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 <linux/kthread.h>
44 #include "ofd_internal.h"
46 struct ofd_inconsistency_item {
47 struct list_head oii_list;
48 struct ofd_object *oii_obj;
49 struct lu_fid oii_pfid;
53 * Verify single object for parent FID consistency.
55 * Part of LFSCK processing which checks single object PFID stored in extended
56 * attribute (XATTR) against real FID of MDT parent object received by LFSCK.
57 * This verifies that the OST object is being referenced by only a single MDT
60 * \param[in] env execution environment
61 * \param[in] ofd OFD device
62 * \param[in] oii object-related local data
63 * \param[in] lr LFSCK request data
65 static void ofd_inconsistency_verify_one(const struct lu_env *env,
66 struct ofd_device *ofd,
67 struct ofd_inconsistency_item *oii,
68 struct lfsck_request *lr)
70 struct ofd_object *fo = oii->oii_obj;
71 struct lu_fid *pfid = &fo->ofo_pfid;
74 LASSERT(fo->ofo_pfid_checking);
75 LASSERT(!fo->ofo_pfid_verified);
77 lr->lr_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
78 lr->lr_fid2 = oii->oii_pfid; /* client given PFID. */
79 lr->lr_fid3 = *pfid; /* OST local stored PFID. */
81 rc = lfsck_in_notify(env, ofd->ofd_osd, lr, NULL);
82 ofd_write_lock(env, fo);
83 switch (lr->lr_status) {
88 CDEBUG(D_LFSCK, "%s: fail to verify OST local stored "
89 "PFID xattr for "DFID", the client given PFID "
90 DFID", OST local stored PFID "DFID": rc = %d\n",
91 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
92 PFID(&oii->oii_pfid), PFID(pfid), rc);
94 fo->ofo_pfid_verified = 1;
96 case LPVS_INCONSISTENT:
99 ofd->ofd_inconsistency_self_detected++;
101 CDEBUG(D_LFSCK, "%s: fail to verify the client given "
102 "PFID for "DFID", the client given PFID "DFID
103 ", local stored PFID "DFID": rc = %d\n",
104 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
105 PFID(&oii->oii_pfid), PFID(pfid), rc);
107 CDEBUG(D_LFSCK, "%s: both the client given PFID and "
108 "the OST local stored PFID are stale for the "
109 "OST-object "DFID", client given PFID is "DFID
110 ", local stored PFID is "DFID"\n",
111 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
112 PFID(&oii->oii_pfid), PFID(pfid));
114 case LPVS_INCONSISTENT_TOFIX:
115 ofd->ofd_inconsistency_self_detected++;
117 ofd->ofd_inconsistency_self_repaired++;
118 CDEBUG(D_LFSCK, "%s: fixed the staled OST PFID xattr "
119 "for "DFID", with the client given PFID "DFID
120 ", the old stored PFID "DFID"\n",
121 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
122 PFID(&oii->oii_pfid), PFID(pfid));
124 CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
125 "for "DFID", client given PFID "DFID", local "
126 "stored PFID "DFID": rc = %d\n",
127 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
128 PFID(&oii->oii_pfid), PFID(pfid), rc);
130 *pfid = oii->oii_pfid;
131 fo->ofo_pfid_verified = 1;
136 fo->ofo_pfid_checking = 0;
137 ofd_write_unlock(env, fo);
139 lu_object_put(env, &fo->ofo_obj.do_lu);
144 * Verification thread to check parent FID consistency.
146 * Kernel thread to check consistency of parent FID for any
147 * new item added for checking by ofd_add_inconsistency_item().
149 * \param[in] args OFD device
151 * \retval 0 on successful thread termination
152 * \retval negative value if thread can't start
154 static int ofd_inconsistency_verification_main(void *args)
157 struct ofd_device *ofd = args;
158 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
159 struct ofd_inconsistency_item *oii;
160 struct lfsck_request *lr = NULL;
161 struct l_wait_info lwi = { 0 };
165 rc = lu_env_init(&env, LCT_DT_THREAD);
166 spin_lock(&ofd->ofd_inconsistency_lock);
167 thread_set_flags(thread, rc != 0 ? SVC_STOPPED : SVC_RUNNING);
168 wake_up_all(&thread->t_ctl_waitq);
169 spin_unlock(&ofd->ofd_inconsistency_lock);
174 if (unlikely(lr == NULL))
175 GOTO(out, rc = -ENOMEM);
177 lr->lr_event = LE_PAIRS_VERIFY;
178 lr->lr_active = LFSCK_TYPE_LAYOUT;
180 spin_lock(&ofd->ofd_inconsistency_lock);
182 if (unlikely(!thread_is_running(thread)))
185 while (!list_empty(&ofd->ofd_inconsistency_list)) {
186 oii = list_entry(ofd->ofd_inconsistency_list.next,
187 struct ofd_inconsistency_item,
189 list_del_init(&oii->oii_list);
190 spin_unlock(&ofd->ofd_inconsistency_lock);
191 ofd_inconsistency_verify_one(&env, ofd, oii, lr);
192 spin_lock(&ofd->ofd_inconsistency_lock);
195 spin_unlock(&ofd->ofd_inconsistency_lock);
196 l_wait_event(thread->t_ctl_waitq,
197 !list_empty(&ofd->ofd_inconsistency_list) ||
198 !thread_is_running(thread),
200 spin_lock(&ofd->ofd_inconsistency_lock);
203 while (!list_empty(&ofd->ofd_inconsistency_list)) {
204 struct ofd_object *fo;
206 oii = list_entry(ofd->ofd_inconsistency_list.next,
207 struct ofd_inconsistency_item,
209 list_del_init(&oii->oii_list);
211 spin_unlock(&ofd->ofd_inconsistency_lock);
213 ofd_write_lock(&env, fo);
214 fo->ofo_pfid_checking = 0;
215 ofd_write_unlock(&env, fo);
217 lu_object_put(&env, &fo->ofo_obj.do_lu);
219 spin_lock(&ofd->ofd_inconsistency_lock);
227 thread_set_flags(thread, SVC_STOPPED);
228 wake_up_all(&thread->t_ctl_waitq);
229 spin_unlock(&ofd->ofd_inconsistency_lock);
236 * Start parent FID verification thread.
238 * See ofd_inconsistency_verification_main().
240 * \param[in] ofd OFD device
242 * \retval 0 on successful start of thread
243 * \retval negative value on error
245 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
247 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
248 struct l_wait_info lwi = { 0 };
249 struct task_struct *task;
252 spin_lock(&ofd->ofd_inconsistency_lock);
253 if (unlikely(thread_is_running(thread))) {
254 spin_unlock(&ofd->ofd_inconsistency_lock);
259 thread_set_flags(thread, 0);
260 spin_unlock(&ofd->ofd_inconsistency_lock);
261 task = kthread_run(ofd_inconsistency_verification_main, ofd,
262 "inconsistency_verification");
265 CERROR("%s: cannot start self_repair thread: rc = %d\n",
269 l_wait_event(thread->t_ctl_waitq,
270 thread_is_running(thread) ||
271 thread_is_stopped(thread),
279 * Stop parent FID verification thread.
281 * \param[in] ofd OFD device
283 * \retval 0 on successful start of thread
284 * \retval -EALREADY if thread is already stopped
286 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
288 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
289 struct l_wait_info lwi = { 0 };
291 spin_lock(&ofd->ofd_inconsistency_lock);
292 if (thread_is_init(thread) || thread_is_stopped(thread)) {
293 spin_unlock(&ofd->ofd_inconsistency_lock);
298 thread_set_flags(thread, SVC_STOPPING);
299 spin_unlock(&ofd->ofd_inconsistency_lock);
300 wake_up_all(&thread->t_ctl_waitq);
301 l_wait_event(thread->t_ctl_waitq,
302 thread_is_stopped(thread),
309 * Add new item for parent FID verification.
311 * Prepare new verification item and pass it to the dedicated
312 * verification thread for further processing.
314 * \param[in] env execution environment
315 * \param[in] fo OFD object
316 * \param[in] oa OBDO structure with PFID
318 static void ofd_add_inconsistency_item(const struct lu_env *env,
319 struct ofd_object *fo, struct obdo *oa)
321 struct ofd_device *ofd = ofd_obj2dev(fo);
322 struct ofd_inconsistency_item *oii;
329 INIT_LIST_HEAD(&oii->oii_list);
330 lu_object_get(&fo->ofo_obj.do_lu);
332 oii->oii_pfid.f_seq = oa->o_parent_seq;
333 oii->oii_pfid.f_oid = oa->o_parent_oid;
334 oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
336 spin_lock(&ofd->ofd_inconsistency_lock);
337 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
338 spin_unlock(&ofd->ofd_inconsistency_lock);
344 fo->ofo_pfid_checking = 1;
345 if (list_empty(&ofd->ofd_inconsistency_list))
347 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
348 spin_unlock(&ofd->ofd_inconsistency_lock);
350 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
352 /* XXX: When the found inconsistency exceeds some threshold,
353 * we can trigger the LFSCK to scan part of the system
354 * or the whole system, which depends on how to define
355 * the threshold, a simple way maybe like that: define
356 * the absolute value of how many inconsisteny allowed
357 * to be repaired via self detect/repair mechanism, if
358 * exceeded, then trigger the LFSCK to scan the layout
359 * inconsistency within the whole system. */
363 * Verify parent FID of an object.
365 * Check the parent FID is sane and start extended
366 * verification procedure otherwise.
368 * \param[in] env execution environment
369 * \param[in] fo OFD object
370 * \param[in] oa OBDO structure with PFID
372 * \retval 0 on successful verification
373 * \retval -EINPROGRESS if PFID is being repaired
374 * \retval -EPERM if PFID was verified but still insane
376 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
379 struct lu_fid *pfid = &fo->ofo_pfid;
383 if (fid_is_sane(pfid)) {
384 if (likely(oa->o_parent_seq == pfid->f_seq &&
385 oa->o_parent_oid == pfid->f_oid &&
386 oa->o_stripe_idx == pfid->f_stripe_idx))
389 if (fo->ofo_pfid_verified)
393 /* The OST-object may be inconsistent, and we need further verification.
394 * To avoid block the RPC service thread, return -EINPROGRESS to client
395 * and make it retry later. */
396 if (fo->ofo_pfid_checking)
397 RETURN(-EINPROGRESS);
399 rc = ofd_object_ff_load(env, fo);
406 if (likely(oa->o_parent_seq == pfid->f_seq &&
407 oa->o_parent_oid == pfid->f_oid &&
408 oa->o_stripe_idx == pfid->f_stripe_idx))
411 /* Push it to the dedicated thread for further verification. */
412 ofd_add_inconsistency_item(env, fo, oa);
414 RETURN(-EINPROGRESS);
418 * Prepare buffers for read request processing.
420 * This function converts remote buffers from client to local buffers
421 * and prepares the latter.
423 * \param[in] env execution environment
424 * \param[in] exp OBD export of client
425 * \param[in] ofd OFD device
426 * \param[in] fid FID of object
427 * \param[in] la object attributes
428 * \param[in] oa OBDO structure from client
429 * \param[in] niocount number of remote buffers
430 * \param[in] rnb remote buffers
431 * \param[in] nr_local number of local buffers
432 * \param[in] lnb local buffers
433 * \param[in] jobid job ID name
435 * \retval 0 on successful prepare
436 * \retval negative value on error
438 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
439 struct ofd_device *ofd, const struct lu_fid *fid,
440 struct lu_attr *la, struct obdo *oa, int niocount,
441 struct niobuf_remote *rnb, int *nr_local,
442 struct niobuf_local *lnb, char *jobid)
444 struct ofd_object *fo;
445 int i, j, rc, tot_bytes = 0;
448 LASSERT(env != NULL);
450 fo = ofd_object_find(env, ofd, fid);
455 ofd_read_lock(env, fo);
456 if (!ofd_object_exists(fo))
457 GOTO(unlock, rc = -ENOENT);
459 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
460 rc = ofd_verify_ff(env, fo, oa);
466 for (i = 0, j = 0; i < niocount; i++) {
467 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
469 if (unlikely(rc < 0))
471 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
472 /* correct index for local buffers to continue with */
475 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
476 tot_bytes += rnb[i].rnb_len;
479 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
480 rc = dt_attr_get(env, ofd_object_child(fo), la);
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),
623 rnb + i, lnb + j, 1);
624 if (unlikely(rc < 0))
626 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
627 /* correct index for local buffers to continue with */
628 for (k = 0; k < rc; k++) {
629 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
630 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
631 lnb[j+k].lnb_rc = -ENOSPC;
633 /* remote client can't break through quota */
634 if (exp_connect_rmtclient(exp))
635 lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
639 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
640 tot_bytes += rnb[i].rnb_len;
642 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
644 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
645 if (unlikely(rc != 0))
648 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
651 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
652 ofd_read_unlock(env, fo);
653 ofd_object_put(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
681 * \retval 0 on successful prepare
682 * \retval negative value on error
684 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
685 struct obdo *oa, int objcount, struct obd_ioobj *obj,
686 struct niobuf_remote *rnb, int *nr_local,
687 struct niobuf_local *lnb, struct obd_trans_info *oti)
689 struct tgt_session_info *tsi = tgt_ses_info(env);
690 struct ofd_device *ofd = ofd_exp(exp);
691 struct ofd_thread_info *info;
693 const struct lu_fid *fid = &oa->o_oi.oi_fid;
696 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
697 CERROR("%s: bulk has too many pages %d, which exceeds the"
698 "maximum pages per RPC of %d\n",
699 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
703 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
704 LASSERT(oti != NULL);
705 info = ofd_info_init(env, exp);
706 ofd_oti2info(info, oti);
709 info = tsi2ofd_info(tsi);
710 jobid = tsi->tsi_jobid;
715 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
716 struct ofd_seq *oseq;
718 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
720 CERROR("%s: Can not find seq for "DOSTID
721 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
726 if (oseq->os_destroys_in_progress == 0) {
727 /* don't fail lookups for orphan recovery, it causes
728 * later LBUGs when objects still exist during
730 ofd_seq_put(env, oseq);
733 ofd_seq_put(env, oseq);
736 LASSERT(objcount == 1);
737 LASSERT(obj->ioo_bufcnt > 0);
739 if (cmd == OBD_BRW_WRITE) {
740 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
741 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
742 objcount, obj, rnb, nr_local, lnb, jobid);
743 } else if (cmd == OBD_BRW_READ) {
744 ofd_grant_prepare_read(env, exp, oa);
745 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
746 obj->ioo_bufcnt, rnb, nr_local, lnb,
748 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
750 CERROR("%s: wrong cmd %d received!\n",
751 exp->exp_obd->obd_name, cmd);
758 * Drop reference on local buffers for read bulk IO.
760 * This will free all local buffers use by this read request.
762 * \param[in] env execution environment
763 * \param[in] ofd OFD device
764 * \param[in] fid object FID
765 * \param[in] objcount always 1
766 * \param[in] niocount number of local buffers
767 * \param[in] lnb local buffers
769 * \retval 0 on successful execution
770 * \retval negative value on error
773 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
774 const struct lu_fid *fid, int objcount, int niocount,
775 struct niobuf_local *lnb)
777 struct ofd_object *fo;
781 LASSERT(niocount > 0);
783 fo = ofd_object_find(env, ofd, fid);
787 LASSERT(ofd_object_exists(fo));
788 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
790 ofd_read_unlock(env, fo);
791 ofd_object_put(env, fo);
792 /* second put is pair to object_get in ofd_preprw_read */
793 ofd_object_put(env, fo);
799 * Set attributes of object during write bulk IO processing.
801 * Change object attributes and write parent FID into extended
802 * attributes when needed.
804 * \param[in] env execution environment
805 * \param[in] ofd OFD device
806 * \param[in] ofd_obj OFD object
807 * \param[in] la object attributes
808 * \param[in] ff parent FID
810 * \retval 0 on successful attributes update
811 * \retval negative value on error
814 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
815 struct ofd_object *ofd_obj, struct lu_attr *la,
816 struct filter_fid *ff)
818 struct ofd_thread_info *info = ofd_info(env);
819 __u64 valid = la->la_valid;
822 struct dt_object *dt_obj;
829 dt_obj = ofd_object_child(ofd_obj);
830 LASSERT(dt_obj != NULL);
832 la->la_valid &= LA_UID | LA_GID;
834 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
839 rc = ofd_object_ff_load(env, ofd_obj);
846 if (!la->la_valid && !ff_needed)
847 /* no attributes to set */
850 th = ofd_trans_create(env, ofd);
852 GOTO(out, rc = PTR_ERR(th));
855 rc = dt_declare_attr_set(env, dt_obj, la, th);
861 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
862 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
863 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
864 ff->ff_parent.f_oid =
865 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
867 info->fti_buf.lb_buf = ff;
868 info->fti_buf.lb_len = sizeof(*ff);
869 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
870 XATTR_NAME_FID, 0, th);
875 /* We don't need a transno for this operation which will be re-executed
876 * anyway when the OST_WRITE (with a transno assigned) is replayed */
877 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
883 rc = dt_attr_set(env, dt_obj, la, th);
888 /* set filter fid EA */
890 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
893 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
896 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
897 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
898 /* Currently, the filter_fid::ff_parent::f_ver is not
899 * the real parent MDT-object's FID::f_ver, instead it
900 * is the OST-object index in its parent MDT-object's
902 ofd_obj->ofo_pfid.f_stripe_idx =
903 le32_to_cpu(ff->ff_parent.f_stripe_idx);
910 dt_trans_stop(env, ofd->ofd_osd, th);
912 la->la_valid = valid;
916 struct ofd_soft_sync_callback {
917 struct dt_txn_commit_cb ossc_cb;
918 struct obd_export *ossc_exp;
922 * Callback function for "soft sync" update.
924 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
925 * See ofd_soft_sync_cb_add() below for more details on soft sync.
927 * \param[in] env execution environment
928 * \param[in] th transaction handle
929 * \param[in] cb callback data
930 * \param[in] err error code
932 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
933 struct dt_txn_commit_cb *cb, int err)
935 struct ofd_soft_sync_callback *ossc;
937 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
939 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
940 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
942 class_export_cb_put(ossc->ossc_exp);
947 * Add callback for "soft sync" processing.
949 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
950 * flag is set in client buffers. The intention is for this operation to
951 * commit pages belonging to a client which has "too many" outstanding
952 * unstable pages in its cache. See LU-2139 for details.
954 * This function adds callback to be called when commit is done.
956 * \param[in] th transaction handle
957 * \param[in] exp OBD export of client
959 * \retval 0 on successful callback adding
960 * \retval negative value on error
962 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
964 struct ofd_soft_sync_callback *ossc;
965 struct dt_txn_commit_cb *dcb;
972 ossc->ossc_exp = class_export_cb_get(exp);
974 dcb = &ossc->ossc_cb;
975 dcb->dcb_func = ofd_cb_soft_sync;
976 INIT_LIST_HEAD(&dcb->dcb_linkage);
977 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
979 rc = dt_trans_cb_add(th, dcb);
981 class_export_cb_put(exp);
989 * Commit bulk IO buffers to the storage.
991 * This function finalizes write IO processing by writing data to the disk.
992 * That write can be synchronous or asynchronous depending on buffers flags.
994 * \param[in] env execution environment
995 * \param[in] exp OBD export of client
996 * \param[in] ofd OFD device
997 * \param[in] fid FID of object
998 * \param[in] la object attributes
999 * \param[in] ff parent FID of object
1000 * \param[in] objcount always 1
1001 * \param[in] niocount number of local buffers
1002 * \param[in] lnb local buffers
1003 * \param[in] old_rc result of processing at this point
1005 * \retval 0 on successful commit
1006 * \retval negative value on error
1009 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1010 struct ofd_device *ofd, const struct lu_fid *fid,
1011 struct lu_attr *la, struct filter_fid *ff, int objcount,
1012 int niocount, struct niobuf_local *lnb, int old_rc)
1014 struct ofd_thread_info *info = ofd_info(env);
1015 struct ofd_object *fo;
1016 struct dt_object *o;
1021 struct filter_export_data *fed = &exp->exp_filter_data;
1022 bool soft_sync = false;
1023 bool cb_registered = false;
1027 LASSERT(objcount == 1);
1029 fo = ofd_object_find(env, ofd, fid);
1030 LASSERT(fo != NULL);
1031 LASSERT(ofd_object_exists(fo));
1033 o = ofd_object_child(fo);
1037 GOTO(out, rc = old_rc);
1040 * The first write to each object must set some attributes. It is
1041 * important to set the uid/gid before calling
1042 * dt_declare_write_commit() since quota enforcement is now handled in
1045 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
1049 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1052 th = ofd_trans_create(env, ofd);
1054 GOTO(out, rc = PTR_ERR(th));
1056 th->th_sync |= ofd->ofd_syncjournal;
1057 if (th->th_sync == 0) {
1058 for (i = 0; i < niocount; i++) {
1059 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1063 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1068 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1069 GOTO(out_stop, rc = -EINPROGRESS);
1071 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1076 /* update [mac]time if needed */
1077 rc = dt_declare_attr_set(env, o, la, th);
1082 rc = ofd_trans_start(env, ofd, fo, th);
1086 rc = dt_write_commit(env, o, lnb, niocount, th);
1091 rc = dt_attr_set(env, o, la, th);
1096 /* get attr to return */
1097 rc = dt_attr_get(env, o, la);
1100 /* Force commit to make the just-deleted blocks
1101 * reusable. LU-456 */
1105 /* do this before trans stop in case commit has finished */
1106 if (!th->th_sync && soft_sync && !cb_registered) {
1107 ofd_soft_sync_cb_add(th, exp);
1108 cb_registered = true;
1111 ofd_trans_stop(env, ofd, th, rc);
1112 if (rc == -ENOSPC && retries++ < 3) {
1113 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1119 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1120 atomic_set(&fed->fed_soft_sync_count, 0);
1121 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1122 ofd->ofd_soft_sync_limit)
1123 dt_commit_async(env, ofd->ofd_osd);
1126 dt_bufs_put(env, o, lnb, niocount);
1127 ofd_read_unlock(env, fo);
1128 ofd_object_put(env, fo);
1129 /* second put is pair to object_get in ofd_preprw_write */
1130 ofd_object_put(env, fo);
1131 ofd_grant_commit(env, info->fti_exp, old_rc);
1136 * Commit bulk IO to the storage.
1138 * This is companion function to the ofd_preprw(). It finishes bulk IO
1139 * request processing by committing buffers to the storage (WRITE) and/or
1140 * freeing those buffers (read/write). See ofd_commitrw_read() and
1141 * ofd_commitrw_write() for details about each type of IO.
1143 * \param[in] env execution environment
1144 * \param[in] cmd IO type (READ/WRITE)
1145 * \param[in] exp OBD export of client
1146 * \param[in] oa OBDO structure from client
1147 * \param[in] objcount always 1
1148 * \param[in] obj object data
1149 * \param[in] rnb remote buffers
1150 * \param[in] npages number of local buffers
1151 * \param[in] lnb local buffers
1152 * \param[in] oti request data from OST
1153 * \param[in] old_rc result of processing at this point
1155 * \retval 0 on successful commit
1156 * \retval negative value on error
1158 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1159 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1160 struct niobuf_remote *rnb, int npages,
1161 struct niobuf_local *lnb, struct obd_trans_info *oti,
1164 struct ofd_thread_info *info = ofd_info(env);
1165 struct ofd_mod_data *fmd;
1167 struct ofd_device *ofd = ofd_exp(exp);
1168 struct filter_fid *ff = NULL;
1169 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1172 LASSERT(npages > 0);
1174 if (cmd == OBD_BRW_WRITE) {
1175 /* Don't update timestamps if this write is older than a
1176 * setattr which modifies the timestamps. b=10150 */
1178 /* XXX when we start having persistent reservations this needs
1179 * to be changed to ofd_fmd_get() to create the fmd if it
1180 * doesn't already exist so we can store the reservation handle
1182 valid = OBD_MD_FLUID | OBD_MD_FLGID;
1183 fmd = ofd_fmd_find(exp, fid);
1184 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1185 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1187 ofd_fmd_put(exp, fmd);
1188 la_from_obdo(&info->fti_attr, oa, valid);
1190 if (oa->o_valid & OBD_MD_FLFID) {
1191 ff = &info->fti_mds_fid;
1192 ofd_prepare_fidea(ff, oa);
1195 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1196 ff, objcount, npages, lnb, old_rc);
1198 obdo_from_la(oa, &info->fti_attr,
1199 OFD_VALID_FLAGS | LA_GID | LA_UID);
1201 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
1203 /* don't report overquota flag if we failed before reaching
1205 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1206 /* return the overquota flags to client */
1207 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1208 if (oa->o_valid & OBD_MD_FLFLAGS)
1209 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1211 oa->o_flags = OBD_FL_NO_USRQUOTA;
1214 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1215 if (oa->o_valid & OBD_MD_FLFLAGS)
1216 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1218 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1221 oa->o_valid |= OBD_MD_FLFLAGS;
1222 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1224 } else if (cmd == OBD_BRW_READ) {
1225 struct ldlm_namespace *ns = ofd->ofd_namespace;
1227 /* If oa != NULL then ofd_preprw_read updated the inode
1228 * atime and we should update the lvb so that other glimpses
1229 * will also get the updated value. bug 5972 */
1230 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1231 struct ldlm_resource *rs = NULL;
1233 ost_fid_build_resid(fid, &info->fti_resid);
1234 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1237 ldlm_res_lvbo_update(rs, NULL, 1);
1238 ldlm_resource_putref(rs);
1241 rc = ofd_commitrw_read(env, ofd, fid, objcount,
1251 ofd_info2oti(info, oti);