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 LASSERT(obd->obd_uses_nid_stats);
63 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
64 /* Self-export gets no proc entry */
67 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
69 /* Mask error for already created
79 stats = exp->exp_nid_stats;
80 LASSERT(stats != NULL);
82 num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
84 stats->nid_stats = lprocfs_alloc_stats(num_stats,
85 LPROCFS_STATS_FLAG_NOPERCPU);
86 if (stats->nid_stats == NULL)
89 lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
90 ofd_stats_counter_init(stats->nid_stats);
91 rc = lprocfs_register_stats(stats->nid_proc, "stats",
96 rc = lprocfs_nid_ldlm_stats_init(stats);
98 lprocfs_free_stats(&stats->nid_stats);
107 static int ofd_parse_connect_data(const struct lu_env *env,
108 struct obd_export *exp,
109 struct obd_connect_data *data,
112 struct ofd_device *ofd = ofd_exp(exp);
113 struct filter_export_data *fed = &exp->exp_filter_data;
118 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
119 " ocd_version: %x ocd_grant: %d ocd_index: %u"
121 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
122 data->ocd_connect_flags, data->ocd_version,
123 data->ocd_grant, data->ocd_index, data->ocd_group);
125 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
126 CWARN("!!! This export (nid %s) used object group %d "
127 "earlier; now it's trying to use group %d! This could "
128 "be a bug in the MDS. Please report to "
129 "http://bugs.whamcloud.com/\n",
130 obd_export_nid2str(exp), fed->fed_group,
134 fed->fed_group = data->ocd_group;
136 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
137 data->ocd_version = LUSTRE_VERSION_CODE;
139 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
140 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
141 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
142 exp->exp_obd->obd_name, data->ocd_group);
143 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
146 if (ofd_grant_param_supp(exp)) {
147 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
148 /* ocd_{blocksize,inodespace} are log2 values */
149 data->ocd_blocksize = ofd->ofd_blockbits;
150 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
151 /* ocd_grant_extent is in 1K blocks */
152 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
155 if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
156 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
159 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
160 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
161 int index = lsd->lsd_osd_index;
163 if (index != data->ocd_index) {
164 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
165 " %u doesn't match actual OST index"
166 " %u in last_rcvd file, bad "
168 obd_export_nid2str(exp), index,
172 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
173 /* this will only happen on the first connect */
174 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
175 /* sync is not needed here as lut_client_add will
176 * set exp_need_sync flag */
177 tgt_server_data_update(env, &ofd->ofd_lut, 0);
180 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
181 data->ocd_brw_size = 65536;
182 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
183 data->ocd_brw_size = min(data->ocd_brw_size,
184 (__u32)DT_MAX_BRW_SIZE);
185 if (data->ocd_brw_size == 0) {
186 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
187 " ocd_version: %x ocd_grant: %d ocd_index: %u "
188 "ocd_brw_size is unexpectedly zero, "
189 "network data corruption?"
190 "Refusing connection of this client\n",
191 exp->exp_obd->obd_name,
192 exp->exp_client_uuid.uuid,
193 exp, data->ocd_connect_flags, data->ocd_version,
194 data->ocd_grant, data->ocd_index);
199 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
200 __u32 cksum_types = data->ocd_cksum_types;
202 /* The client set in ocd_cksum_types the checksum types it
203 * supports. We have to mask off the algorithms that we don't
205 data->ocd_cksum_types &= cksum_types_supported_server();
207 if (unlikely(data->ocd_cksum_types == 0)) {
208 CERROR("%s: Connect with checksum support but no "
209 "ocd_cksum_types is set\n",
210 exp->exp_obd->obd_name);
214 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
215 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
216 cksum_types, data->ocd_cksum_types);
218 /* This client does not support OBD_CONNECT_CKSUM
219 * fall back to CRC32 */
220 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
221 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
222 exp->exp_obd->obd_name, obd_export_nid2str(exp));
225 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
226 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
228 if (OCD_HAS_FLAG(data, PINGLESS)) {
229 if (ptlrpc_pinger_suppress_pings()) {
230 spin_lock(&exp->exp_obd->obd_dev_lock);
231 list_del_init(&exp->exp_obd_chain_timed);
232 spin_unlock(&exp->exp_obd->obd_dev_lock);
234 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
241 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
242 struct obd_device *obd, struct obd_uuid *cluuid,
243 struct obd_connect_data *data, void *localdata)
245 struct ofd_device *ofd;
250 if (exp == NULL || obd == NULL || cluuid == NULL)
253 ofd = ofd_dev(obd->obd_lu_dev);
255 rc = ofd_parse_connect_data(env, exp, data, false);
257 ofd_export_stats_init(ofd, exp, localdata);
262 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
263 struct obd_device *obd, struct obd_uuid *cluuid,
264 struct obd_connect_data *data, void *localdata)
266 struct obd_export *exp;
267 struct ofd_device *ofd;
268 struct lustre_handle conn = { 0 };
272 if (_exp == NULL || obd == NULL || cluuid == NULL)
275 ofd = ofd_dev(obd->obd_lu_dev);
277 rc = class_connect(&conn, obd, cluuid);
281 exp = class_conn2export(&conn);
282 LASSERT(exp != NULL);
284 rc = ofd_parse_connect_data(env, exp, data, true);
288 if (obd->obd_replayable) {
289 struct tg_export_data *ted = &exp->exp_target_data;
291 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
292 sizeof(ted->ted_lcd->lcd_uuid));
293 rc = tgt_client_new(env, exp);
296 ofd_export_stats_init(ofd, exp, localdata);
299 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
300 data ? data->ocd_group : -1);
304 class_disconnect(exp);
312 int ofd_obd_disconnect(struct obd_export *exp)
314 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
321 class_export_get(exp);
323 if (!(exp->exp_flags & OBD_OPT_FORCE))
324 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
326 rc = server_disconnect_export(exp);
328 ofd_grant_discard(exp);
330 /* Do not erase record for recoverable client. */
331 if (exp->exp_obd->obd_replayable &&
332 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
333 rc = lu_env_init(&env, LCT_DT_THREAD);
337 tgt_client_del(&env, exp);
341 class_export_put(exp);
345 static int ofd_init_export(struct obd_export *exp)
349 spin_lock_init(&exp->exp_filter_data.fed_lock);
350 CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
351 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
352 spin_lock(&exp->exp_lock);
353 exp->exp_connecting = 1;
354 spin_unlock(&exp->exp_lock);
356 /* self-export doesn't need client data and ldlm initialization */
357 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
358 &exp->exp_client_uuid)))
361 rc = tgt_client_alloc(exp);
363 ldlm_init_export(exp);
365 CERROR("%s: Can't initialize export: rc %d\n",
366 exp->exp_obd->obd_name, rc);
370 static int ofd_destroy_export(struct obd_export *exp)
372 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
374 if (exp->exp_filter_data.fed_pending)
375 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
376 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
377 exp, exp->exp_filter_data.fed_pending);
379 target_destroy_export(exp);
381 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
382 &exp->exp_client_uuid)))
385 ldlm_destroy_export(exp);
386 tgt_client_free(exp);
388 ofd_fmd_cleanup(exp);
391 * discard grants once we're sure no more
392 * interaction with the client is possible
394 ofd_grant_discard(exp);
395 ofd_fmd_cleanup(exp);
397 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
398 if (ofd->ofd_tot_granted_clients > 0)
399 ofd->ofd_tot_granted_clients --;
402 if (!(exp->exp_flags & OBD_OPT_FORCE))
403 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
405 LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
409 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
411 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
413 CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
414 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
417 int ofd_obd_postrecov(struct obd_device *obd)
420 struct lu_device *ldev = obd->obd_lu_dev;
425 rc = lu_env_init(&env, LCT_DT_THREAD);
428 ofd_info_init(&env, obd->obd_self_export);
430 rc = ofd_postrecov(&env, ofd_dev(ldev));
436 /* This is not called from request handler, check ofd_set_info_hdl() instead
437 * this OBD functions is only used by class_notify_sptlrpc_conf() locally
438 * by direct obd_set_info_async() call */
439 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
440 __u32 keylen, void *key, __u32 vallen, void *val,
441 struct ptlrpc_request_set *set)
447 if (exp->exp_obd == NULL) {
448 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
452 if (KEY_IS(KEY_SPTLRPC_CONF)) {
453 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
455 CERROR("%s: Unsupported key %s\n",
456 exp->exp_obd->obd_name, (char*)key);
462 /* used by nrs_orr_range_fill_physical() in ptlrpc, see LU-3239 */
463 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
464 __u32 keylen, void *key, __u32 *vallen, void *val,
465 struct lov_stripe_md *lsm)
467 struct ofd_thread_info *info;
468 struct ofd_device *ofd;
469 struct ll_fiemap_info_key *fm_key = key;
470 struct ll_user_fiemap *fiemap = val;
475 if (exp->exp_obd == NULL) {
476 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
482 if (KEY_IS(KEY_FIEMAP)) {
483 info = ofd_info_init(env, exp);
485 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi,
486 ofd->ofd_lut.lut_lsd.lsd_osd_index);
490 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
492 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
499 /** helper function for statfs, also used by grant code */
500 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
501 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
506 down_read(&ofd->ofd_lastid_rwsem);
507 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
508 * we may do that in the future.
509 * Return -ENOSPC until the LAST_ID rebuilt. */
510 if (unlikely(ofd->ofd_lastid_rebuilding))
511 GOTO(out, rc = -ENOSPC);
513 spin_lock(&ofd->ofd_osfs_lock);
514 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
517 /* statfs data are too old, get up-to-date one.
518 * we must be cautious here since multiple threads might be
519 * willing to update statfs data concurrently and we must
520 * grant that cached statfs data are always consistent */
522 if (ofd->ofd_statfs_inflight == 0)
523 /* clear inflight counter if no users, although it would
524 * take a while to overflow this 64-bit counter ... */
525 ofd->ofd_osfs_inflight = 0;
526 /* notify ofd_grant_commit() that we want to track writes
527 * completed as of now */
528 ofd->ofd_statfs_inflight++;
529 /* record value of inflight counter before running statfs to
530 * compute the diff once statfs is completed */
531 unstable = ofd->ofd_osfs_inflight;
532 spin_unlock(&ofd->ofd_osfs_lock);
534 /* statfs can sleep ... hopefully not for too long since we can
535 * call it fairly often as space fills up */
536 rc = dt_statfs(env, ofd->ofd_osd, osfs);
540 spin_lock(&ofd->ofd_grant_lock);
541 spin_lock(&ofd->ofd_osfs_lock);
542 /* calculate how much space was written while we released the
544 unstable = ofd->ofd_osfs_inflight - unstable;
545 ofd->ofd_osfs_unstable = 0;
547 /* some writes completed while we were running statfs
548 * w/o the ofd_osfs_lock. Those ones got added to
549 * the cached statfs data that we are about to crunch.
550 * Take them into account in the new statfs data */
551 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
552 unstable >> ofd->ofd_blockbits);
553 /* However, we don't really know if those writes got
554 * accounted in the statfs call, so tell
555 * ofd_grant_space_left() there is some uncertainty
556 * on the accounting of those writes.
557 * The purpose is to prevent spurious error messages in
558 * ofd_grant_space_left() since those writes might be
559 * accounted twice. */
560 ofd->ofd_osfs_unstable += unstable;
562 /* similarly, there is some uncertainty on write requests
563 * between prepare & commit */
564 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
565 spin_unlock(&ofd->ofd_grant_lock);
567 /* finally udpate cached statfs data */
568 ofd->ofd_osfs = *osfs;
569 ofd->ofd_osfs_age = cfs_time_current_64();
571 ofd->ofd_statfs_inflight--; /* stop tracking */
572 if (ofd->ofd_statfs_inflight == 0)
573 ofd->ofd_osfs_inflight = 0;
574 spin_unlock(&ofd->ofd_osfs_lock);
579 /* use cached statfs data */
580 *osfs = ofd->ofd_osfs;
581 spin_unlock(&ofd->ofd_osfs_lock);
589 up_read(&ofd->ofd_lastid_rwsem);
594 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
595 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
597 struct obd_device *obd = class_exp2obd(exp);
598 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
603 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
607 /* at least try to account for cached pages. its still racy and
608 * might be under-reporting if clients haven't announced their
609 * caches with brw recently */
611 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
612 " pending "LPU64" free "LPU64" avail "LPU64"\n",
613 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
614 osfs->os_bfree << ofd->ofd_blockbits,
615 osfs->os_bavail << ofd->ofd_blockbits);
617 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
618 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
619 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
621 /* The QoS code on the MDS does not care about space reserved for
622 * precreate, so take it out. */
623 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
624 struct filter_export_data *fed;
626 fed = &obd->obd_self_export->exp_filter_data;
627 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
628 fed->fed_grant >> ofd->ofd_blockbits);
631 ofd_grant_sanity_check(obd, __FUNCTION__);
632 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
633 LPU64" objects: "LPU64" free; state %x\n",
634 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
635 osfs->os_files, osfs->os_ffree, osfs->os_state);
637 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
638 ofd->ofd_lut.lut_lsd.lsd_osd_index))
641 /* OS_STATE_READONLY can be set by OSD already */
642 if (ofd->ofd_raid_degraded)
643 osfs->os_state |= OS_STATE_DEGRADED;
645 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
646 /* clients which don't support OBD_CONNECT_GRANT_PARAM
647 * should not see a block size > page size, otherwise
648 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
649 * block size which is the biggest block size known to work
650 * with all client's page size. */
651 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
652 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
653 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
654 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
657 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
658 ofd->ofd_lut.lut_lsd.lsd_osd_index))
659 osfs->os_bfree = osfs->os_bavail = 2;
666 /* needed by echo client only for now, RPC handler uses ofd_setattr_hdl() */
667 int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
668 struct obd_info *oinfo, struct obd_trans_info *oti)
670 struct ofd_thread_info *info;
671 struct ofd_device *ofd = ofd_exp(exp);
672 struct ldlm_namespace *ns = ofd->ofd_namespace;
673 struct ldlm_resource *res;
674 struct ofd_object *fo;
675 struct obdo *oa = oinfo->oi_oa;
676 struct lu_fid *fid = &oa->o_oi.oi_fid;
677 struct filter_fid *ff = NULL;
682 info = ofd_info_init(env, exp);
684 ost_fid_build_resid(fid, &info->fti_resid);
686 /* This would be very bad - accidentally truncating a file when
687 * changing the time or similar - bug 12203. */
688 if (oa->o_valid & OBD_MD_FLSIZE &&
689 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
690 static char mdsinum[48];
692 if (oa->o_valid & OBD_MD_FLFID)
693 snprintf(mdsinum, sizeof(mdsinum) - 1,
694 "of parent "DFID, oa->o_parent_seq,
695 oa->o_parent_oid, 0);
699 CERROR("%s: setattr from %s trying to truncate object "DFID
700 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
702 GOTO(out, rc = -EPERM);
705 fo = ofd_object_find_exists(env, ofd, fid);
707 CERROR("%s: can't find object "DFID"\n",
708 ofd_name(ofd), PFID(fid));
709 GOTO(out, rc = PTR_ERR(fo));
712 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
713 info->fti_attr.la_valid &= ~LA_TYPE;
715 if (oa->o_valid & OBD_MD_FLFID) {
716 ff = &info->fti_mds_fid;
717 ofd_prepare_fidea(ff, oa);
720 /* setting objects attributes (including owner/group) */
721 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
723 GOTO(out_unlock, rc);
725 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
728 ofd_object_put(env, fo);
731 /* we do not call this before to avoid lu_object_find() in
732 * ->lvbo_update() holding another reference on the object.
733 * otherwise concurrent destroy can make the object unavailable
734 * for 2nd lu_object_find() waiting for the first reference
735 * to go... deadlock! */
736 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
738 ldlm_res_lvbo_update(res, NULL, 0);
739 ldlm_resource_putref(res);
746 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
747 const struct lu_fid *fid, int orphan)
749 struct ofd_thread_info *info = ofd_info(env);
750 struct lustre_handle lockh;
751 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
753 ldlm_policy_data_t policy = {
754 .l_extent = { 0, OBD_OBJECT_EOF }
756 struct ofd_object *fo;
760 fo = ofd_object_find_exists(env, ofd, fid);
764 /* Tell the clients that the object is gone now and that they should
765 * throw away any cached pages. */
766 ost_fid_build_resid(fid, &info->fti_resid);
767 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
768 LDLM_EXTENT, &policy, LCK_PW, &flags,
769 ldlm_blocking_ast, ldlm_completion_ast,
770 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
772 /* We only care about the side-effects, just drop the lock. */
774 ldlm_lock_decref(&lockh, LCK_PW);
778 rc = ofd_object_destroy(env, fo, orphan);
781 ofd_object_put(env, fo);
785 /* needed by echo client only for now, RPC handler uses ofd_destroy_hdl() */
786 int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
787 struct obdo *oa, struct lov_stripe_md *md,
788 struct obd_trans_info *oti, struct obd_export *md_exp,
791 struct ofd_device *ofd = ofd_exp(exp);
792 struct lu_fid *fid = &oa->o_oi.oi_fid;
797 ofd_info_init(env, exp);
799 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
801 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
803 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
804 ofd_name(ofd), PFID(fid));
806 } else if (rc != 0) {
807 CERROR("%s: error destroying object "DFID": %d\n",
808 ofd_name(ofd), PFID(fid), rc);
816 /* needed by echo client only for now, RPC handler uses ofd_create_hdl()
817 * It is much simpler and just create objects */
818 int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
819 struct obdo *oa, struct lov_stripe_md **ea,
820 struct obd_trans_info *oti)
822 struct ofd_device *ofd = ofd_exp(exp);
823 struct ofd_thread_info *info;
824 obd_seq seq = ostid_seq(&oa->o_oi);
825 struct ofd_seq *oseq;
826 int rc = 0, diff = 1;
832 info = ofd_info_init(env, exp);
834 LASSERT(seq == FID_SEQ_ECHO);
835 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
837 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
839 down_read(&ofd->ofd_lastid_rwsem);
840 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
841 * we may do that in the future.
842 * Return -ENOSPC until the LAST_ID rebuilt. */
843 if (unlikely(ofd->ofd_lastid_rebuilding))
844 GOTO(out_sem, rc = -ENOSPC);
846 oseq = ofd_seq_load(env, ofd, seq);
848 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
849 ofd_name(ofd), seq, PTR_ERR(oseq));
850 GOTO(out_sem, rc = -EINVAL);
853 mutex_lock(&oseq->os_create_lock);
854 rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
856 CDEBUG(D_HA, "%s: failed to acquire grant space for "
857 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
862 next_id = ofd_seq_last_oid(oseq) + 1;
863 count = ofd_precreate_batch(ofd, diff);
865 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
867 CERROR("%s: unable to precreate: rc = %d\n",
870 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
871 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
875 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
877 mutex_unlock(&oseq->os_create_lock);
878 if (rc == 0 && ea != NULL) {
879 struct lov_stripe_md *lsm = *ea;
881 lsm->lsm_oi = oa->o_oi;
883 ofd_seq_put(env, oseq);
886 up_read(&ofd->ofd_lastid_rwsem);
890 /* needed by echo client only for now, RPC handler uses ofd_getattr_hdl() */
891 int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
892 struct obd_info *oinfo)
894 struct ofd_device *ofd = ofd_exp(exp);
895 struct ofd_thread_info *info;
896 struct lu_fid *fid = &oinfo->oi_oa->o_oi.oi_fid;
897 struct ofd_object *fo;
902 info = ofd_info_init(env, exp);
904 fo = ofd_object_find_exists(env, ofd, fid);
906 GOTO(out, rc = PTR_ERR(fo));
909 rc = ofd_attr_get(env, fo, &info->fti_attr);
910 oinfo->oi_oa->o_valid = OBD_MD_FLID;
914 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
915 OFD_VALID_FLAGS | LA_UID | LA_GID);
917 /* Store object version in reply */
918 curr_version = dt_version_get(env, ofd_object_child(fo));
919 if ((__s64)curr_version != -EOPNOTSUPP) {
920 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
921 oinfo->oi_oa->o_data_version = curr_version;
925 ofd_object_put(env, fo);
930 static int ofd_ioc_get_obj_version(const struct lu_env *env,
931 struct ofd_device *ofd, void *karg)
933 struct obd_ioctl_data *data = karg;
935 struct ofd_object *fo;
936 dt_obj_version_t version;
941 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
942 GOTO(out, rc = -EINVAL);
944 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
945 fid = *(struct lu_fid *)data->ioc_inlbuf1;
946 } else if (data->ioc_inlbuf3 != NULL &&
947 data->ioc_inllen3 == sizeof(__u64) &&
948 data->ioc_inlbuf4 != NULL &&
949 data->ioc_inllen4 == sizeof(__u64)) {
952 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
953 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
954 rc = ostid_to_fid(&fid, &ostid,
955 ofd->ofd_lut.lut_lsd.lsd_osd_index);
959 GOTO(out, rc = -EINVAL);
962 if (!fid_is_sane(&fid))
963 GOTO(out, rc = -EINVAL);
965 fo = ofd_object_find(env, ofd, &fid);
967 GOTO(out, rc = PTR_ERR(fo));
969 if (!ofd_object_exists(fo))
970 GOTO(out_fo, rc = -ENOENT);
972 if (lu_object_remote(&fo->ofo_obj.do_lu))
973 GOTO(out_fo, rc = -EREMOTE);
975 version = dt_version_get(env, ofd_object_child(fo));
977 GOTO(out_fo, rc = -EIO);
979 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
983 ofd_object_put(env, fo);
988 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
989 void *karg, void *uarg)
992 struct ofd_device *ofd = ofd_exp(exp);
993 struct obd_device *obd = ofd_obd(ofd);
998 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
999 rc = lu_env_init(&env, LCT_DT_THREAD);
1004 case OBD_IOC_ABORT_RECOVERY:
1005 CERROR("%s: aborting recovery\n", obd->obd_name);
1006 target_stop_recovery_thread(obd);
1009 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1010 rc = dt_sync(&env, ofd->ofd_osd);
1012 case OBD_IOC_SET_READONLY:
1013 rc = dt_sync(&env, ofd->ofd_osd);
1015 rc = dt_ro(&env, ofd->ofd_osd);
1017 case OBD_IOC_START_LFSCK: {
1018 struct obd_ioctl_data *data = karg;
1019 struct lfsck_start_param lsp;
1021 if (unlikely(data == NULL)) {
1026 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1027 lsp.lsp_index_valid = 0;
1028 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1031 case OBD_IOC_STOP_LFSCK: {
1032 struct obd_ioctl_data *data = karg;
1033 struct lfsck_stop stop;
1035 stop.ls_status = LS_STOPPED;
1036 /* Old lfsck utils may pass NULL @stop. */
1037 if (data->ioc_inlbuf1 == NULL)
1041 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1043 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1046 case OBD_IOC_GET_OBJ_VERSION:
1047 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1050 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1058 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1065 case OBD_CLEANUP_EARLY:
1067 case OBD_CLEANUP_EXPORTS:
1068 target_cleanup_recovery(obd);
1074 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1076 ofd_fmd_expire(exp);
1080 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1082 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1083 struct ofd_thread_info *info;
1085 #ifdef USE_HEALTH_CHECK_WRITE
1090 /* obd_proc_read_health pass NULL env, we need real one */
1091 rc = lu_env_init(&env, LCT_DT_THREAD);
1095 info = ofd_info_init(&env, NULL);
1096 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1100 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1101 GOTO(out, rc = -EROFS);
1103 #ifdef USE_HEALTH_CHECK_WRITE
1104 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1105 if (info->fti_buf.lb_buf == NULL)
1106 GOTO(out, rc = -ENOMEM);
1108 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1111 th = dt_trans_create(&env, ofd->ofd_osd);
1113 GOTO(out, rc = PTR_ERR(th));
1115 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1116 &info->fti_buf, info->fti_off, th);
1118 th->th_sync = 1; /* sync IO is needed */
1119 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1121 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1122 &info->fti_buf, &info->fti_off,
1125 dt_trans_stop(&env, ofd->ofd_osd, th);
1127 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1129 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1137 * Handle quota control requests to consult current usage/limit.
1139 * \param obd - is the obd device associated with the ofd
1140 * \param exp - is the client's export
1141 * \param oqctl - is the obd_quotactl request to be processed
1143 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1144 struct obd_quotactl *oqctl)
1146 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1151 /* report success for quota on/off for interoperability with current MDT
1153 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1156 rc = lu_env_init(&env, LCT_DT_THREAD);
1160 rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1166 struct obd_ops ofd_obd_ops = {
1167 .o_owner = THIS_MODULE,
1168 .o_connect = ofd_obd_connect,
1169 .o_reconnect = ofd_obd_reconnect,
1170 .o_disconnect = ofd_obd_disconnect,
1171 .o_create = ofd_echo_create,
1172 .o_setattr = ofd_echo_setattr,
1173 .o_preprw = ofd_preprw,
1174 .o_commitrw = ofd_commitrw,
1175 .o_destroy = ofd_echo_destroy,
1176 .o_init_export = ofd_init_export,
1177 .o_destroy_export = ofd_destroy_export,
1178 .o_postrecov = ofd_obd_postrecov,
1179 .o_getattr = ofd_echo_getattr,
1180 .o_iocontrol = ofd_iocontrol,
1181 .o_precleanup = ofd_precleanup,
1183 .o_health_check = ofd_health_check,
1184 .o_quotactl = ofd_quotactl,
1185 .o_set_info_async = ofd_set_info_async,
1186 .o_get_info = ofd_get_info,