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_obd(ofd)->obd_name,
78 PFID(&fo->ofo_header.loh_fid),
79 PFID(&oii->oii_pfid), PFID(pfid), rc);
81 fo->ofo_pfid_verified = 1;
83 case LPVS_INCONSISTENT:
86 ofd->ofd_inconsistency_self_detected++;
88 CDEBUG(D_LFSCK, "%s: fail to verify the client given "
89 "PFID for "DFID", the client given PFID "DFID
90 ", local stored PFID "DFID": rc = %d\n",
91 ofd_obd(ofd)->obd_name,
92 PFID(&fo->ofo_header.loh_fid),
93 PFID(&oii->oii_pfid), PFID(pfid), rc);
95 CDEBUG(D_LFSCK, "%s: both the client given PFID and "
96 "the OST local stored PFID are stale for the "
97 "OST-object "DFID", client given PFID is "DFID
98 ", local stored PFID is "DFID"\n",
99 ofd_obd(ofd)->obd_name,
100 PFID(&fo->ofo_header.loh_fid),
101 PFID(&oii->oii_pfid), PFID(pfid));
103 case LPVS_INCONSISTENT_TOFIX:
104 ofd->ofd_inconsistency_self_detected++;
106 ofd->ofd_inconsistency_self_repaired++;
107 CDEBUG(D_LFSCK, "%s: fixed the staled OST PFID xattr "
108 "for "DFID", with the client given PFID "DFID
109 ", the old stored PFID "DFID"\n",
110 ofd_obd(ofd)->obd_name,
111 PFID(&fo->ofo_header.loh_fid),
112 PFID(&oii->oii_pfid), PFID(pfid));
114 CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
115 "for "DFID", client given PFID "DFID", local "
116 "stored PFID "DFID": rc = %d\n",
117 ofd_obd(ofd)->obd_name,
118 PFID(&fo->ofo_header.loh_fid),
119 PFID(&oii->oii_pfid), PFID(pfid), rc);
121 *pfid = oii->oii_pfid;
122 fo->ofo_pfid_verified = 1;
127 fo->ofo_pfid_checking = 0;
128 ofd_write_unlock(env, fo);
130 lu_object_put(env, &fo->ofo_obj.do_lu);
134 static int ofd_inconsistency_verification_main(void *args)
137 struct ofd_device *ofd = args;
138 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
139 struct ofd_inconsistency_item *oii;
140 struct lfsck_request *lr = NULL;
141 struct l_wait_info lwi = { 0 };
145 rc = lu_env_init(&env, LCT_DT_THREAD);
146 spin_lock(&ofd->ofd_inconsistency_lock);
147 thread_set_flags(thread, rc != 0 ? SVC_STOPPED : SVC_RUNNING);
148 wake_up_all(&thread->t_ctl_waitq);
149 spin_unlock(&ofd->ofd_inconsistency_lock);
154 if (unlikely(lr == NULL))
155 GOTO(out, rc = -ENOMEM);
157 lr->lr_event = LE_PAIRS_VERIFY;
158 lr->lr_active = LT_LAYOUT;
160 spin_lock(&ofd->ofd_inconsistency_lock);
162 if (unlikely(!thread_is_running(thread)))
165 while (!list_empty(&ofd->ofd_inconsistency_list)) {
166 oii = list_entry(ofd->ofd_inconsistency_list.next,
167 struct ofd_inconsistency_item,
169 list_del_init(&oii->oii_list);
170 spin_unlock(&ofd->ofd_inconsistency_lock);
171 ofd_inconsistency_verify_one(&env, ofd, oii, lr);
172 spin_lock(&ofd->ofd_inconsistency_lock);
175 spin_unlock(&ofd->ofd_inconsistency_lock);
176 l_wait_event(thread->t_ctl_waitq,
177 !list_empty(&ofd->ofd_inconsistency_list) ||
178 !thread_is_running(thread),
180 spin_lock(&ofd->ofd_inconsistency_lock);
183 while (!list_empty(&ofd->ofd_inconsistency_list)) {
184 struct ofd_object *fo;
186 oii = list_entry(ofd->ofd_inconsistency_list.next,
187 struct ofd_inconsistency_item,
189 list_del_init(&oii->oii_list);
191 spin_unlock(&ofd->ofd_inconsistency_lock);
193 ofd_write_lock(&env, fo);
194 fo->ofo_pfid_checking = 0;
195 ofd_write_unlock(&env, fo);
197 lu_object_put(&env, &fo->ofo_obj.do_lu);
199 spin_lock(&ofd->ofd_inconsistency_lock);
207 thread_set_flags(thread, SVC_STOPPED);
208 wake_up_all(&thread->t_ctl_waitq);
209 spin_unlock(&ofd->ofd_inconsistency_lock);
215 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
217 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
218 struct l_wait_info lwi = { 0 };
219 struct task_struct *task;
222 spin_lock(&ofd->ofd_inconsistency_lock);
223 if (unlikely(thread_is_running(thread))) {
224 spin_unlock(&ofd->ofd_inconsistency_lock);
229 thread_set_flags(thread, 0);
230 spin_unlock(&ofd->ofd_inconsistency_lock);
231 task = kthread_run(ofd_inconsistency_verification_main, ofd,
232 "inconsistency_verification");
235 CERROR("%s: cannot start self_repair thread: rc = %d\n",
236 ofd_obd(ofd)->obd_name, rc);
239 l_wait_event(thread->t_ctl_waitq,
240 thread_is_running(thread) ||
241 thread_is_stopped(thread),
248 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
250 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
251 struct l_wait_info lwi = { 0 };
253 spin_lock(&ofd->ofd_inconsistency_lock);
254 if (thread_is_init(thread) || thread_is_stopped(thread)) {
255 spin_unlock(&ofd->ofd_inconsistency_lock);
260 thread_set_flags(thread, SVC_STOPPING);
261 spin_unlock(&ofd->ofd_inconsistency_lock);
262 wake_up_all(&thread->t_ctl_waitq);
263 l_wait_event(thread->t_ctl_waitq,
264 thread_is_stopped(thread),
270 static void ofd_add_inconsistency_item(const struct lu_env *env,
271 struct ofd_object *fo, struct obdo *oa)
273 struct ofd_device *ofd = ofd_obj2dev(fo);
274 struct ofd_inconsistency_item *oii;
279 CERROR("%s: cannot alloc memory for verify OST-object "
280 "consistency for "DFID", client given PFID "DFID
281 ", local stored PFID "DFID"\n",
282 ofd_obd(ofd)->obd_name, PFID(&fo->ofo_header.loh_fid),
283 oa->o_parent_seq, oa->o_parent_oid, oa->o_stripe_idx,
284 PFID(&fo->ofo_pfid));
289 INIT_LIST_HEAD(&oii->oii_list);
290 lu_object_get(&fo->ofo_obj.do_lu);
292 oii->oii_pfid.f_seq = oa->o_parent_seq;
293 oii->oii_pfid.f_oid = oa->o_parent_oid;
294 oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
296 spin_lock(&ofd->ofd_inconsistency_lock);
297 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
298 spin_unlock(&ofd->ofd_inconsistency_lock);
304 fo->ofo_pfid_checking = 1;
305 if (list_empty(&ofd->ofd_inconsistency_list))
307 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
308 spin_unlock(&ofd->ofd_inconsistency_lock);
310 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
312 /* XXX: When the found inconsistency exceeds some threshold,
313 * we can trigger the LFSCK to scan part of the system
314 * or the whole system, which depends on how to define
315 * the threshold, a simple way maybe like that: define
316 * the absolute value of how many inconsisteny allowed
317 * to be repaired via self detect/repair mechanism, if
318 * exceeded, then trigger the LFSCK to scan the layout
319 * inconsistency within the whole system. */
322 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
325 struct lu_fid *pfid = &fo->ofo_pfid;
329 if (fid_is_sane(pfid)) {
330 if (likely(oa->o_parent_seq == pfid->f_seq &&
331 oa->o_parent_oid == pfid->f_oid &&
332 oa->o_stripe_idx == pfid->f_stripe_idx))
335 if (fo->ofo_pfid_verified)
339 /* The OST-object may be inconsistent, and we need further verification.
340 * To avoid block the RPC service thread, return -EINPROGRESS to client
341 * and make it retry later. */
342 if (fo->ofo_pfid_checking)
343 RETURN(-EINPROGRESS);
345 rc = ofd_object_ff_load(env, fo);
352 if (likely(oa->o_parent_seq == pfid->f_seq &&
353 oa->o_parent_oid == pfid->f_oid &&
354 oa->o_stripe_idx == pfid->f_stripe_idx))
357 /* Push it to the dedicated thread for further verification. */
358 ofd_add_inconsistency_item(env, fo, oa);
360 RETURN(-EINPROGRESS);
363 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
364 struct ofd_device *ofd, const struct lu_fid *fid,
365 struct lu_attr *la, struct obdo *oa, int niocount,
366 struct niobuf_remote *rnb, int *nr_local,
367 struct niobuf_local *lnb, char *jobid)
369 struct ofd_object *fo;
370 int i, j, rc, tot_bytes = 0;
373 LASSERT(env != NULL);
375 fo = ofd_object_find(env, ofd, fid);
380 ofd_read_lock(env, fo);
381 if (!ofd_object_exists(fo))
382 GOTO(unlock, rc = -ENOENT);
384 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
385 rc = ofd_verify_ff(env, fo, oa);
390 /* parse remote buffers to local buffers and prepare the latter */
392 for (i = 0, j = 0; i < niocount; i++) {
393 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
394 lnb + j, 0, ofd_object_capa(env, fo));
395 if (unlikely(rc < 0))
397 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
398 /* correct index for local buffers to continue with */
401 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
402 tot_bytes += rnb[i].rnb_len;
405 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
406 rc = dt_attr_get(env, ofd_object_child(fo), la,
407 ofd_object_capa(env, fo));
411 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
415 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
419 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
421 ofd_read_unlock(env, fo);
422 ofd_object_put(env, fo);
426 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
427 struct ofd_device *ofd, const struct lu_fid *fid,
428 struct lu_attr *la, struct obdo *oa,
429 int objcount, struct obd_ioobj *obj,
430 struct niobuf_remote *rnb, int *nr_local,
431 struct niobuf_local *lnb, char *jobid)
433 struct ofd_object *fo;
434 int i, j, k, rc = 0, tot_bytes = 0;
437 LASSERT(env != NULL);
438 LASSERT(objcount == 1);
440 if (unlikely(exp->exp_obd->obd_recovering)) {
441 obd_seq seq = fid_seq(fid);
442 obd_id oid = fid_oid(fid);
443 struct ofd_seq *oseq;
445 oseq = ofd_seq_load(env, ofd, seq);
447 CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
448 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
449 GOTO(out, rc = -EINVAL);
452 if (oid > ofd_seq_last_oid(oseq)) {
456 mutex_lock(&oseq->os_create_lock);
457 diff = oid - ofd_seq_last_oid(oseq);
459 /* Do sync create if the seq is about to used up */
460 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
461 if (unlikely(oid >= IDIF_MAX_OID - 1))
463 } else if (fid_seq_is_norm(seq)) {
465 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
468 CERROR("%s : invalid o_seq "DOSTID"\n",
469 ofd_name(ofd), POSTID(&oa->o_oi));
470 mutex_unlock(&oseq->os_create_lock);
471 ofd_seq_put(env, oseq);
472 GOTO(out, rc = -EINVAL);
476 obd_id next_id = ofd_seq_last_oid(oseq) + 1;
477 int count = ofd_precreate_batch(ofd, diff);
479 rc = ofd_precreate_objects(env, ofd, next_id,
482 mutex_unlock(&oseq->os_create_lock);
483 ofd_seq_put(env, oseq);
490 mutex_unlock(&oseq->os_create_lock);
493 ofd_seq_put(env, oseq);
496 fo = ofd_object_find(env, ofd, fid);
498 GOTO(out, rc = PTR_ERR(fo));
501 ofd_read_lock(env, fo);
502 if (!ofd_object_exists(fo)) {
503 CERROR("%s: BRW to missing obj "DOSTID"\n",
504 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
505 ofd_read_unlock(env, fo);
506 ofd_object_put(env, fo);
507 GOTO(out, rc = -ENOENT);
510 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
511 rc = ofd_verify_ff(env, fo, oa);
513 ofd_read_unlock(env, fo);
514 ofd_object_put(env, fo);
519 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
520 * space back if possible */
521 ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
523 /* parse remote buffers to local buffers and prepare the latter */
525 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
526 rc = dt_bufs_get(env, ofd_object_child(fo),
528 ofd_object_capa(env, fo));
529 if (unlikely(rc < 0))
531 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
532 /* correct index for local buffers to continue with */
533 for (k = 0; k < rc; k++) {
534 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
535 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
536 lnb[j+k].lnb_rc = -ENOSPC;
538 /* remote client can't break through quota */
539 if (exp_connect_rmtclient(exp))
540 lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
544 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
545 tot_bytes += rnb[i].rnb_len;
547 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
549 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
550 if (unlikely(rc != 0))
553 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
556 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
557 ofd_read_unlock(env, fo);
558 /* ofd_grant_prepare_write() was called, so we must commit */
559 ofd_grant_commit(env, exp, rc);
561 /* let's still process incoming grant information packed in the oa,
562 * but without enforcing grant since we won't proceed with the write.
563 * Just like a read request actually. */
564 ofd_grant_prepare_read(env, exp, oa);
568 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
569 struct obdo *oa, int objcount, struct obd_ioobj *obj,
570 struct niobuf_remote *rnb, int *nr_local,
571 struct niobuf_local *lnb, struct obd_trans_info *oti,
572 struct lustre_capa *capa)
574 struct tgt_session_info *tsi = tgt_ses_info(env);
575 struct ofd_device *ofd = ofd_exp(exp);
576 struct ofd_thread_info *info;
578 const struct lu_fid *fid = &oa->o_oi.oi_fid;
581 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
582 CERROR("%s: bulk has too many pages %d, which exceeds the"
583 "maximum pages per RPC of %d\n",
584 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
588 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
589 LASSERT(oti != NULL);
590 info = ofd_info_init(env, exp);
591 ofd_oti2info(info, oti);
592 jobid = oti->oti_jobid;
594 info = tsi2ofd_info(tsi);
595 jobid = tsi->tsi_jobid;
600 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
601 struct ofd_seq *oseq;
603 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
605 CERROR("%s: Can not find seq for "DOSTID
606 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
611 if (oseq->os_destroys_in_progress == 0) {
612 /* don't fail lookups for orphan recovery, it causes
613 * later LBUGs when objects still exist during
615 ofd_seq_put(env, oseq);
618 ofd_seq_put(env, oseq);
621 LASSERT(objcount == 1);
622 LASSERT(obj->ioo_bufcnt > 0);
624 if (cmd == OBD_BRW_WRITE) {
625 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
626 capa, CAPA_OPC_OSS_WRITE);
628 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
629 rc = ofd_preprw_write(env, exp, ofd, fid,
630 &info->fti_attr, oa, objcount,
631 obj, rnb, nr_local, lnb, jobid);
633 } else if (cmd == OBD_BRW_READ) {
634 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
635 capa, CAPA_OPC_OSS_READ);
637 ofd_grant_prepare_read(env, exp, oa);
638 rc = ofd_preprw_read(env, exp, ofd, fid,
640 obj->ioo_bufcnt, rnb, nr_local,
642 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
645 CERROR("%s: wrong cmd %d received!\n",
646 exp->exp_obd->obd_name, cmd);
653 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
654 const struct lu_fid *fid, int objcount, int niocount,
655 struct niobuf_local *lnb)
657 struct ofd_object *fo;
661 LASSERT(niocount > 0);
663 fo = ofd_object_find(env, ofd, fid);
667 LASSERT(ofd_object_exists(fo));
668 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
670 ofd_read_unlock(env, fo);
671 ofd_object_put(env, fo);
672 /* second put is pair to object_get in ofd_preprw_read */
673 ofd_object_put(env, fo);
679 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
680 struct ofd_object *ofd_obj, struct lu_attr *la,
681 struct filter_fid *ff)
683 struct ofd_thread_info *info = ofd_info(env);
684 __u64 valid = la->la_valid;
687 struct dt_object *dt_obj;
694 dt_obj = ofd_object_child(ofd_obj);
695 LASSERT(dt_obj != NULL);
697 la->la_valid &= LA_UID | LA_GID;
699 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
704 rc = ofd_object_ff_load(env, ofd_obj);
711 if (!la->la_valid && !ff_needed)
712 /* no attributes to set */
715 th = ofd_trans_create(env, ofd);
717 GOTO(out, rc = PTR_ERR(th));
720 rc = dt_declare_attr_set(env, dt_obj, la, th);
726 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
727 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
728 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
729 ff->ff_parent.f_oid =
730 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
732 info->fti_buf.lb_buf = ff;
733 info->fti_buf.lb_len = sizeof(*ff);
734 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
735 XATTR_NAME_FID, 0, th);
740 /* We don't need a transno for this operation which will be re-executed
741 * anyway when the OST_WRITE (with a transno assigned) is replayed */
742 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
748 rc = dt_attr_set(env, dt_obj, la, th,
749 ofd_object_capa(env, ofd_obj));
754 /* set filter fid EA */
756 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
759 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
762 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
763 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
764 /* Currently, the filter_fid::ff_parent::f_ver is not
765 * the real parent MDT-object's FID::f_ver, instead it
766 * is the OST-object index in its parent MDT-object's
768 ofd_obj->ofo_pfid.f_stripe_idx =
769 le32_to_cpu(ff->ff_parent.f_stripe_idx);
776 dt_trans_stop(env, ofd->ofd_osd, th);
778 la->la_valid = valid;
782 struct ofd_soft_sync_callback {
783 struct dt_txn_commit_cb ossc_cb;
784 struct obd_export *ossc_exp;
787 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
788 struct dt_txn_commit_cb *cb, int err)
790 struct ofd_soft_sync_callback *ossc;
792 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
794 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
795 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
797 class_export_cb_put(ossc->ossc_exp);
801 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
803 struct ofd_soft_sync_callback *ossc;
804 struct dt_txn_commit_cb *dcb;
811 ossc->ossc_exp = class_export_cb_get(exp);
813 dcb = &ossc->ossc_cb;
814 dcb->dcb_func = ofd_cb_soft_sync;
815 CFS_INIT_LIST_HEAD(&dcb->dcb_linkage);
816 strncpy(dcb->dcb_name, "ofd_cb_soft_sync", MAX_COMMIT_CB_STR_LEN);
817 dcb->dcb_name[MAX_COMMIT_CB_STR_LEN - 1] = '\0';
819 rc = dt_trans_cb_add(th, dcb);
821 class_export_cb_put(exp);
829 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
830 struct ofd_device *ofd, const struct lu_fid *fid,
831 struct lu_attr *la, struct filter_fid *ff, int objcount,
832 int niocount, struct niobuf_local *lnb, int old_rc)
834 struct ofd_thread_info *info = ofd_info(env);
835 struct ofd_object *fo;
841 struct filter_export_data *fed = &exp->exp_filter_data;
842 bool soft_sync = false;
843 bool cb_registered = false;
847 LASSERT(objcount == 1);
849 fo = ofd_object_find(env, ofd, fid);
851 LASSERT(ofd_object_exists(fo));
853 o = ofd_object_child(fo);
857 GOTO(out, rc = old_rc);
860 * The first write to each object must set some attributes. It is
861 * important to set the uid/gid before calling
862 * dt_declare_write_commit() since quota enforcement is now handled in
865 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
869 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
872 th = ofd_trans_create(env, ofd);
874 GOTO(out, rc = PTR_ERR(th));
876 th->th_sync |= ofd->ofd_syncjournal;
877 if (th->th_sync == 0) {
878 for (i = 0; i < niocount; i++) {
879 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
883 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
888 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
889 GOTO(out_stop, rc = -EINPROGRESS);
891 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
896 /* update [mac]time if needed */
897 rc = dt_declare_attr_set(env, o, la, th);
902 rc = ofd_trans_start(env, ofd, fo, th);
906 rc = dt_write_commit(env, o, lnb, niocount, th);
911 rc = dt_attr_set(env, o, la, th, ofd_object_capa(env, fo));
916 /* get attr to return */
917 rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo));
920 /* Force commit to make the just-deleted blocks
921 * reusable. LU-456 */
925 /* do this before trans stop in case commit has finished */
926 if (!th->th_sync && soft_sync && !cb_registered) {
927 ofd_soft_sync_cb_add(th, exp);
928 cb_registered = true;
931 ofd_trans_stop(env, ofd, th, rc);
932 if (rc == -ENOSPC && retries++ < 3) {
933 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
939 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
940 atomic_set(&fed->fed_soft_sync_count, 0);
941 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
942 ofd->ofd_soft_sync_limit)
943 dt_commit_async(env, ofd->ofd_osd);
946 dt_bufs_put(env, o, lnb, niocount);
947 ofd_read_unlock(env, fo);
948 ofd_object_put(env, fo);
949 /* second put is pair to object_get in ofd_preprw_write */
950 ofd_object_put(env, fo);
951 ofd_grant_commit(env, info->fti_exp, old_rc);
955 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
956 struct obdo *oa, int objcount, struct obd_ioobj *obj,
957 struct niobuf_remote *rnb, int npages,
958 struct niobuf_local *lnb, struct obd_trans_info *oti,
961 struct ofd_thread_info *info = ofd_info(env);
962 struct ofd_mod_data *fmd;
964 struct ofd_device *ofd = ofd_exp(exp);
965 struct filter_fid *ff = NULL;
966 const struct lu_fid *fid = &oa->o_oi.oi_fid;
971 if (cmd == OBD_BRW_WRITE) {
972 /* Don't update timestamps if this write is older than a
973 * setattr which modifies the timestamps. b=10150 */
975 /* XXX when we start having persistent reservations this needs
976 * to be changed to ofd_fmd_get() to create the fmd if it
977 * doesn't already exist so we can store the reservation handle
979 valid = OBD_MD_FLUID | OBD_MD_FLGID;
980 fmd = ofd_fmd_find(exp, fid);
981 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
982 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
984 ofd_fmd_put(exp, fmd);
985 la_from_obdo(&info->fti_attr, oa, valid);
987 if (oa->o_valid & OBD_MD_FLFID) {
988 ff = &info->fti_mds_fid;
989 ofd_prepare_fidea(ff, oa);
992 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
993 ff, objcount, npages, lnb, old_rc);
995 obdo_from_la(oa, &info->fti_attr,
996 OFD_VALID_FLAGS | LA_GID | LA_UID);
998 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
1000 /* don't report overquota flag if we failed before reaching
1002 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1003 /* return the overquota flags to client */
1004 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1005 if (oa->o_valid & OBD_MD_FLFLAGS)
1006 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1008 oa->o_flags = OBD_FL_NO_USRQUOTA;
1011 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1012 if (oa->o_valid & OBD_MD_FLFLAGS)
1013 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1015 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1018 oa->o_valid |= OBD_MD_FLFLAGS;
1019 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1021 } else if (cmd == OBD_BRW_READ) {
1022 struct ldlm_namespace *ns = ofd->ofd_namespace;
1024 /* If oa != NULL then ofd_preprw_read updated the inode
1025 * atime and we should update the lvb so that other glimpses
1026 * will also get the updated value. bug 5972 */
1027 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1028 struct ldlm_resource *rs = NULL;
1030 ost_fid_build_resid(fid, &info->fti_resid);
1031 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1034 ns->ns_lvbo->lvbo_update(rs, NULL, 1);
1035 ldlm_resource_putref(rs);
1038 rc = ofd_commitrw_read(env, ofd, fid, objcount,
1048 ofd_info2oti(info, oti);