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, 2017, 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"
45 #include <lustre_nodemap.h>
47 struct ofd_inconsistency_item {
48 struct list_head oii_list;
49 struct ofd_object *oii_obj;
50 struct filter_fid oii_ff;
54 * Verify single object for parent FID consistency.
56 * Part of LFSCK processing which checks single object PFID stored in extended
57 * attribute (XATTR) against real FID of MDT parent object received by LFSCK.
58 * This verifies that the OST object is being referenced by only a single MDT
61 * \param[in] env execution environment
62 * \param[in] ofd OFD device
63 * \param[in] oii object-related local data
64 * \param[in] lrl LFSCK request data
66 static void ofd_inconsistency_verify_one(const struct lu_env *env,
67 struct ofd_device *ofd,
68 struct ofd_inconsistency_item *oii,
69 struct lfsck_req_local *lrl)
71 struct ofd_object *fo = oii->oii_obj;
72 struct filter_fid *client_ff = &oii->oii_ff;
73 struct filter_fid *local_ff = &fo->ofo_ff;
76 LASSERT(fo->ofo_pfid_checking);
77 LASSERT(!fo->ofo_pfid_verified);
79 lrl->lrl_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
80 lrl->lrl_ff_client = *client_ff; /* client given PFID. */
81 lrl->lrl_ff_local = *local_ff; /* OST local stored PFID. */
83 rc = lfsck_in_notify_local(env, ofd->ofd_osd, lrl, NULL);
84 ofd_write_lock(env, fo);
85 switch (lrl->lrl_status) {
90 CDEBUG(D_LFSCK, "%s: fail to verify OST local stored "
91 "PFID xattr for "DFID", the client given PFID "
92 DFID", OST local stored PFID "DFID": rc = %d\n",
93 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
94 PFID(&client_ff->ff_parent),
95 PFID(&local_ff->ff_parent), rc);
97 fo->ofo_pfid_verified = 1;
99 case LPVS_INCONSISTENT:
102 ofd->ofd_inconsistency_self_detected++;
104 CDEBUG(D_LFSCK, "%s: fail to verify the client given "
105 "PFID for "DFID", the client given PFID "DFID
106 ", local stored PFID "DFID": rc = %d\n",
107 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
108 PFID(&client_ff->ff_parent),
109 PFID(&local_ff->ff_parent), rc);
111 CDEBUG(D_LFSCK, "%s: both the client given PFID and "
112 "the OST local stored PFID are stale for the "
113 "OST-object "DFID", client given PFID is "DFID
114 ", local stored PFID is "DFID"\n",
115 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
116 PFID(&client_ff->ff_parent),
117 PFID(&local_ff->ff_parent));
119 case LPVS_INCONSISTENT_TOFIX:
120 ofd->ofd_inconsistency_self_detected++;
122 ofd->ofd_inconsistency_self_repaired++;
123 CDEBUG(D_LFSCK, "%s: fixed the staled OST PFID xattr "
124 "for "DFID", with the client given PFID "DFID
125 ", the old stored PFID "DFID"\n",
126 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
127 PFID(&client_ff->ff_parent),
128 PFID(&local_ff->ff_parent));
130 CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
131 "for "DFID", client given PFID "DFID", local "
132 "stored PFID "DFID": rc = %d\n",
133 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
134 PFID(&client_ff->ff_parent),
135 PFID(&local_ff->ff_parent), rc);
137 local_ff->ff_parent = client_ff->ff_parent;
138 fo->ofo_pfid_verified = 1;
143 fo->ofo_pfid_checking = 0;
144 ofd_write_unlock(env, fo);
146 ofd_object_put(env, fo);
151 * Verification thread to check parent FID consistency.
153 * Kernel thread to check consistency of parent FID for any
154 * new item added for checking by ofd_add_inconsistency_item().
156 * \param[in] args OFD device
158 * \retval 0 on successful thread termination
159 * \retval negative value if thread can't start
161 static int ofd_inconsistency_verification_main(void *args)
164 struct ofd_device *ofd = args;
165 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
166 struct ofd_inconsistency_item *oii;
167 struct lfsck_req_local *lrl = NULL;
168 struct l_wait_info lwi = { 0 };
172 rc = lu_env_init(&env, LCT_DT_THREAD);
173 spin_lock(&ofd->ofd_inconsistency_lock);
174 thread_set_flags(thread, rc ? SVC_STOPPED : SVC_RUNNING);
175 wake_up_all(&thread->t_ctl_waitq);
176 spin_unlock(&ofd->ofd_inconsistency_lock);
182 GOTO(out_unlocked, rc = -ENOMEM);
184 lrl->lrl_event = LEL_PAIRS_VERIFY_LOCAL;
185 lrl->lrl_active = LFSCK_TYPE_LAYOUT;
187 spin_lock(&ofd->ofd_inconsistency_lock);
189 if (unlikely(!thread_is_running(thread)))
192 while (!list_empty(&ofd->ofd_inconsistency_list)) {
193 oii = list_entry(ofd->ofd_inconsistency_list.next,
194 struct ofd_inconsistency_item,
196 list_del_init(&oii->oii_list);
197 spin_unlock(&ofd->ofd_inconsistency_lock);
198 ofd_inconsistency_verify_one(&env, ofd, oii, lrl);
199 spin_lock(&ofd->ofd_inconsistency_lock);
202 spin_unlock(&ofd->ofd_inconsistency_lock);
203 l_wait_event(thread->t_ctl_waitq,
204 !list_empty(&ofd->ofd_inconsistency_list) ||
205 !thread_is_running(thread),
207 spin_lock(&ofd->ofd_inconsistency_lock);
210 while (!list_empty(&ofd->ofd_inconsistency_list)) {
211 struct ofd_object *fo;
213 oii = list_entry(ofd->ofd_inconsistency_list.next,
214 struct ofd_inconsistency_item,
216 list_del_init(&oii->oii_list);
218 spin_unlock(&ofd->ofd_inconsistency_lock);
220 ofd_write_lock(&env, fo);
221 fo->ofo_pfid_checking = 0;
222 ofd_write_unlock(&env, fo);
224 ofd_object_put(&env, fo);
226 spin_lock(&ofd->ofd_inconsistency_lock);
234 spin_lock(&ofd->ofd_inconsistency_lock);
236 thread_set_flags(thread, SVC_STOPPED);
237 wake_up_all(&thread->t_ctl_waitq);
238 spin_unlock(&ofd->ofd_inconsistency_lock);
245 * Start parent FID verification thread.
247 * See ofd_inconsistency_verification_main().
249 * \param[in] ofd OFD device
251 * \retval 0 on successful start of thread
252 * \retval negative value on error
254 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
256 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
257 struct l_wait_info lwi = { 0 };
258 struct task_struct *task;
261 spin_lock(&ofd->ofd_inconsistency_lock);
262 if (unlikely(thread_is_running(thread))) {
263 spin_unlock(&ofd->ofd_inconsistency_lock);
268 thread_set_flags(thread, 0);
269 spin_unlock(&ofd->ofd_inconsistency_lock);
270 task = kthread_run(ofd_inconsistency_verification_main, ofd,
271 "inconsistency_verification");
274 CERROR("%s: cannot start self_repair thread: rc = %d\n",
278 l_wait_event(thread->t_ctl_waitq,
279 thread_is_running(thread) ||
280 thread_is_stopped(thread),
288 * Stop parent FID verification thread.
290 * \param[in] ofd OFD device
292 * \retval 0 on successful start of thread
293 * \retval -EALREADY if thread is already stopped
295 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
297 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
298 struct l_wait_info lwi = { 0 };
300 spin_lock(&ofd->ofd_inconsistency_lock);
301 if (thread_is_init(thread) || thread_is_stopped(thread)) {
302 spin_unlock(&ofd->ofd_inconsistency_lock);
307 thread_set_flags(thread, SVC_STOPPING);
308 spin_unlock(&ofd->ofd_inconsistency_lock);
309 wake_up_all(&thread->t_ctl_waitq);
310 l_wait_event(thread->t_ctl_waitq,
311 thread_is_stopped(thread),
318 * Add new item for parent FID verification.
320 * Prepare new verification item and pass it to the dedicated
321 * verification thread for further processing.
323 * \param[in] env execution environment
324 * \param[in] fo OFD object
325 * \param[in] oa OBDO structure with PFID
327 static void ofd_add_inconsistency_item(const struct lu_env *env,
328 struct ofd_object *fo, struct obdo *oa)
330 struct ofd_device *ofd = ofd_obj2dev(fo);
331 struct ofd_inconsistency_item *oii;
332 struct filter_fid *ff;
339 INIT_LIST_HEAD(&oii->oii_list);
340 lu_object_get(&fo->ofo_obj.do_lu);
343 ff->ff_parent.f_seq = oa->o_parent_seq;
344 ff->ff_parent.f_oid = oa->o_parent_oid;
345 ff->ff_parent.f_stripe_idx = oa->o_stripe_idx;
346 ff->ff_layout = oa->o_layout;
348 spin_lock(&ofd->ofd_inconsistency_lock);
349 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
350 spin_unlock(&ofd->ofd_inconsistency_lock);
356 fo->ofo_pfid_checking = 1;
357 if (list_empty(&ofd->ofd_inconsistency_list))
359 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
360 spin_unlock(&ofd->ofd_inconsistency_lock);
362 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
364 /* XXX: When the found inconsistency exceeds some threshold,
365 * we can trigger the LFSCK to scan part of the system
366 * or the whole system, which depends on how to define
367 * the threshold, a simple way maybe like that: define
368 * the absolute value of how many inconsisteny allowed
369 * to be repaired via self detect/repair mechanism, if
370 * exceeded, then trigger the LFSCK to scan the layout
371 * inconsistency within the whole system. */
375 * Verify parent FID of an object.
377 * Check the parent FID is sane and start extended
378 * verification procedure otherwise.
380 * \param[in] env execution environment
381 * \param[in] fo OFD object
382 * \param[in] oa OBDO structure with PFID
384 * \retval 0 on successful verification
385 * \retval -EINPROGRESS if PFID is being repaired
386 * \retval -EPERM if PFID was verified but still insane
388 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
391 struct lu_fid *pfid = &fo->ofo_ff.ff_parent;
395 if (fid_is_sane(pfid)) {
396 if (likely(oa->o_parent_seq == pfid->f_seq &&
397 oa->o_parent_oid == pfid->f_oid &&
398 oa->o_stripe_idx == pfid->f_stripe_idx))
401 if (fo->ofo_pfid_verified)
405 /* The OST-object may be inconsistent, and we need further verification.
406 * To avoid block the RPC service thread, return -EINPROGRESS to client
407 * and make it retry later. */
408 if (fo->ofo_pfid_checking)
409 RETURN(-EINPROGRESS);
411 rc = ofd_object_ff_load(env, fo);
418 if (likely(oa->o_parent_seq == pfid->f_seq &&
419 oa->o_parent_oid == pfid->f_oid &&
420 oa->o_stripe_idx == pfid->f_stripe_idx))
423 /* Push it to the dedicated thread for further verification. */
424 ofd_add_inconsistency_item(env, fo, oa);
426 RETURN(-EINPROGRESS);
430 * FLR: verify the layout version of object.
432 * \param[in] env execution environment
433 * \param[in] fo OFD object
434 * \param[in] oa OBDO structure with layout version
436 * \retval 0 on successful verification
437 * \retval -EINPROGRESS layout version is in transfer
438 * \retval -ESTALE the layout version on client is stale
440 int ofd_verify_layout_version(const struct lu_env *env,
441 struct ofd_object *fo, const struct obdo *oa)
443 __u32 layout_version;
447 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_OST_SKIP_LV_CHECK)))
450 rc = ofd_object_ff_load(env, fo);
457 layout_version = fo->ofo_ff.ff_layout_version;
458 if (oa->o_layout_version >= layout_version &&
459 oa->o_layout_version <= layout_version + fo->ofo_ff.ff_range)
462 /* normal traffic, decide if to return ESTALE or EINPROGRESS */
463 layout_version &= ~LU_LAYOUT_RESYNC;
465 /* this update is not legitimate */
466 if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) <= layout_version)
467 GOTO(out, rc = -ESTALE);
469 /* layout version may not be transmitted yet */
470 if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) > layout_version)
471 GOTO(out, rc = -EINPROGRESS);
476 CDEBUG(D_INODE, DFID " verify layout version: %u vs. %u/%u, rc: %d\n",
477 PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
478 oa->o_layout_version, fo->ofo_ff.ff_layout_version,
479 fo->ofo_ff.ff_range, rc);
485 * Prepare buffers for read request processing.
487 * This function converts remote buffers from client to local buffers
488 * and prepares the latter.
490 * \param[in] env execution environment
491 * \param[in] exp OBD export of client
492 * \param[in] ofd OFD device
493 * \param[in] fid FID of object
494 * \param[in] la object attributes
495 * \param[in] oa OBDO structure from client
496 * \param[in] niocount number of remote buffers
497 * \param[in] rnb remote buffers
498 * \param[in] nr_local number of local buffers
499 * \param[in] lnb local buffers
500 * \param[in] jobid job ID name
502 * \retval 0 on successful prepare
503 * \retval negative value on error
505 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
506 struct ofd_device *ofd, const struct lu_fid *fid,
507 struct lu_attr *la, struct obdo *oa, int niocount,
508 struct niobuf_remote *rnb, int *nr_local,
509 struct niobuf_local *lnb, char *jobid)
511 struct ofd_object *fo;
512 int i, j, rc, tot_bytes = 0;
513 enum dt_bufs_type dbt = DT_BUFS_TYPE_READ;
516 LASSERT(env != NULL);
518 fo = ofd_object_find(env, ofd, fid);
523 ofd_read_lock(env, fo);
524 if (!ofd_object_exists(fo))
525 GOTO(unlock, rc = -ENOENT);
527 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
528 rc = ofd_verify_ff(env, fo, oa);
533 if (ptlrpc_connection_is_local(exp->exp_connection))
534 dbt |= DT_BUFS_TYPE_LOCAL;
536 for (*nr_local = 0, i = 0, j = 0; i < niocount; i++) {
537 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
539 if (unlikely(rc < 0))
541 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
542 /* correct index for local buffers to continue with */
545 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
546 tot_bytes += rnb[i].rnb_len;
549 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
550 rc = dt_attr_get(env, ofd_object_child(fo), la);
554 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
558 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
562 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
564 ofd_read_unlock(env, fo);
565 ofd_object_put(env, fo);
570 * Prepare buffers for write request processing.
572 * This function converts remote buffers from client to local buffers
573 * and prepares the latter. If there is recovery in progress and required
574 * object is missing then it can be re-created before write.
576 * \param[in] env execution environment
577 * \param[in] exp OBD export of client
578 * \param[in] ofd OFD device
579 * \param[in] fid FID of object
580 * \param[in] la object attributes
581 * \param[in] oa OBDO structure from client
582 * \param[in] objcount always 1
583 * \param[in] obj object data
584 * \param[in] rnb remote buffers
585 * \param[in] nr_local number of local buffers
586 * \param[in] lnb local buffers
587 * \param[in] jobid job ID name
589 * \retval 0 on successful prepare
590 * \retval negative value on error
592 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
593 struct ofd_device *ofd, const struct lu_fid *fid,
594 struct lu_attr *la, struct obdo *oa,
595 int objcount, struct obd_ioobj *obj,
596 struct niobuf_remote *rnb, int *nr_local,
597 struct niobuf_local *lnb, char *jobid)
599 struct ofd_object *fo;
600 int i, j, k, rc = 0, tot_bytes = 0;
601 enum dt_bufs_type dbt = DT_BUFS_TYPE_WRITE;
604 LASSERT(env != NULL);
605 LASSERT(objcount == 1);
607 if (unlikely(exp->exp_obd->obd_recovering)) {
608 u64 seq = fid_seq(fid);
609 u64 oid = fid_oid(fid);
610 struct ofd_seq *oseq;
612 oseq = ofd_seq_load(env, ofd, seq);
614 CERROR("%s: Can't find FID Sequence %#llx: rc = %d\n",
615 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
616 GOTO(out, rc = -EINVAL);
619 if (oid > ofd_seq_last_oid(oseq)) {
623 mutex_lock(&oseq->os_create_lock);
624 diff = oid - ofd_seq_last_oid(oseq);
626 /* Do sync create if the seq is about to used up */
627 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
628 if (unlikely(oid >= IDIF_MAX_OID - 1))
630 } else if (fid_seq_is_norm(seq)) {
632 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
635 CERROR("%s : invalid o_seq "DOSTID"\n",
636 ofd_name(ofd), POSTID(&oa->o_oi));
637 mutex_unlock(&oseq->os_create_lock);
638 ofd_seq_put(env, oseq);
639 GOTO(out, rc = -EINVAL);
643 u64 next_id = ofd_seq_last_oid(oseq) + 1;
644 int count = ofd_precreate_batch(ofd, diff);
646 rc = ofd_precreate_objects(env, ofd, next_id,
649 mutex_unlock(&oseq->os_create_lock);
650 ofd_seq_put(env, oseq);
657 mutex_unlock(&oseq->os_create_lock);
660 ofd_seq_put(env, oseq);
663 fo = ofd_object_find(env, ofd, fid);
665 GOTO(out, rc = PTR_ERR(fo));
668 ofd_read_lock(env, fo);
669 if (!ofd_object_exists(fo)) {
670 CERROR("%s: BRW to missing obj "DOSTID"\n",
671 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
672 ofd_read_unlock(env, fo);
673 ofd_object_put(env, fo);
674 GOTO(out, rc = -ENOENT);
677 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
678 rc = ofd_verify_ff(env, fo, oa);
680 ofd_read_unlock(env, fo);
681 ofd_object_put(env, fo);
686 /* need to verify layout version */
687 if (oa->o_valid & OBD_MD_LAYOUT_VERSION) {
688 rc = ofd_verify_layout_version(env, fo, oa);
690 ofd_read_unlock(env, fo);
691 ofd_object_put(env, fo);
695 oa->o_valid &= ~OBD_MD_LAYOUT_VERSION;
698 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
699 * space back if possible */
700 tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
702 if (ptlrpc_connection_is_local(exp->exp_connection))
703 dbt |= DT_BUFS_TYPE_LOCAL;
705 /* parse remote buffers to local buffers and prepare the latter */
706 for (*nr_local = 0, i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
707 rc = dt_bufs_get(env, ofd_object_child(fo),
708 rnb + i, lnb + j, dbt);
709 if (unlikely(rc < 0))
711 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
712 /* correct index for local buffers to continue with */
713 for (k = 0; k < rc; k++) {
714 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
715 lnb[j+k].lnb_flags &= ~OBD_BRW_LOCALS;
716 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
717 lnb[j+k].lnb_rc = -ENOSPC;
721 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
722 tot_bytes += rnb[i].rnb_len;
724 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
726 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
727 if (unlikely(rc != 0))
730 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
733 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
734 ofd_read_unlock(env, fo);
735 ofd_object_put(env, fo);
736 /* tgt_grant_prepare_write() was called, so we must commit */
737 tgt_grant_commit(exp, oa->o_grant_used, rc);
739 /* let's still process incoming grant information packed in the oa,
740 * but without enforcing grant since we won't proceed with the write.
741 * Just like a read request actually. */
742 tgt_grant_prepare_read(env, exp, oa);
747 * Prepare bulk IO requests for processing.
749 * This function does initial checks of IO and calls corresponding
750 * functions for read/write processing.
752 * \param[in] env execution environment
753 * \param[in] cmd IO type (read/write)
754 * \param[in] exp OBD export of client
755 * \param[in] oa OBDO structure from request
756 * \param[in] objcount always 1
757 * \param[in] obj object data
758 * \param[in] rnb remote buffers
759 * \param[in] nr_local number of local buffers
760 * \param[in] lnb local buffers
762 * \retval 0 on successful prepare
763 * \retval negative value on error
765 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
766 struct obdo *oa, int objcount, struct obd_ioobj *obj,
767 struct niobuf_remote *rnb, int *nr_local,
768 struct niobuf_local *lnb)
770 struct tgt_session_info *tsi = tgt_ses_info(env);
771 struct ofd_device *ofd = ofd_exp(exp);
772 struct ofd_thread_info *info;
774 const struct lu_fid *fid = &oa->o_oi.oi_fid;
777 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
778 CERROR("%s: bulk has too many pages %d, which exceeds the"
779 "maximum pages per RPC of %d\n",
780 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
784 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
785 info = ofd_info_init(env, exp);
788 info = tsi2ofd_info(tsi);
789 jobid = tsi->tsi_jobid;
794 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
795 struct ofd_seq *oseq;
797 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
799 CERROR("%s: Can not find seq for "DOSTID
800 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
805 if (oseq->os_destroys_in_progress == 0) {
806 /* don't fail lookups for orphan recovery, it causes
807 * later LBUGs when objects still exist during
809 ofd_seq_put(env, oseq);
812 ofd_seq_put(env, oseq);
815 LASSERT(objcount == 1);
816 LASSERT(obj->ioo_bufcnt > 0);
818 if (cmd == OBD_BRW_WRITE) {
819 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
820 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
821 objcount, obj, rnb, nr_local, lnb, jobid);
822 } else if (cmd == OBD_BRW_READ) {
823 tgt_grant_prepare_read(env, exp, oa);
824 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
825 obj->ioo_bufcnt, rnb, nr_local, lnb,
827 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
829 CERROR("%s: wrong cmd %d received!\n",
830 exp->exp_obd->obd_name, cmd);
837 * Drop reference on local buffers for read bulk IO.
839 * This will free all local buffers use by this read request.
841 * \param[in] env execution environment
842 * \param[in] ofd OFD device
843 * \param[in] fid object FID
844 * \param[in] objcount always 1
845 * \param[in] niocount number of local buffers
846 * \param[in] lnb local buffers
848 * \retval 0 on successful execution
849 * \retval negative value on error
852 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
853 const struct lu_fid *fid, int objcount, int niocount,
854 struct niobuf_local *lnb)
856 struct ofd_object *fo;
860 LASSERT(niocount > 0);
862 fo = ofd_object_find(env, ofd, fid);
866 LASSERT(ofd_object_exists(fo));
867 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
869 ofd_read_unlock(env, fo);
870 ofd_object_put(env, fo);
871 /* second put is pair to object_get in ofd_preprw_read */
872 ofd_object_put(env, fo);
878 * Set attributes of object during write bulk IO processing.
880 * Change object attributes and write parent FID into extended
881 * attributes when needed.
883 * \param[in] env execution environment
884 * \param[in] ofd OFD device
885 * \param[in] ofd_obj OFD object
886 * \param[in] la object attributes
889 * \retval 0 on successful attributes update
890 * \retval negative value on error
893 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
894 struct ofd_object *ofd_obj, struct lu_attr *la,
897 struct ofd_thread_info *info = ofd_info(env);
898 struct filter_fid *ff = &info->fti_mds_fid;
899 __u64 valid = la->la_valid;
901 struct dt_object *dt_obj;
909 dt_obj = ofd_object_child(ofd_obj);
910 LASSERT(dt_obj != NULL);
912 la->la_valid &= LA_UID | LA_GID | LA_PROJID;
914 rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
918 fl = ofd_object_ff_update(env, ofd_obj, oa, ff);
922 if (!la->la_valid && !fl)
923 /* no attributes to set */
926 th = ofd_trans_create(env, ofd);
928 GOTO(out, rc = PTR_ERR(th));
931 rc = dt_declare_attr_set(env, dt_obj, la, th);
937 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
938 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
939 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
940 le32_add_cpu(&ff->ff_parent.f_oid, -1);
942 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
943 XATTR_NAME_FID, 0, th);
948 /* We don't need a transno for this operation which will be re-executed
949 * anyway when the OST_WRITE (with a transno assigned) is replayed */
950 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
954 /* set uid/gid/projid */
956 rc = dt_attr_set(env, dt_obj, la, th);
961 /* set filter fid EA.
962 * FIXME: it holds read lock of ofd object to modify the XATTR_NAME_FID
963 * while the write lock should be held. However, it should work because
964 * write RPCs only modify ff_{parent,layout} and those information will
965 * be the same from all the write RPCs. The reason that fl is not used
966 * in dt_xattr_set() is to allow this race. */
968 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
971 info->fti_buf.lb_buf = ff;
972 info->fti_buf.lb_len = sizeof(*ff);
973 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
976 filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
982 dt_trans_stop(env, ofd->ofd_osd, th);
984 la->la_valid = valid;
988 struct ofd_soft_sync_callback {
989 struct dt_txn_commit_cb ossc_cb;
990 struct obd_export *ossc_exp;
994 * Callback function for "soft sync" update.
996 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
997 * See ofd_soft_sync_cb_add() below for more details on soft sync.
999 * \param[in] env execution environment
1000 * \param[in] th transaction handle
1001 * \param[in] cb callback data
1002 * \param[in] err error code
1004 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
1005 struct dt_txn_commit_cb *cb, int err)
1007 struct ofd_soft_sync_callback *ossc;
1009 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
1011 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
1012 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
1014 class_export_cb_put(ossc->ossc_exp);
1019 * Add callback for "soft sync" processing.
1021 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
1022 * flag is set in client buffers. The intention is for this operation to
1023 * commit pages belonging to a client which has "too many" outstanding
1024 * unstable pages in its cache. See LU-2139 for details.
1026 * This function adds callback to be called when commit is done.
1028 * \param[in] th transaction handle
1029 * \param[in] exp OBD export of client
1031 * \retval 0 on successful callback adding
1032 * \retval negative value on error
1034 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
1036 struct ofd_soft_sync_callback *ossc;
1037 struct dt_txn_commit_cb *dcb;
1040 OBD_ALLOC_PTR(ossc);
1044 ossc->ossc_exp = class_export_cb_get(exp);
1046 dcb = &ossc->ossc_cb;
1047 dcb->dcb_func = ofd_cb_soft_sync;
1048 INIT_LIST_HEAD(&dcb->dcb_linkage);
1049 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
1051 rc = dt_trans_cb_add(th, dcb);
1053 class_export_cb_put(exp);
1061 * Commit bulk IO buffers to the storage.
1063 * This function finalizes write IO processing by writing data to the disk.
1064 * That write can be synchronous or asynchronous depending on buffers flags.
1066 * \param[in] env execution environment
1067 * \param[in] exp OBD export of client
1068 * \param[in] ofd OFD device
1069 * \param[in] fid FID of object
1070 * \param[in] la object attributes
1071 * \param[in] ff parent FID of object
1072 * \param[in] objcount always 1
1073 * \param[in] niocount number of local buffers
1074 * \param[in] lnb local buffers
1075 * \param[in] granted grant space consumed for the bulk I/O
1076 * \param[in] old_rc result of processing at this point
1078 * \retval 0 on successful commit
1079 * \retval negative value on error
1082 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1083 struct ofd_device *ofd, const struct lu_fid *fid,
1084 struct lu_attr *la, struct obdo *oa, int objcount,
1085 int niocount, struct niobuf_local *lnb,
1086 unsigned long granted, int old_rc)
1088 struct filter_export_data *fed = &exp->exp_filter_data;
1089 struct ofd_object *fo;
1090 struct dt_object *o;
1096 bool soft_sync = false;
1097 bool cb_registered = false;
1098 bool fake_write = false;
1102 LASSERT(objcount == 1);
1104 fo = ofd_object_find(env, ofd, fid);
1105 LASSERT(fo != NULL);
1106 LASSERT(ofd_object_exists(fo));
1108 o = ofd_object_child(fo);
1112 GOTO(out, rc = old_rc);
1115 * The first write to each object must set some attributes. It is
1116 * important to set the uid/gid before calling
1117 * dt_declare_write_commit() since quota enforcement is now handled in
1120 rc = ofd_write_attr_set(env, ofd, fo, la, oa);
1124 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1126 /* do fake write, to simulate the write case for performance testing */
1127 if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) {
1128 struct niobuf_local *last = &lnb[niocount - 1];
1129 __u64 file_size = last->lnb_file_offset + last->lnb_len;
1130 __u64 valid = la->la_valid;
1132 la->la_valid = LA_SIZE;
1134 rc = dt_attr_get(env, o, la);
1135 if (rc < 0 && rc != -ENOENT)
1138 if (file_size < la->la_size)
1139 file_size = la->la_size;
1141 /* dirty inode by setting file size */
1142 la->la_valid = valid | LA_SIZE;
1143 la->la_size = file_size;
1149 th = ofd_trans_create(env, ofd);
1151 GOTO(out, rc = PTR_ERR(th));
1153 th->th_sync |= ofd->ofd_sync_journal;
1154 if (th->th_sync == 0) {
1155 for (i = 0; i < niocount; i++) {
1156 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1160 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1165 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1166 GOTO(out_stop, rc = -EINPROGRESS);
1168 if (likely(!fake_write)) {
1169 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1175 /* update [mac]time if needed */
1176 rc = dt_declare_attr_set(env, o, la, th);
1181 rc = ofd_trans_start(env, ofd, fo, th);
1185 if (likely(!fake_write)) {
1186 rc = dt_write_commit(env, o, lnb, niocount, th);
1192 rc = dt_attr_set(env, o, la, th);
1197 /* get attr to return */
1198 rc = dt_attr_get(env, o, la);
1201 /* Force commit to make the just-deleted blocks
1202 * reusable. LU-456 */
1206 /* do this before trans stop in case commit has finished */
1207 if (!th->th_sync && soft_sync && !cb_registered) {
1208 ofd_soft_sync_cb_add(th, exp);
1209 cb_registered = true;
1212 if (rc == 0 && granted > 0) {
1213 if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
1217 rc2 = ofd_trans_stop(env, ofd, th, rc);
1220 if (rc == -ENOSPC && retries++ < 3) {
1221 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1227 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1228 atomic_set(&fed->fed_soft_sync_count, 0);
1229 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1230 ofd->ofd_soft_sync_limit)
1231 dt_commit_async(env, ofd->ofd_osd);
1234 dt_bufs_put(env, o, lnb, niocount);
1235 ofd_read_unlock(env, fo);
1236 ofd_object_put(env, fo);
1237 /* second put is pair to object_get in ofd_preprw_write */
1238 ofd_object_put(env, fo);
1240 tgt_grant_commit(exp, granted, old_rc);
1245 * Commit bulk IO to the storage.
1247 * This is companion function to the ofd_preprw(). It finishes bulk IO
1248 * request processing by committing buffers to the storage (WRITE) and/or
1249 * freeing those buffers (read/write). See ofd_commitrw_read() and
1250 * ofd_commitrw_write() for details about each type of IO.
1252 * \param[in] env execution environment
1253 * \param[in] cmd IO type (READ/WRITE)
1254 * \param[in] exp OBD export of client
1255 * \param[in] oa OBDO structure from client
1256 * \param[in] objcount always 1
1257 * \param[in] obj object data
1258 * \param[in] rnb remote buffers
1259 * \param[in] npages number of local buffers
1260 * \param[in] lnb local buffers
1261 * \param[in] old_rc result of processing at this point
1263 * \retval 0 on successful commit
1264 * \retval negative value on error
1266 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1267 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1268 struct niobuf_remote *rnb, int npages,
1269 struct niobuf_local *lnb, int old_rc)
1271 struct ofd_thread_info *info = ofd_info(env);
1272 struct ofd_mod_data *fmd;
1274 struct ofd_device *ofd = ofd_exp(exp);
1275 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1278 LASSERT(npages > 0);
1280 if (cmd == OBD_BRW_WRITE) {
1281 struct lu_nodemap *nodemap;
1283 /* Don't update timestamps if this write is older than a
1284 * setattr which modifies the timestamps. b=10150 */
1286 /* XXX when we start having persistent reservations this needs
1287 * to be changed to ofd_fmd_get() to create the fmd if it
1288 * doesn't already exist so we can store the reservation handle
1290 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID;
1291 fmd = ofd_fmd_find(exp, fid);
1292 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1293 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1295 ofd_fmd_put(exp, fmd);
1296 la_from_obdo(&info->fti_attr, oa, valid);
1298 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1299 oa, objcount, npages, lnb,
1300 oa->o_grant_used, old_rc);
1302 obdo_from_la(oa, &info->fti_attr,
1303 OFD_VALID_FLAGS | LA_GID | LA_UID |
1306 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID |
1309 /* don't report overquota flag if we failed before reaching
1311 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1312 /* return the overquota flags to client */
1313 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1314 if (oa->o_valid & OBD_MD_FLFLAGS)
1315 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1317 oa->o_flags = OBD_FL_NO_USRQUOTA;
1320 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1321 if (oa->o_valid & OBD_MD_FLFLAGS)
1322 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1324 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1326 if (lnb[0].lnb_flags & OBD_BRW_OVER_PRJQUOTA) {
1327 if (oa->o_valid & OBD_MD_FLFLAGS)
1328 oa->o_flags |= OBD_FL_NO_PRJQUOTA;
1330 oa->o_flags = OBD_FL_NO_PRJQUOTA;
1333 oa->o_valid |= OBD_MD_FLFLAGS;
1334 oa->o_valid |= OBD_MD_FLALLQUOTA;
1337 /* Convert back to client IDs. LU-9671.
1338 * nodemap_get_from_exp() may fail due to nodemap deactivated,
1339 * server ID will be returned back to client in that case. */
1340 nodemap = nodemap_get_from_exp(exp);
1341 if (nodemap != NULL && !IS_ERR(nodemap)) {
1342 oa->o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1343 NODEMAP_FS_TO_CLIENT,
1345 oa->o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1346 NODEMAP_FS_TO_CLIENT,
1348 nodemap_putref(nodemap);
1350 } else if (cmd == OBD_BRW_READ) {
1351 struct ldlm_namespace *ns = ofd->ofd_namespace;
1353 /* If oa != NULL then ofd_preprw_read updated the inode
1354 * atime and we should update the lvb so that other glimpses
1355 * will also get the updated value. bug 5972 */
1356 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1357 struct ldlm_resource *rs = NULL;
1359 ost_fid_build_resid(fid, &info->fti_resid);
1360 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1363 ldlm_res_lvbo_update(env, rs, NULL, 1);
1364 ldlm_resource_putref(rs);
1367 rc = ofd_commitrw_read(env, ofd, fid, objcount,