exp OBD export of OFD device
611 * \param[in] keylen length of \a key
612 * \param[in] key key name
613 * \param[in] vallen length of \a val
614 * \param[in] val the \a key value
615 * \param[in] set not used in OFD
617 * \retval 0 if successful
618 * \retval negative value on error
620 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
621 __u32 keylen, void *key, __u32 vallen, void *val,
622 struct ptlrpc_request_set *set)
628 if (exp->exp_obd == NULL) {
629 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
633 if (KEY_IS(KEY_SPTLRPC_CONF)) {
634 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
636 CERROR("%s: Unsupported key %s\n",
637 exp->exp_obd->obd_name, (char*)key);
644 * Implementation of obd_ops::o_get_info.
646 * This function is not called from request handler, it is only used by
647 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
649 * \see ofd_get_info_hdl() for request handler function.
651 * \param[in] env execution environment
652 * \param[in] exp OBD export of OFD device
653 * \param[in] keylen length of \a key
654 * \param[in] key key name
655 * \param[out] vallen length of key value
656 * \param[out] val the key value to return
658 * \retval 0 if successful
659 * \retval negative value on error
661 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
662 __u32 keylen, void *key, __u32 *vallen, void *val)
664 struct ofd_thread_info *info;
665 struct ofd_device *ofd;
666 struct ll_fiemap_info_key *fm_key = key;
667 struct fiemap *fiemap = val;
672 if (exp->exp_obd == NULL) {
673 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
679 if (KEY_IS(KEY_FIEMAP)) {
680 info = ofd_info_init(env, exp);
682 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
683 ofd->ofd_lut.lut_lsd.lsd_osd_index);
687 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
689 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
697 * Implementation of obd_ops::o_statfs.
699 * This function returns information about a storage file system.
700 * It is called from several places by using the OBD API as well as
701 * by direct call, e.g. from request handler.
703 * \see ofd_statfs_hdl() for request handler function.
705 * Report also the state of the OST to the caller in osfs->os_state
706 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
708 * \param[in] env execution environment
709 * \param[in] exp OBD export of OFD device
710 * \param[out] osfs statistic data to return
711 * \param[in] max_age maximum age for cached data
712 * \param[in] flags not used in OFD
714 * \retval 0 if successful
715 * \retval negative value on error
717 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
718 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
720 struct obd_device *obd = class_exp2obd(exp);
721 struct ofd_device *ofd = ofd_exp(exp);
722 struct tg_grants_data *tgd = &ofd->ofd_lut.lut_tgd;
727 rc = tgt_statfs_internal(env, &ofd->ofd_lut, osfs, max_age, NULL);
731 /* at least try to account for cached pages. its still racy and
732 * might be under-reporting if clients haven't announced their
733 * caches with brw recently */
735 CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
736 " pending %llu free %llu avail %llu\n",
737 tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
738 tgd->tgd_tot_pending,
739 osfs->os_bfree << tgd->tgd_blockbits,
740 osfs->os_bavail << tgd->tgd_blockbits);
742 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
743 ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
744 osfs->os_bsize - 1) >> tgd->tgd_blockbits));
746 /* The QoS code on the MDS does not care about space reserved for
747 * precreate, so take it out. */
748 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
749 struct tg_export_data *ted;
751 ted = &obd->obd_self_export->exp_target_data;
752 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
753 ted->ted_grant >> tgd->tgd_blockbits);
756 tgt_grant_sanity_check(obd, __func__);
757 CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
758 "%llu objects: %llu free; state %x\n",
759 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
760 osfs->os_files, osfs->os_ffree, osfs->os_state);
762 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
763 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
764 /* Reduce free inode count to zero, but keep "used" intact */
765 osfs->os_files -= osfs->os_ffree;
766 osfs->os_ffree -= osfs->os_ffree;
769 /* OS_STATE_READONLY can be set by OSD already */
770 if (ofd->ofd_raid_degraded)
771 osfs->os_state |= OS_STATE_DEGRADED;
773 if (obd->obd_self_export != exp && !exp_grant_param_supp(exp) &&
774 tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
775 /* clients which don't support OBD_CONNECT_GRANT_PARAM
776 * should not see a block size > page size, otherwise
777 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
778 * block size which is the biggest block size known to work
779 * with all client's page size. */
780 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
781 osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
782 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
783 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
786 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
787 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
788 /* Reduce free blocks count near zero, but keep "used" intact */
789 osfs->os_bavail -= osfs->os_bavail - 2;
790 osfs->os_blocks -= osfs->os_bfree - 2;
791 osfs->os_bfree -= osfs->os_bfree - 2;
800 * Implementation of obd_ops::o_setattr.
802 * This function is only used by ECHO client when it is run on top of OFD,
803 * \see ofd_setattr_hdl() for request handler function.
805 * \param[in] env execution environment
806 * \param[in] exp OBD export of OFD device
807 * \param[in] oa setattr parameters
809 * \retval 0 if successful
810 * \retval negative value on error
812 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
815 struct ofd_thread_info *info;
816 struct ofd_device *ofd = ofd_exp(exp);
817 struct ldlm_namespace *ns = ofd->ofd_namespace;
818 struct ldlm_resource *res;
819 struct ofd_object *fo;
820 struct lu_fid *fid = &oa->o_oi.oi_fid;
821 struct filter_fid *ff = NULL;
826 info = ofd_info_init(env, exp);
828 ost_fid_build_resid(fid, &info->fti_resid);
830 /* This would be very bad - accidentally truncating a file when
831 * changing the time or similar - bug 12203. */
832 if (oa->o_valid & OBD_MD_FLSIZE) {
833 static char mdsinum[48];
835 if (oa->o_valid & OBD_MD_FLFID)
836 snprintf(mdsinum, sizeof(mdsinum) - 1,
837 "of parent "DFID, oa->o_parent_seq,
838 oa->o_parent_oid, 0);
842 CERROR("%s: setattr from %s trying to truncate object "DFID
843 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
845 GOTO(out, rc = -EPERM);
848 fo = ofd_object_find_exists(env, ofd, fid);
850 CERROR("%s: can't find object "DFID"\n",
851 ofd_name(ofd), PFID(fid));
852 GOTO(out, rc = PTR_ERR(fo));
855 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
856 info->fti_attr.la_valid &= ~LA_TYPE;
858 if (oa->o_valid & OBD_MD_FLFID) {
859 ff = &info->fti_mds_fid;
860 ofd_prepare_fidea(ff, oa);
863 /* setting objects attributes (including owner/group) */
864 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
866 GOTO(out_unlock, rc);
868 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
871 ofd_object_put(env, fo);
874 /* we do not call this before to avoid lu_object_find() in
875 * ->lvbo_update() holding another reference on the object.
876 * otherwise concurrent destroy can make the object unavailable
877 * for 2nd lu_object_find() waiting for the first reference
878 * to go... deadlock! */
879 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
881 ldlm_res_lvbo_update(res, NULL, 0);
882 ldlm_resource_putref(res);
890 * Destroy OFD object by its FID.
892 * Supplemental function to destroy object by FID, it is used by request
893 * handler and by ofd_echo_destroy() below to find object by FID, lock it
894 * and call ofd_destroy() finally.
896 * \param[in] env execution environment
897 * \param[in] ofd OFD device
898 * \param[in] fid FID of object
899 * \param[in] orphan set if object being destroyed is an orphan
901 * \retval 0 if successful
902 * \retval negative value on error
904 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
905 const struct lu_fid *fid, int orphan)
907 struct ofd_thread_info *info = ofd_info(env);
908 struct lustre_handle lockh;
909 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
910 struct ofd_object *fo;
911 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
916 fo = ofd_object_find_exists(env, ofd, fid);
920 /* Tell the clients that the object is gone now and that they should
921 * throw away any cached pages. */
922 ost_fid_build_resid(fid, &info->fti_resid);
923 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
924 LDLM_EXTENT, &policy, LCK_PW, &flags,
925 ldlm_blocking_ast, ldlm_completion_ast,
926 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
928 /* We only care about the side-effects, just drop the lock. */
930 ldlm_lock_decref(&lockh, LCK_PW);
934 rc = ofd_destroy(env, fo, orphan);
937 ofd_object_put(env, fo);
942 * Implementation of obd_ops::o_destroy.
944 * This function is only used by ECHO client when it is run on top of OFD,
945 * \see ofd_destroy_hdl() for request handler function.
947 * \param[in] env execution environment
948 * \param[in] exp OBD export of OFD device
949 * \param[in] oa obdo structure with FID
951 * Note: this is OBD API method which is common API for server OBDs and
952 * client OBDs. Thus some parameters used in client OBDs may not be used
953 * on server OBDs and vice versa.
955 * \retval 0 if successful
956 * \retval negative value on error
958 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
961 struct ofd_device *ofd = ofd_exp(exp);
962 struct lu_fid *fid = &oa->o_oi.oi_fid;
967 ofd_info_init(env, exp);
969 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
973 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
975 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
977 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
978 ofd_name(ofd), PFID(fid));
980 } else if (rc != 0) {
981 CERROR("%s: error destroying object "DFID": %d\n",
982 ofd_name(ofd), PFID(fid), rc);
991 * Implementation of obd_ops::o_create.
993 * This function is only used by ECHO client when it is run on top of OFD
994 * and just creates an object.
995 * \see ofd_create_hdl() for request handler function.
997 * \param[in] env execution environment
998 * \param[in] exp OBD export of OFD device
999 * \param[in] oa obdo structure with FID sequence to use
1001 * Note: this is OBD API method which is common API for server OBDs and
1002 * client OBDs. Thus some parameters used in client OBDs may not be used
1003 * on server OBDs and vice versa.
1005 * \retval 0 if successful
1006 * \retval negative value on error
1008 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1011 struct ofd_device *ofd = ofd_exp(exp);
1012 u64 seq = ostid_seq(&oa->o_oi);
1013 struct ofd_seq *oseq;
1022 ofd_info_init(env, exp);
1024 LASSERT(seq == FID_SEQ_ECHO);
1025 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1027 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1029 down_read(&ofd->ofd_lastid_rwsem);
1030 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1031 * we may do that in the future.
1032 * Return -ENOSPC until the LAST_ID rebuilt. */
1033 if (unlikely(ofd->ofd_lastid_rebuilding))
1034 GOTO(out_sem, rc = -ENOSPC);
1036 rc = ofd_validate_seq(exp, seq);
1040 oseq = ofd_seq_load(env, ofd, seq);
1042 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1043 ofd_name(ofd), seq, PTR_ERR(oseq));
1044 GOTO(out_sem, rc = -EINVAL);
1047 mutex_lock(&oseq->os_create_lock);
1048 granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1052 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1053 "precreate (%lld): rc = %d\n", ofd_name(ofd), diff, rc);
1058 next_id = ofd_seq_last_oid(oseq) + 1;
1059 count = ofd_precreate_batch(ofd, (int)diff);
1061 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1063 CERROR("%s: unable to precreate: rc = %d\n",
1066 rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1068 CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n",
1070 (unsigned long long)ofd_seq_last_oid(oseq),
1071 POSTID(&oa->o_oi), rc);
1073 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1076 tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1078 mutex_unlock(&oseq->os_create_lock);
1079 ofd_seq_put(env, oseq);
1082 up_read(&ofd->ofd_lastid_rwsem);
1087 * Implementation of obd_ops::o_getattr.
1089 * This function is only used by ECHO client when it is run on top of OFD
1090 * and returns attributes of object.
1091 * \see ofd_getattr_hdl() for request handler function.
1093 * \param[in] env execution environment
1094 * \param[in] exp OBD export of OFD device
1095 * \param[in,out] oa contains FID of object to get attributes from and
1096 * is used to return attributes back
1098 * \retval 0 if successful
1099 * \retval negative value on error
1101 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1104 struct ofd_device *ofd = ofd_exp(exp);
1105 struct ofd_thread_info *info;
1106 struct lu_fid *fid = &oa->o_oi.oi_fid;
1107 struct ofd_object *fo;
1112 info = ofd_info_init(env, exp);
1114 fo = ofd_object_find_exists(env, ofd, fid);
1116 GOTO(out, rc = PTR_ERR(fo));
1118 LASSERT(fo != NULL);
1119 rc = ofd_attr_get(env, fo, &info->fti_attr);
1120 oa->o_valid = OBD_MD_FLID;
1124 obdo_from_la(oa, &info->fti_attr,
1125 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1127 /* Store object version in reply */
1128 curr_version = dt_version_get(env, ofd_object_child(fo));
1129 if ((__s64)curr_version != -EOPNOTSUPP) {
1130 oa->o_valid |= OBD_MD_FLDATAVERSION;
1131 oa->o_data_version = curr_version;
1135 ofd_object_put(env, fo);
1141 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1143 * This is supplemental function for ofd_iocontrol() to return object
1144 * version for lctl tool.
1146 * \param[in] env execution environment
1147 * \param[in] ofd OFD device
1148 * \param[out] karg ioctl data
1150 * \retval 0 if successful
1151 * \retval negative value on error
1153 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1154 struct ofd_device *ofd, void *karg)
1156 struct obd_ioctl_data *data = karg;
1158 struct ofd_object *fo;
1159 dt_obj_version_t version;
1164 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1165 GOTO(out, rc = -EINVAL);
1167 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1168 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1169 } else if (data->ioc_inlbuf3 != NULL &&
1170 data->ioc_inllen3 == sizeof(__u64) &&
1171 data->ioc_inlbuf4 != NULL &&
1172 data->ioc_inllen4 == sizeof(__u64)) {
1173 struct ost_id ostid;
1175 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1176 rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1179 rc = ostid_to_fid(&fid, &ostid,
1180 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1184 GOTO(out, rc = -EINVAL);
1187 if (!fid_is_sane(&fid))
1188 GOTO(out, rc = -EINVAL);
1190 fo = ofd_object_find(env, ofd, &fid);
1192 GOTO(out, rc = PTR_ERR(fo));
1194 if (!ofd_object_exists(fo))
1195 GOTO(out_fo, rc = -ENOENT);
1197 if (lu_object_remote(&fo->ofo_obj.do_lu))
1198 GOTO(out_fo, rc = -EREMOTE);
1200 version = dt_version_get(env, ofd_object_child(fo));
1202 GOTO(out_fo, rc = -EIO);
1204 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1208 ofd_object_put(env, fo);
1214 * Implementation of obd_ops::o_iocontrol.
1216 * This is OFD ioctl handling function which is primary interface for
1217 * Lustre tools like lfs, lctl and lfsck.
1219 * \param[in] cmd ioctl command
1220 * \param[in] exp OBD export of OFD
1221 * \param[in] len not used
1222 * \param[in,out] karg buffer with data
1223 * \param[in] uarg not used
1225 * \retval 0 if successful
1226 * \retval negative value on error
1228 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1229 void *karg, void __user *uarg)
1232 struct ofd_device *ofd = ofd_exp(exp);
1233 struct obd_device *obd = ofd_obd(ofd);
1238 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1239 rc = lu_env_init(&env, LCT_DT_THREAD);
1244 case OBD_IOC_ABORT_RECOVERY:
1245 CERROR("%s: aborting recovery\n", obd->obd_name);
1246 obd->obd_abort_recovery = 1;
1247 target_stop_recovery_thread(obd);
1250 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1251 rc = dt_sync(&env, ofd->ofd_osd);
1253 case OBD_IOC_SET_READONLY:
1254 rc = dt_sync(&env, ofd->ofd_osd);
1256 rc = dt_ro(&env, ofd->ofd_osd);
1258 case OBD_IOC_START_LFSCK: {
1259 struct obd_ioctl_data *data = karg;
1260 struct lfsck_start_param lsp;
1262 if (unlikely(data == NULL)) {
1267 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1268 lsp.lsp_index_valid = 0;
1269 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1272 case OBD_IOC_STOP_LFSCK: {
1273 struct obd_ioctl_data *data = karg;
1274 struct lfsck_stop stop;
1276 stop.ls_status = LS_STOPPED;
1277 /* Old lfsck utils may pass NULL @stop. */
1278 if (data->ioc_inlbuf1 == NULL)
1282 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1284 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1287 case OBD_IOC_GET_OBJ_VERSION:
1288 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1291 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1300 * Implementation of obd_ops::o_precleanup.
1302 * This function stops device activity before shutting it down. It is called
1303 * from a cleanup function upon forceful device cleanup. For OFD there are no
1304 * special actions, it just invokes target_recovery_cleanup().
1306 * \param[in] obd OBD device of OFD
1310 static int ofd_precleanup(struct obd_device *obd)
1313 target_cleanup_recovery(obd);
1318 * Implementation of obd_ops::o_ping.
1320 * This is OFD-specific part of OBD_PING request handling.
1321 * It controls Filter Modification Data (FMD) expiration each time PING is
1324 * \see ofd_fmd_expire() and ofd_fmd.c for details
1326 * \param[in] env execution environment
1327 * \param[in] exp OBD export of client
1331 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1333 ofd_fmd_expire(exp);
1338 * Implementation of obd_ops::o_health_check.
1340 * This function checks the OFD device health - ability to respond on
1341 * incoming requests. There are two health_check methods:
1342 * - get statfs from the OSD. It checks just responsiveness of
1344 * - do write attempt on bottom device to check it is fully operational and
1345 * is not stuck. This is expensive method and requires special configuration
1346 * option --enable-health-write while building Lustre, it is turned off
1349 * \param[in] nul not used
1350 * \param[in] obd OBD device of OFD
1352 * \retval 0 if successful
1353 * \retval negative value in case of error
1355 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1357 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1358 struct ofd_thread_info *info;
1360 #ifdef USE_HEALTH_CHECK_WRITE
1365 /* obd_proc_read_health pass NULL env, we need real one */
1366 rc = lu_env_init(&env, LCT_DT_THREAD);
1370 info = ofd_info_init(&env, NULL);
1371 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1375 if (info->fti_u.osfs.os_state & OS_STATE_READONLY)
1376 GOTO(out, rc = -EROFS);
1378 #ifdef USE_HEALTH_CHECK_WRITE
1379 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1380 if (info->fti_buf.lb_buf == NULL)
1381 GOTO(out, rc = -ENOMEM);
1383 info->fti_buf.lb_len = PAGE_SIZE;
1386 th = dt_trans_create(&env, ofd->ofd_osd);
1388 GOTO(out, rc = PTR_ERR(th));
1390 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1391 &info->fti_buf, info->fti_off, th);
1393 th->th_sync = 1; /* sync IO is needed */
1394 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1396 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1397 &info->fti_buf, &info->fti_off,
1400 dt_trans_stop(&env, ofd->ofd_osd, th);
1402 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1404 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1411 struct obd_ops ofd_obd_ops = {
1412 .o_owner = THIS_MODULE,
1413 .o_connect = ofd_obd_connect,
1414 .o_reconnect = ofd_obd_reconnect,
1415 .o_disconnect = ofd_obd_disconnect,
1416 .o_create = ofd_echo_create,
1417 .o_statfs = ofd_statfs,
1418 .o_setattr = ofd_echo_setattr,
1419 .o_preprw = ofd_preprw,
1420 .o_commitrw = ofd_commitrw,
1421 .o_destroy = ofd_echo_destroy,
1422 .o_init_export = ofd_init_export,
1423 .o_destroy_export = ofd_destroy_export,
1424 .o_postrecov = ofd_obd_postrecov,
1425 .o_getattr = ofd_echo_getattr,
1426 .o_iocontrol = ofd_iocontrol,
1427 .o_precleanup = ofd_precleanup,
1429 .o_health_check = ofd_health_check,
1430 .o_set_info_async = ofd_set_info_async,
1431 .o_get_info = ofd_get_info,