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 };
221 spin_lock(&ofd->ofd_inconsistency_lock);
222 if (unlikely(thread_is_running(thread))) {
223 spin_unlock(&ofd->ofd_inconsistency_lock);
228 thread_set_flags(thread, 0);
229 spin_unlock(&ofd->ofd_inconsistency_lock);
230 rc = PTR_ERR(kthread_run(ofd_inconsistency_verification_main, ofd,
231 "inconsistency_verification"));
232 if (IS_ERR_VALUE(rc)) {
233 CERROR("%s: cannot start self_repair thread: rc = %ld\n",
234 ofd_obd(ofd)->obd_name, rc);
237 l_wait_event(thread->t_ctl_waitq,
238 thread_is_running(thread) ||
239 thread_is_stopped(thread),
246 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
248 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
249 struct l_wait_info lwi = { 0 };
251 spin_lock(&ofd->ofd_inconsistency_lock);
252 if (thread_is_init(thread) || thread_is_stopped(thread)) {
253 spin_unlock(&ofd->ofd_inconsistency_lock);
258 thread_set_flags(thread, SVC_STOPPING);
259 spin_unlock(&ofd->ofd_inconsistency_lock);
260 wake_up_all(&thread->t_ctl_waitq);
261 l_wait_event(thread->t_ctl_waitq,
262 thread_is_stopped(thread),
268 static void ofd_add_inconsistency_item(const struct lu_env *env,
269 struct ofd_object *fo, struct obdo *oa)
271 struct ofd_device *ofd = ofd_obj2dev(fo);
272 struct ofd_inconsistency_item *oii;
277 CERROR("%s: cannot alloc memory for verify OST-object "
278 "consistency for "DFID", client given PFID "DFID
279 ", local stored PFID "DFID"\n",
280 ofd_obd(ofd)->obd_name, PFID(&fo->ofo_header.loh_fid),
281 oa->o_parent_seq, oa->o_parent_oid, oa->o_stripe_idx,
282 PFID(&fo->ofo_pfid));
287 INIT_LIST_HEAD(&oii->oii_list);
288 lu_object_get(&fo->ofo_obj.do_lu);
290 oii->oii_pfid.f_seq = oa->o_parent_seq;
291 oii->oii_pfid.f_oid = oa->o_parent_oid;
292 oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
294 spin_lock(&ofd->ofd_inconsistency_lock);
295 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
296 spin_unlock(&ofd->ofd_inconsistency_lock);
302 fo->ofo_pfid_checking = 1;
303 if (list_empty(&ofd->ofd_inconsistency_list))
305 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
306 spin_unlock(&ofd->ofd_inconsistency_lock);
308 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
310 /* XXX: When the found inconsistency exceeds some threshold,
311 * we can trigger the LFSCK to scan part of the system
312 * or the whole system, which depends on how to define
313 * the threshold, a simple way maybe like that: define
314 * the absolute value of how many inconsisteny allowed
315 * to be repaired via self detect/repair mechanism, if
316 * exceeded, then trigger the LFSCK to scan the layout
317 * inconsistency within the whole system. */
320 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
323 struct lu_fid *pfid = &fo->ofo_pfid;
327 if (fid_is_sane(pfid)) {
328 if (likely(oa->o_parent_seq == pfid->f_seq &&
329 oa->o_parent_oid == pfid->f_oid &&
330 oa->o_stripe_idx == pfid->f_stripe_idx))
333 if (fo->ofo_pfid_verified)
337 /* The OST-object may be inconsistent, and we need further verification.
338 * To avoid block the RPC service thread, return -EINPROGRESS to client
339 * and make it retry later. */
340 if (fo->ofo_pfid_checking)
341 RETURN(-EINPROGRESS);
343 rc = ofd_object_ff_load(env, fo);
350 if (likely(oa->o_parent_seq == pfid->f_seq &&
351 oa->o_parent_oid == pfid->f_oid &&
352 oa->o_stripe_idx == pfid->f_stripe_idx))
355 /* Push it to the dedicated thread for further verification. */
356 ofd_add_inconsistency_item(env, fo, oa);
358 RETURN(-EINPROGRESS);
361 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
362 struct ofd_device *ofd, const struct lu_fid *fid,
363 struct lu_attr *la, struct obdo *oa, int niocount,
364 struct niobuf_remote *rnb, int *nr_local,
365 struct niobuf_local *lnb, char *jobid)
367 struct ofd_object *fo;
368 int i, j, rc, tot_bytes = 0;
371 LASSERT(env != NULL);
373 fo = ofd_object_find(env, ofd, fid);
378 ofd_read_lock(env, fo);
379 if (!ofd_object_exists(fo))
380 GOTO(unlock, rc = -ENOENT);
382 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
383 rc = ofd_verify_ff(env, fo, oa);
388 /* parse remote buffers to local buffers and prepare the latter */
390 for (i = 0, j = 0; i < niocount; i++) {
391 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
392 lnb + j, 0, ofd_object_capa(env, fo));
393 if (unlikely(rc < 0))
395 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
396 /* correct index for local buffers to continue with */
399 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
400 tot_bytes += rnb[i].rnb_len;
403 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
404 rc = dt_attr_get(env, ofd_object_child(fo), la,
405 ofd_object_capa(env, fo));
409 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
413 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
417 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
419 ofd_read_unlock(env, fo);
420 ofd_object_put(env, fo);
424 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
425 struct ofd_device *ofd, const struct lu_fid *fid,
426 struct lu_attr *la, struct obdo *oa,
427 int objcount, struct obd_ioobj *obj,
428 struct niobuf_remote *rnb, int *nr_local,
429 struct niobuf_local *lnb, char *jobid)
431 struct ofd_object *fo;
432 int i, j, k, rc = 0, tot_bytes = 0;
435 LASSERT(env != NULL);
436 LASSERT(objcount == 1);
438 if (unlikely(exp->exp_obd->obd_recovering)) {
439 obd_seq seq = fid_seq(fid);
440 obd_id oid = fid_oid(fid);
441 struct ofd_seq *oseq;
443 oseq = ofd_seq_load(env, ofd, seq);
445 CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
446 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
447 GOTO(out, rc = -EINVAL);
450 if (oid > ofd_seq_last_oid(oseq)) {
454 mutex_lock(&oseq->os_create_lock);
455 diff = oid - ofd_seq_last_oid(oseq);
457 /* Do sync create if the seq is about to used up */
458 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
459 if (unlikely(oid >= IDIF_MAX_OID - 1))
461 } else if (fid_seq_is_norm(seq)) {
463 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
466 CERROR("%s : invalid o_seq "DOSTID"\n",
467 ofd_name(ofd), POSTID(&oa->o_oi));
468 mutex_unlock(&oseq->os_create_lock);
469 ofd_seq_put(env, oseq);
470 GOTO(out, rc = -EINVAL);
474 obd_id next_id = ofd_seq_last_oid(oseq) + 1;
475 int count = ofd_precreate_batch(ofd, diff);
477 rc = ofd_precreate_objects(env, ofd, next_id,
480 mutex_unlock(&oseq->os_create_lock);
481 ofd_seq_put(env, oseq);
488 mutex_unlock(&oseq->os_create_lock);
491 ofd_seq_put(env, oseq);
494 fo = ofd_object_find(env, ofd, fid);
496 GOTO(out, rc = PTR_ERR(fo));
499 ofd_read_lock(env, fo);
500 if (!ofd_object_exists(fo)) {
501 CERROR("%s: BRW to missing obj "DOSTID"\n",
502 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
503 ofd_read_unlock(env, fo);
504 ofd_object_put(env, fo);
505 GOTO(out, rc = -ENOENT);
508 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
509 rc = ofd_verify_ff(env, fo, oa);
511 ofd_read_unlock(env, fo);
512 ofd_object_put(env, fo);
517 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
518 * space back if possible */
519 ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
521 /* parse remote buffers to local buffers and prepare the latter */
523 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
524 rc = dt_bufs_get(env, ofd_object_child(fo),
526 ofd_object_capa(env, fo));
527 if (unlikely(rc < 0))
529 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
530 /* correct index for local buffers to continue with */
531 for (k = 0; k < rc; k++) {
532 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
533 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
534 lnb[j+k].lnb_rc = -ENOSPC;
536 /* remote client can't break through quota */
537 if (exp_connect_rmtclient(exp))
538 lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
542 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
543 tot_bytes += rnb[i].rnb_len;
545 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
547 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
548 if (unlikely(rc != 0))
551 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
554 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
555 ofd_read_unlock(env, fo);
556 /* ofd_grant_prepare_write() was called, so we must commit */
557 ofd_grant_commit(env, exp, rc);
559 /* let's still process incoming grant information packed in the oa,
560 * but without enforcing grant since we won't proceed with the write.
561 * Just like a read request actually. */
562 ofd_grant_prepare_read(env, exp, oa);
566 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
567 struct obdo *oa, int objcount, struct obd_ioobj *obj,
568 struct niobuf_remote *rnb, int *nr_local,
569 struct niobuf_local *lnb, struct obd_trans_info *oti,
570 struct lustre_capa *capa)
572 struct tgt_session_info *tsi = tgt_ses_info(env);
573 struct ofd_device *ofd = ofd_exp(exp);
574 struct ofd_thread_info *info;
576 const struct lu_fid *fid = &oa->o_oi.oi_fid;
579 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
580 CERROR("%s: bulk has too many pages %d, which exceeds the"
581 "maximum pages per RPC of %d\n",
582 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
586 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
587 LASSERT(oti != NULL);
588 info = ofd_info_init(env, exp);
589 ofd_oti2info(info, oti);
590 jobid = oti->oti_jobid;
592 info = tsi2ofd_info(tsi);
593 jobid = tsi->tsi_jobid;
598 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
599 struct ofd_seq *oseq;
601 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
603 CERROR("%s: Can not find seq for "DOSTID
604 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
609 if (oseq->os_destroys_in_progress == 0) {
610 /* don't fail lookups for orphan recovery, it causes
611 * later LBUGs when objects still exist during
613 ofd_seq_put(env, oseq);
616 ofd_seq_put(env, oseq);
619 LASSERT(objcount == 1);
620 LASSERT(obj->ioo_bufcnt > 0);
622 if (cmd == OBD_BRW_WRITE) {
623 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
624 capa, CAPA_OPC_OSS_WRITE);
626 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
627 rc = ofd_preprw_write(env, exp, ofd, fid,
628 &info->fti_attr, oa, objcount,
629 obj, rnb, nr_local, lnb, jobid);
631 } else if (cmd == OBD_BRW_READ) {
632 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
633 capa, CAPA_OPC_OSS_READ);
635 ofd_grant_prepare_read(env, exp, oa);
636 rc = ofd_preprw_read(env, exp, ofd, fid,
638 obj->ioo_bufcnt, rnb, nr_local,
640 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
643 CERROR("%s: wrong cmd %d received!\n",
644 exp->exp_obd->obd_name, cmd);
651 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
652 const struct lu_fid *fid, int objcount, int niocount,
653 struct niobuf_local *lnb)
655 struct ofd_object *fo;
659 LASSERT(niocount > 0);
661 fo = ofd_object_find(env, ofd, fid);
665 LASSERT(ofd_object_exists(fo));
666 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
668 ofd_read_unlock(env, fo);
669 ofd_object_put(env, fo);
670 /* second put is pair to object_get in ofd_preprw_read */
671 ofd_object_put(env, fo);
677 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
678 struct ofd_object *ofd_obj, struct lu_attr *la,
679 struct filter_fid *ff)
681 struct ofd_thread_info *info = ofd_info(env);
682 __u64 valid = la->la_valid;
685 struct dt_object *dt_obj;
692 dt_obj = ofd_object_child(ofd_obj);
693 LASSERT(dt_obj != NULL);
695 la->la_valid &= LA_UID | LA_GID;
697 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
702 rc = ofd_object_ff_load(env, ofd_obj);
709 if (!la->la_valid && !ff_needed)
710 /* no attributes to set */
713 th = ofd_trans_create(env, ofd);
715 GOTO(out, rc = PTR_ERR(th));
718 rc = dt_declare_attr_set(env, dt_obj, la, th);
724 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
725 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
726 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
727 ff->ff_parent.f_oid =
728 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
730 info->fti_buf.lb_buf = ff;
731 info->fti_buf.lb_len = sizeof(*ff);
732 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
733 XATTR_NAME_FID, 0, th);
738 /* We don't need a transno for this operation which will be re-executed
739 * anyway when the OST_WRITE (with a transno assigned) is replayed */
740 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
746 rc = dt_attr_set(env, dt_obj, la, th,
747 ofd_object_capa(env, ofd_obj));
752 /* set filter fid EA */
754 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
757 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
760 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
761 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
762 /* Currently, the filter_fid::ff_parent::f_ver is not
763 * the real parent MDT-object's FID::f_ver, instead it
764 * is the OST-object index in its parent MDT-object's
766 ofd_obj->ofo_pfid.f_stripe_idx =
767 le32_to_cpu(ff->ff_parent.f_stripe_idx);
774 dt_trans_stop(env, ofd->ofd_osd, th);
776 la->la_valid = valid;
780 struct ofd_soft_sync_callback {
781 struct dt_txn_commit_cb ossc_cb;
782 struct obd_export *ossc_exp;
785 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
786 struct dt_txn_commit_cb *cb, int err)
788 struct ofd_soft_sync_callback *ossc;
790 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
792 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
793 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
795 class_export_cb_put(ossc->ossc_exp);
799 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
801 struct ofd_soft_sync_callback *ossc;
802 struct dt_txn_commit_cb *dcb;
809 ossc->ossc_exp = class_export_cb_get(exp);
811 dcb = &ossc->ossc_cb;
812 dcb->dcb_func = ofd_cb_soft_sync;
813 CFS_INIT_LIST_HEAD(&dcb->dcb_linkage);
814 strncpy(dcb->dcb_name, "ofd_cb_soft_sync", MAX_COMMIT_CB_STR_LEN);
815 dcb->dcb_name[MAX_COMMIT_CB_STR_LEN - 1] = '\0';
817 rc = dt_trans_cb_add(th, dcb);
819 class_export_cb_put(exp);
827 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
828 struct ofd_device *ofd, const struct lu_fid *fid,
829 struct lu_attr *la, struct filter_fid *ff, int objcount,
830 int niocount, struct niobuf_local *lnb, int old_rc)
832 struct ofd_thread_info *info = ofd_info(env);
833 struct ofd_object *fo;
839 struct filter_export_data *fed = &exp->exp_filter_data;
840 bool soft_sync = false;
841 bool cb_registered = false;
845 LASSERT(objcount == 1);
847 fo = ofd_object_find(env, ofd, fid);
849 LASSERT(ofd_object_exists(fo));
851 o = ofd_object_child(fo);
855 GOTO(out, rc = old_rc);
858 * The first write to each object must set some attributes. It is
859 * important to set the uid/gid before calling
860 * dt_declare_write_commit() since quota enforcement is now handled in
863 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
867 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
870 th = ofd_trans_create(env, ofd);
872 GOTO(out, rc = PTR_ERR(th));
874 th->th_sync |= ofd->ofd_syncjournal;
875 if (th->th_sync == 0) {
876 for (i = 0; i < niocount; i++) {
877 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
881 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
886 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
887 GOTO(out_stop, rc = -EINPROGRESS);
889 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
894 /* update [mac]time if needed */
895 rc = dt_declare_attr_set(env, o, la, th);
900 rc = ofd_trans_start(env, ofd, fo, th);
904 rc = dt_write_commit(env, o, lnb, niocount, th);
909 rc = dt_attr_set(env, o, la, th, ofd_object_capa(env, fo));
914 /* get attr to return */
915 rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo));
918 /* Force commit to make the just-deleted blocks
919 * reusable. LU-456 */
923 /* do this before trans stop in case commit has finished */
924 if (!th->th_sync && soft_sync && !cb_registered) {
925 ofd_soft_sync_cb_add(th, exp);
926 cb_registered = true;
929 ofd_trans_stop(env, ofd, th, rc);
930 if (rc == -ENOSPC && retries++ < 3) {
931 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
937 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
938 atomic_set(&fed->fed_soft_sync_count, 0);
939 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
940 ofd->ofd_soft_sync_limit)
941 dt_commit_async(env, ofd->ofd_osd);
944 dt_bufs_put(env, o, lnb, niocount);
945 ofd_read_unlock(env, fo);
946 ofd_object_put(env, fo);
947 /* second put is pair to object_get in ofd_preprw_write */
948 ofd_object_put(env, fo);
949 ofd_grant_commit(env, info->fti_exp, old_rc);
953 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
954 struct obdo *oa, int objcount, struct obd_ioobj *obj,
955 struct niobuf_remote *rnb, int npages,
956 struct niobuf_local *lnb, struct obd_trans_info *oti,
959 struct ofd_thread_info *info = ofd_info(env);
960 struct ofd_mod_data *fmd;
962 struct ofd_device *ofd = ofd_exp(exp);
963 struct filter_fid *ff = NULL;
964 const struct lu_fid *fid = &oa->o_oi.oi_fid;
969 if (cmd == OBD_BRW_WRITE) {
970 /* Don't update timestamps if this write is older than a
971 * setattr which modifies the timestamps. b=10150 */
973 /* XXX when we start having persistent reservations this needs
974 * to be changed to ofd_fmd_get() to create the fmd if it
975 * doesn't already exist so we can store the reservation handle
977 valid = OBD_MD_FLUID | OBD_MD_FLGID;
978 fmd = ofd_fmd_find(exp, fid);
979 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
980 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
982 ofd_fmd_put(exp, fmd);
983 la_from_obdo(&info->fti_attr, oa, valid);
985 if (oa->o_valid & OBD_MD_FLFID) {
986 ff = &info->fti_mds_fid;
987 ofd_prepare_fidea(ff, oa);
990 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
991 ff, objcount, npages, lnb, old_rc);
993 obdo_from_la(oa, &info->fti_attr,
994 OFD_VALID_FLAGS | LA_GID | LA_UID);
996 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
998 /* don't report overquota flag if we failed before reaching
1000 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1001 /* return the overquota flags to client */
1002 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1003 if (oa->o_valid & OBD_MD_FLFLAGS)
1004 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1006 oa->o_flags = OBD_FL_NO_USRQUOTA;
1009 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1010 if (oa->o_valid & OBD_MD_FLFLAGS)
1011 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1013 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1016 oa->o_valid |= OBD_MD_FLFLAGS;
1017 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1019 } else if (cmd == OBD_BRW_READ) {
1020 struct ldlm_namespace *ns = ofd->ofd_namespace;
1022 /* If oa != NULL then ofd_preprw_read updated the inode
1023 * atime and we should update the lvb so that other glimpses
1024 * will also get the updated value. bug 5972 */
1025 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1026 struct ldlm_resource *rs = NULL;
1028 ost_fid_build_resid(fid, &info->fti_resid);
1029 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1032 ns->ns_lvbo->lvbo_update(rs, NULL, 1);
1033 ldlm_resource_putref(rs);
1036 rc = ofd_commitrw_read(env, ofd, fid, objcount,
1046 ofd_info2oti(info, oti);