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>
49 #include <lustre_lfsck.h>
51 static int ofd_export_stats_init(struct ofd_device *ofd,
52 struct obd_export *exp, void *client_nid)
54 struct obd_device *obd = ofd_obd(ofd);
55 struct nid_stat *stats;
61 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
62 /* Self-export gets no proc entry */
65 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
67 /* Mask error for already created
77 stats = exp->exp_nid_stats;
78 LASSERT(stats != NULL);
80 num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
83 stats->nid_stats = lprocfs_alloc_stats(num_stats,
84 LPROCFS_STATS_FLAG_NOPERCPU);
85 if (stats->nid_stats == NULL)
88 lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
89 lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
90 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
91 lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
92 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
94 rc = lprocfs_register_stats(stats->nid_proc, "stats",
99 rc = lprocfs_nid_ldlm_stats_init(stats);
101 lprocfs_free_stats(&stats->nid_stats);
110 static int ofd_parse_connect_data(const struct lu_env *env,
111 struct obd_export *exp,
112 struct obd_connect_data *data,
115 struct ofd_device *ofd = ofd_exp(exp);
116 struct filter_export_data *fed = &exp->exp_filter_data;
121 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
122 " ocd_version: %x ocd_grant: %d ocd_index: %u"
124 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
125 data->ocd_connect_flags, data->ocd_version,
126 data->ocd_grant, data->ocd_index, data->ocd_group);
128 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
129 CWARN("!!! This export (nid %s) used object group %d "
130 "earlier; now it's trying to use group %d! This could "
131 "be a bug in the MDS. Please report to "
132 "http://bugs.whamcloud.com/\n",
133 obd_export_nid2str(exp), fed->fed_group,
137 fed->fed_group = data->ocd_group;
139 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
140 exp->exp_connect_data = *data;
141 data->ocd_version = LUSTRE_VERSION_CODE;
143 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
144 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
145 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
146 exp->exp_obd->obd_name, data->ocd_group);
147 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
150 if (ofd_grant_param_supp(exp)) {
151 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
152 /* ocd_{blocksize,inodespace} are log2 values */
153 data->ocd_blocksize = ofd->ofd_blockbits;
154 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
155 /* ocd_grant_extent is in 1K blocks */
156 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
159 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
160 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
163 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
164 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
165 int index = lsd->lsd_osd_index;
167 if (index != data->ocd_index) {
168 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
169 " %u doesn't match actual OST index"
170 " %u in last_rcvd file, bad "
172 obd_export_nid2str(exp), index,
176 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
177 /* this will only happen on the first connect */
178 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
179 /* sync is not needed here as lut_client_add will
180 * set exp_need_sync flag */
181 tgt_server_data_update(env, &ofd->ofd_lut, 0);
184 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
185 data->ocd_brw_size = 65536;
186 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
187 data->ocd_brw_size = min(data->ocd_brw_size,
188 (__u32)DT_MAX_BRW_SIZE);
189 if (data->ocd_brw_size == 0) {
190 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
191 " ocd_version: %x ocd_grant: %d ocd_index: %u "
192 "ocd_brw_size is unexpectedly zero, "
193 "network data corruption?"
194 "Refusing connection of this client\n",
195 exp->exp_obd->obd_name,
196 exp->exp_client_uuid.uuid,
197 exp, data->ocd_connect_flags, data->ocd_version,
198 data->ocd_grant, data->ocd_index);
203 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
204 __u32 cksum_types = data->ocd_cksum_types;
206 /* The client set in ocd_cksum_types the checksum types it
207 * supports. We have to mask off the algorithms that we don't
209 data->ocd_cksum_types &= cksum_types_supported_server();
211 if (unlikely(data->ocd_cksum_types == 0)) {
212 CERROR("%s: Connect with checksum support but no "
213 "ocd_cksum_types is set\n",
214 exp->exp_obd->obd_name);
218 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
219 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
220 cksum_types, data->ocd_cksum_types);
222 /* This client does not support OBD_CONNECT_CKSUM
223 * fall back to CRC32 */
224 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
225 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
226 exp->exp_obd->obd_name, obd_export_nid2str(exp));
229 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
230 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
232 if (OCD_HAS_FLAG(data, PINGLESS)) {
233 if (ptlrpc_pinger_suppress_pings()) {
234 spin_lock(&exp->exp_obd->obd_dev_lock);
235 list_del_init(&exp->exp_obd_chain_timed);
236 spin_unlock(&exp->exp_obd->obd_dev_lock);
238 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
245 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
246 struct obd_device *obd, struct obd_uuid *cluuid,
247 struct obd_connect_data *data, void *localdata)
249 struct ofd_device *ofd;
254 if (exp == NULL || obd == NULL || cluuid == NULL)
257 ofd = ofd_dev(obd->obd_lu_dev);
259 rc = lu_env_refill((struct lu_env *)env);
261 CERROR("Failure to refill session: '%d'\n", rc);
265 ofd_info_init(env, exp);
266 rc = ofd_parse_connect_data(env, exp, data, false);
268 ofd_export_stats_init(ofd, exp, localdata);
273 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
274 struct obd_device *obd, struct obd_uuid *cluuid,
275 struct obd_connect_data *data, void *localdata)
277 struct obd_export *exp;
278 struct ofd_device *ofd;
279 struct lustre_handle conn = { 0 };
283 if (_exp == NULL || obd == NULL || cluuid == NULL)
286 ofd = ofd_dev(obd->obd_lu_dev);
288 rc = class_connect(&conn, obd, cluuid);
292 exp = class_conn2export(&conn);
293 LASSERT(exp != NULL);
295 rc = lu_env_refill((struct lu_env *)env);
297 CERROR("Failure to refill session: '%d'\n", rc);
301 ofd_info_init(env, exp);
303 rc = ofd_parse_connect_data(env, exp, data, true);
307 if (obd->obd_replayable) {
308 struct tg_export_data *ted = &exp->exp_target_data;
310 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
311 sizeof(ted->ted_lcd->lcd_uuid));
312 rc = tgt_client_new(env, exp);
315 ofd_export_stats_init(ofd, exp, localdata);
318 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
323 class_disconnect(exp);
331 static int ofd_obd_disconnect(struct obd_export *exp)
333 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
340 class_export_get(exp);
342 if (!(exp->exp_flags & OBD_OPT_FORCE))
343 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
345 rc = server_disconnect_export(exp);
347 ofd_grant_discard(exp);
349 rc = lu_env_init(&env, LCT_DT_THREAD);
353 /* Do not erase record for recoverable client. */
354 if (exp->exp_obd->obd_replayable &&
355 (!exp->exp_obd->obd_fail || exp->exp_failed))
356 tgt_client_del(&env, exp);
359 class_export_put(exp);
363 static int ofd_init_export(struct obd_export *exp)
367 spin_lock_init(&exp->exp_filter_data.fed_lock);
368 CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
369 spin_lock(&exp->exp_lock);
370 exp->exp_connecting = 1;
371 spin_unlock(&exp->exp_lock);
373 /* self-export doesn't need client data and ldlm initialization */
374 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
375 &exp->exp_client_uuid)))
378 rc = tgt_client_alloc(exp);
380 ldlm_init_export(exp);
382 CERROR("%s: Can't initialize export: rc %d\n",
383 exp->exp_obd->obd_name, rc);
387 static int ofd_destroy_export(struct obd_export *exp)
389 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
391 if (exp->exp_filter_data.fed_pending)
392 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
393 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
394 exp, exp->exp_filter_data.fed_pending);
396 target_destroy_export(exp);
398 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
399 &exp->exp_client_uuid)))
402 ldlm_destroy_export(exp);
403 tgt_client_free(exp);
405 ofd_fmd_cleanup(exp);
408 * discard grants once we're sure no more
409 * interaction with the client is possible
411 ofd_grant_discard(exp);
412 ofd_fmd_cleanup(exp);
414 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
415 if (ofd->ofd_tot_granted_clients > 0)
416 ofd->ofd_tot_granted_clients --;
419 if (!(exp->exp_flags & OBD_OPT_FORCE))
420 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
422 LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
426 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
428 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
430 CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
431 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
434 int ofd_obd_postrecov(struct obd_device *obd)
437 struct lu_device *ldev = obd->obd_lu_dev;
442 rc = lu_env_init(&env, LCT_DT_THREAD);
445 ofd_info_init(&env, obd->obd_self_export);
447 rc = ofd_postrecov(&env, ofd_dev(ldev));
453 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
454 struct obd_device *obd, int initial)
456 struct filter_obd *fo = &obd->u.filter;
457 struct sptlrpc_rule_set tmp_rset;
460 sptlrpc_rule_set_init(&tmp_rset);
461 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
463 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
468 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
470 write_lock(&fo->fo_sptlrpc_lock);
471 sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
472 fo->fo_sptlrpc_rset = tmp_rset;
473 write_unlock(&fo->fo_sptlrpc_lock);
478 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
484 LCONSOLE_WARN("%s: received MDS connection from %s\n",
485 exp->exp_obd->obd_name, obd_export_nid2str(exp));
489 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
490 __u32 keylen, void *key, __u32 vallen, void *val,
491 struct ptlrpc_request_set *set)
493 struct ofd_device *ofd;
498 if (exp->exp_obd == NULL) {
499 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
505 if (KEY_IS(KEY_CAPA_KEY)) {
506 rc = ofd_update_capa_key(ofd, val);
508 CERROR("%s: update capability key failed: rc = %d\n",
509 exp->exp_obd->obd_name, rc);
510 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
511 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
512 } else if (KEY_IS(KEY_MDS_CONN)) {
513 rc = ofd_set_mds_conn(exp, val);
514 } else if (KEY_IS(KEY_GRANT_SHRINK)) {
515 struct ost_body *body = val;
517 ofd_info_init(env, exp);
518 /** handle grant shrink, similar to a read request */
519 ofd_grant_prepare_read(env, exp, &body->oa);
521 CERROR("%s: Unsupported key %s\n",
522 exp->exp_obd->obd_name, (char*)key);
529 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
530 __u32 keylen, void *key, __u32 *vallen, void *val,
531 struct lov_stripe_md *lsm)
533 struct ofd_device *ofd;
538 if (exp->exp_obd == NULL) {
539 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
543 /* Because ofd_get_info might be called from
544 * handle_request_in as well(see LU-3239), where env might
545 * not be initilaized correctly, and le_ses might be in
546 * an un-initialized state, so only refill le_ctx here */
547 rc = lu_env_refill((struct lu_env *)env);
553 if (KEY_IS(KEY_BLOCKSIZE)) {
554 __u32 *blocksize = val;
556 if (*vallen < sizeof(*blocksize))
557 GOTO(out, rc = -EOVERFLOW);
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))
565 GOTO(out, rc = -EOVERFLOW);
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);
584 GOTO(out, rc = -EOVERFLOW);
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_device *ofd = ofd_exp(exp);
592 struct ofd_object *fo;
593 struct ll_fiemap_info_key *fm_key = key;
597 *vallen = fiemap_count_to_size(
598 fm_key->fiemap.fm_extent_count);
602 rc = ostid_to_fid(&fid, &fm_key->oa.o_oi, 0);
605 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
608 fo = ofd_object_find(env, ofd, &fid);
610 CERROR("%s: error finding object "DFID"\n",
611 exp->exp_obd->obd_name, PFID(&fid));
614 struct ll_user_fiemap *fiemap = val;
616 ofd_read_lock(env, fo);
617 if (ofd_object_exists(fo)) {
618 *fiemap = fm_key->fiemap;
619 rc = dt_fiemap_get(env,
620 ofd_object_child(fo),
625 ofd_read_unlock(env, fo);
626 ofd_object_put(env, fo);
628 } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
629 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
630 *vallen = sizeof(__u32);
631 } else if (KEY_IS(KEY_LAST_FID)) {
632 struct ofd_device *ofd = ofd_exp(exp);
633 struct ofd_seq *oseq;
634 struct lu_fid *fid = val;
638 *vallen = sizeof(struct lu_fid);
642 if (*vallen < sizeof(*fid))
643 GOTO(out, rc = -EOVERFLOW);
645 ofd_info_init(env, exp);
647 fid_le_to_cpu(fid, fid);
649 oseq = ofd_seq_load(env, ofd,
650 ostid_seq((struct ost_id *)fid));
652 GOTO(out, rc = PTR_ERR(oseq));
654 rc = ostid_to_fid(fid, &oseq->os_oi,
655 ofd->ofd_lut.lut_lsd.lsd_osd_index);
659 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
661 *vallen = sizeof(*fid);
663 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 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
846 ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
847 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
848 oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
852 /* This would be very bad - accidentally truncating a file when
853 * changing the time or similar - bug 12203. */
854 if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
855 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
856 static char mdsinum[48];
858 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
859 snprintf(mdsinum, sizeof(mdsinum) - 1,
860 "of parent "DFID, oinfo->oi_oa->o_parent_seq,
861 oinfo->oi_oa->o_parent_oid, 0);
865 CERROR("%s: setattr from %s trying to truncate object "DFID
866 " %s\n", exp->exp_obd->obd_name,
867 obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
868 GOTO(out, rc = -EPERM);
871 fo = ofd_object_find(env, ofd, &info->fti_fid);
873 CERROR("%s: can't find object "DFID"\n",
874 exp->exp_obd->obd_name, PFID(&info->fti_fid));
875 GOTO(out, rc = PTR_ERR(fo));
878 la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
879 info->fti_attr.la_valid &= ~LA_TYPE;
881 if (oa->o_valid & OBD_MD_FLFID) {
882 ff = &info->fti_mds_fid;
883 ofd_prepare_fidea(ff, oa);
886 /* setting objects attributes (including owner/group) */
887 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
889 GOTO(out_unlock, rc);
891 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
893 ldlm_res_lvbo_update(res, NULL, 0);
894 ldlm_resource_putref(res);
897 oinfo->oi_oa->o_valid = OBD_MD_FLID;
899 /* Quota release needs uid/gid info */
900 rc = ofd_attr_get(env, fo, &info->fti_attr);
901 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
902 OFD_VALID_FLAGS | LA_UID | LA_GID);
903 ofd_info2oti(info, oti);
905 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
908 ofd_object_put(env, fo);
913 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
914 struct obd_info *oinfo, struct obd_trans_info *oti,
915 struct ptlrpc_request_set *rqset)
917 struct ofd_thread_info *info;
918 struct ofd_device *ofd = ofd_exp(exp);
919 struct ldlm_namespace *ns = ofd->ofd_namespace;
920 struct ldlm_resource *res;
921 struct ofd_object *fo;
922 struct filter_fid *ff = NULL;
927 info = ofd_info_init(env, exp);
928 ofd_oti2info(info, oti);
930 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
933 ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
935 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
936 ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
937 oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
938 oinfo->oi_policy.l_extent.end);
940 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
941 oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
945 fo = ofd_object_find(env, ofd, &info->fti_fid);
947 CERROR("%s: error finding object "DFID": rc = %ld\n",
948 exp->exp_obd->obd_name, PFID(&info->fti_fid),
950 GOTO(out_env, rc = PTR_ERR(fo));
953 LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
954 if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
956 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
957 } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
958 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
961 la_from_obdo(&info->fti_attr, oinfo->oi_oa,
962 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
963 info->fti_attr.la_valid &= ~LA_TYPE;
964 info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
965 info->fti_attr.la_valid |= LA_SIZE;
967 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
968 ff = &info->fti_mds_fid;
969 ofd_prepare_fidea(ff, oinfo->oi_oa);
972 rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
973 oinfo->oi_policy.l_extent.end, &info->fti_attr,
978 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
980 ldlm_res_lvbo_update(res, NULL, 0);
981 ldlm_resource_putref(res);
984 oinfo->oi_oa->o_valid = OBD_MD_FLID;
985 /* Quota release needs uid/gid info */
986 rc = ofd_attr_get(env, fo, &info->fti_attr);
987 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
988 OFD_VALID_FLAGS | LA_UID | LA_GID);
989 ofd_info2oti(info, oti);
991 ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
994 ofd_object_put(env, fo);
999 static int ofd_destroy_by_fid(const struct lu_env *env,
1000 struct ofd_device *ofd,
1001 const struct lu_fid *fid, int orphan)
1003 struct ofd_thread_info *info = ofd_info(env);
1004 struct lustre_handle lockh;
1005 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
1007 ldlm_policy_data_t policy = {
1008 .l_extent = { 0, OBD_OBJECT_EOF }
1010 struct ofd_object *fo;
1014 fo = ofd_object_find(env, ofd, fid);
1016 RETURN(PTR_ERR(fo));
1017 if (!ofd_object_exists(fo))
1018 GOTO(out, rc = -ENOENT);
1020 /* Tell the clients that the object is gone now and that they should
1021 * throw away any cached pages. */
1022 ost_fid_build_resid(fid, &info->fti_resid);
1023 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1024 LDLM_EXTENT, &policy, LCK_PW, &flags,
1025 ldlm_blocking_ast, ldlm_completion_ast,
1026 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1028 /* We only care about the side-effects, just drop the lock. */
1030 ldlm_lock_decref(&lockh, LCK_PW);
1032 LASSERT(fo != NULL);
1034 rc = ofd_object_destroy(env, fo, orphan);
1037 ofd_object_put(env, fo);
1041 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1042 struct obdo *oa, struct lov_stripe_md *md,
1043 struct obd_trans_info *oti, struct obd_export *md_exp,
1046 struct ofd_device *ofd = ofd_exp(exp);
1047 struct ofd_thread_info *info;
1053 info = ofd_info_init(env, exp);
1054 ofd_oti2info(info, oti);
1056 if (!(oa->o_valid & OBD_MD_FLGROUP))
1057 ostid_set_seq_mdt0(&oa->o_oi);
1059 /* check that o_misc makes sense */
1060 if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1063 count = 1; /* default case - single destroy */
1066 * There can be sequence of objects to destroy. Therefore this request
1067 * may have multiple transaction involved in. It is OK, we need only
1068 * the highest used transno to be reported back in reply but not for
1069 * replays, they must report their transno
1071 if (info->fti_transno == 0) /* not replay */
1072 info->fti_mult_trans = 1;
1074 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1075 POSTID(&oa->o_oi), count);
1079 lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
1085 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1086 if (lrc == -ENOENT) {
1088 "%s: destroying non-existent object "DFID"\n",
1089 ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1090 /* rewrite rc with -ENOENT only if it is 0 */
1093 } else if (lrc != 0) {
1094 CERROR("%s: error destroying object "DFID": %d\n",
1095 ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1100 ostid_inc_id(&oa->o_oi);
1103 /* if we have transaction then there were some deletions, we don't
1104 * need to return ENOENT in that case because it will not wait
1105 * for commit of these deletions. The ENOENT must be returned only
1106 * if there were no transations.
1108 if (rc == -ENOENT) {
1109 if (info->fti_transno != 0)
1111 } else if (rc != 0) {
1113 * If we have at least one transaction then llog record
1114 * on server will be removed upon commit, so for rc != 0
1115 * we return no transno and llog record will be reprocessed.
1117 info->fti_transno = 0;
1119 ofd_info2oti(info, oti);
1124 static int ofd_orphans_destroy(const struct lu_env *env,
1125 struct obd_export *exp, struct ofd_device *ofd,
1128 struct ofd_thread_info *info = ofd_info(env);
1132 struct ost_id oi = oa->o_oi;
1133 __u64 end_id = ostid_id(&oa->o_oi);
1134 struct ofd_seq *oseq;
1138 oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
1140 CERROR("%s: Can not find seq for "DOSTID"\n",
1141 ofd_name(ofd), POSTID(&oa->o_oi));
1145 LASSERT(exp != NULL);
1146 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1148 last = ofd_seq_last_oid(oseq);
1149 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1150 " to "DOSTID"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
1151 end_id + 1, ostid_seq(&oa->o_oi), last);
1153 for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
1154 ostid_dec_id(&oi)) {
1155 rc = ostid_to_fid(&info->fti_fid, &oi, 0);
1158 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1159 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1160 CEMERG("%s: error destroying precreated id "DOSTID
1161 ": rc = %d\n", ofd_name(ofd), POSTID(&oi), rc);
1163 ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
1164 /* update last_id on disk periodically so that if we
1165 * restart * we don't need to re-scan all of the just
1166 * deleted objects. */
1167 if ((ostid_id(&oi) & 511) == 0)
1168 ofd_seq_last_oid_write(env, ofd, oseq);
1171 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1172 ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
1174 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1176 /* don't reuse orphan object, return last used objid */
1177 ostid_set_id(&oa->o_oi, last);
1181 ofd_seq_put(env, oseq);
1185 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1186 struct obdo *oa, struct lov_stripe_md **ea,
1187 struct obd_trans_info *oti)
1189 struct ofd_device *ofd = ofd_exp(exp);
1190 struct ofd_thread_info *info;
1191 obd_seq seq = ostid_seq(&oa->o_oi);
1192 struct ofd_seq *oseq;
1198 info = ofd_info_init(env, exp);
1199 ofd_oti2info(info, oti);
1201 LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1202 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1204 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1206 oseq = ofd_seq_load(env, ofd, seq);
1208 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1209 ofd_name(ofd), seq, PTR_ERR(oseq));
1213 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1214 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1215 if (!ofd_obd(ofd)->obd_recovering ||
1216 ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1217 CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1218 POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1219 GOTO(out_nolock, rc = -EINVAL);
1221 /* do nothing because we create objects during first write */
1222 GOTO(out_nolock, rc = 0);
1224 /* former ofd_handle_precreate */
1225 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1226 (oa->o_flags & OBD_FL_DELORPHAN)) {
1227 /* destroy orphans */
1228 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1229 CERROR("%s: dropping old orphan cleanup request\n",
1231 GOTO(out_nolock, rc = 0);
1233 /* This causes inflight precreates to abort and drop lock */
1234 oseq->os_destroys_in_progress = 1;
1235 mutex_lock(&oseq->os_create_lock);
1236 if (!oseq->os_destroys_in_progress) {
1237 CERROR("%s:["LPU64"] destroys_in_progress already"
1238 " cleared\n", exp->exp_obd->obd_name,
1239 ostid_seq(&oa->o_oi));
1240 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1243 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1244 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1245 ofd_seq_last_oid(oseq), diff);
1246 if (-diff > OST_MAX_PRECREATE) {
1247 /* FIXME: should reset precreate_next_id on MDS */
1249 } else if (diff < 0) {
1250 rc = ofd_orphans_destroy(env, exp, ofd, oa);
1251 oseq->os_destroys_in_progress = 0;
1253 /* XXX: Used by MDS for the first time! */
1254 oseq->os_destroys_in_progress = 0;
1257 mutex_lock(&oseq->os_create_lock);
1258 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1259 CERROR("%s: dropping old precreate request\n",
1260 ofd_obd(ofd)->obd_name);
1263 /* only precreate if seq is 0, IDIF or normal and also o_id
1264 * must be specfied */
1265 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1266 !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1267 !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1268 ostid_id(&oa->o_oi) == 0) {
1269 diff = 1; /* shouldn't we create this right now? */
1271 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1272 /* Do sync create if the seq is about to used up */
1273 if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1274 fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1275 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1277 } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1278 if (unlikely(ostid_id(&oa->o_oi) >=
1279 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1282 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1283 ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1284 GOTO(out, rc = -EINVAL);
1289 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1294 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1295 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1296 /* don't enforce grant during orphan recovery */
1297 rc = ofd_grant_create(env,
1298 ofd_obd(ofd)->obd_self_export,
1301 CDEBUG(D_HA, "%s: failed to acquire grant "
1302 "space for precreate (%d): rc = %d\n",
1303 ofd_name(ofd), diff, rc);
1309 next_id = ofd_seq_last_oid(oseq) + 1;
1310 count = ofd_precreate_batch(ofd, diff);
1312 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1313 " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1314 count, ostid_seq(&oa->o_oi), next_id);
1316 if (cfs_time_after(jiffies, enough_time)) {
1317 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1318 " created at a rate of %d/s\n",
1319 ofd_obd(ofd)->obd_name,
1320 created, diff + created,
1321 created / DISK_TIMEOUT);
1325 rc = ofd_precreate_objects(env, ofd, next_id,
1326 oseq, count, sync_trans);
1330 } else if (rc < 0) {
1335 /* some objects got created, we can return
1336 * them, even if last creation failed */
1339 CERROR("%s: unable to precreate: rc = %d\n",
1342 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1343 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1345 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1346 !(oa->o_flags & OBD_FL_DELORPHAN))
1347 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1351 ofd_info2oti(info, oti);
1353 mutex_unlock(&oseq->os_create_lock);
1355 if (rc == 0 && ea != NULL) {
1356 struct lov_stripe_md *lsm = *ea;
1358 lsm->lsm_oi = oa->o_oi;
1360 ofd_seq_put(env, oseq);
1364 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1365 struct obd_info *oinfo)
1367 struct ofd_device *ofd = ofd_exp(exp);
1368 struct ofd_thread_info *info;
1369 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 if (!ofd_object_exists(fo))
1388 GOTO(out_put, rc = -ENOENT);
1390 LASSERT(fo != NULL);
1391 rc = ofd_attr_get(env, fo, &info->fti_attr);
1392 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1396 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1397 OFD_VALID_FLAGS | LA_UID | LA_GID);
1399 /* Store object version in reply */
1400 curr_version = dt_version_get(env, ofd_object_child(fo));
1401 if ((__s64)curr_version != -EOPNOTSUPP) {
1402 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1403 oinfo->oi_oa->o_data_version = curr_version;
1408 ofd_object_put(env, fo);
1413 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1414 struct obd_info *oinfo, obd_size start, obd_size end,
1415 struct ptlrpc_request_set *set)
1417 struct ofd_device *ofd = ofd_exp(exp);
1418 struct ofd_thread_info *info;
1419 struct ofd_object *fo;
1424 /* if no objid is specified, it means "sync whole filesystem" */
1425 if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1426 rc = dt_sync(env, ofd->ofd_osd);
1430 info = ofd_info_init(env, exp);
1431 rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1435 rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1436 oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1440 fo = ofd_object_find(env, ofd, &info->fti_fid);
1442 CERROR("%s: error finding object "DFID": rc = %ld\n",
1443 exp->exp_obd->obd_name, PFID(&info->fti_fid),
1445 GOTO(out, rc = PTR_ERR(fo));
1448 if (!ofd_object_exists(fo))
1449 GOTO(put, rc = -ENOENT);
1451 if (dt_version_get(env, ofd_object_child(fo)) >
1452 ofd_obd(ofd)->obd_last_committed) {
1453 rc = dt_object_sync(env, ofd_object_child(fo));
1458 oinfo->oi_oa->o_valid = OBD_MD_FLID;
1459 rc = ofd_attr_get(env, fo, &info->fti_attr);
1460 obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1462 ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1465 ofd_object_put(env, fo);
1470 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1471 struct ofd_device *ofd, void *karg)
1473 struct obd_ioctl_data *data = karg;
1475 struct ofd_object *fo;
1476 dt_obj_version_t version;
1481 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1482 GOTO(out, rc = -EINVAL);
1484 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1485 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1486 } else if (data->ioc_inlbuf3 != NULL &&
1487 data->ioc_inllen3 == sizeof(__u64) &&
1488 data->ioc_inlbuf4 != NULL &&
1489 data->ioc_inllen4 == sizeof(__u64)) {
1490 struct ost_id ostid;
1492 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1493 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1494 rc = ostid_to_fid(&fid, &ostid, 0);
1498 GOTO(out, rc = -EINVAL);
1501 if (!fid_is_sane(&fid))
1502 GOTO(out, rc = -EINVAL);
1504 fo = ofd_object_find(env, ofd, &fid);
1506 GOTO(out, rc = PTR_ERR(fo));
1508 if (!ofd_object_exists(fo))
1509 GOTO(out_fo, rc = -ENOENT);
1511 if (lu_object_remote(&fo->ofo_obj.do_lu))
1512 GOTO(out_fo, rc = -EREMOTE);
1514 version = dt_version_get(env, ofd_object_child(fo));
1516 GOTO(out_fo, rc = -EIO);
1518 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1522 ofd_object_put(env, fo);
1527 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1528 void *karg, void *uarg)
1531 struct ofd_device *ofd = ofd_exp(exp);
1532 struct obd_device *obd = ofd_obd(ofd);
1537 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1538 rc = lu_env_init(&env, LCT_DT_THREAD);
1543 case OBD_IOC_ABORT_RECOVERY:
1544 CERROR("%s: aborting recovery\n", obd->obd_name);
1545 target_stop_recovery_thread(obd);
1548 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1549 rc = dt_sync(&env, ofd->ofd_osd);
1551 case OBD_IOC_SET_READONLY:
1552 rc = dt_sync(&env, ofd->ofd_osd);
1554 rc = dt_ro(&env, ofd->ofd_osd);
1556 case OBD_IOC_START_LFSCK: {
1557 struct obd_ioctl_data *data = karg;
1558 struct lfsck_start_param lsp;
1560 if (unlikely(data == NULL)) {
1565 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1566 lsp.lsp_namespace = ofd->ofd_namespace;
1567 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1570 case OBD_IOC_STOP_LFSCK: {
1571 rc = lfsck_stop(&env, ofd->ofd_osd, false);
1574 case OBD_IOC_GET_OBJ_VERSION:
1575 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1578 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1586 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1593 case OBD_CLEANUP_EARLY:
1595 case OBD_CLEANUP_EXPORTS:
1596 target_cleanup_recovery(obd);
1602 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1604 ofd_fmd_expire(exp);
1608 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1610 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1611 struct ofd_thread_info *info;
1613 #ifdef USE_HEALTH_CHECK_WRITE
1618 /* obd_proc_read_health pass NULL env, we need real one */
1619 rc = lu_env_init(&env, LCT_DT_THREAD);
1623 info = ofd_info_init(&env, NULL);
1624 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1628 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1629 GOTO(out, rc = -EROFS);
1631 #ifdef USE_HEALTH_CHECK_WRITE
1632 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1633 if (info->fti_buf.lb_buf == NULL)
1634 GOTO(out, rc = -ENOMEM);
1636 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1639 th = dt_trans_create(&env, ofd->ofd_osd);
1641 GOTO(out, rc = PTR_ERR(th));
1643 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1644 info->fti_buf.lb_len, info->fti_off, th);
1646 th->th_sync = 1; /* sync IO is needed */
1647 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1649 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1650 &info->fti_buf, &info->fti_off,
1653 dt_trans_stop(&env, ofd->ofd_osd, th);
1655 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1657 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1665 * Handle quota control requests to consult current usage/limit.
1667 * \param obd - is the obd device associated with the ofd
1668 * \param exp - is the client's export
1669 * \param oqctl - is the obd_quotactl request to be processed
1671 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1672 struct obd_quotactl *oqctl)
1674 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1679 /* report success for quota on/off for interoperability with current MDT
1681 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1684 rc = lu_env_init(&env, LCT_DT_THREAD);
1688 rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1694 struct obd_ops ofd_obd_ops = {
1695 .o_owner = THIS_MODULE,
1696 .o_connect = ofd_obd_connect,
1697 .o_reconnect = ofd_obd_reconnect,
1698 .o_disconnect = ofd_obd_disconnect,
1699 .o_set_info_async = ofd_set_info_async,
1700 .o_get_info = ofd_get_info,
1701 .o_create = ofd_create,
1702 .o_statfs = ofd_statfs,
1703 .o_setattr = ofd_setattr,
1704 .o_preprw = ofd_preprw,
1705 .o_commitrw = ofd_commitrw,
1706 .o_destroy = ofd_destroy,
1707 .o_init_export = ofd_init_export,
1708 .o_destroy_export = ofd_destroy_export,
1709 .o_postrecov = ofd_obd_postrecov,
1710 .o_punch = ofd_punch,
1711 .o_getattr = ofd_getattr,
1713 .o_iocontrol = ofd_iocontrol,
1714 .o_precleanup = ofd_precleanup,
1716 .o_health_check = ofd_health_check,
1717 .o_quotactl = ofd_quotactl,