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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2014 Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * This file provides functions to handle IO requests from clients and
35 * also LFSCK routines to check parent file identifier (PFID) consistency.
37 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
38 * Author: Fan Yong <fan.yong@intel.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;
52 * Verify single object for parent FID consistency.
54 * Part of LFSCK processing which checks single object PFID stored in extended
55 * attribute (XATTR) against real FID of MDT parent object received by LFSCK.
56 * This verifies that the OST object is being referenced by only a single MDT
59 * \param[in] env execution environment
60 * \param[in] ofd OFD device
61 * \param[in] oii object-related local data
62 * \param[in] lr LFSCK request data
64 static void ofd_inconsistency_verify_one(const struct lu_env *env,
65 struct ofd_device *ofd,
66 struct ofd_inconsistency_item *oii,
67 struct lfsck_request *lr)
69 struct ofd_object *fo = oii->oii_obj;
70 struct lu_fid *pfid = &fo->ofo_pfid;
73 LASSERT(fo->ofo_pfid_checking);
74 LASSERT(!fo->ofo_pfid_verified);
76 lr->lr_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
77 lr->lr_fid2 = oii->oii_pfid; /* client given PFID. */
78 lr->lr_fid3 = *pfid; /* OST local stored PFID. */
80 rc = lfsck_in_notify(env, ofd->ofd_osd, lr, NULL);
81 ofd_write_lock(env, fo);
82 switch (lr->lr_status) {
87 CDEBUG(D_LFSCK, "%s: fail to verify OST local stored "
88 "PFID xattr for "DFID", the client given PFID "
89 DFID", OST 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 fo->ofo_pfid_verified = 1;
95 case LPVS_INCONSISTENT:
98 ofd->ofd_inconsistency_self_detected++;
100 CDEBUG(D_LFSCK, "%s: fail to verify the client given "
101 "PFID for "DFID", the client given PFID "DFID
102 ", local stored PFID "DFID": rc = %d\n",
103 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
104 PFID(&oii->oii_pfid), PFID(pfid), rc);
106 CDEBUG(D_LFSCK, "%s: both the client given PFID and "
107 "the OST local stored PFID are stale for the "
108 "OST-object "DFID", client given PFID is "DFID
109 ", local stored PFID is "DFID"\n",
110 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
111 PFID(&oii->oii_pfid), PFID(pfid));
113 case LPVS_INCONSISTENT_TOFIX:
114 ofd->ofd_inconsistency_self_detected++;
116 ofd->ofd_inconsistency_self_repaired++;
117 CDEBUG(D_LFSCK, "%s: fixed the staled OST PFID xattr "
118 "for "DFID", with the client given PFID "DFID
119 ", the old stored PFID "DFID"\n",
120 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
121 PFID(&oii->oii_pfid), PFID(pfid));
123 CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
124 "for "DFID", client given PFID "DFID", local "
125 "stored PFID "DFID": rc = %d\n",
126 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
127 PFID(&oii->oii_pfid), PFID(pfid), rc);
129 *pfid = oii->oii_pfid;
130 fo->ofo_pfid_verified = 1;
135 fo->ofo_pfid_checking = 0;
136 ofd_write_unlock(env, fo);
138 lu_object_put(env, &fo->ofo_obj.do_lu);
143 * Verification thread to check parent FID consistency.
145 * Kernel thread to check consistency of parent FID for any
146 * new item added for checking by ofd_add_inconsistency_item().
148 * \param[in] args OFD device
150 * \retval 0 on successful thread termination
151 * \retval negative value if thread can't start
153 static int ofd_inconsistency_verification_main(void *args)
156 struct ofd_device *ofd = args;
157 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
158 struct ofd_inconsistency_item *oii;
159 struct lfsck_request *lr = NULL;
160 struct l_wait_info lwi = { 0 };
164 rc = lu_env_init(&env, LCT_DT_THREAD);
165 spin_lock(&ofd->ofd_inconsistency_lock);
166 thread_set_flags(thread, rc != 0 ? SVC_STOPPED : SVC_RUNNING);
167 wake_up_all(&thread->t_ctl_waitq);
168 spin_unlock(&ofd->ofd_inconsistency_lock);
173 if (unlikely(lr == NULL))
174 GOTO(out, rc = -ENOMEM);
176 lr->lr_event = LE_PAIRS_VERIFY;
177 lr->lr_active = LFSCK_TYPE_LAYOUT;
179 spin_lock(&ofd->ofd_inconsistency_lock);
181 if (unlikely(!thread_is_running(thread)))
184 while (!list_empty(&ofd->ofd_inconsistency_list)) {
185 oii = list_entry(ofd->ofd_inconsistency_list.next,
186 struct ofd_inconsistency_item,
188 list_del_init(&oii->oii_list);
189 spin_unlock(&ofd->ofd_inconsistency_lock);
190 ofd_inconsistency_verify_one(&env, ofd, oii, lr);
191 spin_lock(&ofd->ofd_inconsistency_lock);
194 spin_unlock(&ofd->ofd_inconsistency_lock);
195 l_wait_event(thread->t_ctl_waitq,
196 !list_empty(&ofd->ofd_inconsistency_list) ||
197 !thread_is_running(thread),
199 spin_lock(&ofd->ofd_inconsistency_lock);
202 while (!list_empty(&ofd->ofd_inconsistency_list)) {
203 struct ofd_object *fo;
205 oii = list_entry(ofd->ofd_inconsistency_list.next,
206 struct ofd_inconsistency_item,
208 list_del_init(&oii->oii_list);
210 spin_unlock(&ofd->ofd_inconsistency_lock);
212 ofd_write_lock(&env, fo);
213 fo->ofo_pfid_checking = 0;
214 ofd_write_unlock(&env, fo);
216 lu_object_put(&env, &fo->ofo_obj.do_lu);
218 spin_lock(&ofd->ofd_inconsistency_lock);
226 thread_set_flags(thread, SVC_STOPPED);
227 wake_up_all(&thread->t_ctl_waitq);
228 spin_unlock(&ofd->ofd_inconsistency_lock);
235 * Start parent FID verification thread.
237 * See ofd_inconsistency_verification_main().
239 * \param[in] ofd OFD device
241 * \retval 0 on successful start of thread
242 * \retval negative value on error
244 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
246 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
247 struct l_wait_info lwi = { 0 };
248 struct task_struct *task;
251 spin_lock(&ofd->ofd_inconsistency_lock);
252 if (unlikely(thread_is_running(thread))) {
253 spin_unlock(&ofd->ofd_inconsistency_lock);
258 thread_set_flags(thread, 0);
259 spin_unlock(&ofd->ofd_inconsistency_lock);
260 task = kthread_run(ofd_inconsistency_verification_main, ofd,
261 "inconsistency_verification");
264 CERROR("%s: cannot start self_repair thread: rc = %d\n",
268 l_wait_event(thread->t_ctl_waitq,
269 thread_is_running(thread) ||
270 thread_is_stopped(thread),
278 * Stop parent FID verification thread.
280 * \param[in] ofd OFD device
282 * \retval 0 on successful start of thread
283 * \retval -EALREADY if thread is already stopped
285 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
287 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
288 struct l_wait_info lwi = { 0 };
290 spin_lock(&ofd->ofd_inconsistency_lock);
291 if (thread_is_init(thread) || thread_is_stopped(thread)) {
292 spin_unlock(&ofd->ofd_inconsistency_lock);
297 thread_set_flags(thread, SVC_STOPPING);
298 spin_unlock(&ofd->ofd_inconsistency_lock);
299 wake_up_all(&thread->t_ctl_waitq);
300 l_wait_event(thread->t_ctl_waitq,
301 thread_is_stopped(thread),
308 * Add new item for parent FID verification.
310 * Prepare new verification item and pass it to the dedicated
311 * verification thread for further processing.
313 * \param[in] env execution environment
314 * \param[in] fo OFD object
315 * \param[in] oa OBDO structure with PFID
317 static void ofd_add_inconsistency_item(const struct lu_env *env,
318 struct ofd_object *fo, struct obdo *oa)
320 struct ofd_device *ofd = ofd_obj2dev(fo);
321 struct ofd_inconsistency_item *oii;
328 INIT_LIST_HEAD(&oii->oii_list);
329 lu_object_get(&fo->ofo_obj.do_lu);
331 oii->oii_pfid.f_seq = oa->o_parent_seq;
332 oii->oii_pfid.f_oid = oa->o_parent_oid;
333 oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
335 spin_lock(&ofd->ofd_inconsistency_lock);
336 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
337 spin_unlock(&ofd->ofd_inconsistency_lock);
343 fo->ofo_pfid_checking = 1;
344 if (list_empty(&ofd->ofd_inconsistency_list))
346 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
347 spin_unlock(&ofd->ofd_inconsistency_lock);
349 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
351 /* XXX: When the found inconsistency exceeds some threshold,
352 * we can trigger the LFSCK to scan part of the system
353 * or the whole system, which depends on how to define
354 * the threshold, a simple way maybe like that: define
355 * the absolute value of how many inconsisteny allowed
356 * to be repaired via self detect/repair mechanism, if
357 * exceeded, then trigger the LFSCK to scan the layout
358 * inconsistency within the whole system. */
362 * Verify parent FID of an object.
364 * Check the parent FID is sane and start extended
365 * verification procedure otherwise.
367 * \param[in] env execution environment
368 * \param[in] fo OFD object
369 * \param[in] oa OBDO structure with PFID
371 * \retval 0 on successful verification
372 * \retval -EINPROGRESS if PFID is being repaired
373 * \retval -EPERM if PFID was verified but still insane
375 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
378 struct lu_fid *pfid = &fo->ofo_pfid;
382 if (fid_is_sane(pfid)) {
383 if (likely(oa->o_parent_seq == pfid->f_seq &&
384 oa->o_parent_oid == pfid->f_oid &&
385 oa->o_stripe_idx == pfid->f_stripe_idx))
388 if (fo->ofo_pfid_verified)
392 /* The OST-object may be inconsistent, and we need further verification.
393 * To avoid block the RPC service thread, return -EINPROGRESS to client
394 * and make it retry later. */
395 if (fo->ofo_pfid_checking)
396 RETURN(-EINPROGRESS);
398 rc = ofd_object_ff_load(env, fo);
405 if (likely(oa->o_parent_seq == pfid->f_seq &&
406 oa->o_parent_oid == pfid->f_oid &&
407 oa->o_stripe_idx == pfid->f_stripe_idx))
410 /* Push it to the dedicated thread for further verification. */
411 ofd_add_inconsistency_item(env, fo, oa);
413 RETURN(-EINPROGRESS);
417 * Prepare buffers for read request processing.
419 * This function converts remote buffers from client to local buffers
420 * and prepares the latter.
422 * \param[in] env execution environment
423 * \param[in] exp OBD export of client
424 * \param[in] ofd OFD device
425 * \param[in] fid FID of object
426 * \param[in] la object attributes
427 * \param[in] oa OBDO structure from client
428 * \param[in] niocount number of remote buffers
429 * \param[in] rnb remote buffers
430 * \param[in] nr_local number of local buffers
431 * \param[in] lnb local buffers
432 * \param[in] jobid job ID name
434 * \retval 0 on successful prepare
435 * \retval negative value on error
437 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
438 struct ofd_device *ofd, const struct lu_fid *fid,
439 struct lu_attr *la, struct obdo *oa, int niocount,
440 struct niobuf_remote *rnb, int *nr_local,
441 struct niobuf_local *lnb, char *jobid)
443 struct ofd_object *fo;
444 int i, j, rc, tot_bytes = 0;
447 LASSERT(env != NULL);
449 fo = ofd_object_find(env, ofd, fid);
454 ofd_read_lock(env, fo);
455 if (!ofd_object_exists(fo))
456 GOTO(unlock, rc = -ENOENT);
458 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
459 rc = ofd_verify_ff(env, fo, oa);
465 for (i = 0, j = 0; i < niocount; i++) {
466 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
467 lnb + j, 0, ofd_object_capa(env, fo));
468 if (unlikely(rc < 0))
470 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
471 /* correct index for local buffers to continue with */
474 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
475 tot_bytes += rnb[i].rnb_len;
478 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
479 rc = dt_attr_get(env, ofd_object_child(fo), la,
480 ofd_object_capa(env, fo));
484 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
488 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
492 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
494 ofd_read_unlock(env, fo);
495 ofd_object_put(env, fo);
500 * Prepare buffers for write request processing.
502 * This function converts remote buffers from client to local buffers
503 * and prepares the latter. If there is recovery in progress and required
504 * object is missing then it can be re-created before write.
506 * \param[in] env execution environment
507 * \param[in] exp OBD export of client
508 * \param[in] ofd OFD device
509 * \param[in] fid FID of object
510 * \param[in] la object attributes
511 * \param[in] oa OBDO structure from client
512 * \param[in] objcount always 1
513 * \param[in] obj object data
514 * \param[in] rnb remote buffers
515 * \param[in] nr_local number of local buffers
516 * \param[in] lnb local buffers
517 * \param[in] jobid job ID name
519 * \retval 0 on successful prepare
520 * \retval negative value on error
522 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
523 struct ofd_device *ofd, const struct lu_fid *fid,
524 struct lu_attr *la, struct obdo *oa,
525 int objcount, struct obd_ioobj *obj,
526 struct niobuf_remote *rnb, int *nr_local,
527 struct niobuf_local *lnb, char *jobid)
529 struct ofd_object *fo;
530 int i, j, k, rc = 0, tot_bytes = 0;
533 LASSERT(env != NULL);
534 LASSERT(objcount == 1);
536 if (unlikely(exp->exp_obd->obd_recovering)) {
537 u64 seq = fid_seq(fid);
538 u64 oid = fid_oid(fid);
539 struct ofd_seq *oseq;
541 oseq = ofd_seq_load(env, ofd, seq);
543 CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
544 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
545 GOTO(out, rc = -EINVAL);
548 if (oid > ofd_seq_last_oid(oseq)) {
552 mutex_lock(&oseq->os_create_lock);
553 diff = oid - ofd_seq_last_oid(oseq);
555 /* Do sync create if the seq is about to used up */
556 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
557 if (unlikely(oid >= IDIF_MAX_OID - 1))
559 } else if (fid_seq_is_norm(seq)) {
561 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
564 CERROR("%s : invalid o_seq "DOSTID"\n",
565 ofd_name(ofd), POSTID(&oa->o_oi));
566 mutex_unlock(&oseq->os_create_lock);
567 ofd_seq_put(env, oseq);
568 GOTO(out, rc = -EINVAL);
572 u64 next_id = ofd_seq_last_oid(oseq) + 1;
573 int count = ofd_precreate_batch(ofd, diff);
575 rc = ofd_precreate_objects(env, ofd, next_id,
578 mutex_unlock(&oseq->os_create_lock);
579 ofd_seq_put(env, oseq);
586 mutex_unlock(&oseq->os_create_lock);
589 ofd_seq_put(env, oseq);
592 fo = ofd_object_find(env, ofd, fid);
594 GOTO(out, rc = PTR_ERR(fo));
597 ofd_read_lock(env, fo);
598 if (!ofd_object_exists(fo)) {
599 CERROR("%s: BRW to missing obj "DOSTID"\n",
600 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
601 ofd_read_unlock(env, fo);
602 ofd_object_put(env, fo);
603 GOTO(out, rc = -ENOENT);
606 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
607 rc = ofd_verify_ff(env, fo, oa);
609 ofd_read_unlock(env, fo);
610 ofd_object_put(env, fo);
615 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
616 * space back if possible */
617 ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
619 /* parse remote buffers to local buffers and prepare the latter */
621 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
622 rc = dt_bufs_get(env, ofd_object_child(fo),
624 ofd_object_capa(env, fo));
625 if (unlikely(rc < 0))
627 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
628 /* correct index for local buffers to continue with */
629 for (k = 0; k < rc; k++) {
630 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
631 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
632 lnb[j+k].lnb_rc = -ENOSPC;
634 /* remote client can't break through quota */
635 if (exp_connect_rmtclient(exp))
636 lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
640 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
641 tot_bytes += rnb[i].rnb_len;
643 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
645 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
646 if (unlikely(rc != 0))
649 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
652 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
653 ofd_read_unlock(env, fo);
654 /* ofd_grant_prepare_write() was called, so we must commit */
655 ofd_grant_commit(env, exp, rc);
657 /* let's still process incoming grant information packed in the oa,
658 * but without enforcing grant since we won't proceed with the write.
659 * Just like a read request actually. */
660 ofd_grant_prepare_read(env, exp, oa);
665 * Prepare bulk IO requests for processing.
667 * This function does initial checks of IO and calls corresponding
668 * functions for read/write processing.
670 * \param[in] env execution environment
671 * \param[in] cmd IO type (read/write)
672 * \param[in] exp OBD export of client
673 * \param[in] oa OBDO structure from request
674 * \param[in] objcount always 1
675 * \param[in] obj object data
676 * \param[in] rnb remote buffers
677 * \param[in] nr_local number of local buffers
678 * \param[in] lnb local buffers
679 * \param[in] oti request data from OST
680 * \param[in] capa capability
682 * \retval 0 on successful prepare
683 * \retval negative value on error
685 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
686 struct obdo *oa, int objcount, struct obd_ioobj *obj,
687 struct niobuf_remote *rnb, int *nr_local,
688 struct niobuf_local *lnb, struct obd_trans_info *oti,
689 struct lustre_capa *capa)
691 struct tgt_session_info *tsi = tgt_ses_info(env);
692 struct ofd_device *ofd = ofd_exp(exp);
693 struct ofd_thread_info *info;
695 const struct lu_fid *fid = &oa->o_oi.oi_fid;
698 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
699 CERROR("%s: bulk has too many pages %d, which exceeds the"
700 "maximum pages per RPC of %d\n",
701 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
705 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
706 LASSERT(oti != NULL);
707 info = ofd_info_init(env, exp);
708 ofd_oti2info(info, oti);
711 info = tsi2ofd_info(tsi);
712 jobid = tsi->tsi_jobid;
717 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
718 struct ofd_seq *oseq;
720 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
722 CERROR("%s: Can not find seq for "DOSTID
723 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
728 if (oseq->os_destroys_in_progress == 0) {
729 /* don't fail lookups for orphan recovery, it causes
730 * later LBUGs when objects still exist during
732 ofd_seq_put(env, oseq);
735 ofd_seq_put(env, oseq);
738 LASSERT(objcount == 1);
739 LASSERT(obj->ioo_bufcnt > 0);
741 if (cmd == OBD_BRW_WRITE) {
742 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
743 capa, CAPA_OPC_OSS_WRITE);
745 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
746 rc = ofd_preprw_write(env, exp, ofd, fid,
747 &info->fti_attr, oa, objcount,
748 obj, rnb, nr_local, lnb, jobid);
750 } else if (cmd == OBD_BRW_READ) {
751 rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
752 capa, CAPA_OPC_OSS_READ);
754 ofd_grant_prepare_read(env, exp, oa);
755 rc = ofd_preprw_read(env, exp, ofd, fid,
757 obj->ioo_bufcnt, rnb, nr_local,
759 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
762 CERROR("%s: wrong cmd %d received!\n",
763 exp->exp_obd->obd_name, cmd);
770 * Drop reference on local buffers for read bulk IO.
772 * This will free all local buffers use by this read request.
774 * \param[in] env execution environment
775 * \param[in] ofd OFD device
776 * \param[in] fid object FID
777 * \param[in] objcount always 1
778 * \param[in] niocount number of local buffers
779 * \param[in] lnb local buffers
781 * \retval 0 on successful execution
782 * \retval negative value on error
785 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
786 const struct lu_fid *fid, int objcount, int niocount,
787 struct niobuf_local *lnb)
789 struct ofd_object *fo;
793 LASSERT(niocount > 0);
795 fo = ofd_object_find(env, ofd, fid);
799 LASSERT(ofd_object_exists(fo));
800 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
802 ofd_read_unlock(env, fo);
803 ofd_object_put(env, fo);
804 /* second put is pair to object_get in ofd_preprw_read */
805 ofd_object_put(env, fo);
811 * Set attributes of object during write bulk IO processing.
813 * Change object attributes and write parent FID into extended
814 * attributes when needed.
816 * \param[in] env execution environment
817 * \param[in] ofd OFD device
818 * \param[in] ofd_obj OFD object
819 * \param[in] la object attributes
820 * \param[in] ff parent FID
822 * \retval 0 on successful attributes update
823 * \retval negative value on error
826 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
827 struct ofd_object *ofd_obj, struct lu_attr *la,
828 struct filter_fid *ff)
830 struct ofd_thread_info *info = ofd_info(env);
831 __u64 valid = la->la_valid;
834 struct dt_object *dt_obj;
841 dt_obj = ofd_object_child(ofd_obj);
842 LASSERT(dt_obj != NULL);
844 la->la_valid &= LA_UID | LA_GID;
846 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
851 rc = ofd_object_ff_load(env, ofd_obj);
858 if (!la->la_valid && !ff_needed)
859 /* no attributes to set */
862 th = ofd_trans_create(env, ofd);
864 GOTO(out, rc = PTR_ERR(th));
867 rc = dt_declare_attr_set(env, dt_obj, la, th);
873 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
874 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
875 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
876 ff->ff_parent.f_oid =
877 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
879 info->fti_buf.lb_buf = ff;
880 info->fti_buf.lb_len = sizeof(*ff);
881 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
882 XATTR_NAME_FID, 0, th);
887 /* We don't need a transno for this operation which will be re-executed
888 * anyway when the OST_WRITE (with a transno assigned) is replayed */
889 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
895 rc = dt_attr_set(env, dt_obj, la, th,
896 ofd_object_capa(env, ofd_obj));
901 /* set filter fid EA */
903 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
906 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
909 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
910 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
911 /* Currently, the filter_fid::ff_parent::f_ver is not
912 * the real parent MDT-object's FID::f_ver, instead it
913 * is the OST-object index in its parent MDT-object's
915 ofd_obj->ofo_pfid.f_stripe_idx =
916 le32_to_cpu(ff->ff_parent.f_stripe_idx);
923 dt_trans_stop(env, ofd->ofd_osd, th);
925 la->la_valid = valid;
929 struct ofd_soft_sync_callback {
930 struct dt_txn_commit_cb ossc_cb;
931 struct obd_export *ossc_exp;
935 * Callback function for "soft sync" update.
937 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
938 * See ofd_soft_sync_cb_add() below for more details on soft sync.
940 * \param[in] env execution environment
941 * \param[in] th transaction handle
942 * \param[in] cb callback data
943 * \param[in] err error code
945 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
946 struct dt_txn_commit_cb *cb, int err)
948 struct ofd_soft_sync_callback *ossc;
950 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
952 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
953 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
955 class_export_cb_put(ossc->ossc_exp);
960 * Add callback for "soft sync" processing.
962 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
963 * flag is set in client buffers. The intention is for this operation to
964 * commit pages belonging to a client which has "too many" outstanding
965 * unstable pages in its cache. See LU-2139 for details.
967 * This function adds callback to be called when commit is done.
969 * \param[in] th transaction handle
970 * \param[in] exp OBD export of client
972 * \retval 0 on successful callback adding
973 * \retval negative value on error
975 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
977 struct ofd_soft_sync_callback *ossc;
978 struct dt_txn_commit_cb *dcb;
985 ossc->ossc_exp = class_export_cb_get(exp);
987 dcb = &ossc->ossc_cb;
988 dcb->dcb_func = ofd_cb_soft_sync;
989 INIT_LIST_HEAD(&dcb->dcb_linkage);
990 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
992 rc = dt_trans_cb_add(th, dcb);
994 class_export_cb_put(exp);
1002 * Commit bulk IO buffers to the storage.
1004 * This function finalizes write IO processing by writing data to the disk.
1005 * That write can be synchronous or asynchronous depending on buffers flags.
1007 * \param[in] env execution environment
1008 * \param[in] exp OBD export of client
1009 * \param[in] ofd OFD device
1010 * \param[in] fid FID of object
1011 * \param[in] la object attributes
1012 * \param[in] ff parent FID of object
1013 * \param[in] objcount always 1
1014 * \param[in] niocount number of local buffers
1015 * \param[in] lnb local buffers
1016 * \param[in] old_rc result of processing at this point
1018 * \retval 0 on successful commit
1019 * \retval negative value on error
1022 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1023 struct ofd_device *ofd, const struct lu_fid *fid,
1024 struct lu_attr *la, struct filter_fid *ff, int objcount,
1025 int niocount, struct niobuf_local *lnb, int old_rc)
1027 struct ofd_thread_info *info = ofd_info(env);
1028 struct ofd_object *fo;
1029 struct dt_object *o;
1034 struct filter_export_data *fed = &exp->exp_filter_data;
1035 bool soft_sync = false;
1036 bool cb_registered = false;
1040 LASSERT(objcount == 1);
1042 fo = ofd_object_find(env, ofd, fid);
1043 LASSERT(fo != NULL);
1044 LASSERT(ofd_object_exists(fo));
1046 o = ofd_object_child(fo);
1050 GOTO(out, rc = old_rc);
1053 * The first write to each object must set some attributes. It is
1054 * important to set the uid/gid before calling
1055 * dt_declare_write_commit() since quota enforcement is now handled in
1058 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
1062 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1065 th = ofd_trans_create(env, ofd);
1067 GOTO(out, rc = PTR_ERR(th));
1069 th->th_sync |= ofd->ofd_syncjournal;
1070 if (th->th_sync == 0) {
1071 for (i = 0; i < niocount; i++) {
1072 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1076 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1081 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1082 GOTO(out_stop, rc = -EINPROGRESS);
1084 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1089 /* update [mac]time if needed */
1090 rc = dt_declare_attr_set(env, o, la, th);
1095 rc = ofd_trans_start(env, ofd, fo, th);
1099 rc = dt_write_commit(env, o, lnb, niocount, th);
1104 rc = dt_attr_set(env, o, la, th, ofd_object_capa(env, fo));
1109 /* get attr to return */
1110 rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo));
1113 /* Force commit to make the just-deleted blocks
1114 * reusable. LU-456 */
1118 /* do this before trans stop in case commit has finished */
1119 if (!th->th_sync && soft_sync && !cb_registered) {
1120 ofd_soft_sync_cb_add(th, exp);
1121 cb_registered = true;
1124 ofd_trans_stop(env, ofd, th, rc);
1125 if (rc == -ENOSPC && retries++ < 3) {
1126 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1132 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1133 atomic_set(&fed->fed_soft_sync_count, 0);
1134 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1135 ofd->ofd_soft_sync_limit)
1136 dt_commit_async(env, ofd->ofd_osd);
1139 dt_bufs_put(env, o, lnb, niocount);
1140 ofd_read_unlock(env, fo);
1141 ofd_object_put(env, fo);
1142 /* second put is pair to object_get in ofd_preprw_write */
1143 ofd_object_put(env, fo);
1144 ofd_grant_commit(env, info->fti_exp, old_rc);
1149 * Commit bulk IO to the storage.
1151 * This is companion function to the ofd_preprw(). It finishes bulk IO
1152 * request processing by committing buffers to the storage (WRITE) and/or
1153 * freeing those buffers (read/write). See ofd_commitrw_read() and
1154 * ofd_commitrw_write() for details about each type of IO.
1156 * \param[in] env execution environment
1157 * \param[in] cmd IO type (READ/WRITE)
1158 * \param[in] exp OBD export of client
1159 * \param[in] oa OBDO structure from client
1160 * \param[in] objcount always 1
1161 * \param[in] obj object data
1162 * \param[in] rnb remote buffers
1163 * \param[in] npages number of local buffers
1164 * \param[in] lnb local buffers
1165 * \param[in] oti request data from OST
1166 * \param[in] old_rc result of processing at this point
1168 * \retval 0 on successful commit
1169 * \retval negative value on error
1171 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1172 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1173 struct niobuf_remote *rnb, int npages,
1174 struct niobuf_local *lnb, struct obd_trans_info *oti,
1177 struct ofd_thread_info *info = ofd_info(env);
1178 struct ofd_mod_data *fmd;
1180 struct ofd_device *ofd = ofd_exp(exp);
1181 struct filter_fid *ff = NULL;
1182 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1185 LASSERT(npages > 0);
1187 if (cmd == OBD_BRW_WRITE) {
1188 /* Don't update timestamps if this write is older than a
1189 * setattr which modifies the timestamps. b=10150 */
1191 /* XXX when we start having persistent reservations this needs
1192 * to be changed to ofd_fmd_get() to create the fmd if it
1193 * doesn't already exist so we can store the reservation handle
1195 valid = OBD_MD_FLUID | OBD_MD_FLGID;
1196 fmd = ofd_fmd_find(exp, fid);
1197 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1198 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1200 ofd_fmd_put(exp, fmd);
1201 la_from_obdo(&info->fti_attr, oa, valid);
1203 if (oa->o_valid & OBD_MD_FLFID) {
1204 ff = &info->fti_mds_fid;
1205 ofd_prepare_fidea(ff, oa);
1208 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1209 ff, objcount, npages, lnb, old_rc);
1211 obdo_from_la(oa, &info->fti_attr,
1212 OFD_VALID_FLAGS | LA_GID | LA_UID);
1214 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
1216 /* don't report overquota flag if we failed before reaching
1218 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1219 /* return the overquota flags to client */
1220 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1221 if (oa->o_valid & OBD_MD_FLFLAGS)
1222 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1224 oa->o_flags = OBD_FL_NO_USRQUOTA;
1227 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1228 if (oa->o_valid & OBD_MD_FLFLAGS)
1229 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1231 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1234 oa->o_valid |= OBD_MD_FLFLAGS;
1235 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1237 } else if (cmd == OBD_BRW_READ) {
1238 struct ldlm_namespace *ns = ofd->ofd_namespace;
1240 /* If oa != NULL then ofd_preprw_read updated the inode
1241 * atime and we should update the lvb so that other glimpses
1242 * will also get the updated value. bug 5972 */
1243 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1244 struct ldlm_resource *rs = NULL;
1246 ost_fid_build_resid(fid, &info->fti_resid);
1247 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1250 ldlm_res_lvbo_update(rs, NULL, 1);
1251 ldlm_resource_putref(rs);
1254 rc = ofd_commitrw_read(env, ofd, fid, objcount,
1264 ofd_info2oti(info, oti);