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_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;
635 /* remote client can't break through quota */
636 if (exp_connect_rmtclient(exp))
637 lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
641 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
642 tot_bytes += rnb[i].rnb_len;
644 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
646 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
647 if (unlikely(rc != 0))
650 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
653 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
654 ofd_read_unlock(env, fo);
655 ofd_object_put(env, fo);
656 /* ofd_grant_prepare_write() was called, so we must commit */
657 ofd_grant_commit(env, exp, rc);
659 /* let's still process incoming grant information packed in the oa,
660 * but without enforcing grant since we won't proceed with the write.
661 * Just like a read request actually. */
662 ofd_grant_prepare_read(env, exp, oa);
667 * Prepare bulk IO requests for processing.
669 * This function does initial checks of IO and calls corresponding
670 * functions for read/write processing.
672 * \param[in] env execution environment
673 * \param[in] cmd IO type (read/write)
674 * \param[in] exp OBD export of client
675 * \param[in] oa OBDO structure from request
676 * \param[in] objcount always 1
677 * \param[in] obj object data
678 * \param[in] rnb remote buffers
679 * \param[in] nr_local number of local buffers
680 * \param[in] lnb local buffers
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)
690 struct tgt_session_info *tsi = tgt_ses_info(env);
691 struct ofd_device *ofd = ofd_exp(exp);
692 struct ofd_thread_info *info;
694 const struct lu_fid *fid = &oa->o_oi.oi_fid;
697 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
698 CERROR("%s: bulk has too many pages %d, which exceeds the"
699 "maximum pages per RPC of %d\n",
700 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
704 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
705 info = ofd_info_init(env, exp);
708 info = tsi2ofd_info(tsi);
709 jobid = tsi->tsi_jobid;
714 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
715 struct ofd_seq *oseq;
717 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
719 CERROR("%s: Can not find seq for "DOSTID
720 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
725 if (oseq->os_destroys_in_progress == 0) {
726 /* don't fail lookups for orphan recovery, it causes
727 * later LBUGs when objects still exist during
729 ofd_seq_put(env, oseq);
732 ofd_seq_put(env, oseq);
735 LASSERT(objcount == 1);
736 LASSERT(obj->ioo_bufcnt > 0);
738 if (cmd == OBD_BRW_WRITE) {
739 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
740 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
741 objcount, obj, rnb, nr_local, lnb, jobid);
742 } else if (cmd == OBD_BRW_READ) {
743 ofd_grant_prepare_read(env, exp, oa);
744 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
745 obj->ioo_bufcnt, rnb, nr_local, lnb,
747 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
749 CERROR("%s: wrong cmd %d received!\n",
750 exp->exp_obd->obd_name, cmd);
757 * Drop reference on local buffers for read bulk IO.
759 * This will free all local buffers use by this read request.
761 * \param[in] env execution environment
762 * \param[in] ofd OFD device
763 * \param[in] fid object FID
764 * \param[in] objcount always 1
765 * \param[in] niocount number of local buffers
766 * \param[in] lnb local buffers
768 * \retval 0 on successful execution
769 * \retval negative value on error
772 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
773 const struct lu_fid *fid, int objcount, int niocount,
774 struct niobuf_local *lnb)
776 struct ofd_object *fo;
780 LASSERT(niocount > 0);
782 fo = ofd_object_find(env, ofd, fid);
786 LASSERT(ofd_object_exists(fo));
787 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
789 ofd_read_unlock(env, fo);
790 ofd_object_put(env, fo);
791 /* second put is pair to object_get in ofd_preprw_read */
792 ofd_object_put(env, fo);
798 * Set attributes of object during write bulk IO processing.
800 * Change object attributes and write parent FID into extended
801 * attributes when needed.
803 * \param[in] env execution environment
804 * \param[in] ofd OFD device
805 * \param[in] ofd_obj OFD object
806 * \param[in] la object attributes
807 * \param[in] ff parent FID
809 * \retval 0 on successful attributes update
810 * \retval negative value on error
813 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
814 struct ofd_object *ofd_obj, struct lu_attr *la,
815 struct filter_fid *ff)
817 struct ofd_thread_info *info = ofd_info(env);
818 __u64 valid = la->la_valid;
821 struct dt_object *dt_obj;
828 dt_obj = ofd_object_child(ofd_obj);
829 LASSERT(dt_obj != NULL);
831 la->la_valid &= LA_UID | LA_GID;
833 rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
838 rc = ofd_object_ff_load(env, ofd_obj);
845 if (!la->la_valid && !ff_needed)
846 /* no attributes to set */
849 th = ofd_trans_create(env, ofd);
851 GOTO(out, rc = PTR_ERR(th));
854 rc = dt_declare_attr_set(env, dt_obj, la, th);
860 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
861 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
862 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
863 ff->ff_parent.f_oid =
864 cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
866 info->fti_buf.lb_buf = ff;
867 info->fti_buf.lb_len = sizeof(*ff);
868 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
869 XATTR_NAME_FID, 0, th);
874 /* We don't need a transno for this operation which will be re-executed
875 * anyway when the OST_WRITE (with a transno assigned) is replayed */
876 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
882 rc = dt_attr_set(env, dt_obj, la, th);
887 /* set filter fid EA */
889 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
892 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
895 ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
896 ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
897 /* Currently, the filter_fid::ff_parent::f_ver is not
898 * the real parent MDT-object's FID::f_ver, instead it
899 * is the OST-object index in its parent MDT-object's
901 ofd_obj->ofo_pfid.f_stripe_idx =
902 le32_to_cpu(ff->ff_parent.f_stripe_idx);
909 dt_trans_stop(env, ofd->ofd_osd, th);
911 la->la_valid = valid;
915 struct ofd_soft_sync_callback {
916 struct dt_txn_commit_cb ossc_cb;
917 struct obd_export *ossc_exp;
921 * Callback function for "soft sync" update.
923 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
924 * See ofd_soft_sync_cb_add() below for more details on soft sync.
926 * \param[in] env execution environment
927 * \param[in] th transaction handle
928 * \param[in] cb callback data
929 * \param[in] err error code
931 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
932 struct dt_txn_commit_cb *cb, int err)
934 struct ofd_soft_sync_callback *ossc;
936 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
938 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
939 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
941 class_export_cb_put(ossc->ossc_exp);
946 * Add callback for "soft sync" processing.
948 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
949 * flag is set in client buffers. The intention is for this operation to
950 * commit pages belonging to a client which has "too many" outstanding
951 * unstable pages in its cache. See LU-2139 for details.
953 * This function adds callback to be called when commit is done.
955 * \param[in] th transaction handle
956 * \param[in] exp OBD export of client
958 * \retval 0 on successful callback adding
959 * \retval negative value on error
961 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
963 struct ofd_soft_sync_callback *ossc;
964 struct dt_txn_commit_cb *dcb;
971 ossc->ossc_exp = class_export_cb_get(exp);
973 dcb = &ossc->ossc_cb;
974 dcb->dcb_func = ofd_cb_soft_sync;
975 INIT_LIST_HEAD(&dcb->dcb_linkage);
976 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
978 rc = dt_trans_cb_add(th, dcb);
980 class_export_cb_put(exp);
988 * Commit bulk IO buffers to the storage.
990 * This function finalizes write IO processing by writing data to the disk.
991 * That write can be synchronous or asynchronous depending on buffers flags.
993 * \param[in] env execution environment
994 * \param[in] exp OBD export of client
995 * \param[in] ofd OFD device
996 * \param[in] fid FID of object
997 * \param[in] la object attributes
998 * \param[in] ff parent FID of object
999 * \param[in] objcount always 1
1000 * \param[in] niocount number of local buffers
1001 * \param[in] lnb local buffers
1002 * \param[in] old_rc result of processing at this point
1004 * \retval 0 on successful commit
1005 * \retval negative value on error
1008 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1009 struct ofd_device *ofd, const struct lu_fid *fid,
1010 struct lu_attr *la, struct filter_fid *ff, int objcount,
1011 int niocount, struct niobuf_local *lnb, int old_rc)
1013 struct ofd_thread_info *info = ofd_info(env);
1014 struct ofd_object *fo;
1015 struct dt_object *o;
1020 struct filter_export_data *fed = &exp->exp_filter_data;
1021 bool soft_sync = false;
1022 bool cb_registered = false;
1026 LASSERT(objcount == 1);
1028 fo = ofd_object_find(env, ofd, fid);
1029 LASSERT(fo != NULL);
1030 LASSERT(ofd_object_exists(fo));
1032 o = ofd_object_child(fo);
1036 GOTO(out, rc = old_rc);
1039 * The first write to each object must set some attributes. It is
1040 * important to set the uid/gid before calling
1041 * dt_declare_write_commit() since quota enforcement is now handled in
1044 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
1048 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1051 th = ofd_trans_create(env, ofd);
1053 GOTO(out, rc = PTR_ERR(th));
1055 th->th_sync |= ofd->ofd_syncjournal;
1056 if (th->th_sync == 0) {
1057 for (i = 0; i < niocount; i++) {
1058 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1062 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1067 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1068 GOTO(out_stop, rc = -EINPROGRESS);
1070 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1075 /* update [mac]time if needed */
1076 rc = dt_declare_attr_set(env, o, la, th);
1081 rc = ofd_trans_start(env, ofd, fo, th);
1085 rc = dt_write_commit(env, o, lnb, niocount, th);
1090 rc = dt_attr_set(env, o, la, th);
1095 /* get attr to return */
1096 rc = dt_attr_get(env, o, la);
1099 /* Force commit to make the just-deleted blocks
1100 * reusable. LU-456 */
1104 /* do this before trans stop in case commit has finished */
1105 if (!th->th_sync && soft_sync && !cb_registered) {
1106 ofd_soft_sync_cb_add(th, exp);
1107 cb_registered = true;
1110 ofd_trans_stop(env, ofd, th, rc);
1111 if (rc == -ENOSPC && retries++ < 3) {
1112 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1118 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1119 atomic_set(&fed->fed_soft_sync_count, 0);
1120 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1121 ofd->ofd_soft_sync_limit)
1122 dt_commit_async(env, ofd->ofd_osd);
1125 dt_bufs_put(env, o, lnb, niocount);
1126 ofd_read_unlock(env, fo);
1127 ofd_object_put(env, fo);
1128 /* second put is pair to object_get in ofd_preprw_write */
1129 ofd_object_put(env, fo);
1130 ofd_grant_commit(env, info->fti_exp, old_rc);
1135 * Commit bulk IO to the storage.
1137 * This is companion function to the ofd_preprw(). It finishes bulk IO
1138 * request processing by committing buffers to the storage (WRITE) and/or
1139 * freeing those buffers (read/write). See ofd_commitrw_read() and
1140 * ofd_commitrw_write() for details about each type of IO.
1142 * \param[in] env execution environment
1143 * \param[in] cmd IO type (READ/WRITE)
1144 * \param[in] exp OBD export of client
1145 * \param[in] oa OBDO structure from client
1146 * \param[in] objcount always 1
1147 * \param[in] obj object data
1148 * \param[in] rnb remote buffers
1149 * \param[in] npages number of local buffers
1150 * \param[in] lnb local buffers
1151 * \param[in] old_rc result of processing at this point
1153 * \retval 0 on successful commit
1154 * \retval negative value on error
1156 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1157 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1158 struct niobuf_remote *rnb, int npages,
1159 struct niobuf_local *lnb, int old_rc)
1161 struct ofd_thread_info *info = ofd_info(env);
1162 struct ofd_mod_data *fmd;
1164 struct ofd_device *ofd = ofd_exp(exp);
1165 struct filter_fid *ff = NULL;
1166 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1169 LASSERT(npages > 0);
1171 if (cmd == OBD_BRW_WRITE) {
1172 /* Don't update timestamps if this write is older than a
1173 * setattr which modifies the timestamps. b=10150 */
1175 /* XXX when we start having persistent reservations this needs
1176 * to be changed to ofd_fmd_get() to create the fmd if it
1177 * doesn't already exist so we can store the reservation handle
1179 valid = OBD_MD_FLUID | OBD_MD_FLGID;
1180 fmd = ofd_fmd_find(exp, fid);
1181 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1182 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1184 ofd_fmd_put(exp, fmd);
1185 la_from_obdo(&info->fti_attr, oa, valid);
1187 if (oa->o_valid & OBD_MD_FLFID) {
1188 ff = &info->fti_mds_fid;
1189 ofd_prepare_fidea(ff, oa);
1192 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1193 ff, objcount, npages, lnb, old_rc);
1195 obdo_from_la(oa, &info->fti_attr,
1196 OFD_VALID_FLAGS | LA_GID | LA_UID);
1198 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
1200 /* don't report overquota flag if we failed before reaching
1202 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1203 /* return the overquota flags to client */
1204 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1205 if (oa->o_valid & OBD_MD_FLFLAGS)
1206 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1208 oa->o_flags = OBD_FL_NO_USRQUOTA;
1211 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1212 if (oa->o_valid & OBD_MD_FLFLAGS)
1213 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1215 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1218 oa->o_valid |= OBD_MD_FLFLAGS;
1219 oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
1221 } else if (cmd == OBD_BRW_READ) {
1222 struct ldlm_namespace *ns = ofd->ofd_namespace;
1224 /* If oa != NULL then ofd_preprw_read updated the inode
1225 * atime and we should update the lvb so that other glimpses
1226 * will also get the updated value. bug 5972 */
1227 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1228 struct ldlm_resource *rs = NULL;
1230 ost_fid_build_resid(fid, &info->fti_resid);
1231 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1234 ldlm_res_lvbo_update(rs, NULL, 1);
1235 ldlm_resource_putref(rs);
1238 rc = ofd_commitrw_read(env, ofd, fid, objcount,