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 static 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)
505 spin_lock(&ofd->ofd_osfs_lock);
506 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
509 /* statfs data are too old, get up-to-date one.
510 * we must be cautious here since multiple threads might be
511 * willing to update statfs data concurrently and we must
512 * grant that cached statfs data are always consistent */
514 if (ofd->ofd_statfs_inflight == 0)
515 /* clear inflight counter if no users, although it would
516 * take a while to overflow this 64-bit counter ... */
517 ofd->ofd_osfs_inflight = 0;
518 /* notify ofd_grant_commit() that we want to track writes
519 * completed as of now */
520 ofd->ofd_statfs_inflight++;
521 /* record value of inflight counter before running statfs to
522 * compute the diff once statfs is completed */
523 unstable = ofd->ofd_osfs_inflight;
524 spin_unlock(&ofd->ofd_osfs_lock);
526 /* statfs can sleep ... hopefully not for too long since we can
527 * call it fairly often as space fills up */
528 rc = dt_statfs(env, ofd->ofd_osd, osfs);
532 spin_lock(&ofd->ofd_grant_lock);
533 spin_lock(&ofd->ofd_osfs_lock);
534 /* calculate how much space was written while we released the
536 unstable = ofd->ofd_osfs_inflight - unstable;
537 ofd->ofd_osfs_unstable = 0;
539 /* some writes completed while we were running statfs
540 * w/o the ofd_osfs_lock. Those ones got added to
541 * the cached statfs data that we are about to crunch.
542 * Take them into account in the new statfs data */
543 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
544 unstable >> ofd->ofd_blockbits);
545 /* However, we don't really know if those writes got
546 * accounted in the statfs call, so tell
547 * ofd_grant_space_left() there is some uncertainty
548 * on the accounting of those writes.
549 * The purpose is to prevent spurious error messages in
550 * ofd_grant_space_left() since those writes might be
551 * accounted twice. */
552 ofd->ofd_osfs_unstable += unstable;
554 /* similarly, there is some uncertainty on write requests
555 * between prepare & commit */
556 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
557 spin_unlock(&ofd->ofd_grant_lock);
559 /* finally udpate cached statfs data */
560 ofd->ofd_osfs = *osfs;
561 ofd->ofd_osfs_age = cfs_time_current_64();
563 ofd->ofd_statfs_inflight--; /* stop tracking */
564 if (ofd->ofd_statfs_inflight == 0)
565 ofd->ofd_osfs_inflight = 0;
566 spin_unlock(&ofd->ofd_osfs_lock);
571 /* use cached statfs data */
572 *osfs = ofd->ofd_osfs;
573 spin_unlock(&ofd->ofd_osfs_lock);
580 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
581 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
583 struct obd_device *obd = class_exp2obd(exp);
584 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
589 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
593 /* at least try to account for cached pages. its still racy and
594 * might be under-reporting if clients haven't announced their
595 * caches with brw recently */
597 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
598 " pending "LPU64" free "LPU64" avail "LPU64"\n",
599 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
600 osfs->os_bfree << ofd->ofd_blockbits,
601 osfs->os_bavail << ofd->ofd_blockbits);
603 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
604 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
605 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
607 /* The QoS code on the MDS does not care about space reserved for
608 * precreate, so take it out. */
609 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
610 struct filter_export_data *fed;
612 fed = &obd->obd_self_export->exp_filter_data;
613 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
614 fed->fed_grant >> ofd->ofd_blockbits);
617 ofd_grant_sanity_check(obd, __FUNCTION__);
618 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
619 LPU64" objects: "LPU64" free; state %x\n",
620 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
621 osfs->os_files, osfs->os_ffree, osfs->os_state);
623 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
624 ofd->ofd_lut.lut_lsd.lsd_osd_index))
627 /* OS_STATE_READONLY can be set by OSD already */
628 if (ofd->ofd_raid_degraded)
629 osfs->os_state |= OS_STATE_DEGRADED;
631 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
632 /* clients which don't support OBD_CONNECT_GRANT_PARAM
633 * should not see a block size > page size, otherwise
634 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
635 * block size which is the biggest block size known to work
636 * with all client's page size. */
637 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
638 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
639 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
640 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
643 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
644 ofd->ofd_lut.lut_lsd.lsd_osd_index))
645 osfs->os_bfree = osfs->os_bavail = 2;
652 /* needed by echo client only for now, RPC handler uses ofd_setattr_hdl() */
653 int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
654 struct obd_info *oinfo, struct obd_trans_info *oti)
656 struct ofd_thread_info *info;
657 struct ofd_device *ofd = ofd_exp(exp);
658 struct ldlm_namespace *ns = ofd->ofd_namespace;
659 struct ldlm_resource *res;
660 struct ofd_object *fo;
661 struct obdo *oa = oinfo->oi_oa;
662 struct lu_fid *fid = &oa->o_oi.oi_fid;
663 struct filter_fid *ff = NULL;
668 info = ofd_info_init(env, exp);
670 ost_fid_build_resid(fid, &info->fti_resid);
672 /* This would be very bad - accidentally truncating a file when
673 * changing the time or similar - bug 12203. */
674 if (oa->o_valid & OBD_MD_FLSIZE &&
675 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
676 static char mdsinum[48];
678 if (oa->o_valid & OBD_MD_FLFID)
679 snprintf(mdsinum, sizeof(mdsinum) - 1,
680 "of parent "DFID, oa->o_parent_seq,
681 oa->o_parent_oid, 0);
685 CERROR("%s: setattr from %s trying to truncate object "DFID
686 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
688 GOTO(out, rc = -EPERM);
691 fo = ofd_object_find_exists(env, ofd, fid);
693 CERROR("%s: can't find object "DFID"\n",
694 ofd_name(ofd), PFID(fid));
695 GOTO(out, rc = PTR_ERR(fo));
698 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
699 info->fti_attr.la_valid &= ~LA_TYPE;
701 if (oa->o_valid & OBD_MD_FLFID) {
702 ff = &info->fti_mds_fid;
703 ofd_prepare_fidea(ff, oa);
706 /* setting objects attributes (including owner/group) */
707 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
709 GOTO(out_unlock, rc);
711 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
714 ofd_object_put(env, fo);
717 /* we do not call this before to avoid lu_object_find() in
718 * ->lvbo_update() holding another reference on the object.
719 * otherwise concurrent destroy can make the object unavailable
720 * for 2nd lu_object_find() waiting for the first reference
721 * to go... deadlock! */
722 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
724 ldlm_res_lvbo_update(res, NULL, 0);
725 ldlm_resource_putref(res);
732 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
733 const struct lu_fid *fid, int orphan)
735 struct ofd_thread_info *info = ofd_info(env);
736 struct lustre_handle lockh;
737 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
739 ldlm_policy_data_t policy = {
740 .l_extent = { 0, OBD_OBJECT_EOF }
742 struct ofd_object *fo;
746 fo = ofd_object_find_exists(env, ofd, fid);
750 /* Tell the clients that the object is gone now and that they should
751 * throw away any cached pages. */
752 ost_fid_build_resid(fid, &info->fti_resid);
753 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
754 LDLM_EXTENT, &policy, LCK_PW, &flags,
755 ldlm_blocking_ast, ldlm_completion_ast,
756 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
758 /* We only care about the side-effects, just drop the lock. */
760 ldlm_lock_decref(&lockh, LCK_PW);
764 rc = ofd_object_destroy(env, fo, orphan);
767 ofd_object_put(env, fo);
771 /* needed by echo client only for now, RPC handler uses ofd_destroy_hdl() */
772 int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
773 struct obdo *oa, struct lov_stripe_md *md,
774 struct obd_trans_info *oti, struct obd_export *md_exp,
777 struct ofd_device *ofd = ofd_exp(exp);
778 struct lu_fid *fid = &oa->o_oi.oi_fid;
783 ofd_info_init(env, exp);
785 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
787 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
789 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
790 ofd_name(ofd), PFID(fid));
792 } else if (rc != 0) {
793 CERROR("%s: error destroying object "DFID": %d\n",
794 ofd_name(ofd), PFID(fid), rc);
802 /* needed by echo client only for now, RPC handler uses ofd_create_hdl()
803 * It is much simpler and just create objects */
804 int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
805 struct obdo *oa, struct lov_stripe_md **ea,
806 struct obd_trans_info *oti)
808 struct ofd_device *ofd = ofd_exp(exp);
809 struct ofd_thread_info *info;
810 obd_seq seq = ostid_seq(&oa->o_oi);
811 struct ofd_seq *oseq;
812 int rc = 0, diff = 1;
818 info = ofd_info_init(env, exp);
820 LASSERT(seq == FID_SEQ_ECHO);
821 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
823 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
825 oseq = ofd_seq_load(env, ofd, seq);
827 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
828 ofd_name(ofd), seq, PTR_ERR(oseq));
832 mutex_lock(&oseq->os_create_lock);
833 rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
835 CDEBUG(D_HA, "%s: failed to acquire grant space for "
836 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
841 next_id = ofd_seq_last_oid(oseq) + 1;
842 count = ofd_precreate_batch(ofd, diff);
844 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
846 CERROR("%s: unable to precreate: rc = %d\n",
849 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
850 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
854 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
856 mutex_unlock(&oseq->os_create_lock);
857 if (rc == 0 && ea != NULL) {
858 struct lov_stripe_md *lsm = *ea;
860 lsm->lsm_oi = oa->o_oi;
862 ofd_seq_put(env, oseq);
866 /* needed by echo client only for now, RPC handler uses ofd_getattr_hdl() */
867 int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
868 struct obd_info *oinfo)
870 struct ofd_device *ofd = ofd_exp(exp);
871 struct ofd_thread_info *info;
872 struct lu_fid *fid = &oinfo->oi_oa->o_oi.oi_fid;
873 struct ofd_object *fo;
878 info = ofd_info_init(env, exp);
880 fo = ofd_object_find_exists(env, ofd, fid);
882 GOTO(out, rc = PTR_ERR(fo));
885 rc = ofd_attr_get(env, fo, &info->fti_attr);
886 oinfo->oi_oa->o_valid = OBD_MD_FLID;
890 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
891 OFD_VALID_FLAGS | LA_UID | LA_GID);
893 /* Store object version in reply */
894 curr_version = dt_version_get(env, ofd_object_child(fo));
895 if ((__s64)curr_version != -EOPNOTSUPP) {
896 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
897 oinfo->oi_oa->o_data_version = curr_version;
901 ofd_object_put(env, fo);
906 static int ofd_ioc_get_obj_version(const struct lu_env *env,
907 struct ofd_device *ofd, void *karg)
909 struct obd_ioctl_data *data = karg;
911 struct ofd_object *fo;
912 dt_obj_version_t version;
917 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
918 GOTO(out, rc = -EINVAL);
920 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
921 fid = *(struct lu_fid *)data->ioc_inlbuf1;
922 } else if (data->ioc_inlbuf3 != NULL &&
923 data->ioc_inllen3 == sizeof(__u64) &&
924 data->ioc_inlbuf4 != NULL &&
925 data->ioc_inllen4 == sizeof(__u64)) {
928 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
929 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
930 rc = ostid_to_fid(&fid, &ostid,
931 ofd->ofd_lut.lut_lsd.lsd_osd_index);
935 GOTO(out, rc = -EINVAL);
938 if (!fid_is_sane(&fid))
939 GOTO(out, rc = -EINVAL);
941 fo = ofd_object_find(env, ofd, &fid);
943 GOTO(out, rc = PTR_ERR(fo));
945 if (!ofd_object_exists(fo))
946 GOTO(out_fo, rc = -ENOENT);
948 if (lu_object_remote(&fo->ofo_obj.do_lu))
949 GOTO(out_fo, rc = -EREMOTE);
951 version = dt_version_get(env, ofd_object_child(fo));
953 GOTO(out_fo, rc = -EIO);
955 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
959 ofd_object_put(env, fo);
964 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
965 void *karg, void *uarg)
968 struct ofd_device *ofd = ofd_exp(exp);
969 struct obd_device *obd = ofd_obd(ofd);
974 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
975 rc = lu_env_init(&env, LCT_DT_THREAD);
980 case OBD_IOC_ABORT_RECOVERY:
981 CERROR("%s: aborting recovery\n", obd->obd_name);
982 target_stop_recovery_thread(obd);
985 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
986 rc = dt_sync(&env, ofd->ofd_osd);
988 case OBD_IOC_SET_READONLY:
989 rc = dt_sync(&env, ofd->ofd_osd);
991 rc = dt_ro(&env, ofd->ofd_osd);
993 case OBD_IOC_START_LFSCK: {
994 struct obd_ioctl_data *data = karg;
995 struct lfsck_start_param lsp;
997 if (unlikely(data == NULL)) {
1002 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1003 lsp.lsp_namespace = ofd->ofd_namespace;
1004 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1007 case OBD_IOC_STOP_LFSCK: {
1008 rc = lfsck_stop(&env, ofd->ofd_osd, false);
1011 case OBD_IOC_GET_OBJ_VERSION:
1012 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1015 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1023 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1030 case OBD_CLEANUP_EARLY:
1032 case OBD_CLEANUP_EXPORTS:
1033 target_cleanup_recovery(obd);
1039 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1041 ofd_fmd_expire(exp);
1045 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1047 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1048 struct ofd_thread_info *info;
1050 #ifdef USE_HEALTH_CHECK_WRITE
1055 /* obd_proc_read_health pass NULL env, we need real one */
1056 rc = lu_env_init(&env, LCT_DT_THREAD);
1060 info = ofd_info_init(&env, NULL);
1061 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1065 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1066 GOTO(out, rc = -EROFS);
1068 #ifdef USE_HEALTH_CHECK_WRITE
1069 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1070 if (info->fti_buf.lb_buf == NULL)
1071 GOTO(out, rc = -ENOMEM);
1073 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1076 th = dt_trans_create(&env, ofd->ofd_osd);
1078 GOTO(out, rc = PTR_ERR(th));
1080 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1081 info->fti_buf.lb_len, info->fti_off, th);
1083 th->th_sync = 1; /* sync IO is needed */
1084 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1086 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1087 &info->fti_buf, &info->fti_off,
1090 dt_trans_stop(&env, ofd->ofd_osd, th);
1092 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1094 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1102 * Handle quota control requests to consult current usage/limit.
1104 * \param obd - is the obd device associated with the ofd
1105 * \param exp - is the client's export
1106 * \param oqctl - is the obd_quotactl request to be processed
1108 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1109 struct obd_quotactl *oqctl)
1111 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1116 /* report success for quota on/off for interoperability with current MDT
1118 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1121 rc = lu_env_init(&env, LCT_DT_THREAD);
1125 rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1131 struct obd_ops ofd_obd_ops = {
1132 .o_owner = THIS_MODULE,
1133 .o_connect = ofd_obd_connect,
1134 .o_reconnect = ofd_obd_reconnect,
1135 .o_disconnect = ofd_obd_disconnect,
1136 .o_create = ofd_echo_create,
1137 .o_setattr = ofd_echo_setattr,
1138 .o_preprw = ofd_preprw,
1139 .o_commitrw = ofd_commitrw,
1140 .o_destroy = ofd_echo_destroy,
1141 .o_init_export = ofd_init_export,
1142 .o_destroy_export = ofd_destroy_export,
1143 .o_postrecov = ofd_obd_postrecov,
1144 .o_getattr = ofd_echo_getattr,
1145 .o_iocontrol = ofd_iocontrol,
1146 .o_precleanup = ofd_precleanup,
1148 .o_health_check = ofd_health_check,
1149 .o_quotactl = ofd_quotactl,
1150 .o_set_info_async = ofd_set_info_async,
1151 .o_get_info = ofd_get_info,