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, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/ofd/ofd_obd.c
33 * This file contains OBD API methods for OBD Filter Device (OFD) which are
34 * used for export handling, configuration purposes and recovery.
35 * Several methods are used by ECHO client only since it still uses OBD API.
36 * Such methods have _echo_ prefix in name.
38 * Author: Andreas Dilger <andreas.dilger@intel.com>
39 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
40 * Author: Mikhail Pershin <mike.pershin@intel.com>
43 #define DEBUG_SUBSYSTEM S_FILTER
45 #include "ofd_internal.h"
46 #include <obd_cksum.h>
47 #include <uapi/linux/lustre/lustre_ioctl.h>
48 #include <lustre_quota.h>
49 #include <lustre_lfsck.h>
50 #include <lustre_nodemap.h>
53 * Initialize OFD per-export statistics.
55 * This function sets up procfs entries for various OFD export counters. These
56 * counters are for per-client statistics tracked on the server.
58 * \param[in] ofd OFD device
59 * \param[in] exp OBD export
60 * \param[in] client_nid NID of client
62 * \retval 0 if successful
63 * \retval negative value on error
65 static int ofd_export_stats_init(struct ofd_device *ofd,
66 struct obd_export *exp,
67 lnet_nid_t *client_nid4)
69 struct obd_device *obd = ofd_obd(ofd);
70 struct nid_stat *stats;
75 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
76 /* Self-export gets no proc entry */
80 struct lnet_nid client_nid;
82 lnet_nid4_to_nid(*client_nid4, &client_nid);
83 rc = lprocfs_exp_setup(exp, &client_nid);
85 rc = lprocfs_exp_setup(exp, NULL);
87 /* Mask error for already created /proc entries */
88 RETURN(rc == -EALREADY ? 0 : rc);
90 stats = exp->exp_nid_stats;
91 stats->nid_stats = lprocfs_stats_alloc(LPROC_OFD_STATS_LAST,
92 LPROCFS_STATS_FLAG_NOPERCPU);
93 if (!stats->nid_stats)
96 ofd_stats_counter_init(stats->nid_stats, 0, LPROCFS_CNTR_HISTOGRAM);
98 rc = lprocfs_stats_register(stats->nid_proc, "stats", stats->nid_stats);
100 lprocfs_stats_free(&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;
153 "%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_group %u\n",
154 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
155 data->ocd_connect_flags, data->ocd_version,
156 data->ocd_grant, data->ocd_index, data->ocd_group);
158 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
159 CWARN("!!! This export (nid %s) used object group %d earlier; now it's trying to use group %d! This could be a bug in the MDS. Please report to https://jira.whamcloud.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;
168 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
169 data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2;
171 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
172 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
173 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
174 exp->exp_obd->obd_name, data->ocd_group);
175 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
178 /* Determine optimal brw size before calculating grant */
179 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
180 data->ocd_brw_size = 65536;
181 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
182 if (data->ocd_brw_size > ofd->ofd_brw_size)
183 data->ocd_brw_size = ofd->ofd_brw_size;
184 if (data->ocd_brw_size == 0) {
185 CERROR("%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_brw_size is unexpectedly zero, network data corruption? Refusing connection of this client\n",
186 exp->exp_obd->obd_name,
187 exp->exp_client_uuid.uuid,
188 exp, data->ocd_connect_flags, data->ocd_version,
189 data->ocd_grant, data->ocd_index);
194 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
195 struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf;
197 /* client is reporting its page size, for future use */
198 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
199 data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits;
201 * ddp_inodespace may not be power-of-two value, eg. for ldiskfs
202 * it's LDISKFS_DIR_REC_LEN(20) = 28.
204 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
205 /* ocd_grant_tax_kb is in 1K byte blocks */
206 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
207 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
211 * Save connect_data we have so far because tgt_grant_connect()
212 * uses it to calculate grant, and we want to save the client
213 * version before it is overwritten by LUSTRE_VERSION_CODE.
215 exp->exp_connect_data = *data;
216 if (OCD_HAS_FLAG(data, GRANT))
217 tgt_grant_connect(env, exp, data, new_connection);
219 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
220 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
221 int index = lsd->lsd_osd_index;
223 if (index != data->ocd_index) {
224 LCONSOLE_ERROR_MSG(0x136,
225 "Connection from %s to index %u doesn't match actual OST index %u in last_rcvd file, bad configuration?\n",
226 obd_export_nid2str(exp), index,
230 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
231 /* this will only happen on the first connect */
232 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
234 * sync is not needed here as tgt_client_new will
235 * set exp_need_sync flag
237 tgt_server_data_update(env, &ofd->ofd_lut, 0);
241 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
242 __u32 cksum_types = data->ocd_cksum_types;
244 tgt_mask_cksum_types(&ofd->ofd_lut, &data->ocd_cksum_types);
246 if (unlikely(data->ocd_cksum_types == 0)) {
247 CERROR("%s: Connect with checksum support but no ocd_cksum_types is set\n",
248 exp->exp_obd->obd_name);
253 "%s: cli %s supports cksum type %x, return %x\n",
254 exp->exp_obd->obd_name, obd_export_nid2str(exp),
255 cksum_types, data->ocd_cksum_types);
258 * This client does not support OBD_CONNECT_CKSUM.
259 * Report failure to negotiate checksum at connect
262 "%s: cli %s does not support OBD_CONNECT_CKSUM\n",
263 exp->exp_obd->obd_name, obd_export_nid2str(exp));
266 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
267 data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes;
269 data->ocd_version = LUSTRE_VERSION_CODE;
271 if (OCD_HAS_FLAG(data, PINGLESS)) {
272 if (ptlrpc_pinger_suppress_pings()) {
273 spin_lock(&exp->exp_obd->obd_dev_lock);
274 list_del_init(&exp->exp_obd_chain_timed);
275 spin_unlock(&exp->exp_obd->obd_dev_lock);
277 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
281 if (!ofd->ofd_lut.lut_dt_conf.ddp_has_lseek_data_hole)
282 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
288 * Re-initialize export upon client reconnection.
290 * This function parses connection data from reconnect and resets
293 * \param[in] env execution environment
294 * \param[in] exp OBD export
295 * \param[in] obd OFD device
296 * \param[in] cluuid NID of client
297 * \param[in] data connection data from request
298 * \param[in] localdata client NID
300 * \retval 0 if successful
301 * \retval negative value on error
303 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
304 struct obd_device *obd, struct obd_uuid *cluuid,
305 struct obd_connect_data *data,
308 struct ofd_device *ofd;
313 if (!exp || !obd || !cluuid)
316 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
317 if (rc != 0 && rc != -EEXIST)
320 ofd = ofd_dev(obd->obd_lu_dev);
322 rc = ofd_parse_connect_data(env, exp, data, false);
324 ofd_export_stats_init(ofd, exp, client_nid);
326 nodemap_del_member(exp);
332 * Initialize new client connection.
334 * This function handles new connection to the OFD. The new export is
335 * created (in context of class_connect()) and persistent client data is
336 * initialized on storage.
338 * \param[in] env execution environment
339 * \param[out] _exp stores pointer to new export
340 * \param[in] obd OFD device
341 * \param[in] cluuid client UUID
342 * \param[in] data connection data from request
343 * \param[in] localdata client NID
345 * \retval 0 if successful
346 * \retval negative value on error
348 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
349 struct obd_device *obd, struct obd_uuid *cluuid,
350 struct obd_connect_data *data, void *localdata)
352 struct obd_export *exp;
353 struct ofd_device *ofd;
354 struct lustre_handle conn = { 0 };
359 if (!_exp || !obd || !cluuid)
362 ofd = ofd_dev(obd->obd_lu_dev);
364 rc = class_connect(&conn, obd, cluuid);
368 exp = class_conn2export(&conn);
369 LASSERT(exp != NULL);
372 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
373 if (rc != 0 && rc != -EEXIST)
377 "%s: cannot find nodemap for client %s: nid is null\n",
378 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 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
471 spin_lock(&exp->exp_lock);
472 exp->exp_connecting = 1;
473 spin_unlock(&exp->exp_lock);
475 /* self-export doesn't need client data and ldlm initialization */
476 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
477 &exp->exp_client_uuid)))
480 rc = tgt_client_alloc(exp);
482 ldlm_init_export(exp);
484 CERROR("%s: Can't initialize export: rc %d\n",
485 exp->exp_obd->obd_name, rc);
490 * Implementation of obd_ops::o_destroy_export.
492 * This function is called from class_export_destroy() to cleanup
493 * the OFD-specific data for export being destroyed.
495 * \param[in] exp OBD export
497 * \retval 0 if successful
498 * \retval negative value on error
500 static int ofd_destroy_export(struct obd_export *exp)
502 struct ofd_device *ofd = ofd_exp(exp);
504 if (exp->exp_target_data.ted_pending)
505 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
506 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
507 exp, exp->exp_target_data.ted_pending);
509 target_destroy_export(exp);
511 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
512 &exp->exp_client_uuid)))
515 ldlm_destroy_export(exp);
516 tgt_client_free(exp);
519 * discard grants once we're sure no more
520 * interaction with the client is possible
522 tgt_grant_discard(exp);
524 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
525 ofd->ofd_lut.lut_tgd.tgd_tot_granted_clients--;
527 if (!(exp->exp_flags & OBD_OPT_FORCE))
528 tgt_grant_sanity_check(exp->exp_obd, __func__);
534 * Notify all devices in server stack about recovery completion.
536 * This function calls ldo_recovery_complete() for all lower devices in the
537 * server stack so they will be prepared for normal operations.
539 * \param[in] env execution environment
540 * \param[in] ofd OFD device
542 * \retval 0 if successful
543 * \retval negative value on error
545 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
547 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
550 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
552 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
553 struct lfsck_start_param lsp;
555 lsp.lsp_start = NULL;
556 lsp.lsp_index_valid = 0;
557 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
558 if (rc != 0 && rc != -EALREADY)
559 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
563 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
567 * Implementation of obd_ops::o_postrecov.
569 * This function is called from target_finish_recovery() upon recovery
572 * \param[in] obd OBD device of OFD
574 * \retval 0 if successful
575 * \retval negative value on error
577 static int ofd_obd_postrecov(struct obd_device *obd)
580 struct lu_device *ldev = obd->obd_lu_dev;
585 rc = lu_env_init(&env, LCT_DT_THREAD);
588 ofd_info_init(&env, obd->obd_self_export);
590 rc = ofd_postrecov(&env, ofd_dev(ldev));
597 * Implementation of obd_ops::o_set_info_async.
599 * This function is not called from request handler, it is only used by
600 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
601 * \see ofd_set_info_hdl() for request handler function.
603 * \param[in] env execution environment
604 * \param[in] exp OBD export of OFD device
605 * \param[in] keylen length of \a key
606 * \param[in] key key name
607 * \param[in] vallen length of \a val
608 * \param[in] val the \a key value
609 * \param[in] set not used in OFD
611 * \retval 0 if successful
612 * \retval negative value on error
614 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
615 __u32 keylen, void *key, __u32 vallen, void *val,
616 struct ptlrpc_request_set *set)
623 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
627 if (KEY_IS(KEY_SPTLRPC_CONF)) {
628 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
630 CERROR("%s: Unsupported key %s\n",
631 exp->exp_obd->obd_name, (char *)key);
638 * Implementation of obd_ops::o_get_info.
640 * This function is not called from request handler, it is only used by
641 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
643 * \see ofd_get_info_hdl() for request handler function.
645 * \param[in] env execution environment
646 * \param[in] exp OBD export of OFD device
647 * \param[in] keylen length of \a key
648 * \param[in] key key name
649 * \param[out] vallen length of key value
650 * \param[out] val the key value to return
652 * \retval 0 if successful
653 * \retval negative value on error
655 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
656 __u32 keylen, void *key, __u32 *vallen, void *val)
658 struct ofd_thread_info *info;
659 struct ofd_device *ofd;
660 struct ll_fiemap_info_key *fm_key = key;
661 struct fiemap *fiemap = val;
667 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
673 if (KEY_IS(KEY_FIEMAP)) {
674 info = ofd_info_init(env, exp);
676 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
677 ofd->ofd_lut.lut_lsd.lsd_osd_index);
681 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
683 CERROR("%s: not supported key %s\n",
684 ofd_name(ofd), (char *)key);
692 * Implementation of obd_ops::o_statfs.
694 * This function returns information about a storage file system.
695 * It is called from several places by using the OBD API as well as
696 * by direct call, e.g. from request handler.
698 * \see ofd_statfs_hdl() for request handler function.
700 * Report also the state of the OST to the caller in osfs->os_state
701 * (OS_STATFS_READONLY, OS_STATFS_DEGRADED).
703 * \param[in] env execution environment
704 * \param[in] exp OBD export of OFD device
705 * \param[out] osfs statistic data to return
706 * \param[in] max_age maximum age for cached data
707 * \param[in] flags not used in OFD
709 * \retval 0 if successful
710 * \retval negative value on error
712 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
713 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
715 struct obd_device *obd = class_exp2obd(exp);
716 struct ofd_device *ofd = ofd_exp(exp);
717 struct tg_grants_data *tgd = &ofd->ofd_lut.lut_tgd;
718 int current_blockbits;
723 rc = tgt_statfs_internal(env, &ofd->ofd_lut, osfs, max_age, NULL);
727 /* tgd_blockbit is recordsize bits set during mkfs.
728 * This once set does not change. However, 'zfs set'
729 * can be used to change the OST blocksize. Instead
730 * of using cached value of 'tgd_blockbit' always
731 * calculate the blocksize bits which may have
734 current_blockbits = fls64(osfs->os_bsize) - 1;
737 * at least try to account for cached pages. its still racy and
738 * might be under-reporting if clients haven't announced their
739 * caches with brw recently
741 CDEBUG(D_SUPER | D_CACHE,
742 "blocks cached %llu granted %llu pending %llu free %llu avail %llu\n",
743 tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
744 tgd->tgd_tot_pending,
745 osfs->os_bfree << current_blockbits,
746 osfs->os_bavail << current_blockbits);
748 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
749 ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
750 osfs->os_bsize - 1) >> current_blockbits));
753 * The QoS code on the MDS does not care about space reserved for
754 * precreate, so take it out.
756 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
757 struct tg_export_data *ted;
759 ted = &obd->obd_self_export->exp_target_data;
760 osfs->os_granted = min_t(u64, osfs->os_bavail,
761 ted->ted_grant >> current_blockbits);
762 osfs->os_bavail -= osfs->os_granted;
765 tgt_grant_sanity_check(obd, __func__);
766 CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
767 "%llu objects: %llu free; state %x\n",
768 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
769 osfs->os_files, osfs->os_ffree, osfs->os_state);
771 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
772 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
773 /* Reduce free inode count to zero, but keep "used" intact */
774 osfs->os_files -= osfs->os_ffree;
775 osfs->os_ffree -= osfs->os_ffree;
778 /* OS_STATFS_READONLY can be set by OSD already */
779 if (ofd->ofd_raid_degraded)
780 osfs->os_state |= OS_STATFS_DEGRADED;
782 if (ofd->ofd_no_precreate)
783 osfs->os_state |= OS_STATFS_NOPRECREATE;
785 if (obd->obd_self_export != exp && !exp_grant_param_supp(exp) &&
786 current_blockbits > COMPAT_BSIZE_SHIFT) {
788 * clients which don't support OBD_CONNECT_GRANT_PARAM
789 * should not see a block size > page size, otherwise
790 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
791 * block size which is the biggest block size known to work
792 * with all client's page size.
794 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
795 osfs->os_bfree <<= current_blockbits - COMPAT_BSIZE_SHIFT;
796 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
797 osfs->os_granted <<= current_blockbits - COMPAT_BSIZE_SHIFT;
798 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
801 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
802 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
803 /* Reduce free blocks count near zero, but keep "used" intact */
804 osfs->os_bavail -= osfs->os_bavail - 2;
805 osfs->os_blocks -= osfs->os_bfree - 2;
806 osfs->os_bfree -= osfs->os_bfree - 2;
815 * Implementation of obd_ops::o_setattr.
817 * This function is only used by ECHO client when it is run on top of OFD,
818 * \see ofd_setattr_hdl() for request handler function.
820 * \param[in] env execution environment
821 * \param[in] exp OBD export of OFD device
822 * \param[in] oa setattr parameters
824 * \retval 0 if successful
825 * \retval negative value on error
827 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
830 struct ofd_thread_info *info;
831 struct ofd_device *ofd = ofd_exp(exp);
832 struct ldlm_namespace *ns = ofd->ofd_namespace;
833 struct ldlm_resource *res;
834 struct ofd_object *fo;
835 struct lu_fid *fid = &oa->o_oi.oi_fid;
836 ktime_t kstart = ktime_get();
841 info = ofd_info_init(env, exp);
843 ost_fid_build_resid(fid, &info->fti_resid);
846 * This would be very bad - accidentally truncating a file when
847 * changing the time or similar - bug 12203.
849 if (oa->o_valid & OBD_MD_FLSIZE) {
850 static char mdsinum[48];
852 if (oa->o_valid & OBD_MD_FLFID)
853 snprintf(mdsinum, sizeof(mdsinum) - 1,
854 "of parent "DFID, oa->o_parent_seq,
855 oa->o_parent_oid, 0);
859 CERROR("%s: setattr from %s trying to truncate object "DFID
860 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
862 GOTO(out, rc = -EPERM);
865 fo = ofd_object_find_exists(env, ofd, fid);
867 CERROR("%s: can't find object "DFID"\n",
868 ofd_name(ofd), PFID(fid));
869 GOTO(out, rc = PTR_ERR(fo));
872 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
873 info->fti_attr.la_valid &= ~LA_TYPE;
875 /* setting objects attributes (including owner/group) */
876 rc = ofd_attr_set(env, fo, &info->fti_attr, oa);
878 GOTO(out_unlock, rc);
880 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL,
881 ktime_us_delta(ktime_get(), kstart));
884 ofd_object_put(env, fo);
888 * we do not call this before to avoid lu_object_find() in
889 * ->lvbo_update() holding another reference on the object.
890 * otherwise concurrent destroy can make the object unavailable
891 * for 2nd lu_object_find() waiting for the first reference
894 res = ldlm_resource_get(ns, &info->fti_resid, LDLM_EXTENT, 0);
896 ldlm_res_lvbo_update(res, NULL, 0);
897 ldlm_resource_putref(res);
905 * Destroy OFD object by its FID.
907 * Supplemental function to destroy object by FID, it is used by request
908 * handler and by ofd_echo_destroy() below to find object by FID, lock it
909 * and call ofd_destroy() finally.
911 * \param[in] env execution environment
912 * \param[in] ofd OFD device
913 * \param[in] fid FID of object
914 * \param[in] orphan set if object being destroyed is an orphan
916 * \retval 0 if successful
917 * \retval negative value on error
919 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
920 const struct lu_fid *fid, int orphan)
922 struct ofd_thread_info *info = ofd_info(env);
923 struct lustre_handle lockh;
924 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
925 struct ofd_object *fo;
926 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
931 fo = ofd_object_find_exists(env, ofd, fid);
936 * Tell the clients that the object is gone now and that they should
937 * throw away any cached pages.
939 ost_fid_build_resid(fid, &info->fti_resid);
940 rc = ldlm_cli_enqueue_local(env, ofd->ofd_namespace, &info->fti_resid,
941 LDLM_EXTENT, &policy, LCK_PW, &flags,
942 ldlm_blocking_ast, ldlm_completion_ast,
943 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
945 /* We only care about the side-effects, just drop the lock. */
947 ldlm_lock_decref(&lockh, LCK_PW);
951 rc = ofd_destroy(env, fo, orphan);
954 ofd_object_put(env, fo);
959 * Implementation of obd_ops::o_destroy.
961 * This function is only used by ECHO client when it is run on top of OFD,
962 * \see ofd_destroy_hdl() for request handler function.
964 * \param[in] env execution environment
965 * \param[in] exp OBD export of OFD device
966 * \param[in] oa obdo structure with FID
968 * Note: this is OBD API method which is common API for server OBDs and
969 * client OBDs. Thus some parameters used in client OBDs may not be used
970 * on server OBDs and vice versa.
972 * \retval 0 if successful
973 * \retval negative value on error
975 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
978 struct ofd_device *ofd = ofd_exp(exp);
979 struct lu_fid *fid = &oa->o_oi.oi_fid;
984 ofd_info_init(env, exp);
986 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
990 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
992 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
994 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
995 ofd_name(ofd), PFID(fid));
997 } else if (rc != 0) {
998 CERROR("%s: error destroying object "DFID": %d\n",
999 ofd_name(ofd), PFID(fid), rc);
1008 * Implementation of obd_ops::o_create.
1010 * This function is only used by ECHO client when it is run on top of OFD
1011 * and just creates an object.
1012 * \see ofd_create_hdl() for request handler function.
1014 * \param[in] env execution environment
1015 * \param[in] exp OBD export of OFD device
1016 * \param[in] oa obdo structure with FID sequence to use
1018 * Note: this is OBD API method which is common API for server OBDs and
1019 * client OBDs. Thus some parameters used in client OBDs may not be used
1020 * on server OBDs and vice versa.
1022 * \retval 0 if successful
1023 * \retval negative value on error
1025 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1028 struct ofd_device *ofd = ofd_exp(exp);
1029 u64 seq = ostid_seq(&oa->o_oi);
1030 struct ofd_seq *oseq;
1039 if (ofd->ofd_no_precreate)
1042 ofd_info_init(env, exp);
1044 LASSERT(seq == FID_SEQ_ECHO);
1045 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1047 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1049 down_read(&ofd->ofd_lastid_rwsem);
1051 * Currently, for safe, we do not distinguish which LAST_ID is broken,
1052 * we may do that in the future.
1053 * Return -ENOSPC until the LAST_ID rebuilt.
1055 if (unlikely(ofd->ofd_lastid_rebuilding))
1056 GOTO(out_sem, rc = -ENOSPC);
1058 rc = ofd_validate_seq(exp, seq);
1062 oseq = ofd_seq_load(env, ofd, seq);
1064 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1065 ofd_name(ofd), seq, PTR_ERR(oseq));
1066 GOTO(out_sem, rc = -EINVAL);
1069 mutex_lock(&oseq->os_create_lock);
1070 granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1075 "%s: failed to acquire grant space for precreate (%lld): rc = %d\n",
1076 ofd_name(ofd), diff, rc);
1081 next_id = ofd_seq_last_oid(oseq) + 1;
1082 count = ofd_precreate_batch(ofd, (int)diff);
1084 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1086 CERROR("%s: unable to precreate: rc = %d\n",
1089 rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1091 CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n",
1093 (unsigned long long)ofd_seq_last_oid(oseq),
1094 POSTID(&oa->o_oi), rc);
1096 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1099 tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1101 mutex_unlock(&oseq->os_create_lock);
1102 ofd_seq_put(env, oseq);
1105 up_read(&ofd->ofd_lastid_rwsem);
1110 * Implementation of obd_ops::o_getattr.
1112 * This function is only used by ECHO client when it is run on top of OFD
1113 * and returns attributes of object.
1114 * \see ofd_getattr_hdl() for request handler function.
1116 * \param[in] env execution environment
1117 * \param[in] exp OBD export of OFD device
1118 * \param[in,out] oa contains FID of object to get attributes from and
1119 * is used to return attributes back
1121 * \retval 0 if successful
1122 * \retval negative value on error
1124 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1127 struct ofd_device *ofd = ofd_exp(exp);
1128 struct ofd_thread_info *info;
1129 struct lu_fid *fid = &oa->o_oi.oi_fid;
1130 struct ofd_object *fo;
1135 info = ofd_info_init(env, exp);
1137 fo = ofd_object_find_exists(env, ofd, fid);
1139 GOTO(out, rc = PTR_ERR(fo));
1141 LASSERT(fo != NULL);
1142 rc = ofd_attr_get(env, fo, &info->fti_attr);
1143 oa->o_valid = OBD_MD_FLID;
1147 obdo_from_la(oa, &info->fti_attr,
1148 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1150 /* Store object version in reply */
1151 curr_version = dt_version_get(env, ofd_object_child(fo));
1152 if ((__s64)curr_version != -EOPNOTSUPP) {
1153 oa->o_valid |= OBD_MD_FLDATAVERSION;
1154 oa->o_data_version = curr_version;
1158 ofd_object_put(env, fo);
1164 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1166 * This is supplemental function for ofd_iocontrol() to return object
1167 * version for lctl tool.
1169 * \param[in] env execution environment
1170 * \param[in] ofd OFD device
1171 * \param[out] karg ioctl data
1173 * \retval 0 if successful
1174 * \retval negative value on error
1176 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1177 struct ofd_device *ofd, void *karg)
1179 struct obd_ioctl_data *data = karg;
1181 struct ofd_object *fo;
1182 dt_obj_version_t version;
1187 if (!data->ioc_inlbuf2 || data->ioc_inllen2 != sizeof(version))
1188 GOTO(out, rc = -EINVAL);
1190 if (data->ioc_inlbuf1 && data->ioc_inllen1 == sizeof(fid)) {
1191 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1192 } else if (data->ioc_inlbuf3 &&
1193 data->ioc_inllen3 == sizeof(__u64) &&
1194 data->ioc_inlbuf4 &&
1195 data->ioc_inllen4 == sizeof(__u64)) {
1196 struct ost_id ostid = { };
1198 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1199 rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1202 rc = ostid_to_fid(&fid, &ostid,
1203 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1207 GOTO(out, rc = -EINVAL);
1210 if (!fid_is_sane(&fid))
1211 GOTO(out, rc = -EINVAL);
1213 fo = ofd_object_find(env, ofd, &fid);
1215 GOTO(out, rc = PTR_ERR(fo));
1217 if (!ofd_object_exists(fo))
1218 GOTO(out_fo, rc = -ENOENT);
1220 if (lu_object_remote(&fo->ofo_obj.do_lu))
1221 GOTO(out_fo, rc = -EREMOTE);
1223 version = dt_version_get(env, ofd_object_child(fo));
1225 GOTO(out_fo, rc = -EIO);
1227 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1231 ofd_object_put(env, fo);
1237 * Implementation of obd_ops::o_iocontrol.
1239 * This is OFD ioctl handling function which is primary interface for
1240 * Lustre tools like lfs, lctl and lfsck.
1242 * \param[in] cmd ioctl command
1243 * \param[in] exp OBD export of OFD
1244 * \param[in] len not used
1245 * \param[in,out] karg buffer with data
1246 * \param[in] uarg not used
1248 * \retval 0 if successful
1249 * \retval negative value on error
1251 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1252 void *karg, void __user *uarg)
1255 struct ofd_device *ofd = ofd_exp(exp);
1256 struct obd_device *obd = ofd_obd(ofd);
1260 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
1261 obd->obd_name, cmd, len, karg, uarg);
1262 rc = lu_env_init(&env, LCT_DT_THREAD);
1267 case OBD_IOC_ABORT_RECOVERY:
1268 CWARN("%s: Aborting recovery\n", obd->obd_name);
1269 obd->obd_abort_recovery = 1;
1270 target_stop_recovery_thread(obd);
1273 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1274 rc = dt_sync(&env, ofd->ofd_osd);
1276 case OBD_IOC_SET_READONLY:
1277 rc = dt_sync(&env, ofd->ofd_osd);
1279 rc = dt_ro(&env, ofd->ofd_osd);
1281 case OBD_IOC_START_LFSCK: {
1282 struct obd_ioctl_data *data = karg;
1283 struct lfsck_start_param lsp;
1285 if (unlikely(!data)) {
1290 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1291 lsp.lsp_index_valid = 0;
1292 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1295 case OBD_IOC_STOP_LFSCK: {
1296 struct obd_ioctl_data *data = karg;
1297 struct lfsck_stop stop;
1299 stop.ls_status = LS_STOPPED;
1300 /* Old lfsck utils may pass NULL @stop. */
1301 if (!data->ioc_inlbuf1)
1305 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1307 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1310 case OBD_IOC_GET_OBJ_VERSION:
1311 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1314 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
1323 * Implementation of obd_ops::o_precleanup.
1325 * This function stops device activity before shutting it down. It is called
1326 * from a cleanup function upon forceful device cleanup. For OFD there are no
1327 * special actions, it just invokes target_recovery_cleanup().
1329 * \param[in] obd OBD device of OFD
1333 static int ofd_precleanup(struct obd_device *obd)
1336 target_cleanup_recovery(obd);
1341 * Implementation of obd_ops::o_health_check.
1343 * This function checks the OFD device health - ability to respond on
1344 * incoming requests. There are two health_check methods:
1345 * - get statfs from the OSD. It checks just responsiveness of
1347 * - do write attempt on bottom device to check it is fully operational and
1348 * is not stuck. This is expensive method and requires special configuration
1349 * option --enable-health-write while building Lustre, it is turned off
1352 * \param[in] nul not used
1353 * \param[in] obd OBD device of OFD
1355 * \retval 0 if successful
1356 * \retval negative value in case of error
1358 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1360 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1361 struct ofd_thread_info *info;
1363 #ifdef USE_HEALTH_CHECK_WRITE
1368 /* obd_proc_read_health pass NULL env, we need real one */
1369 rc = lu_env_init(&env, LCT_DT_THREAD);
1373 info = ofd_info_init(&env, NULL);
1374 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1378 if (info->fti_u.osfs.os_state & OS_STATFS_READONLY)
1379 GOTO(out, rc = -EROFS);
1381 #ifdef USE_HEALTH_CHECK_WRITE
1382 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1383 if (!info->fti_buf.lb_buf)
1384 GOTO(out, rc = -ENOMEM);
1386 info->fti_buf.lb_len = PAGE_SIZE;
1389 th = dt_trans_create(&env, ofd->ofd_osd);
1391 GOTO(out, rc = PTR_ERR(th));
1393 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1394 &info->fti_buf, info->fti_off, th);
1396 th->th_sync = 1; /* sync IO is needed */
1397 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1399 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1400 &info->fti_buf, &info->fti_off,
1403 dt_trans_stop(&env, ofd->ofd_osd, th);
1405 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1407 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",
1415 const struct obd_ops ofd_obd_ops = {
1416 .o_owner = THIS_MODULE,
1417 .o_connect = ofd_obd_connect,
1418 .o_reconnect = ofd_obd_reconnect,
1419 .o_disconnect = ofd_obd_disconnect,
1420 .o_create = ofd_echo_create,
1421 .o_statfs = ofd_statfs,
1422 .o_setattr = ofd_echo_setattr,
1423 .o_preprw = ofd_preprw,
1424 .o_commitrw = ofd_commitrw,
1425 .o_destroy = ofd_echo_destroy,
1426 .o_init_export = ofd_init_export,
1427 .o_destroy_export = ofd_destroy_export,
1428 .o_postrecov = ofd_obd_postrecov,
1429 .o_getattr = ofd_echo_getattr,
1430 .o_iocontrol = ofd_iocontrol,
1431 .o_precleanup = ofd_precleanup,
1432 .o_health_check = ofd_health_check,
1433 .o_set_info_async = ofd_set_info_async,
1434 .o_get_info = ofd_get_info,