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, 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 "ofd_internal.h"
46 #include <obd_cksum.h>
47 #include <lustre_quota.h>
49 static int ofd_export_stats_init(struct ofd_device *ofd,
50 struct obd_export *exp, void *client_nid)
52 struct obd_device *obd = ofd_obd(ofd);
53 struct nid_stat *stats;
59 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
60 /* Self-export gets no proc entry */
63 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
65 /* Mask error for already created
75 stats = exp->exp_nid_stats;
76 LASSERT(stats != NULL);
78 OBD_ALLOC(stats->nid_brw_stats, sizeof(struct brw_stats));
79 if (stats->nid_brw_stats == NULL)
80 GOTO(clean, rc = -ENOMEM);
82 for (i = 0; i < BRW_LAST; i++)
83 spin_lock_init(&stats->nid_brw_stats->hist[i].oh_lock);
85 rc = lprocfs_seq_create(stats->nid_proc, "brw_stats", 0644,
86 &ofd_per_nid_stats_fops, stats);
88 CWARN("Error adding the brw_stats file\n");
90 num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
93 stats->nid_stats = lprocfs_alloc_stats(num_stats,
94 LPROCFS_STATS_FLAG_NOPERCPU);
95 if (stats->nid_stats == NULL)
98 lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
99 lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
100 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
101 lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
102 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
104 rc = lprocfs_register_stats(stats->nid_proc, "stats",
109 rc = lprocfs_nid_ldlm_stats_init(stats);
111 lprocfs_free_stats(&stats->nid_stats);
120 static int ofd_parse_connect_data(const struct lu_env *env,
121 struct obd_export *exp,
122 struct obd_connect_data *data)
124 struct ofd_device *ofd = ofd_exp(exp);
125 struct filter_export_data *fed = &exp->exp_filter_data;
130 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
131 " ocd_version: %x ocd_grant: %d ocd_index: %u"
133 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
134 data->ocd_connect_flags, data->ocd_version,
135 data->ocd_grant, data->ocd_index, data->ocd_group);
137 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
138 CWARN("!!! This export (nid %s) used object group %d "
139 "earlier; now it's trying to use group %d! This could "
140 "be a bug in the MDS. Please report to "
141 "http://bugs.whamcloud.com/\n",
142 obd_export_nid2str(exp), fed->fed_group,
146 fed->fed_group = data->ocd_group;
148 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
149 exp->exp_connect_data = *data;
150 data->ocd_version = LUSTRE_VERSION_CODE;
152 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
153 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
154 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
155 exp->exp_obd->obd_name, data->ocd_group);
156 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
159 if (ofd_grant_param_supp(exp)) {
160 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
161 /* ocd_{blocksize,inodespace} are log2 values */
162 data->ocd_blocksize = ofd->ofd_blockbits;
163 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
164 /* ocd_grant_extent is in 1K blocks */
165 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
168 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
169 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant);
171 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
172 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
173 int index = lsd->lsd_osd_index;
175 if (index != data->ocd_index) {
176 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
177 " %u doesn't match actual OST index"
178 " %u in last_rcvd file, bad "
180 obd_export_nid2str(exp), index,
184 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
185 /* this will only happen on the first connect */
186 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
187 /* sync is not needed here as lut_client_add will
188 * set exp_need_sync flag */
189 tgt_server_data_update(env, &ofd->ofd_lut, 0);
192 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
193 data->ocd_brw_size = 65536;
194 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
195 data->ocd_brw_size = min(data->ocd_brw_size,
196 (__u32)DT_MAX_BRW_SIZE);
197 if (data->ocd_brw_size == 0) {
198 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
199 " ocd_version: %x ocd_grant: %d ocd_index: %u "
200 "ocd_brw_size is unexpectedly zero, "
201 "network data corruption?"
202 "Refusing connection of this client\n",
203 exp->exp_obd->obd_name,
204 exp->exp_client_uuid.uuid,
205 exp, data->ocd_connect_flags, data->ocd_version,
206 data->ocd_grant, data->ocd_index);
211 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
212 __u32 cksum_types = data->ocd_cksum_types;
214 /* The client set in ocd_cksum_types the checksum types it
215 * supports. We have to mask off the algorithms that we don't
217 data->ocd_cksum_types &= cksum_types_supported_server();
219 if (unlikely(data->ocd_cksum_types == 0)) {
220 CERROR("%s: Connect with checksum support but no "
221 "ocd_cksum_types is set\n",
222 exp->exp_obd->obd_name);
226 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
227 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
228 cksum_types, data->ocd_cksum_types);
230 /* This client does not support OBD_CONNECT_CKSUM
231 * fall back to CRC32 */
232 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
233 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
234 exp->exp_obd->obd_name, obd_export_nid2str(exp));
237 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
238 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
240 if (data->ocd_connect_flags & OBD_CONNECT_PINGLESS) {
241 if (suppress_pings) {
242 spin_lock(&exp->exp_obd->obd_dev_lock);
243 list_del_init(&exp->exp_obd_chain_timed);
244 spin_unlock(&exp->exp_obd->obd_dev_lock);
246 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
253 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
254 struct obd_device *obd, struct obd_uuid *cluuid,
255 struct obd_connect_data *data, void *localdata)
257 struct ofd_device *ofd;
262 if (exp == NULL || obd == NULL || cluuid == NULL)
265 ofd = ofd_dev(obd->obd_lu_dev);
267 rc = lu_env_refill((struct lu_env *)env);
269 CERROR("Failure to refill session: '%d'\n", rc);
273 ofd_info_init(env, exp);
274 rc = ofd_parse_connect_data(env, exp, data);
276 ofd_export_stats_init(ofd, exp, localdata);
281 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
282 struct obd_device *obd, struct obd_uuid *cluuid,
283 struct obd_connect_data *data, void *localdata)
285 struct obd_export *exp;
286 struct ofd_device *ofd;
287 struct lustre_handle conn = { 0 };
291 if (_exp == NULL || obd == NULL || cluuid == NULL)
294 ofd = ofd_dev(obd->obd_lu_dev);
296 rc = class_connect(&conn, obd, cluuid);
300 exp = class_conn2export(&conn);
301 LASSERT(exp != NULL);
303 rc = lu_env_refill((struct lu_env *)env);
305 CERROR("Failure to refill session: '%d'\n", rc);
309 ofd_info_init(env, exp);
311 rc = ofd_parse_connect_data(env, exp, data);
315 if (obd->obd_replayable) {
316 struct tg_export_data *ted = &exp->exp_target_data;
318 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
319 sizeof(ted->ted_lcd->lcd_uuid));
320 rc = tgt_client_new(env, exp);
323 ofd_export_stats_init(ofd, exp, localdata);
326 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
331 class_disconnect(exp);
339 static int ofd_obd_disconnect(struct obd_export *exp)
341 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
348 class_export_get(exp);
350 if (!(exp->exp_flags & OBD_OPT_FORCE))
351 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
353 rc = server_disconnect_export(exp);
355 ofd_grant_discard(exp);
357 rc = lu_env_init(&env, LCT_DT_THREAD);
361 /* Do not erase record for recoverable client. */
362 if (exp->exp_obd->obd_replayable &&
363 (!exp->exp_obd->obd_fail || exp->exp_failed))
364 tgt_client_del(&env, exp);
367 class_export_put(exp);
371 static int ofd_init_export(struct obd_export *exp)
375 spin_lock_init(&exp->exp_filter_data.fed_lock);
376 CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
377 spin_lock(&exp->exp_lock);
378 exp->exp_connecting = 1;
379 spin_unlock(&exp->exp_lock);
381 /* self-export doesn't need client data and ldlm initialization */
382 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
383 &exp->exp_client_uuid)))
386 rc = tgt_client_alloc(exp);
388 ldlm_init_export(exp);
390 CERROR("%s: Can't initialize export: rc %d\n",
391 exp->exp_obd->obd_name, rc);
395 static int ofd_destroy_export(struct obd_export *exp)
397 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
399 if (exp->exp_filter_data.fed_pending)
400 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
401 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
402 exp, exp->exp_filter_data.fed_pending);
404 target_destroy_export(exp);
406 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
407 &exp->exp_client_uuid)))
410 ldlm_destroy_export(exp);
411 tgt_client_free(exp);
413 ofd_fmd_cleanup(exp);
416 * discard grants once we're sure no more
417 * interaction with the client is possible
419 ofd_grant_discard(exp);
420 ofd_fmd_cleanup(exp);
422 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
423 if (ofd->ofd_tot_granted_clients > 0)
424 ofd->ofd_tot_granted_clients --;
427 if (!(exp->exp_flags & OBD_OPT_FORCE))
428 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
430 LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
434 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
436 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
438 CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
439 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
442 int ofd_obd_postrecov(struct obd_device *obd)
445 struct lu_device *ldev = obd->obd_lu_dev;
450 rc = lu_env_init(&env, LCT_DT_THREAD);
453 ofd_info_init(&env, obd->obd_self_export);
455 rc = ofd_postrecov(&env, ofd_dev(ldev));
461 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
462 struct obd_device *obd, int initial)
464 struct filter_obd *fo = &obd->u.filter;
465 struct sptlrpc_rule_set tmp_rset;
468 sptlrpc_rule_set_init(&tmp_rset);
469 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
471 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
476 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
478 write_lock(&fo->fo_sptlrpc_lock);
479 sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
480 fo->fo_sptlrpc_rset = tmp_rset;
481 write_unlock(&fo->fo_sptlrpc_lock);
486 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
492 LCONSOLE_WARN("%s: received MDS connection from %s\n",
493 exp->exp_obd->obd_name, obd_export_nid2str(exp));
497 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
498 __u32 keylen, void *key, __u32 vallen, void *val,
499 struct ptlrpc_request_set *set)
501 struct ofd_device *ofd;
506 if (exp->exp_obd == NULL) {
507 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
513 if (KEY_IS(KEY_CAPA_KEY)) {
514 rc = ofd_update_capa_key(ofd, val);
516 CERROR("%s: update capability key failed: rc = %d\n",
517 exp->exp_obd->obd_name, rc);
518 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
519 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
520 } else if (KEY_IS(KEY_MDS_CONN)) {
521 rc = ofd_set_mds_conn(exp, val);
522 } else if (KEY_IS(KEY_GRANT_SHRINK)) {
523 struct ost_body *body = val;
525 ofd_info_init(env, exp);
526 /** handle grant shrink, similar to a read request */
527 ofd_grant_prepare_read(env, exp, &body->oa);
529 CERROR("%s: Unsupported key %s\n",
530 exp->exp_obd->obd_name, (char*)key);
537 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
538 __u32 keylen, void *key, __u32 *vallen, void *val,
539 struct lov_stripe_md *lsm)
541 struct ofd_device *ofd;
546 if (exp->exp_obd == NULL) {
547 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
553 if (KEY_IS(KEY_BLOCKSIZE)) {
554 __u32 *blocksize = val;
556 if (*vallen < sizeof(*blocksize))
558 *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
560 *vallen = sizeof(*blocksize);
561 } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
562 __u32 *blocksize_bits = val;
563 if (blocksize_bits) {
564 if (*vallen < sizeof(*blocksize_bits))
566 *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
568 *vallen = sizeof(*blocksize_bits);
569 } else if (KEY_IS(KEY_LAST_ID)) {
570 obd_id *last_id = val;
571 struct ofd_seq *oseq;
574 *vallen = sizeof(obd_id);
577 ofd_info_init(env, exp);
578 oseq = ofd_seq_load(env, ofd,
579 (obd_seq)exp->exp_filter_data.fed_group);
580 LASSERT(!IS_ERR(oseq));
582 if (*vallen < sizeof(*last_id)) {
583 ofd_seq_put(env, oseq);
586 *last_id = ofd_seq_last_oid(oseq);
588 ofd_seq_put(env, oseq);
589 *vallen = sizeof(*last_id);
590 } else if (KEY_IS(KEY_FIEMAP)) {
591 struct ofd_thread_info *info;
592 struct ofd_device *ofd = ofd_exp(exp);
593 struct ofd_object *fo;
594 struct ll_fiemap_info_key *fm_key = key;
597 *vallen = fiemap_count_to_size(
598 fm_key->fiemap.fm_extent_count);
602 info = ofd_info_init(env, exp);
604 fid_ostid_unpack(&info->fti_fid, &fm_key->oa.o_oi, 0);
606 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
607 PFID(&info->fti_fid));
609 fo = ofd_object_find(env, ofd, &info->fti_fid);
611 CERROR("%s: error finding object "DFID"\n",
612 exp->exp_obd->obd_name, PFID(&info->fti_fid));
615 struct ll_user_fiemap *fiemap = val;
617 ofd_read_lock(env, fo);
618 if (ofd_object_exists(fo)) {
619 *fiemap = fm_key->fiemap;
620 rc = dt_fiemap_get(env,
621 ofd_object_child(fo),
626 ofd_read_unlock(env, fo);
627 ofd_object_put(env, fo);
629 } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
630 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
631 *vallen = sizeof(__u32);
632 } else if (KEY_IS(KEY_LAST_FID)) {
634 struct ofd_device *ofd = ofd_exp(exp);
635 struct ofd_seq *oseq;
636 struct lu_fid *last_fid = val;
639 if (last_fid == NULL) {
640 *vallen = sizeof(struct lu_fid);
644 if (*vallen < sizeof(*last_fid))
647 rc = lu_env_init(&env, LCT_DT_THREAD);
650 ofd_info_init(&env, exp);
651 fid_le_to_cpu(last_fid, last_fid);
652 oseq = ofd_seq_load(&env, ofd, fid_seq(last_fid));
654 GOTO(out_fid, rc = PTR_ERR(oseq));
656 last_fid->f_seq = oseq->os_seq;
657 last_fid->f_oid = oseq->os_last_oid;
658 fid_cpu_to_le(last_fid, last_fid);
660 *vallen = sizeof(*last_fid);
661 ofd_seq_put(&env, oseq);
665 CERROR("Not supported key %s\n", (char*)key);
672 /** helper function for statfs, also used by grant code */
673 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
674 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
678 spin_lock(&ofd->ofd_osfs_lock);
679 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
682 /* statfs data are too old, get up-to-date one.
683 * we must be cautious here since multiple threads might be
684 * willing to update statfs data concurrently and we must
685 * grant that cached statfs data are always consistent */
687 if (ofd->ofd_statfs_inflight == 0)
688 /* clear inflight counter if no users, although it would
689 * take a while to overflow this 64-bit counter ... */
690 ofd->ofd_osfs_inflight = 0;
691 /* notify ofd_grant_commit() that we want to track writes
692 * completed as of now */
693 ofd->ofd_statfs_inflight++;
694 /* record value of inflight counter before running statfs to
695 * compute the diff once statfs is completed */
696 unstable = ofd->ofd_osfs_inflight;
697 spin_unlock(&ofd->ofd_osfs_lock);
699 /* statfs can sleep ... hopefully not for too long since we can
700 * call it fairly often as space fills up */
701 rc = dt_statfs(env, ofd->ofd_osd, osfs);
705 spin_lock(&ofd->ofd_grant_lock);
706 spin_lock(&ofd->ofd_osfs_lock);
707 /* calculate how much space was written while we released the
709 unstable = ofd->ofd_osfs_inflight - unstable;
710 ofd->ofd_osfs_unstable = 0;
712 /* some writes completed while we were running statfs
713 * w/o the ofd_osfs_lock. Those ones got added to
714 * the cached statfs data that we are about to crunch.
715 * Take them into account in the new statfs data */
716 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
717 unstable >> ofd->ofd_blockbits);
718 /* However, we don't really know if those writes got
719 * accounted in the statfs call, so tell
720 * ofd_grant_space_left() there is some uncertainty
721 * on the accounting of those writes.
722 * The purpose is to prevent spurious error messages in
723 * ofd_grant_space_left() since those writes might be
724 * accounted twice. */
725 ofd->ofd_osfs_unstable += unstable;
727 /* similarly, there is some uncertainty on write requests
728 * between prepare & commit */
729 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
730 spin_unlock(&ofd->ofd_grant_lock);
732 /* finally udpate cached statfs data */
733 ofd->ofd_osfs = *osfs;
734 ofd->ofd_osfs_age = cfs_time_current_64();
736 ofd->ofd_statfs_inflight--; /* stop tracking */
737 if (ofd->ofd_statfs_inflight == 0)
738 ofd->ofd_osfs_inflight = 0;
739 spin_unlock(&ofd->ofd_osfs_lock);
744 /* use cached statfs data */
745 *osfs = ofd->ofd_osfs;
746 spin_unlock(&ofd->ofd_osfs_lock);
753 static int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
754 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
756 struct obd_device *obd = class_exp2obd(exp);
757 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
762 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
766 /* at least try to account for cached pages. its still racy and
767 * might be under-reporting if clients haven't announced their
768 * caches with brw recently */
770 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
771 " pending "LPU64" free "LPU64" avail "LPU64"\n",
772 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
773 osfs->os_bfree << ofd->ofd_blockbits,
774 osfs->os_bavail << ofd->ofd_blockbits);
776 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
777 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
778 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
780 /* The QoS code on the MDS does not care about space reserved for
781 * precreate, so take it out. */
782 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
783 struct filter_export_data *fed;
785 fed = &obd->obd_self_export->exp_filter_data;
786 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
787 fed->fed_grant >> ofd->ofd_blockbits);
790 ofd_grant_sanity_check(obd, __FUNCTION__);
791 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
792 LPU64" objects: "LPU64" free; state %x\n",
793 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
794 osfs->os_files, osfs->os_ffree, osfs->os_state);
796 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
797 ofd->ofd_lut.lut_lsd.lsd_osd_index))
800 /* OS_STATE_READONLY can be set by OSD already */
801 if (ofd->ofd_raid_degraded)
802 osfs->os_state |= OS_STATE_DEGRADED;
804 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
805 /* clients which don't support OBD_CONNECT_GRANT_PARAM
806 * should not see a block size > page size, otherwise
807 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
808 * block size which is the biggest block size known to work
809 * with all client's page size. */
810 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
811 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
812 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
813 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
816 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
817 ofd->ofd_lut.lut_lsd.lsd_osd_index))
818 osfs->os_bfree = osfs->os_bavail = 2;
825 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
826 struct obd_info *oinfo, struct obd_trans_info *oti)
828 struct ofd_thread_info *info;
829 struct ofd_device *ofd = ofd_exp(exp);
830 struct ldlm_namespace *ns = ofd->ofd_namespace;
831 struct ldlm_resource *res;
832 struct ofd_object *fo;
833 struct obdo *oa = oinfo->oi_oa;
834 struct filter_fid *ff = NULL;
839 info = ofd_info_init(env, exp);
840 ofd_oti2info(info, oti);
842 fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
843 ofd_build_resid(&info->fti_fid, &info->fti_resid);
845 rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq,
846 oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
850 /* This would be very bad - accidentally truncating a file when
851 * changing the time or similar - bug 12203. */
852 if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
853 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
854 static char mdsinum[48];
856 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
857 snprintf(mdsinum, sizeof(mdsinum) - 1,
858 "of parent "DFID, oinfo->oi_oa->o_parent_seq,
859 oinfo->oi_oa->o_parent_oid, 0);
863 CERROR("%s: setattr from %s trying to truncate object "DFID
864 " %s\n", exp->exp_obd->obd_name,
865 obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
866 GOTO(out, rc = -EPERM);
869 fo = ofd_object_find(env, ofd, &info->fti_fid);
871 CERROR("%s: can't find object "DFID"\n",
872 exp->exp_obd->obd_name, PFID(&info->fti_fid));
873 GOTO(out, rc = PTR_ERR(fo));
876 la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
877 info->fti_attr.la_valid &= ~LA_TYPE;
879 if (oa->o_valid & OBD_MD_FLFID) {
880 ff = &info->fti_mds_fid;
881 ofd_prepare_fidea(ff, oa);
884 /* setting objects attributes (including owner/group) */
885 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
887 GOTO(out_unlock, rc);
889 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
891 ldlm_res_lvbo_update(res, NULL, 0);
892 ldlm_resource_putref(res);
895 oinfo->oi_oa->o_valid = OBD_MD_FLID;
897 /* Quota release needs uid/gid info */
898 rc = ofd_attr_get(env, fo, &info->fti_attr);
899 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
900 OFD_VALID_FLAGS | LA_UID | LA_GID);
901 ofd_info2oti(info, oti);
903 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
906 ofd_object_put(env, fo);
911 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
912 struct obd_info *oinfo, struct obd_trans_info *oti,
913 struct ptlrpc_request_set *rqset)
915 struct ofd_thread_info *info;
916 struct ofd_device *ofd = ofd_exp(exp);
917 struct ldlm_namespace *ns = ofd->ofd_namespace;
918 struct ldlm_resource *res;
919 struct ofd_object *fo;
920 struct filter_fid *ff = NULL;
925 info = ofd_info_init(env, exp);
926 ofd_oti2info(info, oti);
928 fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
929 ofd_build_resid(&info->fti_fid, &info->fti_resid);
931 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
932 ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
933 oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
934 oinfo->oi_policy.l_extent.end);
936 rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
937 oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
941 fo = ofd_object_find(env, ofd, &info->fti_fid);
943 CERROR("%s: error finding object "DFID": rc = %ld\n",
944 exp->exp_obd->obd_name, PFID(&info->fti_fid),
946 GOTO(out_env, rc = PTR_ERR(fo));
949 LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
950 if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
952 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
953 } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
954 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
957 la_from_obdo(&info->fti_attr, oinfo->oi_oa,
958 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
959 info->fti_attr.la_valid &= ~LA_TYPE;
960 info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
961 info->fti_attr.la_valid |= LA_SIZE;
963 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
964 ff = &info->fti_mds_fid;
965 ofd_prepare_fidea(ff, oinfo->oi_oa);
968 rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
969 oinfo->oi_policy.l_extent.end, &info->fti_attr,
974 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
976 ldlm_res_lvbo_update(res, NULL, 0);
977 ldlm_resource_putref(res);
980 oinfo->oi_oa->o_valid = OBD_MD_FLID;
981 /* Quota release needs uid/gid info */
982 rc = ofd_attr_get(env, fo, &info->fti_attr);
983 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
984 OFD_VALID_FLAGS | LA_UID | LA_GID);
985 ofd_info2oti(info, oti);
987 ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
990 ofd_object_put(env, fo);
995 static int ofd_destroy_by_fid(const struct lu_env *env,
996 struct ofd_device *ofd,
997 const struct lu_fid *fid, int orphan)
999 struct ofd_thread_info *info = ofd_info(env);
1000 struct lustre_handle lockh;
1001 __u64 flags = LDLM_AST_DISCARD_DATA, rc = 0;
1002 ldlm_policy_data_t policy = {
1003 .l_extent = { 0, OBD_OBJECT_EOF }
1005 struct ofd_object *fo;
1009 fo = ofd_object_find(env, ofd, fid);
1011 RETURN(PTR_ERR(fo));
1013 /* Tell the clients that the object is gone now and that they should
1014 * throw away any cached pages. */
1015 ofd_build_resid(fid, &info->fti_resid);
1016 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1017 LDLM_EXTENT, &policy, LCK_PW, &flags,
1018 ldlm_blocking_ast, ldlm_completion_ast,
1019 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1021 /* We only care about the side-effects, just drop the lock. */
1023 ldlm_lock_decref(&lockh, LCK_PW);
1025 LASSERT(fo != NULL);
1027 rc = ofd_object_destroy(env, fo, orphan);
1029 ofd_object_put(env, fo);
1033 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1034 struct obdo *oa, struct lov_stripe_md *md,
1035 struct obd_trans_info *oti, struct obd_export *md_exp,
1038 struct ofd_device *ofd = ofd_exp(exp);
1039 struct ofd_thread_info *info;
1045 info = ofd_info_init(env, exp);
1046 ofd_oti2info(info, oti);
1048 if (!(oa->o_valid & OBD_MD_FLGROUP))
1051 /* check that o_misc makes sense */
1052 if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1055 count = 1; /* default case - single destroy */
1058 * There can be sequence of objects to destroy. Therefore this request
1059 * may have multiple transaction involved in. It is OK, we need only
1060 * the highest used transno to be reported back in reply but not for
1061 * replays, they must report their transno
1063 if (info->fti_transno == 0) /* not replay */
1064 info->fti_mult_trans = 1;
1068 fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
1069 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1070 if (lrc == -ENOENT) {
1072 "%s: destroying non-existent object "DFID"\n",
1073 ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1074 /* rewrite rc with -ENOENT only if it is 0 */
1077 } else if (lrc != 0) {
1078 CERROR("%s: error destroying object "DFID": %d\n",
1079 ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1087 /* if we have transaction then there were some deletions, we don't
1088 * need to return ENOENT in that case because it will not wait
1089 * for commit of these deletions. The ENOENT must be returned only
1090 * if there were no transations.
1092 if (rc == -ENOENT) {
1093 if (info->fti_transno != 0)
1095 } else if (rc != 0) {
1097 * If we have at least one transaction then llog record
1098 * on server will be removed upon commit, so for rc != 0
1099 * we return no transno and llog record will be reprocessed.
1101 info->fti_transno = 0;
1103 ofd_info2oti(info, oti);
1107 static int ofd_orphans_destroy(const struct lu_env *env,
1108 struct obd_export *exp, struct ofd_device *ofd,
1111 struct ofd_thread_info *info = ofd_info(env);
1115 struct ost_id oi = oa->o_oi;
1116 struct ofd_seq *oseq;
1120 oseq = ofd_seq_get(ofd, oa->o_seq);
1122 CERROR("%s: Can not find seq for "LPU64":"LPU64"\n",
1123 ofd_name(ofd), oa->o_seq, oa->o_id);
1127 LASSERT(exp != NULL);
1128 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1130 last = ofd_seq_last_oid(oseq);
1131 LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1132 " to "LPU64"\n", ofd_name(ofd), oa->o_seq,
1133 oa->o_id + 1, last);
1135 for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
1136 fid_ostid_unpack(&info->fti_fid, &oi, 0);
1137 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1138 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1139 CEMERG("error destroying precreated id "LPU64": %d\n",
1142 ofd_seq_last_oid_set(oseq, oi.oi_id - 1);
1143 /* update last_id on disk periodically so that if we
1144 * restart * we don't need to re-scan all of the just
1145 * deleted objects. */
1146 if ((oi.oi_id & 511) == 0)
1147 ofd_seq_last_oid_write(env, ofd, oseq);
1150 CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
1151 ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
1153 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1155 /* don't reuse orphan object, return last used objid */
1159 ofd_seq_put(env, oseq);
1163 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1164 struct obdo *oa, struct lov_stripe_md **ea,
1165 struct obd_trans_info *oti)
1167 struct ofd_device *ofd = ofd_exp(exp);
1168 struct ofd_thread_info *info;
1169 obd_seq seq = oa->o_seq;
1170 struct ofd_seq *oseq;
1176 info = ofd_info_init(env, exp);
1177 ofd_oti2info(info, oti);
1179 LASSERT(oa->o_seq >= FID_SEQ_OST_MDT0);
1180 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1182 CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
1185 oseq = ofd_seq_load(env, ofd, seq);
1187 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1188 ofd_name(ofd), seq, PTR_ERR(oseq));
1192 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1193 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1194 if (!ofd_obd(ofd)->obd_recovering ||
1195 oa->o_id > ofd_seq_last_oid(oseq)) {
1196 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1197 oa->o_id, ofd_seq_last_oid(oseq));
1198 GOTO(out_nolock, rc = -EINVAL);
1200 /* do nothing because we create objects during first write */
1201 GOTO(out_nolock, rc = 0);
1203 /* former ofd_handle_precreate */
1204 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1205 (oa->o_flags & OBD_FL_DELORPHAN)) {
1206 /* destroy orphans */
1207 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1208 CERROR("%s: dropping old orphan cleanup request\n",
1210 GOTO(out_nolock, rc = 0);
1212 /* This causes inflight precreates to abort and drop lock */
1213 oseq->os_destroys_in_progress = 1;
1214 mutex_lock(&oseq->os_create_lock);
1215 if (!oseq->os_destroys_in_progress) {
1216 CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
1217 exp->exp_obd->obd_name, oa->o_seq);
1220 diff = oa->o_id - ofd_seq_last_oid(oseq);
1221 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1222 ofd_seq_last_oid(oseq), diff);
1223 if (-diff > OST_MAX_PRECREATE) {
1224 /* FIXME: should reset precreate_next_id on MDS */
1226 } else if (diff < 0) {
1227 rc = ofd_orphans_destroy(env, exp, ofd, oa);
1228 oseq->os_destroys_in_progress = 0;
1230 /* XXX: Used by MDS for the first time! */
1231 oseq->os_destroys_in_progress = 0;
1234 mutex_lock(&oseq->os_create_lock);
1235 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1236 CERROR("%s: dropping old precreate request\n",
1237 ofd_obd(ofd)->obd_name);
1240 /* only precreate if seq is 0, IDIF or normal and also o_id
1241 * must be specfied */
1242 if ((!fid_seq_is_mdt(oa->o_seq) &&
1243 !fid_seq_is_norm(oa->o_seq) &&
1244 !fid_seq_is_idif(oa->o_seq)) || oa->o_id == 0) {
1245 diff = 1; /* shouldn't we create this right now? */
1247 diff = oa->o_id - ofd_seq_last_oid(oseq);
1248 /* Do sync create if the seq is about to used up */
1249 if (fid_seq_is_idif(oa->o_seq) ||
1250 fid_seq_is_mdt0(oa->o_seq)) {
1251 if (unlikely(oa->o_id >= IDIF_MAX_OID - 1))
1253 } else if (fid_seq_is_norm(oa->o_seq)) {
1254 if (unlikely(oa->o_id >=
1255 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1258 CERROR("%s : invalid o_seq "LPX64": rc = %d\n",
1259 ofd_name(ofd), oa->o_seq, -EINVAL);
1260 GOTO(out, rc = -EINVAL);
1265 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1270 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1271 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1272 /* don't enforce grant during orphan recovery */
1273 rc = ofd_grant_create(env,
1274 ofd_obd(ofd)->obd_self_export,
1277 CDEBUG(D_HA, "%s: failed to acquire grant space"
1278 "for precreate (%d)\n",
1279 ofd_name(ofd), diff);
1285 next_id = ofd_seq_last_oid(oseq) + 1;
1286 count = ofd_precreate_batch(ofd, diff);
1288 CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
1289 " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1290 count, oa->o_seq, next_id);
1292 if (cfs_time_after(jiffies, enough_time)) {
1293 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1294 " created at a rate of %d/s\n",
1295 ofd_obd(ofd)->obd_name,
1296 created, diff + created,
1297 created / DISK_TIMEOUT);
1301 rc = ofd_precreate_objects(env, ofd, next_id,
1302 oseq, count, sync_trans);
1306 } else if (rc < 0) {
1311 /* some objects got created, we can return
1312 * them, even if last creation failed */
1315 CERROR("unable to precreate: %d\n", rc);
1317 oa->o_id = ofd_seq_last_oid(oseq);
1318 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1320 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1321 !(oa->o_flags & OBD_FL_DELORPHAN))
1322 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1326 ofd_info2oti(info, oti);
1328 mutex_unlock(&oseq->os_create_lock);
1330 if (rc == 0 && ea != NULL) {
1331 struct lov_stripe_md *lsm = *ea;
1333 lsm->lsm_object_id = oa->o_id;
1335 ofd_seq_put(env, oseq);
1339 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1340 struct obd_info *oinfo)
1342 struct ofd_device *ofd = ofd_exp(exp);
1343 struct ofd_thread_info *info;
1344 struct ofd_object *fo;
1350 info = ofd_info_init(env, exp);
1352 fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1353 rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1354 oinfo_capa(oinfo), CAPA_OPC_META_READ);
1358 fo = ofd_object_find(env, ofd, &info->fti_fid);
1360 GOTO(out, rc = PTR_ERR(fo));
1361 LASSERT(fo != NULL);
1362 rc = ofd_attr_get(env, fo, &info->fti_attr);
1363 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1365 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1366 OFD_VALID_FLAGS | LA_UID | LA_GID);
1368 /* Store object version in reply */
1369 curr_version = dt_version_get(env, ofd_object_child(fo));
1370 if ((__s64)curr_version != -EOPNOTSUPP) {
1371 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1372 oinfo->oi_oa->o_data_version = curr_version;
1374 ofd_object_put(env, fo);
1379 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1380 struct obd_info *oinfo, obd_size start, obd_size end,
1381 struct ptlrpc_request_set *set)
1383 struct ofd_device *ofd = ofd_exp(exp);
1384 struct ofd_thread_info *info;
1385 struct ofd_object *fo;
1390 /* if no objid is specified, it means "sync whole filesystem" */
1391 if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1392 rc = dt_sync(env, ofd->ofd_osd);
1396 info = ofd_info_init(env, exp);
1397 fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1399 rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1400 oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1404 fo = ofd_object_find(env, ofd, &info->fti_fid);
1406 CERROR("%s: error finding object "DFID": rc = %ld\n",
1407 exp->exp_obd->obd_name, PFID(&info->fti_fid),
1409 GOTO(out, rc = PTR_ERR(fo));
1412 if (!ofd_object_exists(fo))
1413 GOTO(put, rc = -ENOENT);
1415 if (dt_version_get(env, ofd_object_child(fo)) >
1416 ofd_obd(ofd)->obd_last_committed) {
1417 rc = dt_object_sync(env, ofd_object_child(fo));
1422 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1423 rc = ofd_attr_get(env, fo, &info->fti_attr);
1424 obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1426 ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1429 ofd_object_put(env, fo);
1434 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1435 void *karg, void *uarg)
1438 struct ofd_device *ofd = ofd_exp(exp);
1439 struct obd_device *obd = ofd_obd(ofd);
1444 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1445 rc = lu_env_init(&env, LCT_LOCAL);
1450 case OBD_IOC_ABORT_RECOVERY:
1451 CERROR("%s: aborting recovery\n", obd->obd_name);
1452 target_stop_recovery_thread(obd);
1455 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1456 rc = dt_sync(&env, ofd->ofd_osd);
1458 case OBD_IOC_SET_READONLY:
1459 rc = dt_sync(&env, ofd->ofd_osd);
1461 rc = dt_ro(&env, ofd->ofd_osd);
1464 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1472 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1479 case OBD_CLEANUP_EARLY:
1481 case OBD_CLEANUP_EXPORTS:
1482 target_cleanup_recovery(obd);
1488 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1490 ofd_fmd_expire(exp);
1494 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1496 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1497 struct ofd_thread_info *info;
1499 #ifdef USE_HEALTH_CHECK_WRITE
1504 /* obd_proc_read_health pass NULL env, we need real one */
1505 rc = lu_env_init(&env, LCT_DT_THREAD);
1509 info = ofd_info_init(&env, NULL);
1510 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1514 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1515 GOTO(out, rc = -EROFS);
1517 #ifdef USE_HEALTH_CHECK_WRITE
1518 OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1519 if (info->fti_buf.lb_buf == NULL)
1520 GOTO(out, rc = -ENOMEM);
1522 info->fti_buf.lb_len = CFS_PAGE_SIZE;
1525 th = dt_trans_create(&env, ofd->ofd_osd);
1527 GOTO(out, rc = PTR_ERR(th));
1529 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1530 info->fti_buf.lb_len, info->fti_off, th);
1532 th->th_sync = 1; /* sync IO is needed */
1533 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1535 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1536 &info->fti_buf, &info->fti_off,
1539 dt_trans_stop(&env, ofd->ofd_osd, th);
1541 OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1543 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1551 * Handle quota control requests to consult current usage/limit.
1553 * \param obd - is the obd device associated with the ofd
1554 * \param exp - is the client's export
1555 * \param oqctl - is the obd_quotactl request to be processed
1557 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1558 struct obd_quotactl *oqctl)
1560 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1565 /* report success for quota on/off for interoperability with current MDT
1567 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1570 rc = lu_env_init(&env, LCT_DT_THREAD);
1574 rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1580 struct obd_ops ofd_obd_ops = {
1581 .o_owner = THIS_MODULE,
1582 .o_connect = ofd_obd_connect,
1583 .o_reconnect = ofd_obd_reconnect,
1584 .o_disconnect = ofd_obd_disconnect,
1585 .o_set_info_async = ofd_set_info_async,
1586 .o_get_info = ofd_get_info,
1587 .o_create = ofd_create,
1588 .o_statfs = ofd_statfs,
1589 .o_setattr = ofd_setattr,
1590 .o_preprw = ofd_preprw,
1591 .o_commitrw = ofd_commitrw,
1592 .o_destroy = ofd_destroy,
1593 .o_init_export = ofd_init_export,
1594 .o_destroy_export = ofd_destroy_export,
1595 .o_postrecov = ofd_obd_postrecov,
1596 .o_punch = ofd_punch,
1597 .o_getattr = ofd_getattr,
1599 .o_iocontrol = ofd_iocontrol,
1600 .o_precleanup = ofd_precleanup,
1602 .o_health_check = ofd_health_check,
1603 .o_quotactl = ofd_quotactl,