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_nid)
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 */
79 rc = lprocfs_exp_setup(exp, client_nid);
81 /* Mask error for already created /proc entries */
82 RETURN(rc == -EALREADY ? 0 : rc);
84 stats = exp->exp_nid_stats;
85 stats->nid_stats = lprocfs_alloc_stats(LPROC_OFD_STATS_LAST,
86 LPROCFS_STATS_FLAG_NOPERCPU);
87 if (!stats->nid_stats)
90 ofd_stats_counter_init(stats->nid_stats, 0);
92 rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
94 lprocfs_free_stats(&stats->nid_stats);
98 rc = lprocfs_nid_ldlm_stats_init(stats);
107 * Match client and OST server connection feature flags.
109 * Compute the compatibility flags for a connection request based on
110 * features mutually supported by client and server.
112 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
113 * must not be updated here, otherwise a partially initialized value may
114 * be exposed. After the connection request is successfully processed,
115 * the top-level tgt_connect() request handler atomically updates the export
116 * connect flags from the obd_connect_data::ocd_connect_flags field of the
117 * reply. \see tgt_connect().
119 * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
120 * full struct obd_connect_data. So care must be taken when accessing fields
121 * that are not present in struct obd_connect_data_v1. See LU-16.
123 * \param[in] env execution environment
124 * \param[in] exp the obd_export associated with this
126 * \param[in] data stores data for this connect request
127 * \param[in] new_connection is this connection new or not
129 * \retval 0 if success
130 * \retval -EPROTO client and server feature requirements are
132 * \retval -EBADF OST index in connect request doesn't match
135 static int ofd_parse_connect_data(const struct lu_env *env,
136 struct obd_export *exp,
137 struct obd_connect_data *data,
140 struct ofd_device *ofd = ofd_exp(exp);
141 struct filter_export_data *fed = &exp->exp_filter_data;
147 "%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_group %u\n",
148 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
149 data->ocd_connect_flags, data->ocd_version,
150 data->ocd_grant, data->ocd_index, data->ocd_group);
152 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
153 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",
154 obd_export_nid2str(exp), fed->fed_group,
158 fed->fed_group = data->ocd_group;
160 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
162 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
163 data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2;
165 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
166 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
167 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
168 exp->exp_obd->obd_name, data->ocd_group);
169 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
172 /* Determine optimal brw size before calculating grant */
173 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
174 data->ocd_brw_size = 65536;
175 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
176 if (data->ocd_brw_size > ofd->ofd_brw_size)
177 data->ocd_brw_size = ofd->ofd_brw_size;
178 if (data->ocd_brw_size == 0) {
179 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",
180 exp->exp_obd->obd_name,
181 exp->exp_client_uuid.uuid,
182 exp, data->ocd_connect_flags, data->ocd_version,
183 data->ocd_grant, data->ocd_index);
188 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
189 struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf;
191 /* client is reporting its page size, for future use */
192 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
193 data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits;
195 * ddp_inodespace may not be power-of-two value, eg. for ldiskfs
196 * it's LDISKFS_DIR_REC_LEN(20) = 28.
198 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
199 /* ocd_grant_tax_kb is in 1K byte blocks */
200 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
201 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
205 * Save connect_data we have so far because tgt_grant_connect()
206 * uses it to calculate grant, and we want to save the client
207 * version before it is overwritten by LUSTRE_VERSION_CODE.
209 exp->exp_connect_data = *data;
210 if (OCD_HAS_FLAG(data, GRANT))
211 tgt_grant_connect(env, exp, data, new_connection);
213 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
214 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
215 int index = lsd->lsd_osd_index;
217 if (index != data->ocd_index) {
218 LCONSOLE_ERROR_MSG(0x136,
219 "Connection from %s to index %u doesn't match actual OST index %u in last_rcvd file, bad configuration?\n",
220 obd_export_nid2str(exp), index,
224 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
225 /* this will only happen on the first connect */
226 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
228 * sync is not needed here as tgt_client_new will
229 * set exp_need_sync flag
231 tgt_server_data_update(env, &ofd->ofd_lut, 0);
235 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
236 __u32 cksum_types = data->ocd_cksum_types;
238 tgt_mask_cksum_types(&ofd->ofd_lut, &data->ocd_cksum_types);
240 if (unlikely(data->ocd_cksum_types == 0)) {
241 CERROR("%s: Connect with checksum support but no ocd_cksum_types is set\n",
242 exp->exp_obd->obd_name);
247 "%s: cli %s supports cksum type %x, return %x\n",
248 exp->exp_obd->obd_name, obd_export_nid2str(exp),
249 cksum_types, data->ocd_cksum_types);
252 * This client does not support OBD_CONNECT_CKSUM.
253 * Report failure to negotiate checksum at connect
256 "%s: cli %s does not support OBD_CONNECT_CKSUM\n",
257 exp->exp_obd->obd_name, obd_export_nid2str(exp));
260 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
261 data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes;
263 data->ocd_version = LUSTRE_VERSION_CODE;
265 if (OCD_HAS_FLAG(data, PINGLESS)) {
266 if (ptlrpc_pinger_suppress_pings()) {
267 spin_lock(&exp->exp_obd->obd_dev_lock);
268 list_del_init(&exp->exp_obd_chain_timed);
269 spin_unlock(&exp->exp_obd->obd_dev_lock);
271 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
275 if (!ofd->ofd_lut.lut_dt_conf.ddp_has_lseek_data_hole)
276 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
282 * Re-initialize export upon client reconnection.
284 * This function parses connection data from reconnect and resets
287 * \param[in] env execution environment
288 * \param[in] exp OBD export
289 * \param[in] obd OFD device
290 * \param[in] cluuid NID of client
291 * \param[in] data connection data from request
292 * \param[in] localdata client NID
294 * \retval 0 if successful
295 * \retval negative value on error
297 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
298 struct obd_device *obd, struct obd_uuid *cluuid,
299 struct obd_connect_data *data,
302 struct ofd_device *ofd;
307 if (!exp || !obd || !cluuid)
310 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
311 if (rc != 0 && rc != -EEXIST)
314 ofd = ofd_dev(obd->obd_lu_dev);
316 rc = ofd_parse_connect_data(env, exp, data, false);
318 ofd_export_stats_init(ofd, exp, client_nid);
320 nodemap_del_member(exp);
326 * Initialize new client connection.
328 * This function handles new connection to the OFD. The new export is
329 * created (in context of class_connect()) and persistent client data is
330 * initialized on storage.
332 * \param[in] env execution environment
333 * \param[out] _exp stores pointer to new export
334 * \param[in] obd OFD device
335 * \param[in] cluuid client UUID
336 * \param[in] data connection data from request
337 * \param[in] localdata client NID
339 * \retval 0 if successful
340 * \retval negative value on error
342 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
343 struct obd_device *obd, struct obd_uuid *cluuid,
344 struct obd_connect_data *data, void *localdata)
346 struct obd_export *exp;
347 struct ofd_device *ofd;
348 struct lustre_handle conn = { 0 };
353 if (!_exp || !obd || !cluuid)
356 ofd = ofd_dev(obd->obd_lu_dev);
358 rc = class_connect(&conn, obd, cluuid);
362 exp = class_conn2export(&conn);
363 LASSERT(exp != NULL);
366 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
367 if (rc != 0 && rc != -EEXIST)
371 "%s: cannot find nodemap for client %s: nid is null\n",
372 obd->obd_name, cluuid->uuid);
375 rc = ofd_parse_connect_data(env, exp, data, true);
379 if (obd->obd_replayable) {
380 struct tg_export_data *ted = &exp->exp_target_data;
382 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
383 sizeof(ted->ted_lcd->lcd_uuid));
384 rc = tgt_client_new(env, exp);
387 ofd_export_stats_init(ofd, exp, localdata);
390 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
391 data ? data->ocd_group : -1);
395 class_disconnect(exp);
396 nodemap_del_member(exp);
405 * Disconnect a connected client.
407 * This function terminates the client connection. The client export is
408 * disconnected (cleaned up) and client data on persistent storage is removed.
410 * \param[in] exp OBD export
412 * \retval 0 if successful
413 * \retval negative value on error
415 int ofd_obd_disconnect(struct obd_export *exp)
417 struct ofd_device *ofd = ofd_exp(exp);
424 class_export_get(exp);
426 if (!(exp->exp_flags & OBD_OPT_FORCE))
427 tgt_grant_sanity_check(ofd_obd(ofd), __func__);
429 rc = server_disconnect_export(exp);
431 tgt_grant_discard(exp);
433 /* Do not erase record for recoverable client. */
434 if (exp->exp_obd->obd_replayable &&
435 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
436 rc = lu_env_init(&env, LCT_DT_THREAD);
440 tgt_client_del(&env, exp);
444 nodemap_del_member(exp);
445 class_export_put(exp);
450 * Implementation of obd_ops::o_init_export.
452 * This function is called from class_new_export() and initializes
453 * the OFD-specific data for new export.
455 * \param[in] exp OBD export
457 * \retval 0 if successful
458 * \retval negative value on error
460 static int ofd_init_export(struct obd_export *exp)
464 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
465 spin_lock(&exp->exp_lock);
466 exp->exp_connecting = 1;
467 spin_unlock(&exp->exp_lock);
469 /* self-export doesn't need client data and ldlm initialization */
470 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
471 &exp->exp_client_uuid)))
474 rc = tgt_client_alloc(exp);
476 ldlm_init_export(exp);
478 CERROR("%s: Can't initialize export: rc %d\n",
479 exp->exp_obd->obd_name, rc);
484 * Implementation of obd_ops::o_destroy_export.
486 * This function is called from class_export_destroy() to cleanup
487 * the OFD-specific data for export being destroyed.
489 * \param[in] exp OBD export
491 * \retval 0 if successful
492 * \retval negative value on error
494 static int ofd_destroy_export(struct obd_export *exp)
496 struct ofd_device *ofd = ofd_exp(exp);
498 if (exp->exp_target_data.ted_pending)
499 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
500 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
501 exp, exp->exp_target_data.ted_pending);
503 target_destroy_export(exp);
505 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
506 &exp->exp_client_uuid)))
509 ldlm_destroy_export(exp);
510 tgt_client_free(exp);
513 * discard grants once we're sure no more
514 * interaction with the client is possible
516 tgt_grant_discard(exp);
518 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
519 ofd->ofd_lut.lut_tgd.tgd_tot_granted_clients--;
521 if (!(exp->exp_flags & OBD_OPT_FORCE))
522 tgt_grant_sanity_check(exp->exp_obd, __func__);
528 * Notify all devices in server stack about recovery completion.
530 * This function calls ldo_recovery_complete() for all lower devices in the
531 * server stack so they will be prepared for normal operations.
533 * \param[in] env execution environment
534 * \param[in] ofd OFD device
536 * \retval 0 if successful
537 * \retval negative value on error
539 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
541 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
544 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
546 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
547 struct lfsck_start_param lsp;
549 lsp.lsp_start = NULL;
550 lsp.lsp_index_valid = 0;
551 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
552 if (rc != 0 && rc != -EALREADY)
553 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
557 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
561 * Implementation of obd_ops::o_postrecov.
563 * This function is called from target_finish_recovery() upon recovery
566 * \param[in] obd OBD device of OFD
568 * \retval 0 if successful
569 * \retval negative value on error
571 static int ofd_obd_postrecov(struct obd_device *obd)
574 struct lu_device *ldev = obd->obd_lu_dev;
579 rc = lu_env_init(&env, LCT_DT_THREAD);
582 ofd_info_init(&env, obd->obd_self_export);
584 rc = ofd_postrecov(&env, ofd_dev(ldev));
591 * Implementation of obd_ops::o_set_info_async.
593 * This function is not called from request handler, it is only used by
594 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
595 * \see ofd_set_info_hdl() for request handler function.
597 * \param[in] env execution environment
598 * \param[in] exp OBD export of OFD device
599 * \param[in] keylen length of \a key
600 * \param[in] key key name
601 * \param[in] vallen length of \a val
602 * \param[in] val the \a key value
603 * \param[in] set not used in OFD
605 * \retval 0 if successful
606 * \retval negative value on error
608 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
609 __u32 keylen, void *key, __u32 vallen, void *val,
610 struct ptlrpc_request_set *set)
617 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
621 if (KEY_IS(KEY_SPTLRPC_CONF)) {
622 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
624 CERROR("%s: Unsupported key %s\n",
625 exp->exp_obd->obd_name, (char *)key);
632 * Implementation of obd_ops::o_get_info.
634 * This function is not called from request handler, it is only used by
635 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
637 * \see ofd_get_info_hdl() for request handler function.
639 * \param[in] env execution environment
640 * \param[in] exp OBD export of OFD device
641 * \param[in] keylen length of \a key
642 * \param[in] key key name
643 * \param[out] vallen length of key value
644 * \param[out] val the key value to return
646 * \retval 0 if successful
647 * \retval negative value on error
649 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
650 __u32 keylen, void *key, __u32 *vallen, void *val)
652 struct ofd_thread_info *info;
653 struct ofd_device *ofd;
654 struct ll_fiemap_info_key *fm_key = key;
655 struct fiemap *fiemap = val;
661 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
667 if (KEY_IS(KEY_FIEMAP)) {
668 info = ofd_info_init(env, exp);
670 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
671 ofd->ofd_lut.lut_lsd.lsd_osd_index);
675 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
677 CERROR("%s: not supported key %s\n",
678 ofd_name(ofd), (char *)key);
686 * Implementation of obd_ops::o_statfs.
688 * This function returns information about a storage file system.
689 * It is called from several places by using the OBD API as well as
690 * by direct call, e.g. from request handler.
692 * \see ofd_statfs_hdl() for request handler function.
694 * Report also the state of the OST to the caller in osfs->os_state
695 * (OS_STATFS_READONLY, OS_STATFS_DEGRADED).
697 * \param[in] env execution environment
698 * \param[in] exp OBD export of OFD device
699 * \param[out] osfs statistic data to return
700 * \param[in] max_age maximum age for cached data
701 * \param[in] flags not used in OFD
703 * \retval 0 if successful
704 * \retval negative value on error
706 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
707 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
709 struct obd_device *obd = class_exp2obd(exp);
710 struct ofd_device *ofd = ofd_exp(exp);
711 struct tg_grants_data *tgd = &ofd->ofd_lut.lut_tgd;
712 int current_blockbits;
717 rc = tgt_statfs_internal(env, &ofd->ofd_lut, osfs, max_age, NULL);
721 /* tgd_blockbit is recordsize bits set during mkfs.
722 * This once set does not change. However, 'zfs set'
723 * can be used to change the OST blocksize. Instead
724 * of using cached value of 'tgd_blockbit' always
725 * calculate the blocksize bits which may have
728 current_blockbits = fls64(osfs->os_bsize) - 1;
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,
736 "blocks cached %llu granted %llu 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 << current_blockbits,
740 osfs->os_bavail << current_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) >> current_blockbits));
747 * The QoS code on the MDS does not care about space reserved for
748 * precreate, so take it out.
750 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
751 struct tg_export_data *ted;
753 ted = &obd->obd_self_export->exp_target_data;
754 osfs->os_granted = min_t(u64, osfs->os_bavail,
755 ted->ted_grant >> current_blockbits);
756 osfs->os_bavail -= osfs->os_granted;
759 tgt_grant_sanity_check(obd, __func__);
760 CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
761 "%llu objects: %llu free; state %x\n",
762 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
763 osfs->os_files, osfs->os_ffree, osfs->os_state);
765 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
766 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
767 /* Reduce free inode count to zero, but keep "used" intact */
768 osfs->os_files -= osfs->os_ffree;
769 osfs->os_ffree -= osfs->os_ffree;
772 /* OS_STATFS_READONLY can be set by OSD already */
773 if (ofd->ofd_raid_degraded)
774 osfs->os_state |= OS_STATFS_DEGRADED;
776 if (ofd->ofd_no_precreate)
777 osfs->os_state |= OS_STATFS_NOPRECREATE;
779 if (obd->obd_self_export != exp && !exp_grant_param_supp(exp) &&
780 current_blockbits > COMPAT_BSIZE_SHIFT) {
782 * clients which don't support OBD_CONNECT_GRANT_PARAM
783 * should not see a block size > page size, otherwise
784 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
785 * block size which is the biggest block size known to work
786 * with all client's page size.
788 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
789 osfs->os_bfree <<= current_blockbits - COMPAT_BSIZE_SHIFT;
790 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
791 osfs->os_granted <<= current_blockbits - COMPAT_BSIZE_SHIFT;
792 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
795 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
796 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
797 /* Reduce free blocks count near zero, but keep "used" intact */
798 osfs->os_bavail -= osfs->os_bavail - 2;
799 osfs->os_blocks -= osfs->os_bfree - 2;
800 osfs->os_bfree -= osfs->os_bfree - 2;
809 * Implementation of obd_ops::o_setattr.
811 * This function is only used by ECHO client when it is run on top of OFD,
812 * \see ofd_setattr_hdl() for request handler function.
814 * \param[in] env execution environment
815 * \param[in] exp OBD export of OFD device
816 * \param[in] oa setattr parameters
818 * \retval 0 if successful
819 * \retval negative value on error
821 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
824 struct ofd_thread_info *info;
825 struct ofd_device *ofd = ofd_exp(exp);
826 struct ldlm_namespace *ns = ofd->ofd_namespace;
827 struct ldlm_resource *res;
828 struct ofd_object *fo;
829 struct lu_fid *fid = &oa->o_oi.oi_fid;
830 ktime_t kstart = ktime_get();
835 info = ofd_info_init(env, exp);
837 ost_fid_build_resid(fid, &info->fti_resid);
840 * This would be very bad - accidentally truncating a file when
841 * changing the time or similar - bug 12203.
843 if (oa->o_valid & OBD_MD_FLSIZE) {
844 static char mdsinum[48];
846 if (oa->o_valid & OBD_MD_FLFID)
847 snprintf(mdsinum, sizeof(mdsinum) - 1,
848 "of parent "DFID, oa->o_parent_seq,
849 oa->o_parent_oid, 0);
853 CERROR("%s: setattr from %s trying to truncate object "DFID
854 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
856 GOTO(out, rc = -EPERM);
859 fo = ofd_object_find_exists(env, ofd, fid);
861 CERROR("%s: can't find object "DFID"\n",
862 ofd_name(ofd), PFID(fid));
863 GOTO(out, rc = PTR_ERR(fo));
866 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
867 info->fti_attr.la_valid &= ~LA_TYPE;
869 /* setting objects attributes (including owner/group) */
870 rc = ofd_attr_set(env, fo, &info->fti_attr, oa);
872 GOTO(out_unlock, rc);
874 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL,
875 ktime_us_delta(ktime_get(), kstart));
878 ofd_object_put(env, fo);
882 * we do not call this before to avoid lu_object_find() in
883 * ->lvbo_update() holding another reference on the object.
884 * otherwise concurrent destroy can make the object unavailable
885 * for 2nd lu_object_find() waiting for the first reference
888 res = ldlm_resource_get(ns, NULL, &info->fti_resid,
891 ldlm_res_lvbo_update(res, NULL, 0);
892 ldlm_resource_putref(res);
900 * Destroy OFD object by its FID.
902 * Supplemental function to destroy object by FID, it is used by request
903 * handler and by ofd_echo_destroy() below to find object by FID, lock it
904 * and call ofd_destroy() finally.
906 * \param[in] env execution environment
907 * \param[in] ofd OFD device
908 * \param[in] fid FID of object
909 * \param[in] orphan set if object being destroyed is an orphan
911 * \retval 0 if successful
912 * \retval negative value on error
914 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
915 const struct lu_fid *fid, int orphan)
917 struct ofd_thread_info *info = ofd_info(env);
918 struct lustre_handle lockh;
919 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
920 struct ofd_object *fo;
921 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
926 fo = ofd_object_find_exists(env, ofd, fid);
931 * Tell the clients that the object is gone now and that they should
932 * throw away any cached pages.
934 ost_fid_build_resid(fid, &info->fti_resid);
935 rc = ldlm_cli_enqueue_local(env, ofd->ofd_namespace, &info->fti_resid,
936 LDLM_EXTENT, &policy, LCK_PW, &flags,
937 ldlm_blocking_ast, ldlm_completion_ast,
938 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
940 /* We only care about the side-effects, just drop the lock. */
942 ldlm_lock_decref(&lockh, LCK_PW);
946 rc = ofd_destroy(env, fo, orphan);
949 ofd_object_put(env, fo);
954 * Implementation of obd_ops::o_destroy.
956 * This function is only used by ECHO client when it is run on top of OFD,
957 * \see ofd_destroy_hdl() for request handler function.
959 * \param[in] env execution environment
960 * \param[in] exp OBD export of OFD device
961 * \param[in] oa obdo structure with FID
963 * Note: this is OBD API method which is common API for server OBDs and
964 * client OBDs. Thus some parameters used in client OBDs may not be used
965 * on server OBDs and vice versa.
967 * \retval 0 if successful
968 * \retval negative value on error
970 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
973 struct ofd_device *ofd = ofd_exp(exp);
974 struct lu_fid *fid = &oa->o_oi.oi_fid;
979 ofd_info_init(env, exp);
981 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
985 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
987 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
989 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
990 ofd_name(ofd), PFID(fid));
992 } else if (rc != 0) {
993 CERROR("%s: error destroying object "DFID": %d\n",
994 ofd_name(ofd), PFID(fid), rc);
1003 * Implementation of obd_ops::o_create.
1005 * This function is only used by ECHO client when it is run on top of OFD
1006 * and just creates an object.
1007 * \see ofd_create_hdl() for request handler function.
1009 * \param[in] env execution environment
1010 * \param[in] exp OBD export of OFD device
1011 * \param[in] oa obdo structure with FID sequence to use
1013 * Note: this is OBD API method which is common API for server OBDs and
1014 * client OBDs. Thus some parameters used in client OBDs may not be used
1015 * on server OBDs and vice versa.
1017 * \retval 0 if successful
1018 * \retval negative value on error
1020 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1023 struct ofd_device *ofd = ofd_exp(exp);
1024 u64 seq = ostid_seq(&oa->o_oi);
1025 struct ofd_seq *oseq;
1034 if (ofd->ofd_no_precreate)
1037 ofd_info_init(env, exp);
1039 LASSERT(seq == FID_SEQ_ECHO);
1040 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1042 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1044 down_read(&ofd->ofd_lastid_rwsem);
1046 * Currently, for safe, we do not distinguish which LAST_ID is broken,
1047 * we may do that in the future.
1048 * Return -ENOSPC until the LAST_ID rebuilt.
1050 if (unlikely(ofd->ofd_lastid_rebuilding))
1051 GOTO(out_sem, rc = -ENOSPC);
1053 rc = ofd_validate_seq(exp, seq);
1057 oseq = ofd_seq_load(env, ofd, seq);
1059 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1060 ofd_name(ofd), seq, PTR_ERR(oseq));
1061 GOTO(out_sem, rc = -EINVAL);
1064 mutex_lock(&oseq->os_create_lock);
1065 granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1070 "%s: failed to acquire grant space for precreate (%lld): rc = %d\n",
1071 ofd_name(ofd), diff, rc);
1076 next_id = ofd_seq_last_oid(oseq) + 1;
1077 count = ofd_precreate_batch(ofd, (int)diff);
1079 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1081 CERROR("%s: unable to precreate: rc = %d\n",
1084 rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1086 CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n",
1088 (unsigned long long)ofd_seq_last_oid(oseq),
1089 POSTID(&oa->o_oi), rc);
1091 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1094 tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1096 mutex_unlock(&oseq->os_create_lock);
1097 ofd_seq_put(env, oseq);
1100 up_read(&ofd->ofd_lastid_rwsem);
1105 * Implementation of obd_ops::o_getattr.
1107 * This function is only used by ECHO client when it is run on top of OFD
1108 * and returns attributes of object.
1109 * \see ofd_getattr_hdl() for request handler function.
1111 * \param[in] env execution environment
1112 * \param[in] exp OBD export of OFD device
1113 * \param[in,out] oa contains FID of object to get attributes from and
1114 * is used to return attributes back
1116 * \retval 0 if successful
1117 * \retval negative value on error
1119 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1122 struct ofd_device *ofd = ofd_exp(exp);
1123 struct ofd_thread_info *info;
1124 struct lu_fid *fid = &oa->o_oi.oi_fid;
1125 struct ofd_object *fo;
1130 info = ofd_info_init(env, exp);
1132 fo = ofd_object_find_exists(env, ofd, fid);
1134 GOTO(out, rc = PTR_ERR(fo));
1136 LASSERT(fo != NULL);
1137 rc = ofd_attr_get(env, fo, &info->fti_attr);
1138 oa->o_valid = OBD_MD_FLID;
1142 obdo_from_la(oa, &info->fti_attr,
1143 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1145 /* Store object version in reply */
1146 curr_version = dt_version_get(env, ofd_object_child(fo));
1147 if ((__s64)curr_version != -EOPNOTSUPP) {
1148 oa->o_valid |= OBD_MD_FLDATAVERSION;
1149 oa->o_data_version = curr_version;
1153 ofd_object_put(env, fo);
1159 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1161 * This is supplemental function for ofd_iocontrol() to return object
1162 * version for lctl tool.
1164 * \param[in] env execution environment
1165 * \param[in] ofd OFD device
1166 * \param[out] karg ioctl data
1168 * \retval 0 if successful
1169 * \retval negative value on error
1171 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1172 struct ofd_device *ofd, void *karg)
1174 struct obd_ioctl_data *data = karg;
1176 struct ofd_object *fo;
1177 dt_obj_version_t version;
1182 if (!data->ioc_inlbuf2 || data->ioc_inllen2 != sizeof(version))
1183 GOTO(out, rc = -EINVAL);
1185 if (data->ioc_inlbuf1 && data->ioc_inllen1 == sizeof(fid)) {
1186 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1187 } else if (data->ioc_inlbuf3 &&
1188 data->ioc_inllen3 == sizeof(__u64) &&
1189 data->ioc_inlbuf4 &&
1190 data->ioc_inllen4 == sizeof(__u64)) {
1191 struct ost_id ostid = { };
1193 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1194 rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1197 rc = ostid_to_fid(&fid, &ostid,
1198 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1202 GOTO(out, rc = -EINVAL);
1205 if (!fid_is_sane(&fid))
1206 GOTO(out, rc = -EINVAL);
1208 fo = ofd_object_find(env, ofd, &fid);
1210 GOTO(out, rc = PTR_ERR(fo));
1212 if (!ofd_object_exists(fo))
1213 GOTO(out_fo, rc = -ENOENT);
1215 if (lu_object_remote(&fo->ofo_obj.do_lu))
1216 GOTO(out_fo, rc = -EREMOTE);
1218 version = dt_version_get(env, ofd_object_child(fo));
1220 GOTO(out_fo, rc = -EIO);
1222 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1226 ofd_object_put(env, fo);
1232 * Implementation of obd_ops::o_iocontrol.
1234 * This is OFD ioctl handling function which is primary interface for
1235 * Lustre tools like lfs, lctl and lfsck.
1237 * \param[in] cmd ioctl command
1238 * \param[in] exp OBD export of OFD
1239 * \param[in] len not used
1240 * \param[in,out] karg buffer with data
1241 * \param[in] uarg not used
1243 * \retval 0 if successful
1244 * \retval negative value on error
1246 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1247 void *karg, void __user *uarg)
1250 struct ofd_device *ofd = ofd_exp(exp);
1251 struct obd_device *obd = ofd_obd(ofd);
1256 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1257 rc = lu_env_init(&env, LCT_DT_THREAD);
1262 case OBD_IOC_ABORT_RECOVERY:
1263 CERROR("%s: aborting recovery\n", obd->obd_name);
1264 obd->obd_abort_recovery = 1;
1265 target_stop_recovery_thread(obd);
1268 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1269 rc = dt_sync(&env, ofd->ofd_osd);
1271 case OBD_IOC_SET_READONLY:
1272 rc = dt_sync(&env, ofd->ofd_osd);
1274 rc = dt_ro(&env, ofd->ofd_osd);
1276 case OBD_IOC_START_LFSCK: {
1277 struct obd_ioctl_data *data = karg;
1278 struct lfsck_start_param lsp;
1280 if (unlikely(!data)) {
1285 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1286 lsp.lsp_index_valid = 0;
1287 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1290 case OBD_IOC_STOP_LFSCK: {
1291 struct obd_ioctl_data *data = karg;
1292 struct lfsck_stop stop;
1294 stop.ls_status = LS_STOPPED;
1295 /* Old lfsck utils may pass NULL @stop. */
1296 if (!data->ioc_inlbuf1)
1300 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1302 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1305 case OBD_IOC_GET_OBJ_VERSION:
1306 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1309 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1318 * Implementation of obd_ops::o_precleanup.
1320 * This function stops device activity before shutting it down. It is called
1321 * from a cleanup function upon forceful device cleanup. For OFD there are no
1322 * special actions, it just invokes target_recovery_cleanup().
1324 * \param[in] obd OBD device of OFD
1328 static int ofd_precleanup(struct obd_device *obd)
1331 target_cleanup_recovery(obd);
1336 * Implementation of obd_ops::o_health_check.
1338 * This function checks the OFD device health - ability to respond on
1339 * incoming requests. There are two health_check methods:
1340 * - get statfs from the OSD. It checks just responsiveness of
1342 * - do write attempt on bottom device to check it is fully operational and
1343 * is not stuck. This is expensive method and requires special configuration
1344 * option --enable-health-write while building Lustre, it is turned off
1347 * \param[in] nul not used
1348 * \param[in] obd OBD device of OFD
1350 * \retval 0 if successful
1351 * \retval negative value in case of error
1353 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1355 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1356 struct ofd_thread_info *info;
1358 #ifdef USE_HEALTH_CHECK_WRITE
1363 /* obd_proc_read_health pass NULL env, we need real one */
1364 rc = lu_env_init(&env, LCT_DT_THREAD);
1368 info = ofd_info_init(&env, NULL);
1369 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1373 if (info->fti_u.osfs.os_state & OS_STATFS_READONLY)
1374 GOTO(out, rc = -EROFS);
1376 #ifdef USE_HEALTH_CHECK_WRITE
1377 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1378 if (!info->fti_buf.lb_buf)
1379 GOTO(out, rc = -ENOMEM);
1381 info->fti_buf.lb_len = PAGE_SIZE;
1384 th = dt_trans_create(&env, ofd->ofd_osd);
1386 GOTO(out, rc = PTR_ERR(th));
1388 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1389 &info->fti_buf, info->fti_off, th);
1391 th->th_sync = 1; /* sync IO is needed */
1392 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1394 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1395 &info->fti_buf, &info->fti_off,
1398 dt_trans_stop(&env, ofd->ofd_osd, th);
1400 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1402 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",
1410 const struct obd_ops ofd_obd_ops = {
1411 .o_owner = THIS_MODULE,
1412 .o_connect = ofd_obd_connect,
1413 .o_reconnect = ofd_obd_reconnect,
1414 .o_disconnect = ofd_obd_disconnect,
1415 .o_create = ofd_echo_create,
1416 .o_statfs = ofd_statfs,
1417 .o_setattr = ofd_echo_setattr,
1418 .o_preprw = ofd_preprw,
1419 .o_commitrw = ofd_commitrw,
1420 .o_destroy = ofd_echo_destroy,
1421 .o_init_export = ofd_init_export,
1422 .o_destroy_export = ofd_destroy_export,
1423 .o_postrecov = ofd_obd_postrecov,
1424 .o_getattr = ofd_echo_getattr,
1425 .o_iocontrol = ofd_iocontrol,
1426 .o_precleanup = ofd_precleanup,
1427 .o_health_check = ofd_health_check,
1428 .o_set_info_async = ofd_set_info_async,
1429 .o_get_info = ofd_get_info,