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.
38 * Author: Alex Tomas <bzzz@whamcloud.com>
41 #define DEBUG_SUBSYSTEM S_FILTER
43 #include "ofd_internal.h"
45 struct ofd_inconsistency_item {
46 struct list_head oii_list;
47 struct ofd_object *oii_obj;
48 struct lu_fid oii_pfid;
51 static void ofd_inconsistency_verify_one(const struct lu_env *env,
52 struct ofd_device *ofd,
53 struct ofd_inconsistency_item *oii,
54 struct lfsck_request *lr)
56 struct ofd_object *fo = oii->oii_obj;
57 struct lu_fid *pfid = &fo->ofo_pfid;
60 LASSERT(fo->ofo_pfid_checking);
61 LASSERT(!fo->ofo_pfid_verified);
63 lr->lr_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
64 lr->lr_fid2 = oii->oii_pfid; /* client given PFID. */
65 lr->lr_fid3 = *pfid; /* OST local stored PFID. */
67 rc = lfsck_in_notify(env, ofd->ofd_osd, lr);
68 ofd_write_lock(env, fo);
69 switch (lr->lr_status) {
74 CDEBUG(D_LFSCK, "%s: fail to verify OST local stored "
75 "PFID xattr for "DFID", the client given PFID "
76 DFID", OST local stored PFID "DFID": rc = %d\n",
77 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
78 PFID(&oii->oii_pfid), PFID(pfid), rc);
80 fo->ofo_pfid_verified = 1;
82 case LPVS_INCONSISTENT:
85 ofd->ofd_inconsistency_self_detected++;
87 CDEBUG(D_LFSCK, "%s: fail to verify the client given "
88 "PFID for "DFID", the client given PFID "DFID
89 ", local stored PFID "DFID": rc = %d\n",
90 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
91 PFID(&oii->oii_pfid), PFID(pfid), rc);
93 CDEBUG(D_LFSCK, "%s: both the client given PFID and "
94 "the OST local stored PFID are stale for the "
95 "OST-object "DFID", client given PFID is "DFID
96 ", local stored PFID is "DFID"\n",
97 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
98 PFID(&oii->oii_pfid), PFID(pfid));
100 case LPVS_INCONSISTENT_TOFIX:
101 ofd->ofd_inconsistency_self_detected++;
103 ofd->ofd_inconsistency_self_repaired++;
104 CDEBUG(D_LFSCK, "%s: fixed the staled OST PFID xattr "
105 "for "DFID", with the client given PFID "DFID
106 ", the old stored PFID "DFID"\n",
107 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
108 PFID(&oii->oii_pfid), PFID(pfid));
110 CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
111 "for "DFID", client given PFID "DFID", local "
112 "stored PFID "DFID": rc = %d\n",
113 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
114 PFID(&oii->oii_pfid), PFID(pfid), rc);
116 *pfid = oii->oii_pfid;
117 fo->ofo_pfid_verified = 1;
122 fo->ofo_pfid_checking = 0;
123 ofd_write_unlock(env, fo);
125 lu_object_put(env, &fo->ofo_obj.do_lu);
129 static int ofd_inconsistency_verification_main(void *args)
132 struct ofd_device *ofd = args;
133 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
134 struct ofd_inconsistency_item *oii;
135 struct lfsck_request *lr = NULL;
136 struct l_wait_info lwi = { 0 };
140 rc = lu_env_init(&env, LCT_DT_THREAD);
141 spin_lock(&ofd->ofd_inconsistency_lock);
142 thread_set_flags(thread, rc != 0 ? SVC_STOPPED : SVC_RUNNING);
143 wake_up_all(&thread->t_ctl_waitq);
144 spin_unlock(&ofd->ofd_inconsistency_lock);
149 if (unlikely(lr == NULL))
150 GOTO(out, rc = -ENOMEM);
152 lr->lr_event = LE_PAIRS_VERIFY;
153 lr->lr_active = LFSCK_TYPE_LAYOUT;
155 spin_lock(&ofd->ofd_inconsistency_lock);
157 if (unlikely(!thread_is_running(thread)))
160 while (!list_empty(&ofd->ofd_inconsistency_list)) {
161 oii = list_entry(ofd->ofd_inconsistency_list.next,
162 struct ofd_inconsistency_item,
164 list_del_init(&oii->oii_list);
165 spin_unlock(&ofd->ofd_inconsistency_lock);
166 ofd_inconsistency_verify_one(&env, ofd, oii, lr);
167 spin_lock(&ofd->ofd_inconsistency_lock);
170 spin_unlock(&ofd->ofd_inconsistency_lock);
171 l_wait_event(thread->t_ctl_waitq,
172 !list_empty(&ofd->ofd_inconsistency_list) ||
173 !thread_is_running(thread),
175 spin_lock(&ofd->ofd_inconsistency_lock);
178 while (!list_empty(&ofd->ofd_inconsistency_list)) {
179 struct ofd_object *fo;
181 oii = list_entry(ofd->ofd_inconsistency_list.next,
182 struct ofd_inconsistency_item,
184 list_del_init(&oii->oii_list);
186 spin_unlock(&ofd->ofd_inconsistency_lock);
188 ofd_write_lock(&env, fo);
189 fo->ofo_pfid_checking = 0;
190 ofd_write_unlock(&env, fo);
192 lu_object_put(&env, &fo->ofo_obj.do_lu);
194 spin_lock(&ofd->ofd_inconsistency_lock);
202 thread_set_flags(thread, SVC_STOPPED);
203 wake_up_all(&thread->t_ctl_waitq);
204 spin_unlock(&ofd->ofd_inconsistency_lock);
210 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
212 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
213 struct l_wait_info lwi = { 0 };
214 struct task_struct *task;
217 spin_lock(&ofd->ofd_inconsistency_lock);
218 if (unlikely(thread_is_running(thread))) {
219 spin_unlock(&ofd->ofd_inconsistency_lock);
224 thread_set_flags(thread, 0);
225 spin_unlock(&ofd->ofd_inconsistency_lock);
226 task = kthread_run(ofd_inconsistency_verification_main, ofd,
227 "inconsistency_verification");
230 CERROR("%s: cannot start self_repair thread: rc = %d\n",
234 l_wait_event(thread->t_ctl_waitq,
235 thread_is_running(thread) ||
236 thread_is_stopped(thread),
243 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
245 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
246 struct l_wait_info lwi = { 0 };
248 spin_lock(&ofd->ofd_inconsistency_lock);
249 if (thread_is_init(thread) || thread_is_stopped(thread)) {
250 spin_unlock(&ofd->ofd_inconsistency_lock);
255 thread_set_flags(thread, SVC_STOPPING);
256 spin_unlock(&ofd->ofd_inconsistency_lock);
257 wake_up_all(&thread->t_ctl_waitq);
258 l_wait_event(thread->t_ctl_waitq,
259 thread_is_stopped(thread),
265 static void ofd_add_inconsistency_item(const struct lu_env *env,
266 struct ofd_object *fo, struct obdo *oa)
268 struct ofd_device *ofd = ofd_obj2dev(fo);
269 struct ofd_inconsistency_item *oii;
274 CERROR("%s: cannot alloc memory for verify OST-object "
275 "consistency for "DFID", client given PFID "DFID
276 ", local stored PFID "DFID"\n",
277 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
278 oa->o_parent_seq, oa->o_parent_oid, oa->o_stripe_idx,
279 PFID(&fo->ofo_pfid));
284 INIT_LIST_HEAD(&oii->oii_list);
285 lu_object_get(&fo->ofo_obj.do_lu);
287 oii->oii_pfid.f_seq = oa->o_parent_seq;
288 oii->oii_pfid.f_oid = oa->o_parent_oid;
289 oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
291 spin_lock(&ofd->ofd_inconsistency_lock);
292 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
293 spin_unlock(&ofd->ofd_inconsistency_lock);
299 fo->ofo_pfid_checking = 1;
300 if (list_empty(&ofd->ofd_inconsistency_list))
302 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
303 spin_unlock(&ofd->ofd_inconsistency_lock);
305 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
307 /* XXX: When the found inconsistency exceeds some threshold,
308 * we can trigger the LFSCK to scan part of the system
309 * or the whole system, which depends on how to define
310 * the threshold, a simple way maybe like that: define
311 * the absolute value of how many inconsisteny allowed
312 * to be repaired via self detect/repair mechanism, if
313 * exceeded, then trigger the LFSCK to scan the layout
314 * inconsistency within the whole system. */
317 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
320 struct lu_fid *pfid = &fo->ofo_pfid;
324 if (fid_is_sane(pfid)) {
325 if (likely(oa->o_parent_seq == pfid->f_seq &&
326 oa->o_parent_oid == pfid->f_oid &&
327 oa->o_stripe_idx == pfid->f_stripe_idx))
330 if (fo->ofo_pfid_verified)
334 /* The OST-object may be inconsistent, and we need further verification.
335 * To avoid block the RPC service thread, return -EINPROGRESS to client
336 * and make it retry later. */
337 if (fo->ofo_pfid_checking)
338 RETURN(-EINPROGRESS);
340 rc = ofd_object_ff_load(env, fo);
347 if (likely(oa->o_parent_seq == pfid->f_seq &&
348 oa->o_parent_oid == pfid->f_oid &&
349 oa->o_stripe_idx == pfid->f_stripe_idx))
352 /* Push it to the dedicated thread for further verification. */
353 ofd_add_inconsistency_item(env, fo, oa);
355 RETURN(-EINPROGRESS);
358 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
359 struct ofd_device *ofd, const struct lu_fid *fid,
360 struct lu_attr *la, struct obdo *oa, int niocount,
361 struct niobuf_remote *rnb, int *nr_local,
362 struct niobuf_local *lnb, char *jobid)
364 struct ofd_object *fo;
365 int i, j, rc, tot_bytes = 0;
368 LASSERT(env != NULL);
370 fo = ofd_object_find(env, ofd, fid);
375 ofd_read_lock(env, fo);
376 if (!ofd_object_exists(fo))
377 GOTO(unlock, rc = -ENOENT);
379 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
380 rc = ofd_verify_ff(env, fo, oa);
385 /* parse remote buffers to local buffers and prepare the latter */
387 for (i = 0, j = 0; i < niocount; i++) {
388 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
389 lnb + j, 0, ofd_object_capa(env, fo));
390 if (unlikely(rc < 0))
392 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
393 /* correct index for local buffers to continue with */
396 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
397 tot_bytes += rnb[i].rnb_len;
400 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
401 rc = dt_attr_get(env, ofd_object_child(fo), la,
402 ofd_object_capa(env, fo));
406 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
410 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
414 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
416 ofd_read_unlock(env, fo);
417 ofd_object_put(env, fo);
421 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
422 struct ofd_device *ofd, const struct lu_fid *fid,
423 struct lu_attr *la, struct obdo *oa,
424 int objcount, struct obd_ioobj *obj,
425 struct niobuf_remote *rnb, int *nr_local,
426 struct niobuf_local *lnb, char *jobid)
428 struct ofd_object *fo;
429 int i, j, k, rc = 0, tot_bytes = 0;
432 LASSERT(env != NULL);
433 LASSERT(objcount == 1);
435 if (unlikely(exp->exp_obd->obd_recovering)) {
436 obd_seq seq = fid_seq(fid);
437 obd_id oid = fid_oid(fid);
438 struct ofd_seq *oseq;
440 oseq = ofd_seq_load(env, ofd, seq);
442 CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
443 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
444 GOTO(out, rc = -EINVAL);
447 if (oid > ofd_seq_last_oid(oseq)) {
451 mutex_lock(&oseq->os_create_lock);
452 diff = oid - ofd_seq_last_oid(oseq);
454 /* Do sync create if the seq is about to used up */
455 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
456 if (unlikely(oid >= IDIF_MAX_OID - 1))
458 } else if (fid_seq_is_norm(seq)) {
460 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
463 CERROR("%s : invalid o_seq "DOSTID"\n",
464 ofd_name(ofd), POSTID(&oa->o_oi));
465 mutex_unlock(&oseq->os_create_lock);
466 ofd_seq_put(env, oseq);
467 GOTO(out, rc = -EINVAL);
471 obd_id next_id = ofd_seq_last_oid(oseq) + 1;
472 int count = ofd_precreate_batch(ofd, diff);
474 rc = ofd_precreate_objects(env, ofd, next_id,
477 mutex_unlock(&oseq->os_create_lock);
478 ofd_seq_put(env, oseq);
485 mutex_unlock(&oseq->os_create_lock);
488 ofd_seq_put(env, oseq);
491 fo = ofd_object_find(env, ofd, fid);
493 GOTO(out, rc = PTR_ERR(fo));
496 ofd_read_lock(env, fo);
497 if (!ofd_object_exists(fo)) {
498 CERROR("%s: BRW to missing obj "DOSTID"\n",
499 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
500 ofd_read_unlock(env, fo);
501 ofd_object_put(env, fo);
502 GOTO(out, rc = -ENOENT);
505 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
506 rc = ofd_verify_ff(env, fo, oa);
508 ofd_read_unlock(env, fo);
509 ofd_object_put(env, fo);
514 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
515 * space back if possible */
516 ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
518 /* parse remote buffers to local buffers and prepare the latter */
520 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
521 rc = dt_bufs_get(env, ofd_object_child(fo),
523 ofd_object_capa(env, fo));
524 if (unlikely(rc < 0))
526 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
527 /* correct index for local buffers to continue with */
528 for (k = 0; k < rc; k++) {
529 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
530 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
531 lnb[j+k].lnb_rc = -ENOSPC;
533 /* remote client can't break through quota */
534 if (exp_connect_rmtclient(exp))
535 lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
539 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
540 tot_bytes += rnb[i].rnb_len;
542 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
544 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
545 if (unlikely(rc != 0))
548 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
551 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
552 ofd_read_unlock(env, fo);
553 /* ofd_grant_prepare_write() was called, so we must commit */
554 ofd_grant_commit(env, exp, rc);
556 /* let's still process incoming grant information packed in the oa,
557 * but without enforcing grant since we won't proceed with the write.
558 * Just like a read request actually. */
559 ofd_grant_prepare_read(env, exp, oa);
563 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
564 struct obdo *oa, int objcount, struct obd_ioobj *obj,
565 struct niobuf_remote *rnb, int *nr_local,
566 struct niobuf_local *lnb, struct obd_trans_info *oti,
567 struct lustre_capa *capa)
569 struct tgt_session_info *tsi = tgt_ses_info(env);
570 struct ofd_device *ofd = ofd_exp(exp);
571 struct ofd_thread_info *info;
573 const struct lu_fid *fid = &oa->o_oi.oi_fid;
576 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
577 CERROR("%s: bulk has too many pages %d, which exceeds the"
578 "maximum pages per RPC of %d\n",
579 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
583 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
584 LASSERT(oti != NULL);
585 info = ofd_info_init(env, exp);
586 ofd_oti2info(info, oti);
589 info = tsi2ofd_info(tsi);
590 jobid = tsi->tsi_jobid;
595 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
596 struct ofd_seq *oseq;
598 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
600 CERROR("%s: Can not find seq for "DOSTID
601 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
606 if (oseq->os_destroys_in_progress == 0) {
607 /* don't fail lookups for orphan recovery, it causes
608 * later LBUGs when objects still exist during
610 ofd_seq_put(env, oseq);
613 ofd_seq_put(env, oseq);
616 LASSERT(objcount == 1);
617 LASSERT(obj->ioo_bufcnt > 0);
619 if (cmd == OBD_BRW_WRITE) {
620 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
621 capa, CAPA_OPC_OSS_WRITE);
623 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
624 rc = ofd_preprw_write(env, exp, ofd, fid,
625 &info->fti_attr, oa, objcount,
626 obj, rnb, nr_local, lnb, jobid);
628 } else if (cmd == OBD_BRW_READ) {
629 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
630 capa, CAPA_OPC_OSS_READ);
632 ofd_grant_prepare_read(env, exp, oa);
633 rc = ofd_preprw_read(env, exp, ofd, fid,
635 obj->ioo_bufcnt, rnb, nr_local,
637 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
640 CERROR("%s: wrong cmd %d received!\n",
641 exp->exp_obd->obd_name, cmd);
648 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
649 const struct lu_fid *fid, int objcount, int niocount,
650 struct niobuf_local *lnb)
652 struct ofd_object *fo;
656 LASSERT(niocount > 0);
658 fo = ofd_object_find(env, ofd, fid);
662 LASSERT(ofd_object_exists(fo));
663 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
665 ofd_read_unlock(env, fo);
666 ofd_object_put(env, fo);
667 /* second put is pair to object_get in ofd_preprw_read */
668 ofd_object_put(env, fo);
674 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
675 struct ofd_object *ofd_obj, struct lu_attr *la,
676 struct filter_fid *ff)
678 struct ofd_thread_info *info = ofd_info(env);
679 __u64 valid = la->la_valid;
682 struct dt_object *dt_obj;
689 dt_obj = ofd_object_child(ofd_obj);
690 LASSERT(dt_obj != NULL);
692 la->la_valid &= LA_UID | LA_GID;
694 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
699 rc = ofd_object_ff_load(env, ofd_obj);
706 if (!la->la_valid && !ff_needed)
707 /* no attributes to set */
710 th = ofd_trans_create(env, ofd);
712 GOTO(out, rc = PTR_ERR(th));
715 rc = dt_declare_attr_set(env, dt_obj, la, th);
721 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
722 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
723 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
724 ff->ff_parent.f_oid =
725 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
727 info->fti_buf.lb_buf = ff;
728 info->fti_buf.lb_len = sizeof(*ff);
729 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
730 XATTR_NAME_FID, 0, th);
735 /* We don't need a transno for this operation which will be re-executed
736 * anyway when the OST_WRITE (with a transno assigned) is replayed */
737 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
743 rc = dt_attr_set(env, dt_obj, la, th,
744 ofd_object_capa(env, ofd_obj));
749 /* set filter fid EA */
751 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
754 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
757 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
758 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
759 /* Currently, the filter_fid::ff_parent::f_ver is not
760 * the real parent MDT-object's FID::f_ver, instead it
761 * is the OST-object index in its parent MDT-object's
763 ofd_obj->ofo_pfid.f_stripe_idx =
764 le32_to_cpu(ff->ff_parent.f_stripe_idx);
771 dt_trans_stop(env, ofd->ofd_osd, th);
773 la->la_valid = valid;
777 struct ofd_soft_sync_callback {
778 struct dt_txn_commit_cb ossc_cb;
779 struct obd_export *ossc_exp;
782 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
783 struct dt_txn_commit_cb *cb, int err)
785 struct ofd_soft_sync_callback *ossc;
787 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
789 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
790 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
792 class_export_cb_put(ossc->ossc_exp);
796 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
798 struct ofd_soft_sync_callback *ossc;
799 struct dt_txn_commit_cb *dcb;
806 ossc->ossc_exp = class_export_cb_get(exp);
808 dcb = &ossc->ossc_cb;
809 dcb->dcb_func = ofd_cb_soft_sync;
810 CFS_INIT_LIST_HEAD(&dcb->dcb_linkage);
811 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
813 rc = dt_trans_cb_add(th, dcb);
815 class_export_cb_put(exp);
823 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
824 struct ofd_device *ofd, const struct lu_fid *fid,
825 struct lu_attr *la, struct filter_fid *ff, int objcount,
826 int niocount, struct niobuf_local *lnb, int old_rc)
828 struct ofd_thread_info *info = ofd_info(env);
829 struct ofd_object *fo;
835 struct filter_export_data *fed = &exp->exp_filter_data;
836 bool soft_sync = false;
837 bool cb_registered = false;
841 LASSERT(objcount == 1);
843 fo = ofd_object_find(env, ofd, fid);
845 LASSERT(ofd_object_exists(fo));
847 o = ofd_object_child(fo);
851 GOTO(out, rc = old_rc);
854 * The first write to each object must set some attributes. It is
855 * important to set the uid/gid before calling
856 * dt_declare_write_commit() since quota enforcement is now handled in
859 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
863 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
866 th = ofd_trans_create(env, ofd);
868 GOTO(out, rc = PTR_ERR(th));
870 th->th_sync |= ofd->ofd_syncjournal;
871 if (th->th_sync == 0) {
872 for (i = 0; i < niocount; i++) {
873 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
877 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
882 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
883 GOTO(out_stop, rc = -EINPROGRESS);
885 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
890 /* update [mac]time if needed */
891 rc = dt_declare_attr_set(env, o, la, th);
896 rc = ofd_trans_start(env, ofd, fo, th);
900 rc = dt_write_commit(env, o, lnb, niocount, th);
905 rc = dt_attr_set(env, o, la, th, ofd_object_capa(env, fo));
910 /* get attr to return */
911 rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo));
914 /* Force commit to make the just-deleted blocks
915 * reusable. LU-456 */
919 /* do this before trans stop in case commit has finished */
920 if (!th->th_sync && soft_sync && !cb_registered) {
921 ofd_soft_sync_cb_add(th, exp);
922 cb_registered = true;
925 ofd_trans_stop(env, ofd, th, rc);
926 if (rc == -ENOSPC && retries++ < 3) {
927 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
933 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
934 atomic_set(&fed->fed_soft_sync_count, 0);
935 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
936 ofd->ofd_soft_sync_limit)
937 dt_commit_async(env, ofd->ofd_osd);
940 dt_bufs_put(env, o, lnb, niocount);
941 ofd_read_unlock(env, fo);
942 ofd_object_put(env, fo);
943 /* second put is pair to object_get in ofd_preprw_write */
944 ofd_object_put(env, fo);
945 ofd_grant_commit(env, info->fti_exp, old_rc);
949 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
950 struct obdo *oa, int objcount, struct obd_ioobj *obj,
951 struct niobuf_remote *rnb, int npages,
952 struct niobuf_local *lnb, struct obd_trans_info *oti,
955 struct ofd_thread_info *info = ofd_info(env);
956 struct ofd_mod_data *fmd;
958 struct ofd_device *ofd = ofd_exp(exp);
959 struct filter_fid *ff = NULL;
960 const struct lu_fid *fid = &oa->o_oi.oi_fid;
965 if (cmd == OBD_BRW_WRITE) {
966 /* Don't update timestamps if this write is older than a
967 * setattr which modifies the timestamps. b=10150 */
969 /* XXX when we start having persistent reservations this needs
970 * to be changed to ofd_fmd_get() to create the fmd if it
971 * doesn't already exist so we can store the reservation handle
973 valid = OBD_MD_FLUID | OBD_MD_FLGID;
974 fmd = ofd_fmd_find(exp, fid);
975 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
976 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
978 ofd_fmd_put(exp, fmd);
979 la_from_obdo(&info->fti_attr, oa, valid);
981 if (oa->o_valid & OBD_MD_FLFID) {
982 ff = &info->fti_mds_fid;
983 ofd_prepare_fidea(ff, oa);
986 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
987 ff, objcount, npages, lnb, old_rc);
989 obdo_from_la(oa, &info->fti_attr,
990 OFD_VALID_FLAGS | LA_GID | LA_UID);
992 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
994 /* don't report overquota flag if we failed before reaching
996 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
997 /* return the overquota flags to client */
998 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
999 if (oa->o_valid & OBD_MD_FLFLAGS)
1000 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1002 oa->o_flags = OBD_FL_NO_USRQUOTA;
1005 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1006 if (oa->o_valid & OBD_MD_FLFLAGS)
1007 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1009 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1012 oa->o_valid |= OBD_MD_FLFLAGS;
1013 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1015 } else if (cmd == OBD_BRW_READ) {
1016 struct ldlm_namespace *ns = ofd->ofd_namespace;
1018 /* If oa != NULL then ofd_preprw_read updated the inode
1019 * atime and we should update the lvb so that other glimpses
1020 * will also get the updated value. bug 5972 */
1021 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1022 struct ldlm_resource *rs = NULL;
1024 ost_fid_build_resid(fid, &info->fti_resid);
1025 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1028 ns->ns_lvbo->lvbo_update(rs, NULL, 1);
1029 ldlm_resource_putref(rs);
1032 rc = ofd_commitrw_read(env, ofd, fid, objcount,
1042 ofd_info2oti(info, oti);