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, 2015, 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_unlocked, 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 spin_lock(&ofd->ofd_inconsistency_lock);
229 thread_set_flags(thread, SVC_STOPPED);
230 wake_up_all(&thread->t_ctl_waitq);
231 spin_unlock(&ofd->ofd_inconsistency_lock);
238 * Start parent FID verification thread.
240 * See ofd_inconsistency_verification_main().
242 * \param[in] ofd OFD device
244 * \retval 0 on successful start of thread
245 * \retval negative value on error
247 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
249 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
250 struct l_wait_info lwi = { 0 };
251 struct task_struct *task;
254 spin_lock(&ofd->ofd_inconsistency_lock);
255 if (unlikely(thread_is_running(thread))) {
256 spin_unlock(&ofd->ofd_inconsistency_lock);
261 thread_set_flags(thread, 0);
262 spin_unlock(&ofd->ofd_inconsistency_lock);
263 task = kthread_run(ofd_inconsistency_verification_main, ofd,
264 "inconsistency_verification");
267 CERROR("%s: cannot start self_repair thread: rc = %d\n",
271 l_wait_event(thread->t_ctl_waitq,
272 thread_is_running(thread) ||
273 thread_is_stopped(thread),
281 * Stop parent FID verification thread.
283 * \param[in] ofd OFD device
285 * \retval 0 on successful start of thread
286 * \retval -EALREADY if thread is already stopped
288 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
290 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
291 struct l_wait_info lwi = { 0 };
293 spin_lock(&ofd->ofd_inconsistency_lock);
294 if (thread_is_init(thread) || thread_is_stopped(thread)) {
295 spin_unlock(&ofd->ofd_inconsistency_lock);
300 thread_set_flags(thread, SVC_STOPPING);
301 spin_unlock(&ofd->ofd_inconsistency_lock);
302 wake_up_all(&thread->t_ctl_waitq);
303 l_wait_event(thread->t_ctl_waitq,
304 thread_is_stopped(thread),
311 * Add new item for parent FID verification.
313 * Prepare new verification item and pass it to the dedicated
314 * verification thread for further processing.
316 * \param[in] env execution environment
317 * \param[in] fo OFD object
318 * \param[in] oa OBDO structure with PFID
320 static void ofd_add_inconsistency_item(const struct lu_env *env,
321 struct ofd_object *fo, struct obdo *oa)
323 struct ofd_device *ofd = ofd_obj2dev(fo);
324 struct ofd_inconsistency_item *oii;
331 INIT_LIST_HEAD(&oii->oii_list);
332 lu_object_get(&fo->ofo_obj.do_lu);
334 oii->oii_pfid.f_seq = oa->o_parent_seq;
335 oii->oii_pfid.f_oid = oa->o_parent_oid;
336 oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
338 spin_lock(&ofd->ofd_inconsistency_lock);
339 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
340 spin_unlock(&ofd->ofd_inconsistency_lock);
346 fo->ofo_pfid_checking = 1;
347 if (list_empty(&ofd->ofd_inconsistency_list))
349 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
350 spin_unlock(&ofd->ofd_inconsistency_lock);
352 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
354 /* XXX: When the found inconsistency exceeds some threshold,
355 * we can trigger the LFSCK to scan part of the system
356 * or the whole system, which depends on how to define
357 * the threshold, a simple way maybe like that: define
358 * the absolute value of how many inconsisteny allowed
359 * to be repaired via self detect/repair mechanism, if
360 * exceeded, then trigger the LFSCK to scan the layout
361 * inconsistency within the whole system. */
365 * Verify parent FID of an object.
367 * Check the parent FID is sane and start extended
368 * verification procedure otherwise.
370 * \param[in] env execution environment
371 * \param[in] fo OFD object
372 * \param[in] oa OBDO structure with PFID
374 * \retval 0 on successful verification
375 * \retval -EINPROGRESS if PFID is being repaired
376 * \retval -EPERM if PFID was verified but still insane
378 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
381 struct lu_fid *pfid = &fo->ofo_pfid;
385 if (fid_is_sane(pfid)) {
386 if (likely(oa->o_parent_seq == pfid->f_seq &&
387 oa->o_parent_oid == pfid->f_oid &&
388 oa->o_stripe_idx == pfid->f_stripe_idx))
391 if (fo->ofo_pfid_verified)
395 /* The OST-object may be inconsistent, and we need further verification.
396 * To avoid block the RPC service thread, return -EINPROGRESS to client
397 * and make it retry later. */
398 if (fo->ofo_pfid_checking)
399 RETURN(-EINPROGRESS);
401 rc = ofd_object_ff_load(env, fo);
408 if (likely(oa->o_parent_seq == pfid->f_seq &&
409 oa->o_parent_oid == pfid->f_oid &&
410 oa->o_stripe_idx == pfid->f_stripe_idx))
413 /* Push it to the dedicated thread for further verification. */
414 ofd_add_inconsistency_item(env, fo, oa);
416 RETURN(-EINPROGRESS);
420 * Prepare buffers for read request processing.
422 * This function converts remote buffers from client to local buffers
423 * and prepares the latter.
425 * \param[in] env execution environment
426 * \param[in] exp OBD export of client
427 * \param[in] ofd OFD device
428 * \param[in] fid FID of object
429 * \param[in] la object attributes
430 * \param[in] oa OBDO structure from client
431 * \param[in] niocount number of remote buffers
432 * \param[in] rnb remote buffers
433 * \param[in] nr_local number of local buffers
434 * \param[in] lnb local buffers
435 * \param[in] jobid job ID name
437 * \retval 0 on successful prepare
438 * \retval negative value on error
440 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
441 struct ofd_device *ofd, const struct lu_fid *fid,
442 struct lu_attr *la, struct obdo *oa, int niocount,
443 struct niobuf_remote *rnb, int *nr_local,
444 struct niobuf_local *lnb, char *jobid)
446 struct ofd_object *fo;
447 int i, j, rc, tot_bytes = 0;
450 LASSERT(env != NULL);
452 fo = ofd_object_find(env, ofd, fid);
457 ofd_read_lock(env, fo);
458 if (!ofd_object_exists(fo))
459 GOTO(unlock, rc = -ENOENT);
461 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
462 rc = ofd_verify_ff(env, fo, oa);
468 for (i = 0, j = 0; i < niocount; i++) {
469 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
471 if (unlikely(rc < 0))
473 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
474 /* correct index for local buffers to continue with */
477 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
478 tot_bytes += rnb[i].rnb_len;
481 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
482 rc = dt_attr_get(env, ofd_object_child(fo), la);
486 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
490 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
494 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
496 ofd_read_unlock(env, fo);
497 ofd_object_put(env, fo);
502 * Prepare buffers for write request processing.
504 * This function converts remote buffers from client to local buffers
505 * and prepares the latter. If there is recovery in progress and required
506 * object is missing then it can be re-created before write.
508 * \param[in] env execution environment
509 * \param[in] exp OBD export of client
510 * \param[in] ofd OFD device
511 * \param[in] fid FID of object
512 * \param[in] la object attributes
513 * \param[in] oa OBDO structure from client
514 * \param[in] objcount always 1
515 * \param[in] obj object data
516 * \param[in] rnb remote buffers
517 * \param[in] nr_local number of local buffers
518 * \param[in] lnb local buffers
519 * \param[in] jobid job ID name
521 * \retval 0 on successful prepare
522 * \retval negative value on error
524 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
525 struct ofd_device *ofd, const struct lu_fid *fid,
526 struct lu_attr *la, struct obdo *oa,
527 int objcount, struct obd_ioobj *obj,
528 struct niobuf_remote *rnb, int *nr_local,
529 struct niobuf_local *lnb, char *jobid)
531 struct ofd_object *fo;
532 int i, j, k, rc = 0, tot_bytes = 0;
535 LASSERT(env != NULL);
536 LASSERT(objcount == 1);
538 if (unlikely(exp->exp_obd->obd_recovering)) {
539 u64 seq = fid_seq(fid);
540 u64 oid = fid_oid(fid);
541 struct ofd_seq *oseq;
543 oseq = ofd_seq_load(env, ofd, seq);
545 CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
546 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
547 GOTO(out, rc = -EINVAL);
550 if (oid > ofd_seq_last_oid(oseq)) {
554 mutex_lock(&oseq->os_create_lock);
555 diff = oid - ofd_seq_last_oid(oseq);
557 /* Do sync create if the seq is about to used up */
558 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
559 if (unlikely(oid >= IDIF_MAX_OID - 1))
561 } else if (fid_seq_is_norm(seq)) {
563 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
566 CERROR("%s : invalid o_seq "DOSTID"\n",
567 ofd_name(ofd), POSTID(&oa->o_oi));
568 mutex_unlock(&oseq->os_create_lock);
569 ofd_seq_put(env, oseq);
570 GOTO(out, rc = -EINVAL);
574 u64 next_id = ofd_seq_last_oid(oseq) + 1;
575 int count = ofd_precreate_batch(ofd, diff);
577 rc = ofd_precreate_objects(env, ofd, next_id,
580 mutex_unlock(&oseq->os_create_lock);
581 ofd_seq_put(env, oseq);
588 mutex_unlock(&oseq->os_create_lock);
591 ofd_seq_put(env, oseq);
594 fo = ofd_object_find(env, ofd, fid);
596 GOTO(out, rc = PTR_ERR(fo));
599 ofd_read_lock(env, fo);
600 if (!ofd_object_exists(fo)) {
601 CERROR("%s: BRW to missing obj "DOSTID"\n",
602 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
603 ofd_read_unlock(env, fo);
604 ofd_object_put(env, fo);
605 GOTO(out, rc = -ENOENT);
608 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
609 rc = ofd_verify_ff(env, fo, oa);
611 ofd_read_unlock(env, fo);
612 ofd_object_put(env, fo);
617 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
618 * space back if possible */
619 ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
621 /* parse remote buffers to local buffers and prepare the latter */
623 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
624 rc = dt_bufs_get(env, ofd_object_child(fo),
625 rnb + i, lnb + j, 1);
626 if (unlikely(rc < 0))
628 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
629 /* correct index for local buffers to continue with */
630 for (k = 0; k < rc; k++) {
631 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
632 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
633 lnb[j+k].lnb_rc = -ENOSPC;
637 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
638 tot_bytes += rnb[i].rnb_len;
640 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
642 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
643 if (unlikely(rc != 0))
646 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
649 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
650 ofd_read_unlock(env, fo);
651 ofd_object_put(env, fo);
652 /* ofd_grant_prepare_write() was called, so we must commit */
653 ofd_grant_commit(exp, oa->o_grant_used, rc);
655 /* let's still process incoming grant information packed in the oa,
656 * but without enforcing grant since we won't proceed with the write.
657 * Just like a read request actually. */
658 ofd_grant_prepare_read(env, exp, oa);
663 * Prepare bulk IO requests for processing.
665 * This function does initial checks of IO and calls corresponding
666 * functions for read/write processing.
668 * \param[in] env execution environment
669 * \param[in] cmd IO type (read/write)
670 * \param[in] exp OBD export of client
671 * \param[in] oa OBDO structure from request
672 * \param[in] objcount always 1
673 * \param[in] obj object data
674 * \param[in] rnb remote buffers
675 * \param[in] nr_local number of local buffers
676 * \param[in] lnb local buffers
678 * \retval 0 on successful prepare
679 * \retval negative value on error
681 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
682 struct obdo *oa, int objcount, struct obd_ioobj *obj,
683 struct niobuf_remote *rnb, int *nr_local,
684 struct niobuf_local *lnb)
686 struct tgt_session_info *tsi = tgt_ses_info(env);
687 struct ofd_device *ofd = ofd_exp(exp);
688 struct ofd_thread_info *info;
690 const struct lu_fid *fid = &oa->o_oi.oi_fid;
693 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
694 CERROR("%s: bulk has too many pages %d, which exceeds the"
695 "maximum pages per RPC of %d\n",
696 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
700 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
701 info = ofd_info_init(env, exp);
704 info = tsi2ofd_info(tsi);
705 jobid = tsi->tsi_jobid;
710 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
711 struct ofd_seq *oseq;
713 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
715 CERROR("%s: Can not find seq for "DOSTID
716 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
721 if (oseq->os_destroys_in_progress == 0) {
722 /* don't fail lookups for orphan recovery, it causes
723 * later LBUGs when objects still exist during
725 ofd_seq_put(env, oseq);
728 ofd_seq_put(env, oseq);
731 LASSERT(objcount == 1);
732 LASSERT(obj->ioo_bufcnt > 0);
734 if (cmd == OBD_BRW_WRITE) {
735 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
736 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
737 objcount, obj, rnb, nr_local, lnb, jobid);
738 } else if (cmd == OBD_BRW_READ) {
739 ofd_grant_prepare_read(env, exp, oa);
740 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
741 obj->ioo_bufcnt, rnb, nr_local, lnb,
743 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
745 CERROR("%s: wrong cmd %d received!\n",
746 exp->exp_obd->obd_name, cmd);
753 * Drop reference on local buffers for read bulk IO.
755 * This will free all local buffers use by this read request.
757 * \param[in] env execution environment
758 * \param[in] ofd OFD device
759 * \param[in] fid object FID
760 * \param[in] objcount always 1
761 * \param[in] niocount number of local buffers
762 * \param[in] lnb local buffers
764 * \retval 0 on successful execution
765 * \retval negative value on error
768 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
769 const struct lu_fid *fid, int objcount, int niocount,
770 struct niobuf_local *lnb)
772 struct ofd_object *fo;
776 LASSERT(niocount > 0);
778 fo = ofd_object_find(env, ofd, fid);
782 LASSERT(ofd_object_exists(fo));
783 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
785 ofd_read_unlock(env, fo);
786 ofd_object_put(env, fo);
787 /* second put is pair to object_get in ofd_preprw_read */
788 ofd_object_put(env, fo);
794 * Set attributes of object during write bulk IO processing.
796 * Change object attributes and write parent FID into extended
797 * attributes when needed.
799 * \param[in] env execution environment
800 * \param[in] ofd OFD device
801 * \param[in] ofd_obj OFD object
802 * \param[in] la object attributes
803 * \param[in] ff parent FID
805 * \retval 0 on successful attributes update
806 * \retval negative value on error
809 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
810 struct ofd_object *ofd_obj, struct lu_attr *la,
811 struct filter_fid *ff)
813 struct ofd_thread_info *info = ofd_info(env);
814 __u64 valid = la->la_valid;
817 struct dt_object *dt_obj;
824 dt_obj = ofd_object_child(ofd_obj);
825 LASSERT(dt_obj != NULL);
827 la->la_valid &= LA_UID | LA_GID;
829 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
834 rc = ofd_object_ff_load(env, ofd_obj);
841 if (!la->la_valid && !ff_needed)
842 /* no attributes to set */
845 th = ofd_trans_create(env, ofd);
847 GOTO(out, rc = PTR_ERR(th));
850 rc = dt_declare_attr_set(env, dt_obj, la, th);
856 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
857 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
858 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
859 ff->ff_parent.f_oid =
860 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
862 info->fti_buf.lb_buf = ff;
863 info->fti_buf.lb_len = sizeof(*ff);
864 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
865 XATTR_NAME_FID, 0, th);
870 /* We don't need a transno for this operation which will be re-executed
871 * anyway when the OST_WRITE (with a transno assigned) is replayed */
872 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
878 rc = dt_attr_set(env, dt_obj, la, th);
883 /* set filter fid EA */
885 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
888 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
891 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
892 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
893 /* Currently, the filter_fid::ff_parent::f_ver is not
894 * the real parent MDT-object's FID::f_ver, instead it
895 * is the OST-object index in its parent MDT-object's
897 ofd_obj->ofo_pfid.f_stripe_idx =
898 le32_to_cpu(ff->ff_parent.f_stripe_idx);
905 dt_trans_stop(env, ofd->ofd_osd, th);
907 la->la_valid = valid;
911 struct ofd_soft_sync_callback {
912 struct dt_txn_commit_cb ossc_cb;
913 struct obd_export *ossc_exp;
917 * Callback function for "soft sync" update.
919 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
920 * See ofd_soft_sync_cb_add() below for more details on soft sync.
922 * \param[in] env execution environment
923 * \param[in] th transaction handle
924 * \param[in] cb callback data
925 * \param[in] err error code
927 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
928 struct dt_txn_commit_cb *cb, int err)
930 struct ofd_soft_sync_callback *ossc;
932 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
934 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
935 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
937 class_export_cb_put(ossc->ossc_exp);
942 * Add callback for "soft sync" processing.
944 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
945 * flag is set in client buffers. The intention is for this operation to
946 * commit pages belonging to a client which has "too many" outstanding
947 * unstable pages in its cache. See LU-2139 for details.
949 * This function adds callback to be called when commit is done.
951 * \param[in] th transaction handle
952 * \param[in] exp OBD export of client
954 * \retval 0 on successful callback adding
955 * \retval negative value on error
957 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
959 struct ofd_soft_sync_callback *ossc;
960 struct dt_txn_commit_cb *dcb;
967 ossc->ossc_exp = class_export_cb_get(exp);
969 dcb = &ossc->ossc_cb;
970 dcb->dcb_func = ofd_cb_soft_sync;
971 INIT_LIST_HEAD(&dcb->dcb_linkage);
972 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
974 rc = dt_trans_cb_add(th, dcb);
976 class_export_cb_put(exp);
984 * Commit bulk IO buffers to the storage.
986 * This function finalizes write IO processing by writing data to the disk.
987 * That write can be synchronous or asynchronous depending on buffers flags.
989 * \param[in] env execution environment
990 * \param[in] exp OBD export of client
991 * \param[in] ofd OFD device
992 * \param[in] fid FID of object
993 * \param[in] la object attributes
994 * \param[in] ff parent FID of object
995 * \param[in] objcount always 1
996 * \param[in] niocount number of local buffers
997 * \param[in] lnb local buffers
998 * \param[in] granted grant space consumed for the bulk I/O
999 * \param[in] old_rc result of processing at this point
1001 * \retval 0 on successful commit
1002 * \retval negative value on error
1005 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1006 struct ofd_device *ofd, const struct lu_fid *fid,
1007 struct lu_attr *la, struct filter_fid *ff, int objcount,
1008 int niocount, struct niobuf_local *lnb,
1009 unsigned long granted, int old_rc)
1011 struct filter_export_data *fed = &exp->exp_filter_data;
1012 struct ofd_object *fo;
1013 struct dt_object *o;
1018 bool soft_sync = false;
1019 bool cb_registered = false;
1020 bool fake_write = false;
1024 LASSERT(objcount == 1);
1026 fo = ofd_object_find(env, ofd, fid);
1027 LASSERT(fo != NULL);
1028 LASSERT(ofd_object_exists(fo));
1030 o = ofd_object_child(fo);
1034 GOTO(out, rc = old_rc);
1037 * The first write to each object must set some attributes. It is
1038 * important to set the uid/gid before calling
1039 * dt_declare_write_commit() since quota enforcement is now handled in
1042 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
1046 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1048 /* do fake write, to simulate the write case for performance testing */
1049 if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_WRITE)) {
1050 struct niobuf_local *last = &lnb[niocount - 1];
1051 __u64 file_size = last->lnb_file_offset + last->lnb_len;
1052 __u64 valid = la->la_valid;
1054 la->la_valid = LA_SIZE;
1056 rc = dt_attr_get(env, o, la);
1057 if (rc < 0 && rc != -ENOENT)
1060 if (file_size < la->la_size)
1061 file_size = la->la_size;
1063 /* dirty inode by setting file size */
1064 la->la_valid = valid | LA_SIZE;
1065 la->la_size = file_size;
1071 th = ofd_trans_create(env, ofd);
1073 GOTO(out, rc = PTR_ERR(th));
1075 th->th_sync |= ofd->ofd_syncjournal;
1076 if (th->th_sync == 0) {
1077 for (i = 0; i < niocount; i++) {
1078 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1082 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1087 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1088 GOTO(out_stop, rc = -EINPROGRESS);
1090 if (likely(!fake_write)) {
1091 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1097 /* update [mac]time if needed */
1098 rc = dt_declare_attr_set(env, o, la, th);
1103 rc = ofd_trans_start(env, ofd, fo, th);
1107 if (likely(!fake_write)) {
1108 rc = dt_write_commit(env, o, lnb, niocount, th);
1114 rc = dt_attr_set(env, o, la, th);
1119 /* get attr to return */
1120 rc = dt_attr_get(env, o, la);
1123 /* Force commit to make the just-deleted blocks
1124 * reusable. LU-456 */
1128 /* do this before trans stop in case commit has finished */
1129 if (!th->th_sync && soft_sync && !cb_registered) {
1130 ofd_soft_sync_cb_add(th, exp);
1131 cb_registered = true;
1134 if (rc == 0 && granted > 0) {
1135 if (ofd_grant_commit_cb_add(th, exp, granted) == 0)
1139 ofd_trans_stop(env, ofd, th, rc);
1140 if (rc == -ENOSPC && retries++ < 3) {
1141 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1147 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1148 atomic_set(&fed->fed_soft_sync_count, 0);
1149 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1150 ofd->ofd_soft_sync_limit)
1151 dt_commit_async(env, ofd->ofd_osd);
1154 dt_bufs_put(env, o, lnb, niocount);
1155 ofd_read_unlock(env, fo);
1156 ofd_object_put(env, fo);
1157 /* second put is pair to object_get in ofd_preprw_write */
1158 ofd_object_put(env, fo);
1160 ofd_grant_commit(exp, granted, old_rc);
1165 * Commit bulk IO to the storage.
1167 * This is companion function to the ofd_preprw(). It finishes bulk IO
1168 * request processing by committing buffers to the storage (WRITE) and/or
1169 * freeing those buffers (read/write). See ofd_commitrw_read() and
1170 * ofd_commitrw_write() for details about each type of IO.
1172 * \param[in] env execution environment
1173 * \param[in] cmd IO type (READ/WRITE)
1174 * \param[in] exp OBD export of client
1175 * \param[in] oa OBDO structure from client
1176 * \param[in] objcount always 1
1177 * \param[in] obj object data
1178 * \param[in] rnb remote buffers
1179 * \param[in] npages number of local buffers
1180 * \param[in] lnb local buffers
1181 * \param[in] old_rc result of processing at this point
1183 * \retval 0 on successful commit
1184 * \retval negative value on error
1186 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1187 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1188 struct niobuf_remote *rnb, int npages,
1189 struct niobuf_local *lnb, int old_rc)
1191 struct ofd_thread_info *info = ofd_info(env);
1192 struct ofd_mod_data *fmd;
1194 struct ofd_device *ofd = ofd_exp(exp);
1195 struct filter_fid *ff = NULL;
1196 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1199 LASSERT(npages > 0);
1201 if (cmd == OBD_BRW_WRITE) {
1202 /* Don't update timestamps if this write is older than a
1203 * setattr which modifies the timestamps. b=10150 */
1205 /* XXX when we start having persistent reservations this needs
1206 * to be changed to ofd_fmd_get() to create the fmd if it
1207 * doesn't already exist so we can store the reservation handle
1209 valid = OBD_MD_FLUID | OBD_MD_FLGID;
1210 fmd = ofd_fmd_find(exp, fid);
1211 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1212 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1214 ofd_fmd_put(exp, fmd);
1215 la_from_obdo(&info->fti_attr, oa, valid);
1217 if (oa->o_valid & OBD_MD_FLFID) {
1218 ff = &info->fti_mds_fid;
1219 ofd_prepare_fidea(ff, oa);
1222 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1223 ff, objcount, npages, lnb,
1224 oa->o_grant_used, old_rc);
1226 obdo_from_la(oa, &info->fti_attr,
1227 OFD_VALID_FLAGS | LA_GID | LA_UID);
1229 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
1231 /* don't report overquota flag if we failed before reaching
1233 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1234 /* return the overquota flags to client */
1235 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1236 if (oa->o_valid & OBD_MD_FLFLAGS)
1237 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1239 oa->o_flags = OBD_FL_NO_USRQUOTA;
1242 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1243 if (oa->o_valid & OBD_MD_FLFLAGS)
1244 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1246 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1249 oa->o_valid |= OBD_MD_FLFLAGS;
1250 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1252 } else if (cmd == OBD_BRW_READ) {
1253 struct ldlm_namespace *ns = ofd->ofd_namespace;
1255 /* If oa != NULL then ofd_preprw_read updated the inode
1256 * atime and we should update the lvb so that other glimpses
1257 * will also get the updated value. bug 5972 */
1258 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1259 struct ldlm_resource *rs = NULL;
1261 ost_fid_build_resid(fid, &info->fti_resid);
1262 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1265 ldlm_res_lvbo_update(rs, NULL, 1);
1266 ldlm_resource_putref(rs);
1269 rc = ofd_commitrw_read(env, ofd, fid, objcount,