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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ofd/ofd_obd.c
38 * Author: Andreas Dilger <adilger@whamcloud.com>
39 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40 * Author: Mike Pershin <tappro@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_FILTER
45 #include <lustre/lustre_idl.h>
46 #include "ofd_internal.h"
47 #include <obd_cksum.h>
48 #include <lustre_quota.h>
50 static int ofd_export_stats_init(struct ofd_device *ofd,
51 struct obd_export *exp, void *client_nid)
53 struct obd_device *obd = ofd_obd(ofd);
54 struct nid_stat *stats;
60 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
61 /* Self-export gets no proc entry */
64 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
66 /* Mask error for already created
76 stats = exp->exp_nid_stats;
77 LASSERT(stats != NULL);
79 num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
82 stats->nid_stats = lprocfs_alloc_stats(num_stats,
83 LPROCFS_STATS_FLAG_NOPERCPU);
84 if (stats->nid_stats == NULL)
87 lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
88 lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
89 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
90 lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
91 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
93 rc = lprocfs_register_stats(stats->nid_proc, "stats",
98 rc = lprocfs_nid_ldlm_stats_init(stats);
100 lprocfs_free_stats(&stats->nid_stats);
109 static int ofd_parse_connect_data(const struct lu_env *env,
110 struct obd_export *exp,
111 struct obd_connect_data *data,
114 struct ofd_device *ofd = ofd_exp(exp);
115 struct filter_export_data *fed = &exp->exp_filter_data;
120 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
121 " ocd_version: %x ocd_grant: %d ocd_index: %u"
123 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
124 data->ocd_connect_flags, data->ocd_version,
125 data->ocd_grant, data->ocd_index, data->ocd_group);
127 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
128 CWARN("!!! This export (nid %s) used object group %d "
129 "earlier; now it's trying to use group %d! This could "
130 "be a bug in the MDS. Please report to "
131 "http://bugs.whamcloud.com/\n",
132 obd_export_nid2str(exp), fed->fed_group,
136 fed->fed_group = data->ocd_group;
138 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
139 exp->exp_connect_data = *data;
140 data->ocd_version = LUSTRE_VERSION_CODE;
142 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
143 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
144 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
145 exp->exp_obd->obd_name, data->ocd_group);
146 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
149 if (ofd_grant_param_supp(exp)) {
150 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
151 /* ocd_{blocksize,inodespace} are log2 values */
152 data->ocd_blocksize = ofd->ofd_blockbits;
153 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
154 /* ocd_grant_extent is in 1K blocks */
155 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
158 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
159 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
162 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
163 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
164 int index = lsd->lsd_osd_index;
166 if (index != data->ocd_index) {
167 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
168 " %u doesn't match actual OST index"
169 " %u in last_rcvd file, bad "
171 obd_export_nid2str(exp), index,
175 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
176 /* this will only happen on the first connect */
177 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
178 /* sync is not needed here as lut_client_add will
179 * set exp_need_sync flag */
180 tgt_server_data_update(env, &ofd->ofd_lut, 0);
183 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
184 data->ocd_brw_size = 65536;
185 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
186 data->ocd_brw_size = min(data->ocd_brw_size,
187 (__u32)DT_MAX_BRW_SIZE);
188 if (data->ocd_brw_size == 0) {
189 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
190 " ocd_version: %x ocd_grant: %d ocd_index: %u "
191 "ocd_brw_size is unexpectedly zero, "
192 "network data corruption?"
193 "Refusing connection of this client\n",
194 exp->exp_obd->obd_name,
195 exp->exp_client_uuid.uuid,
196 exp, data->ocd_connect_flags, data->ocd_version,
197 data->ocd_grant, data->ocd_index);
202 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
203 __u32 cksum_types = data->ocd_cksum_types;
205 /* The client set in ocd_cksum_types the checksum types it
206 * supports. We have to mask off the algorithms that we don't
208 data->ocd_cksum_types &= cksum_types_supported_server();
210 if (unlikely(data->ocd_cksum_types == 0)) {
211 CERROR("%s: Connect with checksum support but no "
212 "ocd_cksum_types is set\n",
213 exp->exp_obd->obd_name);
217 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
218 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
219 cksum_types, data->ocd_cksum_types);
221 /* This client does not support OBD_CONNECT_CKSUM
222 * fall back to CRC32 */
223 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
224 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
225 exp->exp_obd->obd_name, obd_export_nid2str(exp));
228 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
229 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
231 if (OCD_HAS_FLAG(data, PINGLESS)) {
232 if (ptlrpc_pinger_suppress_pings()) {
233 spin_lock(&exp->exp_obd->obd_dev_lock);
234 list_del_init(&exp->exp_obd_chain_timed);
235 spin_unlock(&exp->exp_obd->obd_dev_lock);
237 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
244 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
245 struct obd_device *obd, struct obd_uuid *cluuid,
246 struct obd_connect_data *data, void *localdata)
248 struct ofd_device *ofd;
253 if (exp == NULL || obd == NULL || cluuid == NULL)
256 ofd = ofd_dev(obd->obd_lu_dev);
258 rc = lu_env_refill((struct lu_env *)env);
260 CERROR("Failure to refill session: '%d'\n", rc);
264 ofd_info_init(env, exp);
265 rc = ofd_parse_connect_data(env, exp, data, false);
267 ofd_export_stats_init(ofd, exp, localdata);
272 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
273 struct obd_device *obd, struct obd_uuid *cluuid,
274 struct obd_connect_data *data, void *localdata)
276 struct obd_export *exp;
277 struct ofd_device *ofd;
278 struct lustre_handle conn = { 0 };
282 if (_exp == NULL || obd == NULL || cluuid == NULL)
285 ofd = ofd_dev(obd->obd_lu_dev);
287 rc = class_connect(&conn, obd, cluuid);
291 exp = class_conn2export(&conn);
292 LASSERT(exp != NULL);
294 rc = lu_env_refill((struct lu_env *)env);
296 CERROR("Failure to refill session: '%d'\n", rc);
300 ofd_info_init(env, exp);
302 rc = ofd_parse_connect_data(env, exp, data, true);
306 if (obd->obd_replayable) {
307 struct tg_export_data *ted = &exp->exp_target_data;
309 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
310 sizeof(ted->ted_lcd->lcd_uuid));
311 rc = tgt_client_new(env, exp);
314 ofd_export_stats_init(ofd, exp, localdata);
317 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
322 class_disconnect(exp);
330 static int ofd_obd_disconnect(struct obd_export *exp)
332 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
339 class_export_get(exp);
341 if (!(exp->exp_flags & OBD_OPT_FORCE))
342 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
344 rc = server_disconnect_export(exp);
346 ofd_grant_discard(exp);
348 rc = lu_env_init(&env, LCT_DT_THREAD);
352 /* Do not erase record for recoverable client. */
353 if (exp->exp_obd->obd_replayable &&
354 (!exp->exp_obd->obd_fail || exp->exp_failed))
355 tgt_client_del(&env, exp);
358 class_export_put(exp);
362 static int ofd_init_export(struct obd_export *exp)
366 spin_lock_init(&exp->exp_filter_data.fed_lock);
367 CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
368 spin_lock(&exp->exp_lock);
369 exp->exp_connecting = 1;
370 spin_unlock(&exp->exp_lock);
372 /* self-export doesn't need client data and ldlm initialization */
373 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
374 &exp->exp_client_uuid)))
377 rc = tgt_client_alloc(exp);
379 ldlm_init_export(exp);
381 CERROR("%s: Can't initialize export: rc %d\n",
382 exp->exp_obd->obd_name, rc);
386 static int ofd_destroy_export(struct obd_export *exp)
388 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
390 if (exp->exp_filter_data.fed_pending)
391 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
392 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
393 exp, exp->exp_filter_data.fed_pending);
395 target_destroy_export(exp);
397 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
398 &exp->exp_client_uuid)))
401 ldlm_destroy_export(exp);
402 tgt_client_free(exp);
404 ofd_fmd_cleanup(exp);
407 * discard grants once we're sure no more
408 * interaction with the client is possible
410 ofd_grant_discard(exp);
411 ofd_fmd_cleanup(exp);
413 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
414 if (ofd->ofd_tot_granted_clients > 0)
415 ofd->ofd_tot_granted_clients --;
418 if (!(exp->exp_flags & OBD_OPT_FORCE))
419 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
421 LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
425 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
427 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
429 CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
430 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
433 int ofd_obd_postrecov(struct obd_device *obd)
436 struct lu_device *ldev = obd->obd_lu_dev;
441 rc = lu_env_init(&env, LCT_DT_THREAD);
444 ofd_info_init(&env, obd->obd_self_export);
446 rc = ofd_postrecov(&env, ofd_dev(ldev));
452 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
453 struct obd_device *obd, int initial)
455 struct filter_obd *fo = &obd->u.filter;
456 struct sptlrpc_rule_set tmp_rset;
459 sptlrpc_rule_set_init(&tmp_rset);
460 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
462 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
467 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
469 write_lock(&fo->fo_sptlrpc_lock);
470 sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
471 fo->fo_sptlrpc_rset = tmp_rset;
472 write_unlock(&fo->fo_sptlrpc_lock);
477 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
483 LCONSOLE_WARN("%s: received MDS connection from %s\n",
484 exp->exp_obd->obd_name, obd_export_nid2str(exp));
488 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
489 __u32 keylen, void *key, __u32 vallen, void *val,
490 struct ptlrpc_request_set *set)
492 struct ofd_device *ofd;
497 if (exp->exp_obd == NULL) {
498 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
504 if (KEY_IS(KEY_CAPA_KEY)) {
505 rc = ofd_update_capa_key(ofd, val);
507 CERROR("%s: update capability key failed: rc = %d\n",
508 exp->exp_obd->obd_name, rc);
509 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
510 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
511 } else if (KEY_IS(KEY_MDS_CONN)) {
512 rc = ofd_set_mds_conn(exp, val);
513 } else if (KEY_IS(KEY_GRANT_SHRINK)) {
514 struct ost_body *body = val;
516 ofd_info_init(env, exp);
517 /** handle grant shrink, similar to a read request */
518 ofd_grant_prepare_read(env, exp, &body->oa);
520 CERROR("%s: Unsupported key %s\n",
521 exp->exp_obd->obd_name, (char*)key);
528 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
529 __u32 keylen, void *key, __u32 *vallen, void *val,
530 struct lov_stripe_md *lsm)
532 struct ofd_device *ofd;
537 if (exp->exp_obd == NULL) {
538 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
542 /* Because ofd_get_info might be called from
543 * handle_request_in as well(see LU-3239), where env might
544 * not be initilaized correctly, and le_ses might be in
545 * an un-initialized state, so only refill le_ctx here */
546 rc = lu_env_refill((struct lu_env *)env);
552 if (KEY_IS(KEY_BLOCKSIZE)) {
553 __u32 *blocksize = val;
555 if (*vallen < sizeof(*blocksize))
556 GOTO(out, rc = -EOVERFLOW);
557 *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
559 *vallen = sizeof(*blocksize);
560 } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
561 __u32 *blocksize_bits = val;
562 if (blocksize_bits) {
563 if (*vallen < sizeof(*blocksize_bits))
564 GOTO(out, rc = -EOVERFLOW);
565 *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
567 *vallen = sizeof(*blocksize_bits);
568 } else if (KEY_IS(KEY_LAST_ID)) {
569 obd_id *last_id = val;
570 struct ofd_seq *oseq;
573 *vallen = sizeof(obd_id);
576 ofd_info_init(env, exp);
577 oseq = ofd_seq_load(env, ofd,
578 (obd_seq)exp->exp_filter_data.fed_group);
579 LASSERT(!IS_ERR(oseq));
581 if (*vallen < sizeof(*last_id)) {
582 ofd_seq_put(env, oseq);
583 GOTO(out, rc = -EOVERFLOW);
585 *last_id = ofd_seq_last_oid(oseq);
587 ofd_seq_put(env, oseq);
588 *vallen = sizeof(*last_id);
589 } else if (KEY_IS(KEY_FIEMAP)) {
590 struct ofd_device *ofd = ofd_exp(exp);
591 struct ofd_object *fo;
592 struct ll_fiemap_info_key *fm_key = key;
596 *vallen = fiemap_count_to_size(
597 fm_key->fiemap.fm_extent_count);
601 rc = ostid_to_fid(&fid, &fm_key->oa.o_oi, 0);
604 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
607 fo = ofd_object_find(env, ofd, &fid);
609 CERROR("%s: error finding object "DFID"\n",
610 exp->exp_obd->obd_name, PFID(&fid));
613 struct ll_user_fiemap *fiemap = val;
615 ofd_read_lock(env, fo);
616 if (ofd_object_exists(fo)) {
617 *fiemap = fm_key->fiemap;
618 rc = dt_fiemap_get(env,
619 ofd_object_child(fo),
624 ofd_read_unlock(env, fo);
625 ofd_object_put(env, fo);
627 } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
628 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
629 *vallen = sizeof(__u32);
630 } else if (KEY_IS(KEY_LAST_FID)) {
631 struct ofd_device *ofd = ofd_exp(exp);
632 struct ofd_seq *oseq;
633 struct lu_fid *fid = val;
637 *vallen = sizeof(struct lu_fid);
641 if (*vallen < sizeof(*fid))
642 GOTO(out, rc = -EOVERFLOW);
644 ofd_info_init(env, exp);
646 fid_le_to_cpu(fid, fid);
648 oseq = ofd_seq_load(env, ofd,
649 ostid_seq((struct ost_id *)fid));
651 GOTO(out, rc = PTR_ERR(oseq));
653 rc = ostid_to_fid(fid, &oseq->os_oi,
654 ofd->ofd_lut.lut_lsd.lsd_osd_index);
658 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
660 *vallen = sizeof(*fid);
662 ofd_seq_put(env, oseq);
664 CERROR("Not supported key %s\n", (char*)key);
671 /** helper function for statfs, also used by grant code */
672 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
673 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
677 spin_lock(&ofd->ofd_osfs_lock);
678 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
681 /* statfs data are too old, get up-to-date one.
682 * we must be cautious here since multiple threads might be
683 * willing to update statfs data concurrently and we must
684 * grant that cached statfs data are always consistent */
686 if (ofd->ofd_statfs_inflight == 0)
687 /* clear inflight counter if no users, although it would
688 * take a while to overflow this 64-bit counter ... */
689 ofd->ofd_osfs_inflight = 0;
690 /* notify ofd_grant_commit() that we want to track writes
691 * completed as of now */
692 ofd->ofd_statfs_inflight++;
693 /* record value of inflight counter before running statfs to
694 * compute the diff once statfs is completed */
695 unstable = ofd->ofd_osfs_inflight;
696 spin_unlock(&ofd->ofd_osfs_lock);
698 /* statfs can sleep ... hopefully not for too long since we can
699 * call it fairly often as space fills up */
700 rc = dt_statfs(env, ofd->ofd_osd, osfs);
704 spin_lock(&ofd->ofd_grant_lock);
705 spin_lock(&ofd->ofd_osfs_lock);
706 /* calculate how much space was written while we released the
708 unstable = ofd->ofd_osfs_inflight - unstable;
709 ofd->ofd_osfs_unstable = 0;
711 /* some writes completed while we were running statfs
712 * w/o the ofd_osfs_lock. Those ones got added to
713 * the cached statfs data that we are about to crunch.
714 * Take them into account in the new statfs data */
715 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
716 unstable >> ofd->ofd_blockbits);
717 /* However, we don't really know if those writes got
718 * accounted in the statfs call, so tell
719 * ofd_grant_space_left() there is some uncertainty
720 * on the accounting of those writes.
721 * The purpose is to prevent spurious error messages in
722 * ofd_grant_space_left() since those writes might be
723 * accounted twice. */
724 ofd->ofd_osfs_unstable += unstable;
726 /* similarly, there is some uncertainty on write requests
727 * between prepare & commit */
728 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
729 spin_unlock(&ofd->ofd_grant_lock);
731 /* finally udpate cached statfs data */
732 ofd->ofd_osfs = *osfs;
733 ofd->ofd_osfs_age = cfs_time_current_64();
735 ofd->ofd_statfs_inflight--; /* stop tracking */
736 if (ofd->ofd_statfs_inflight == 0)
737 ofd->ofd_osfs_inflight = 0;
738 spin_unlock(&ofd->ofd_osfs_lock);
743 /* use cached statfs data */
744 *osfs = ofd->ofd_osfs;
745 spin_unlock(&ofd->ofd_osfs_lock);
752 static int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
753 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
755 struct obd_device *obd = class_exp2obd(exp);
756 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
761 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
765 /* at least try to account for cached pages. its still racy and
766 * might be under-reporting if clients haven't announced their
767 * caches with brw recently */
769 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
770 " pending "LPU64" free "LPU64" avail "LPU64"\n",
771 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
772 osfs->os_bfree << ofd->ofd_blockbits,
773 osfs->os_bavail << ofd->ofd_blockbits);
775 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
776 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
777 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
779 /* The QoS code on the MDS does not care about space reserved for
780 * precreate, so take it out. */
781 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
782 struct filter_export_data *fed;
784 fed = &obd->obd_self_export->exp_filter_data;
785 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
786 fed->fed_grant >> ofd->ofd_blockbits);
789 ofd_grant_sanity_check(obd, __FUNCTION__);
790 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
791 LPU64" objects: "LPU64" free; state %x\n",
792 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
793 osfs->os_files, osfs->os_ffree, osfs->os_state);
795 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
796 ofd->ofd_lut.lut_lsd.lsd_osd_index))
799 /* OS_STATE_READONLY can be set by OSD already */
800 if (ofd->ofd_raid_degraded)
801 osfs->os_state |= OS_STATE_DEGRADED;
803 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
804 /* clients which don't support OBD_CONNECT_GRANT_PARAM
805 * should not see a block size > page size, otherwise
806 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
807 * block size which is the biggest block size known to work
808 * with all client's page size. */
809 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
810 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
811 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
812 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
815 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
816 ofd->ofd_lut.lut_lsd.lsd_osd_index))
817 osfs->os_bfree = osfs->os_bavail = 2;
824 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
825 struct obd_info *oinfo, struct obd_trans_info *oti)
827 struct ofd_thread_info *info;
828 struct ofd_device *ofd = ofd_exp(exp);
829 struct ldlm_namespace *ns = ofd->ofd_namespace;
830 struct ldlm_resource *res;
831 struct ofd_object *fo;
832 struct obdo *oa = oinfo->oi_oa;
833 struct filter_fid *ff = NULL;
838 info = ofd_info_init(env, exp);
839 ofd_oti2info(info, oti);
841 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
845 ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
846 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
847 oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
851 /* This would be very bad - accidentally truncating a file when
852 * changing the time or similar - bug 12203. */
853 if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
854 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
855 static char mdsinum[48];
857 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
858 snprintf(mdsinum, sizeof(mdsinum) - 1,
859 "of parent "DFID, oinfo->oi_oa->o_parent_seq,
860 oinfo->oi_oa->o_parent_oid, 0);
864 CERROR("%s: setattr from %s trying to truncate object "DFID
865 " %s\n", exp->exp_obd->obd_name,
866 obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
867 GOTO(out, rc = -EPERM);
870 fo = ofd_object_find(env, ofd, &info->fti_fid);
872 CERROR("%s: can't find object "DFID"\n",
873 exp->exp_obd->obd_name, PFID(&info->fti_fid));
874 GOTO(out, rc = PTR_ERR(fo));
877 la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
878 info->fti_attr.la_valid &= ~LA_TYPE;
880 if (oa->o_valid & OBD_MD_FLFID) {
881 ff = &info->fti_mds_fid;
882 ofd_prepare_fidea(ff, oa);
885 /* setting objects attributes (including owner/group) */
886 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
888 GOTO(out_unlock, rc);
890 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
892 ldlm_res_lvbo_update(res, NULL, 0);
893 ldlm_resource_putref(res);
896 oinfo->oi_oa->o_valid = OBD_MD_FLID;
898 /* Quota release needs uid/gid info */
899 rc = ofd_attr_get(env, fo, &info->fti_attr);
900 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
901 OFD_VALID_FLAGS | LA_UID | LA_GID);
902 ofd_info2oti(info, oti);
904 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
907 ofd_object_put(env, fo);
912 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
913 struct obd_info *oinfo, struct obd_trans_info *oti,
914 struct ptlrpc_request_set *rqset)
916 struct ofd_thread_info *info;
917 struct ofd_device *ofd = ofd_exp(exp);
918 struct ldlm_namespace *ns = ofd->ofd_namespace;
919 struct ldlm_resource *res;
920 struct ofd_object *fo;
921 struct filter_fid *ff = NULL;
926 info = ofd_info_init(env, exp);
927 ofd_oti2info(info, oti);
929 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
932 ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
934 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
935 ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
936 oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
937 oinfo->oi_policy.l_extent.end);
939 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
940 oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
944 fo = ofd_object_find(env, ofd, &info->fti_fid);
946 CERROR("%s: error finding object "DFID": rc = %ld\n",
947 exp->exp_obd->obd_name, PFID(&info->fti_fid),
949 GOTO(out_env, rc = PTR_ERR(fo));
952 LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
953 if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
955 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
956 } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
957 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
960 la_from_obdo(&info->fti_attr, oinfo->oi_oa,
961 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
962 info->fti_attr.la_valid &= ~LA_TYPE;
963 info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
964 info->fti_attr.la_valid |= LA_SIZE;
966 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
967 ff = &info->fti_mds_fid;
968 ofd_prepare_fidea(ff, oinfo->oi_oa);
971 rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
972 oinfo->oi_policy.l_extent.end, &info->fti_attr,
977 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
979 ldlm_res_lvbo_update(res, NULL, 0);
980 ldlm_resource_putref(res);
983 oinfo->oi_oa->o_valid = OBD_MD_FLID;
984 /* Quota release needs uid/gid info */
985 rc = ofd_attr_get(env, fo, &info->fti_attr);
986 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
987 OFD_VALID_FLAGS | LA_UID | LA_GID);
988 ofd_info2oti(info, oti);
990 ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
993 ofd_object_put(env, fo);
998 static int ofd_destroy_by_fid(const struct lu_env *env,
999 struct ofd_device *ofd,
1000 const struct lu_fid *fid, int orphan)
1002 struct ofd_thread_info *info = ofd_info(env);
1003 struct lustre_handle lockh;
1004 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
1006 ldlm_policy_data_t policy = {
1007 .l_extent = { 0, OBD_OBJECT_EOF }
1009 struct ofd_object *fo;
1013 fo = ofd_object_find(env, ofd, fid);
1015 RETURN(PTR_ERR(fo));
1016 if (!ofd_object_exists(fo))
1017 GOTO(out, rc = -ENOENT);
1019 /* Tell the clients that the object is gone now and that they should
1020 * throw away any cached pages. */
1021 ost_fid_build_resid(fid, &info->fti_resid);
1022 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1023 LDLM_EXTENT, &policy, LCK_PW, &flags,
1024 ldlm_blocking_ast, ldlm_completion_ast,
1025 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1027 /* We only care about the side-effects, just drop the lock. */
1029 ldlm_lock_decref(&lockh, LCK_PW);
1031 LASSERT(fo != NULL);
1033 rc = ofd_object_destroy(env, fo, orphan);
1036 ofd_object_put(env, fo);
1040 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1041 struct obdo *oa, struct lov_stripe_md *md,
1042 struct obd_trans_info *oti, struct obd_export *md_exp,
1045 struct ofd_device *ofd = ofd_exp(exp);
1046 struct ofd_thread_info *info;
1052 info = ofd_info_init(env, exp);
1053 ofd_oti2info(info, oti);
1055 if (!(oa->o_valid & OBD_MD_FLGROUP))
1056 ostid_set_seq_mdt0(&oa->o_oi);
1058 /* check that o_misc makes sense */
1059 if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1062 count = 1; /* default case - single destroy */
1065 * There can be sequence of objects to destroy. Therefore this request
1066 * may have multiple transaction involved in. It is OK, we need only
1067 * the highest used transno to be reported back in reply but not for
1068 * replays, they must report their transno
1070 if (info->fti_transno == 0) /* not replay */
1071 info->fti_mult_trans = 1;
1073 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1074 POSTID(&oa->o_oi), count);
1078 lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
1084 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1085 if (lrc == -ENOENT) {
1087 "%s: destroying non-existent object "DFID"\n",
1088 ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1089 /* rewrite rc with -ENOENT only if it is 0 */
1092 } else if (lrc != 0) {
1093 CERROR("%s: error destroying object "DFID": %d\n",
1094 ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1099 ostid_inc_id(&oa->o_oi);
1102 /* if we have transaction then there were some deletions, we don't
1103 * need to return ENOENT in that case because it will not wait
1104 * for commit of these deletions. The ENOENT must be returned only
1105 * if there were no transations.
1107 if (rc == -ENOENT) {
1108 if (info->fti_transno != 0)
1110 } else if (rc != 0) {
1112 * If we have at least one transaction then llog record
1113 * on server will be removed upon commit, so for rc != 0
1114 * we return no transno and llog record will be reprocessed.
1116 info->fti_transno = 0;
1118 ofd_info2oti(info, oti);
1123 static int ofd_orphans_destroy(const struct lu_env *env,
1124 struct obd_export *exp, struct ofd_device *ofd,
1127 struct ofd_thread_info *info = ofd_info(env);
1131 struct ost_id oi = oa->o_oi;
1132 __u64 end_id = ostid_id(&oa->o_oi);
1133 struct ofd_seq *oseq;
1137 oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
1139 CERROR("%s: Can not find seq for "DOSTID"\n",
1140 ofd_name(ofd), POSTID(&oa->o_oi));
1144 LASSERT(exp != NULL);
1145 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1147 last = ofd_seq_last_oid(oseq);
1148 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1149 " to "DOSTID"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
1150 end_id + 1, ostid_seq(&oa->o_oi), last);
1152 for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
1153 ostid_dec_id(&oi)) {
1154 rc = ostid_to_fid(&info->fti_fid, &oi, 0);
1157 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1158 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1159 CEMERG("%s: error destroying precreated id "DOSTID
1160 ": rc = %d\n", ofd_name(ofd), POSTID(&oi), rc);
1162 ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
1163 /* update last_id on disk periodically so that if we
1164 * restart * we don't need to re-scan all of the just
1165 * deleted objects. */
1166 if ((ostid_id(&oi) & 511) == 0)
1167 ofd_seq_last_oid_write(env, ofd, oseq);
1170 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1171 ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
1173 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1175 /* don't reuse orphan object, return last used objid */
1176 ostid_set_id(&oa->o_oi, last);
1180 ofd_seq_put(env, oseq);
1184 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1185 struct obdo *oa, struct lov_stripe_md **ea,
1186 struct obd_trans_info *oti)
1188 struct ofd_device *ofd = ofd_exp(exp);
1189 struct ofd_thread_info *info;
1190 obd_seq seq = ostid_seq(&oa->o_oi);
1191 struct ofd_seq *oseq;
1197 info = ofd_info_init(env, exp);
1198 ofd_oti2info(info, oti);
1200 LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1201 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1203 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1205 oseq = ofd_seq_load(env, ofd, seq);
1207 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1208 ofd_name(ofd), seq, PTR_ERR(oseq));
1212 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1213 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1214 if (!ofd_obd(ofd)->obd_recovering ||
1215 ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1216 CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1217 POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1218 GOTO(out_nolock, rc = -EINVAL);
1220 /* do nothing because we create objects during first write */
1221 GOTO(out_nolock, rc = 0);
1223 /* former ofd_handle_precreate */
1224 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1225 (oa->o_flags & OBD_FL_DELORPHAN)) {
1226 /* destroy orphans */
1227 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1228 CERROR("%s: dropping old orphan cleanup request\n",
1230 GOTO(out_nolock, rc = 0);
1232 /* This causes inflight precreates to abort and drop lock */
1233 oseq->os_destroys_in_progress = 1;
1234 mutex_lock(&oseq->os_create_lock);
1235 if (!oseq->os_destroys_in_progress) {
1236 CERROR("%s:["LPU64"] destroys_in_progress already"
1237 " cleared\n", exp->exp_obd->obd_name,
1238 ostid_seq(&oa->o_oi));
1239 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1242 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1243 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1244 ofd_seq_last_oid(oseq), diff);
1245 if (-diff > OST_MAX_PRECREATE) {
1246 /* FIXME: should reset precreate_next_id on MDS */
1248 } else if (diff < 0) {
1249 rc = ofd_orphans_destroy(env, exp, ofd, oa);
1250 oseq->os_destroys_in_progress = 0;
1252 /* XXX: Used by MDS for the first time! */
1253 oseq->os_destroys_in_progress = 0;
1256 mutex_lock(&oseq->os_create_lock);
1257 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1258 CERROR("%s: dropping old precreate request\n",
1259 ofd_obd(ofd)->obd_name);
1262 /* only precreate if seq is 0, IDIF or normal and also o_id
1263 * must be specfied */
1264 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1265 !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1266 !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1267 ostid_id(&oa->o_oi) == 0) {
1268 diff = 1; /* shouldn't we create this right now? */
1270 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1271 /* Do sync create if the seq is about to used up */
1272 if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1273 fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1274 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1276 } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1277 if (unlikely(ostid_id(&oa->o_oi) >=
1278 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1281 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1282 ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1283 GOTO(out, rc = -EINVAL);
1288 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1293 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1294 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1295 /* don't enforce grant during orphan recovery */
1296 rc = ofd_grant_create(env,
1297 ofd_obd(ofd)->obd_self_export,
1300 CDEBUG(D_HA, "%s: failed to acquire grant "
1301 "space for precreate (%d): rc = %d\n",
1302 ofd_name(ofd), diff, rc);
1308 next_id = ofd_seq_last_oid(oseq) + 1;
1309 count = ofd_precreate_batch(ofd, diff);
1311 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1312 " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1313 count, ostid_seq(&oa->o_oi), next_id);
1315 if (cfs_time_after(jiffies, enough_time)) {
1316 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1317 " created at a rate of %d/s\n",
1318 ofd_obd(ofd)->obd_name,
1319 created, diff + created,
1320 created / DISK_TIMEOUT);
1324 rc = ofd_precreate_objects(env, ofd, next_id,
1325 oseq, count, sync_trans);
1329 } else if (rc < 0) {
1334 /* some objects got created, we can return
1335 * them, even if last creation failed */
1338 CERROR("%s: unable to precreate: rc = %d\n",
1341 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1342 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1344 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1345 !(oa->o_flags & OBD_FL_DELORPHAN))
1346 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1350 ofd_info2oti(info, oti);
1352 mutex_unlock(&oseq->os_create_lock);
1354 if (rc == 0 && ea != NULL) {
1355 struct lov_stripe_md *lsm = *ea;
1357 lsm->lsm_oi = oa->o_oi;
1359 ofd_seq_put(env, oseq);
1363 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1364 struct obd_info *oinfo)
1366 struct ofd_device *ofd = ofd_exp(exp);
1367 struct ofd_thread_info *info;
1368 struct ofd_object *fo;
1374 info = ofd_info_init(env, exp);
1376 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1379 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1380 oinfo_capa(oinfo), CAPA_OPC_META_READ);
1384 fo = ofd_object_find(env, ofd, &info->fti_fid);
1386 GOTO(out, rc = PTR_ERR(fo));
1387 LASSERT(fo != NULL);
1388 rc = ofd_attr_get(env, fo, &info->fti_attr);
1389 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1391 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1392 OFD_VALID_FLAGS | LA_UID | LA_GID);
1394 /* Store object version in reply */
1395 curr_version = dt_version_get(env, ofd_object_child(fo));
1396 if ((__s64)curr_version != -EOPNOTSUPP) {
1397 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1398 oinfo->oi_oa->o_data_version = curr_version;
1400 ofd_object_put(env, fo);
1405 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1406 struct obd_info *oinfo, obd_size start, obd_size end,
1407 struct ptlrpc_request_set *set)
1409 struct ofd_device *ofd = ofd_exp(exp);
1410 struct ofd_thread_info *info;
1411 struct ofd_object *fo;
1416 /* if no objid is specified, it means "sync whole filesystem" */
1417 if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1418 rc = dt_sync(env, ofd->ofd_osd);
1422 info = ofd_info_init(env, exp);
1423 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1427 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1428 oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1432 fo = ofd_object_find(env, ofd, &info->fti_fid);
1434 CERROR("%s: error finding object "DFID": rc = %ld\n",
1435 exp->exp_obd->obd_name, PFID(&info->fti_fid),
1437 GOTO(out, rc = PTR_ERR(fo));
1440 if (!ofd_object_exists(fo))
1441 GOTO(put, rc = -ENOENT);
1443 if (dt_version_get(env, ofd_object_child(fo)) >
1444 ofd_obd(ofd)->obd_last_committed) {
1445 rc = dt_object_sync(env, ofd_object_child(fo));
1450 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1451 rc = ofd_attr_get(env, fo, &info->fti_attr);
1452 obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1454 ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1457 ofd_object_put(env, fo);
1462 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1463 struct ofd_device *ofd, void *karg)
1465 struct obd_ioctl_data *data = karg;
1467 struct ofd_object *fo;
1468 dt_obj_version_t version;
1473 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1474 GOTO(out, rc = -EINVAL);
1476 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1477 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1478 } else if (data->ioc_inlbuf3 != NULL &&
1479 data->ioc_inllen3 == sizeof(__u64) &&
1480 data->ioc_inlbuf4 != NULL &&
1481 data->ioc_inllen4 == sizeof(__u64)) {
1482 struct ost_id ostid;
1484 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1485 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1486 rc = ostid_to_fid(&fid, &ostid, 0);
1490 GOTO(out, rc = -EINVAL);
1493 if (!fid_is_sane(&fid))
1494 GOTO(out, rc = -EINVAL);
1496 fo = ofd_object_find(env, ofd, &fid);
1498 GOTO(out, rc = PTR_ERR(fo));
1500 if (!ofd_object_exists(fo))
1501 GOTO(out_fo, rc = -ENOENT);
1503 if (lu_object_remote(&fo->ofo_obj.do_lu))
1504 GOTO(out_fo, rc = -EREMOTE);
1506 version = dt_version_get(env, ofd_object_child(fo));
1508 GOTO(out_fo, rc = -EIO);
1510 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1514 ofd_object_put(env, fo);
1519 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1520 void *karg, void *uarg)
1523 struct ofd_device *ofd = ofd_exp(exp);
1524 struct obd_device *obd = ofd_obd(ofd);
1529 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1530 rc = lu_env_init(&env, LCT_DT_THREAD);
1535 case OBD_IOC_ABORT_RECOVERY:
1536 CERROR("%s: aborting recovery\n", obd->obd_name);
1537 target_stop_recovery_thread(obd);
1540 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1541 rc = dt_sync(&env, ofd->ofd_osd);
1543 case OBD_IOC_SET_READONLY:
1544 rc = dt_sync(&env, ofd->ofd_osd);
1546 rc = dt_ro(&env, ofd->ofd_osd);
1548 case OBD_IOC_GET_OBJ_VERSION:
1549 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1552 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1560 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1567 case OBD_CLEANUP_EARLY:
1569 case OBD_CLEANUP_EXPORTS:
1570 target_cleanup_recovery(obd);
1576 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1578 ofd_fmd_expire(exp);
1582 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1584 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1585 struct ofd_thread_info *info;
1587 #ifdef USE_HEALTH_CHECK_WRITE
1592 /* obd_proc_read_health pass NULL env, we need real one */
1593 rc = lu_env_init(&env, LCT_DT_THREAD);
1597 info = ofd_info_init(&env, NULL);
1598 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1602 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1603 GOTO(out, rc = -EROFS);
1605 #ifdef USE_HEALTH_CHECK_WRITE
1606 OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1607 if (info->fti_buf.lb_buf == NULL)
1608 GOTO(out, rc = -ENOMEM);
1610 info->fti_buf.lb_len = CFS_PAGE_SIZE;
1613 th = dt_trans_create(&env, ofd->ofd_osd);
1615 GOTO(out, rc = PTR_ERR(th));
1617 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1618 info->fti_buf.lb_len, info->fti_off, th);
1620 th->th_sync = 1; /* sync IO is needed */
1621 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1623 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1624 &info->fti_buf, &info->fti_off,
1627 dt_trans_stop(&env, ofd->ofd_osd, th);
1629 OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1631 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1639 * Handle quota control requests to consult current usage/limit.
1641 * \param obd - is the obd device associated with the ofd
1642 * \param exp - is the client's export
1643 * \param oqctl - is the obd_quotactl request to be processed
1645 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1646 struct obd_quotactl *oqctl)
1648 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1653 /* report success for quota on/off for interoperability with current MDT
1655 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1658 rc = lu_env_init(&env, LCT_DT_THREAD);
1662 rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1668 struct obd_ops ofd_obd_ops = {
1669 .o_owner = THIS_MODULE,
1670 .o_connect = ofd_obd_connect,
1671 .o_reconnect = ofd_obd_reconnect,
1672 .o_disconnect = ofd_obd_disconnect,
1673 .o_set_info_async = ofd_set_info_async,
1674 .o_get_info = ofd_get_info,
1675 .o_create = ofd_create,
1676 .o_statfs = ofd_statfs,
1677 .o_setattr = ofd_setattr,
1678 .o_preprw = ofd_preprw,
1679 .o_commitrw = ofd_commitrw,
1680 .o_destroy = ofd_destroy,
1681 .o_init_export = ofd_init_export,
1682 .o_destroy_export = ofd_destroy_export,
1683 .o_postrecov = ofd_obd_postrecov,
1684 .o_punch = ofd_punch,
1685 .o_getattr = ofd_getattr,
1687 .o_iocontrol = ofd_iocontrol,
1688 .o_precleanup = ofd_precleanup,
1690 .o_health_check = ofd_health_check,
1691 .o_quotactl = ofd_quotactl,