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_ioctl.h>
49 #include <lustre_quota.h>
50 #include <lustre_lfsck.h>
52 static int ofd_export_stats_init(struct ofd_device *ofd,
53 struct obd_export *exp, void *client_nid)
55 struct obd_device *obd = ofd_obd(ofd);
56 struct nid_stat *stats;
62 LASSERT(obd->obd_uses_nid_stats);
64 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
65 /* Self-export gets no proc entry */
68 rc = lprocfs_exp_setup(exp, client_nid, &newnid);
70 /* Mask error for already created
80 stats = exp->exp_nid_stats;
81 LASSERT(stats != NULL);
83 num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
85 stats->nid_stats = lprocfs_alloc_stats(num_stats,
86 LPROCFS_STATS_FLAG_NOPERCPU);
87 if (stats->nid_stats == NULL)
90 lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
91 ofd_stats_counter_init(stats->nid_stats);
92 rc = lprocfs_register_stats(stats->nid_proc, "stats",
97 rc = lprocfs_nid_ldlm_stats_init(stats);
99 lprocfs_free_stats(&stats->nid_stats);
108 static int ofd_parse_connect_data(const struct lu_env *env,
109 struct obd_export *exp,
110 struct obd_connect_data *data,
113 struct ofd_device *ofd = ofd_exp(exp);
114 struct filter_export_data *fed = &exp->exp_filter_data;
119 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
120 " ocd_version: %x ocd_grant: %d ocd_index: %u"
122 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
123 data->ocd_connect_flags, data->ocd_version,
124 data->ocd_grant, data->ocd_index, data->ocd_group);
126 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
127 CWARN("!!! This export (nid %s) used object group %d "
128 "earlier; now it's trying to use group %d! This could "
129 "be a bug in the MDS. Please report to "
130 "http://bugs.whamcloud.com/\n",
131 obd_export_nid2str(exp), fed->fed_group,
135 fed->fed_group = data->ocd_group;
137 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
138 data->ocd_version = LUSTRE_VERSION_CODE;
140 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
141 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
142 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
143 exp->exp_obd->obd_name, data->ocd_group);
144 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
147 if (ofd_grant_param_supp(exp)) {
148 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
149 /* ocd_{blocksize,inodespace} are log2 values */
150 data->ocd_blocksize = ofd->ofd_blockbits;
151 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
152 /* ocd_grant_extent is in 1K blocks */
153 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
156 if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
157 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
160 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
161 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
162 int index = lsd->lsd_osd_index;
164 if (index != data->ocd_index) {
165 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
166 " %u doesn't match actual OST index"
167 " %u in last_rcvd file, bad "
169 obd_export_nid2str(exp), index,
173 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
174 /* this will only happen on the first connect */
175 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
176 /* sync is not needed here as lut_client_add will
177 * set exp_need_sync flag */
178 tgt_server_data_update(env, &ofd->ofd_lut, 0);
181 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
182 data->ocd_brw_size = 65536;
183 } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
184 data->ocd_brw_size = min(data->ocd_brw_size,
185 (__u32)DT_MAX_BRW_SIZE);
186 if (data->ocd_brw_size == 0) {
187 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
188 " ocd_version: %x ocd_grant: %d ocd_index: %u "
189 "ocd_brw_size is unexpectedly zero, "
190 "network data corruption?"
191 "Refusing connection of this client\n",
192 exp->exp_obd->obd_name,
193 exp->exp_client_uuid.uuid,
194 exp, data->ocd_connect_flags, data->ocd_version,
195 data->ocd_grant, data->ocd_index);
200 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
201 __u32 cksum_types = data->ocd_cksum_types;
203 /* The client set in ocd_cksum_types the checksum types it
204 * supports. We have to mask off the algorithms that we don't
206 data->ocd_cksum_types &= cksum_types_supported_server();
208 if (unlikely(data->ocd_cksum_types == 0)) {
209 CERROR("%s: Connect with checksum support but no "
210 "ocd_cksum_types is set\n",
211 exp->exp_obd->obd_name);
215 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
216 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
217 cksum_types, data->ocd_cksum_types);
219 /* This client does not support OBD_CONNECT_CKSUM
220 * fall back to CRC32 */
221 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
222 "OBD_CONNECT_CKSUM, CRC32 will be used\n",
223 exp->exp_obd->obd_name, obd_export_nid2str(exp));
226 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
227 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
229 if (OCD_HAS_FLAG(data, PINGLESS)) {
230 if (ptlrpc_pinger_suppress_pings()) {
231 spin_lock(&exp->exp_obd->obd_dev_lock);
232 list_del_init(&exp->exp_obd_chain_timed);
233 spin_unlock(&exp->exp_obd->obd_dev_lock);
235 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
242 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
243 struct obd_device *obd, struct obd_uuid *cluuid,
244 struct obd_connect_data *data, void *localdata)
246 struct ofd_device *ofd;
251 if (exp == NULL || obd == NULL || cluuid == NULL)
254 ofd = ofd_dev(obd->obd_lu_dev);
256 rc = ofd_parse_connect_data(env, exp, data, false);
258 ofd_export_stats_init(ofd, exp, localdata);
263 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
264 struct obd_device *obd, struct obd_uuid *cluuid,
265 struct obd_connect_data *data, void *localdata)
267 struct obd_export *exp;
268 struct ofd_device *ofd;
269 struct lustre_handle conn = { 0 };
273 if (_exp == NULL || obd == NULL || cluuid == NULL)
276 ofd = ofd_dev(obd->obd_lu_dev);
278 rc = class_connect(&conn, obd, cluuid);
282 exp = class_conn2export(&conn);
283 LASSERT(exp != NULL);
285 rc = ofd_parse_connect_data(env, exp, data, true);
289 if (obd->obd_replayable) {
290 struct tg_export_data *ted = &exp->exp_target_data;
292 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
293 sizeof(ted->ted_lcd->lcd_uuid));
294 rc = tgt_client_new(env, exp);
297 ofd_export_stats_init(ofd, exp, localdata);
300 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
301 data ? data->ocd_group : -1);
305 class_disconnect(exp);
313 int ofd_obd_disconnect(struct obd_export *exp)
315 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
322 class_export_get(exp);
324 if (!(exp->exp_flags & OBD_OPT_FORCE))
325 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
327 rc = server_disconnect_export(exp);
329 ofd_grant_discard(exp);
331 /* Do not erase record for recoverable client. */
332 if (exp->exp_obd->obd_replayable &&
333 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
334 rc = lu_env_init(&env, LCT_DT_THREAD);
338 tgt_client_del(&env, exp);
342 class_export_put(exp);
346 static int ofd_init_export(struct obd_export *exp)
350 spin_lock_init(&exp->exp_filter_data.fed_lock);
351 CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
352 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
353 spin_lock(&exp->exp_lock);
354 exp->exp_connecting = 1;
355 spin_unlock(&exp->exp_lock);
357 /* self-export doesn't need client data and ldlm initialization */
358 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
359 &exp->exp_client_uuid)))
362 rc = tgt_client_alloc(exp);
364 ldlm_init_export(exp);
366 CERROR("%s: Can't initialize export: rc %d\n",
367 exp->exp_obd->obd_name, rc);
371 static int ofd_destroy_export(struct obd_export *exp)
373 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
375 if (exp->exp_filter_data.fed_pending)
376 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
377 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
378 exp, exp->exp_filter_data.fed_pending);
380 target_destroy_export(exp);
382 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
383 &exp->exp_client_uuid)))
386 ldlm_destroy_export(exp);
387 tgt_client_free(exp);
389 ofd_fmd_cleanup(exp);
392 * discard grants once we're sure no more
393 * interaction with the client is possible
395 ofd_grant_discard(exp);
396 ofd_fmd_cleanup(exp);
398 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
399 if (ofd->ofd_tot_granted_clients > 0)
400 ofd->ofd_tot_granted_clients --;
403 if (!(exp->exp_flags & OBD_OPT_FORCE))
404 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
406 LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
410 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
412 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
414 CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
415 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
418 int ofd_obd_postrecov(struct obd_device *obd)
421 struct lu_device *ldev = obd->obd_lu_dev;
426 rc = lu_env_init(&env, LCT_DT_THREAD);
429 ofd_info_init(&env, obd->obd_self_export);
431 rc = ofd_postrecov(&env, ofd_dev(ldev));
437 /* This is not called from request handler, check ofd_set_info_hdl() instead
438 * this OBD functions is only used by class_notify_sptlrpc_conf() locally
439 * by direct obd_set_info_async() call */
440 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
441 __u32 keylen, void *key, __u32 vallen, void *val,
442 struct ptlrpc_request_set *set)
448 if (exp->exp_obd == NULL) {
449 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
453 if (KEY_IS(KEY_SPTLRPC_CONF)) {
454 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
456 CERROR("%s: Unsupported key %s\n",
457 exp->exp_obd->obd_name, (char*)key);
463 /* used by nrs_orr_range_fill_physical() in ptlrpc, see LU-3239 */
464 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
465 __u32 keylen, void *key, __u32 *vallen, void *val,
466 struct lov_stripe_md *lsm)
468 struct ofd_thread_info *info;
469 struct ofd_device *ofd;
470 struct ll_fiemap_info_key *fm_key = key;
471 struct ll_user_fiemap *fiemap = val;
476 if (exp->exp_obd == NULL) {
477 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
483 if (KEY_IS(KEY_FIEMAP)) {
484 info = ofd_info_init(env, exp);
486 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi,
487 ofd->ofd_lut.lut_lsd.lsd_osd_index);
491 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
493 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
500 /** helper function for statfs, also used by grant code */
501 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
502 struct obd_statfs *osfs, __u64 max_age, int *from_cache)
507 down_read(&ofd->ofd_lastid_rwsem);
508 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
509 * we may do that in the future.
510 * Return -ENOSPC until the LAST_ID rebuilt. */
511 if (unlikely(ofd->ofd_lastid_rebuilding))
512 GOTO(out, rc = -ENOSPC);
514 spin_lock(&ofd->ofd_osfs_lock);
515 if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
518 /* statfs data are too old, get up-to-date one.
519 * we must be cautious here since multiple threads might be
520 * willing to update statfs data concurrently and we must
521 * grant that cached statfs data are always consistent */
523 if (ofd->ofd_statfs_inflight == 0)
524 /* clear inflight counter if no users, although it would
525 * take a while to overflow this 64-bit counter ... */
526 ofd->ofd_osfs_inflight = 0;
527 /* notify ofd_grant_commit() that we want to track writes
528 * completed as of now */
529 ofd->ofd_statfs_inflight++;
530 /* record value of inflight counter before running statfs to
531 * compute the diff once statfs is completed */
532 unstable = ofd->ofd_osfs_inflight;
533 spin_unlock(&ofd->ofd_osfs_lock);
535 /* statfs can sleep ... hopefully not for too long since we can
536 * call it fairly often as space fills up */
537 rc = dt_statfs(env, ofd->ofd_osd, osfs);
541 spin_lock(&ofd->ofd_grant_lock);
542 spin_lock(&ofd->ofd_osfs_lock);
543 /* calculate how much space was written while we released the
545 unstable = ofd->ofd_osfs_inflight - unstable;
546 ofd->ofd_osfs_unstable = 0;
548 /* some writes completed while we were running statfs
549 * w/o the ofd_osfs_lock. Those ones got added to
550 * the cached statfs data that we are about to crunch.
551 * Take them into account in the new statfs data */
552 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
553 unstable >> ofd->ofd_blockbits);
554 /* However, we don't really know if those writes got
555 * accounted in the statfs call, so tell
556 * ofd_grant_space_left() there is some uncertainty
557 * on the accounting of those writes.
558 * The purpose is to prevent spurious error messages in
559 * ofd_grant_space_left() since those writes might be
560 * accounted twice. */
561 ofd->ofd_osfs_unstable += unstable;
563 /* similarly, there is some uncertainty on write requests
564 * between prepare & commit */
565 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
566 spin_unlock(&ofd->ofd_grant_lock);
568 /* finally udpate cached statfs data */
569 ofd->ofd_osfs = *osfs;
570 ofd->ofd_osfs_age = cfs_time_current_64();
572 ofd->ofd_statfs_inflight--; /* stop tracking */
573 if (ofd->ofd_statfs_inflight == 0)
574 ofd->ofd_osfs_inflight = 0;
575 spin_unlock(&ofd->ofd_osfs_lock);
580 /* use cached statfs data */
581 *osfs = ofd->ofd_osfs;
582 spin_unlock(&ofd->ofd_osfs_lock);
590 up_read(&ofd->ofd_lastid_rwsem);
595 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
596 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
598 struct obd_device *obd = class_exp2obd(exp);
599 struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
604 rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
608 /* at least try to account for cached pages. its still racy and
609 * might be under-reporting if clients haven't announced their
610 * caches with brw recently */
612 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
613 " pending "LPU64" free "LPU64" avail "LPU64"\n",
614 ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
615 osfs->os_bfree << ofd->ofd_blockbits,
616 osfs->os_bavail << ofd->ofd_blockbits);
618 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
619 ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
620 osfs->os_bsize - 1) >> ofd->ofd_blockbits));
622 /* The QoS code on the MDS does not care about space reserved for
623 * precreate, so take it out. */
624 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
625 struct filter_export_data *fed;
627 fed = &obd->obd_self_export->exp_filter_data;
628 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
629 fed->fed_grant >> ofd->ofd_blockbits);
632 ofd_grant_sanity_check(obd, __FUNCTION__);
633 CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
634 LPU64" objects: "LPU64" free; state %x\n",
635 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
636 osfs->os_files, osfs->os_ffree, osfs->os_state);
638 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
639 ofd->ofd_lut.lut_lsd.lsd_osd_index))
642 /* OS_STATE_READONLY can be set by OSD already */
643 if (ofd->ofd_raid_degraded)
644 osfs->os_state |= OS_STATE_DEGRADED;
646 if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
647 /* clients which don't support OBD_CONNECT_GRANT_PARAM
648 * should not see a block size > page size, otherwise
649 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
650 * block size which is the biggest block size known to work
651 * with all client's page size. */
652 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
653 osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
654 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
655 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
658 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
659 ofd->ofd_lut.lut_lsd.lsd_osd_index))
660 osfs->os_bfree = osfs->os_bavail = 2;
667 /* needed by echo client only for now, RPC handler uses ofd_setattr_hdl() */
668 int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
669 struct obd_info *oinfo, struct obd_trans_info *oti)
671 struct ofd_thread_info *info;
672 struct ofd_device *ofd = ofd_exp(exp);
673 struct ldlm_namespace *ns = ofd->ofd_namespace;
674 struct ldlm_resource *res;
675 struct ofd_object *fo;
676 struct obdo *oa = oinfo->oi_oa;
677 struct lu_fid *fid = &oa->o_oi.oi_fid;
678 struct filter_fid *ff = NULL;
683 info = ofd_info_init(env, exp);
685 ost_fid_build_resid(fid, &info->fti_resid);
687 /* This would be very bad - accidentally truncating a file when
688 * changing the time or similar - bug 12203. */
689 if (oa->o_valid & OBD_MD_FLSIZE &&
690 oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
691 static char mdsinum[48];
693 if (oa->o_valid & OBD_MD_FLFID)
694 snprintf(mdsinum, sizeof(mdsinum) - 1,
695 "of parent "DFID, oa->o_parent_seq,
696 oa->o_parent_oid, 0);
700 CERROR("%s: setattr from %s trying to truncate object "DFID
701 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
703 GOTO(out, rc = -EPERM);
706 fo = ofd_object_find_exists(env, ofd, fid);
708 CERROR("%s: can't find object "DFID"\n",
709 ofd_name(ofd), PFID(fid));
710 GOTO(out, rc = PTR_ERR(fo));
713 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
714 info->fti_attr.la_valid &= ~LA_TYPE;
716 if (oa->o_valid & OBD_MD_FLFID) {
717 ff = &info->fti_mds_fid;
718 ofd_prepare_fidea(ff, oa);
721 /* setting objects attributes (including owner/group) */
722 rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
724 GOTO(out_unlock, rc);
726 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
729 ofd_object_put(env, fo);
732 /* we do not call this before to avoid lu_object_find() in
733 * ->lvbo_update() holding another reference on the object.
734 * otherwise concurrent destroy can make the object unavailable
735 * for 2nd lu_object_find() waiting for the first reference
736 * to go... deadlock! */
737 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
739 ldlm_res_lvbo_update(res, NULL, 0);
740 ldlm_resource_putref(res);
747 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
748 const struct lu_fid *fid, int orphan)
750 struct ofd_thread_info *info = ofd_info(env);
751 struct lustre_handle lockh;
752 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
754 ldlm_policy_data_t policy = {
755 .l_extent = { 0, OBD_OBJECT_EOF }
757 struct ofd_object *fo;
761 fo = ofd_object_find_exists(env, ofd, fid);
765 /* Tell the clients that the object is gone now and that they should
766 * throw away any cached pages. */
767 ost_fid_build_resid(fid, &info->fti_resid);
768 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
769 LDLM_EXTENT, &policy, LCK_PW, &flags,
770 ldlm_blocking_ast, ldlm_completion_ast,
771 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
773 /* We only care about the side-effects, just drop the lock. */
775 ldlm_lock_decref(&lockh, LCK_PW);
779 rc = ofd_object_destroy(env, fo, orphan);
782 ofd_object_put(env, fo);
786 /* needed by echo client only for now, RPC handler uses ofd_destroy_hdl() */
787 int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
788 struct obdo *oa, struct lov_stripe_md *md,
789 struct obd_trans_info *oti, struct obd_export *md_exp,
792 struct ofd_device *ofd = ofd_exp(exp);
793 struct lu_fid *fid = &oa->o_oi.oi_fid;
798 ofd_info_init(env, exp);
800 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
802 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
804 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
805 ofd_name(ofd), PFID(fid));
807 } else if (rc != 0) {
808 CERROR("%s: error destroying object "DFID": %d\n",
809 ofd_name(ofd), PFID(fid), rc);
817 /* needed by echo client only for now, RPC handler uses ofd_create_hdl()
818 * It is much simpler and just create objects */
819 int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
820 struct obdo *oa, struct lov_stripe_md **ea,
821 struct obd_trans_info *oti)
823 struct ofd_device *ofd = ofd_exp(exp);
824 struct ofd_thread_info *info;
825 obd_seq seq = ostid_seq(&oa->o_oi);
826 struct ofd_seq *oseq;
827 int rc = 0, diff = 1;
833 info = ofd_info_init(env, exp);
835 LASSERT(seq == FID_SEQ_ECHO);
836 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
838 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
840 down_read(&ofd->ofd_lastid_rwsem);
841 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
842 * we may do that in the future.
843 * Return -ENOSPC until the LAST_ID rebuilt. */
844 if (unlikely(ofd->ofd_lastid_rebuilding))
845 GOTO(out_sem, rc = -ENOSPC);
847 oseq = ofd_seq_load(env, ofd, seq);
849 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
850 ofd_name(ofd), seq, PTR_ERR(oseq));
851 GOTO(out_sem, rc = -EINVAL);
854 mutex_lock(&oseq->os_create_lock);
855 rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
857 CDEBUG(D_HA, "%s: failed to acquire grant space for "
858 "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
863 next_id = ofd_seq_last_oid(oseq) + 1;
864 count = ofd_precreate_batch(ofd, diff);
866 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
868 CERROR("%s: unable to precreate: rc = %d\n",
871 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
872 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
876 ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
878 mutex_unlock(&oseq->os_create_lock);
879 if (rc == 0 && ea != NULL) {
880 struct lov_stripe_md *lsm = *ea;
882 lsm->lsm_oi = oa->o_oi;
884 ofd_seq_put(env, oseq);
887 up_read(&ofd->ofd_lastid_rwsem);
891 /* needed by echo client only for now, RPC handler uses ofd_getattr_hdl() */
892 int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
893 struct obd_info *oinfo)
895 struct ofd_device *ofd = ofd_exp(exp);
896 struct ofd_thread_info *info;
897 struct lu_fid *fid = &oinfo->oi_oa->o_oi.oi_fid;
898 struct ofd_object *fo;
903 info = ofd_info_init(env, exp);
905 fo = ofd_object_find_exists(env, ofd, fid);
907 GOTO(out, rc = PTR_ERR(fo));
910 rc = ofd_attr_get(env, fo, &info->fti_attr);
911 oinfo->oi_oa->o_valid = OBD_MD_FLID;
915 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
916 OFD_VALID_FLAGS | LA_UID | LA_GID);
918 /* Store object version in reply */
919 curr_version = dt_version_get(env, ofd_object_child(fo));
920 if ((__s64)curr_version != -EOPNOTSUPP) {
921 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
922 oinfo->oi_oa->o_data_version = curr_version;
926 ofd_object_put(env, fo);
931 static int ofd_ioc_get_obj_version(const struct lu_env *env,
932 struct ofd_device *ofd, void *karg)
934 struct obd_ioctl_data *data = karg;
936 struct ofd_object *fo;
937 dt_obj_version_t version;
942 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
943 GOTO(out, rc = -EINVAL);
945 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
946 fid = *(struct lu_fid *)data->ioc_inlbuf1;
947 } else if (data->ioc_inlbuf3 != NULL &&
948 data->ioc_inllen3 == sizeof(__u64) &&
949 data->ioc_inlbuf4 != NULL &&
950 data->ioc_inllen4 == sizeof(__u64)) {
953 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
954 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
955 rc = ostid_to_fid(&fid, &ostid,
956 ofd->ofd_lut.lut_lsd.lsd_osd_index);
960 GOTO(out, rc = -EINVAL);
963 if (!fid_is_sane(&fid))
964 GOTO(out, rc = -EINVAL);
966 fo = ofd_object_find(env, ofd, &fid);
968 GOTO(out, rc = PTR_ERR(fo));
970 if (!ofd_object_exists(fo))
971 GOTO(out_fo, rc = -ENOENT);
973 if (lu_object_remote(&fo->ofo_obj.do_lu))
974 GOTO(out_fo, rc = -EREMOTE);
976 version = dt_version_get(env, ofd_object_child(fo));
978 GOTO(out_fo, rc = -EIO);
980 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
984 ofd_object_put(env, fo);
989 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
990 void *karg, void *uarg)
993 struct ofd_device *ofd = ofd_exp(exp);
994 struct obd_device *obd = ofd_obd(ofd);
999 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1000 rc = lu_env_init(&env, LCT_DT_THREAD);
1005 case OBD_IOC_ABORT_RECOVERY:
1006 CERROR("%s: aborting recovery\n", obd->obd_name);
1007 target_stop_recovery_thread(obd);
1010 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1011 rc = dt_sync(&env, ofd->ofd_osd);
1013 case OBD_IOC_SET_READONLY:
1014 rc = dt_sync(&env, ofd->ofd_osd);
1016 rc = dt_ro(&env, ofd->ofd_osd);
1018 case OBD_IOC_START_LFSCK: {
1019 struct obd_ioctl_data *data = karg;
1020 struct lfsck_start_param lsp;
1022 if (unlikely(data == NULL)) {
1027 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1028 lsp.lsp_index_valid = 0;
1029 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1032 case OBD_IOC_STOP_LFSCK: {
1033 struct obd_ioctl_data *data = karg;
1034 struct lfsck_stop stop;
1036 stop.ls_status = LS_STOPPED;
1037 /* Old lfsck utils may pass NULL @stop. */
1038 if (data->ioc_inlbuf1 == NULL)
1042 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1044 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1047 case OBD_IOC_GET_OBJ_VERSION:
1048 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1051 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1059 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1066 case OBD_CLEANUP_EARLY:
1068 case OBD_CLEANUP_EXPORTS:
1069 target_cleanup_recovery(obd);
1075 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1077 ofd_fmd_expire(exp);
1081 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1083 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1084 struct ofd_thread_info *info;
1086 #ifdef USE_HEALTH_CHECK_WRITE
1091 /* obd_proc_read_health pass NULL env, we need real one */
1092 rc = lu_env_init(&env, LCT_DT_THREAD);
1096 info = ofd_info_init(&env, NULL);
1097 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1101 if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1102 GOTO(out, rc = -EROFS);
1104 #ifdef USE_HEALTH_CHECK_WRITE
1105 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1106 if (info->fti_buf.lb_buf == NULL)
1107 GOTO(out, rc = -ENOMEM);
1109 info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1112 th = dt_trans_create(&env, ofd->ofd_osd);
1114 GOTO(out, rc = PTR_ERR(th));
1116 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1117 &info->fti_buf, info->fti_off, th);
1119 th->th_sync = 1; /* sync IO is needed */
1120 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1122 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1123 &info->fti_buf, &info->fti_off,
1126 dt_trans_stop(&env, ofd->ofd_osd, th);
1128 OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1130 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1138 * Handle quota control requests to consult current usage/limit.
1140 * \param obd - is the obd device associated with the ofd
1141 * \param exp - is the client's export
1142 * \param oqctl - is the obd_quotactl request to be processed
1144 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1145 struct obd_quotactl *oqctl)
1147 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1152 /* report success for quota on/off for interoperability with current MDT
1154 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1157 rc = lu_env_init(&env, LCT_DT_THREAD);
1161 rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1167 struct obd_ops ofd_obd_ops = {
1168 .o_owner = THIS_MODULE,
1169 .o_connect = ofd_obd_connect,
1170 .o_reconnect = ofd_obd_reconnect,
1171 .o_disconnect = ofd_obd_disconnect,
1172 .o_create = ofd_echo_create,
1173 .o_statfs = ofd_statfs,
1174 .o_setattr = ofd_echo_setattr,
1175 .o_preprw = ofd_preprw,
1176 .o_commitrw = ofd_commitrw,
1177 .o_destroy = ofd_echo_destroy,
1178 .o_init_export = ofd_init_export,
1179 .o_destroy_export = ofd_destroy_export,
1180 .o_postrecov = ofd_obd_postrecov,
1181 .o_getattr = ofd_echo_getattr,
1182 .o_iocontrol = ofd_iocontrol,
1183 .o_precleanup = ofd_precleanup,
1185 .o_health_check = ofd_health_check,
1186 .o_quotactl = ofd_quotactl,
1187 .o_set_info_async = ofd_set_info_async,
1188 .o_get_info = ofd_get_info,