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, 2016, 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)
446 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_OST_SKIP_LV_CHECK)))
449 rc = ofd_object_ff_load(env, fo);
456 /* this update is not legitimate */
457 if (oa->o_layout_version < fo->ofo_ff.ff_layout_version)
458 GOTO(out, rc = -ESTALE);
460 /* layout version is not transmitted yet */
461 if (oa->o_layout_version >
462 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range)
463 GOTO(out, rc = -EINPROGRESS);
468 CDEBUG(D_INODE, DFID " verify layout version: %u vs. %u, rc: %d\n",
469 PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
470 fo->ofo_ff.ff_layout_version, oa->o_layout_version, rc);
476 * Prepare buffers for read request processing.
478 * This function converts remote buffers from client to local buffers
479 * and prepares the latter.
481 * \param[in] env execution environment
482 * \param[in] exp OBD export of client
483 * \param[in] ofd OFD device
484 * \param[in] fid FID of object
485 * \param[in] la object attributes
486 * \param[in] oa OBDO structure from client
487 * \param[in] niocount number of remote buffers
488 * \param[in] rnb remote buffers
489 * \param[in] nr_local number of local buffers
490 * \param[in] lnb local buffers
491 * \param[in] jobid job ID name
493 * \retval 0 on successful prepare
494 * \retval negative value on error
496 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
497 struct ofd_device *ofd, const struct lu_fid *fid,
498 struct lu_attr *la, struct obdo *oa, int niocount,
499 struct niobuf_remote *rnb, int *nr_local,
500 struct niobuf_local *lnb, char *jobid)
502 struct ofd_object *fo;
503 int i, j, rc, tot_bytes = 0;
504 enum dt_bufs_type dbt = DT_BUFS_TYPE_READ;
507 LASSERT(env != NULL);
509 fo = ofd_object_find(env, ofd, fid);
514 ofd_read_lock(env, fo);
515 if (!ofd_object_exists(fo))
516 GOTO(unlock, rc = -ENOENT);
518 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
519 rc = ofd_verify_ff(env, fo, oa);
524 if (ptlrpc_connection_is_local(exp->exp_connection))
525 dbt |= DT_BUFS_TYPE_LOCAL;
527 for (*nr_local = 0, i = 0, j = 0; i < niocount; i++) {
528 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
530 if (unlikely(rc < 0))
532 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
533 /* correct index for local buffers to continue with */
536 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
537 tot_bytes += rnb[i].rnb_len;
540 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
541 rc = dt_attr_get(env, ofd_object_child(fo), la);
545 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
549 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
553 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
555 ofd_read_unlock(env, fo);
556 ofd_object_put(env, fo);
561 * Prepare buffers for write request processing.
563 * This function converts remote buffers from client to local buffers
564 * and prepares the latter. If there is recovery in progress and required
565 * object is missing then it can be re-created before write.
567 * \param[in] env execution environment
568 * \param[in] exp OBD export of client
569 * \param[in] ofd OFD device
570 * \param[in] fid FID of object
571 * \param[in] la object attributes
572 * \param[in] oa OBDO structure from client
573 * \param[in] objcount always 1
574 * \param[in] obj object data
575 * \param[in] rnb remote buffers
576 * \param[in] nr_local number of local buffers
577 * \param[in] lnb local buffers
578 * \param[in] jobid job ID name
580 * \retval 0 on successful prepare
581 * \retval negative value on error
583 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
584 struct ofd_device *ofd, const struct lu_fid *fid,
585 struct lu_attr *la, struct obdo *oa,
586 int objcount, struct obd_ioobj *obj,
587 struct niobuf_remote *rnb, int *nr_local,
588 struct niobuf_local *lnb, char *jobid)
590 struct ofd_object *fo;
591 int i, j, k, rc = 0, tot_bytes = 0;
592 enum dt_bufs_type dbt = DT_BUFS_TYPE_WRITE;
595 LASSERT(env != NULL);
596 LASSERT(objcount == 1);
598 if (unlikely(exp->exp_obd->obd_recovering)) {
599 u64 seq = fid_seq(fid);
600 u64 oid = fid_oid(fid);
601 struct ofd_seq *oseq;
603 oseq = ofd_seq_load(env, ofd, seq);
605 CERROR("%s: Can't find FID Sequence %#llx: rc = %d\n",
606 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
607 GOTO(out, rc = -EINVAL);
610 if (oid > ofd_seq_last_oid(oseq)) {
614 mutex_lock(&oseq->os_create_lock);
615 diff = oid - ofd_seq_last_oid(oseq);
617 /* Do sync create if the seq is about to used up */
618 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
619 if (unlikely(oid >= IDIF_MAX_OID - 1))
621 } else if (fid_seq_is_norm(seq)) {
623 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
626 CERROR("%s : invalid o_seq "DOSTID"\n",
627 ofd_name(ofd), POSTID(&oa->o_oi));
628 mutex_unlock(&oseq->os_create_lock);
629 ofd_seq_put(env, oseq);
630 GOTO(out, rc = -EINVAL);
634 u64 next_id = ofd_seq_last_oid(oseq) + 1;
635 int count = ofd_precreate_batch(ofd, diff);
637 rc = ofd_precreate_objects(env, ofd, next_id,
640 mutex_unlock(&oseq->os_create_lock);
641 ofd_seq_put(env, oseq);
648 mutex_unlock(&oseq->os_create_lock);
651 ofd_seq_put(env, oseq);
654 fo = ofd_object_find(env, ofd, fid);
656 GOTO(out, rc = PTR_ERR(fo));
659 ofd_read_lock(env, fo);
660 if (!ofd_object_exists(fo)) {
661 CERROR("%s: BRW to missing obj "DOSTID"\n",
662 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
663 ofd_read_unlock(env, fo);
664 ofd_object_put(env, fo);
665 GOTO(out, rc = -ENOENT);
668 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
669 rc = ofd_verify_ff(env, fo, oa);
671 ofd_read_unlock(env, fo);
672 ofd_object_put(env, fo);
677 /* need to verify layout version */
678 if (oa->o_valid & OBD_MD_LAYOUT_VERSION) {
679 rc = ofd_verify_layout_version(env, fo, oa);
681 ofd_read_unlock(env, fo);
682 ofd_object_put(env, fo);
686 oa->o_valid &= ~OBD_MD_LAYOUT_VERSION;
689 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
690 * space back if possible */
691 tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
693 if (ptlrpc_connection_is_local(exp->exp_connection))
694 dbt |= DT_BUFS_TYPE_LOCAL;
696 /* parse remote buffers to local buffers and prepare the latter */
697 for (*nr_local = 0, i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
698 rc = dt_bufs_get(env, ofd_object_child(fo),
699 rnb + i, lnb + j, dbt);
700 if (unlikely(rc < 0))
702 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
703 /* correct index for local buffers to continue with */
704 for (k = 0; k < rc; k++) {
705 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
706 lnb[j+k].lnb_flags &= ~OBD_BRW_LOCALS;
707 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
708 lnb[j+k].lnb_rc = -ENOSPC;
712 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
713 tot_bytes += rnb[i].rnb_len;
715 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
717 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
718 if (unlikely(rc != 0))
721 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
724 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
725 ofd_read_unlock(env, fo);
726 ofd_object_put(env, fo);
727 /* tgt_grant_prepare_write() was called, so we must commit */
728 tgt_grant_commit(exp, oa->o_grant_used, rc);
730 /* let's still process incoming grant information packed in the oa,
731 * but without enforcing grant since we won't proceed with the write.
732 * Just like a read request actually. */
733 tgt_grant_prepare_read(env, exp, oa);
738 * Prepare bulk IO requests for processing.
740 * This function does initial checks of IO and calls corresponding
741 * functions for read/write processing.
743 * \param[in] env execution environment
744 * \param[in] cmd IO type (read/write)
745 * \param[in] exp OBD export of client
746 * \param[in] oa OBDO structure from request
747 * \param[in] objcount always 1
748 * \param[in] obj object data
749 * \param[in] rnb remote buffers
750 * \param[in] nr_local number of local buffers
751 * \param[in] lnb local buffers
753 * \retval 0 on successful prepare
754 * \retval negative value on error
756 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
757 struct obdo *oa, int objcount, struct obd_ioobj *obj,
758 struct niobuf_remote *rnb, int *nr_local,
759 struct niobuf_local *lnb)
761 struct tgt_session_info *tsi = tgt_ses_info(env);
762 struct ofd_device *ofd = ofd_exp(exp);
763 struct ofd_thread_info *info;
765 const struct lu_fid *fid = &oa->o_oi.oi_fid;
768 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
769 CERROR("%s: bulk has too many pages %d, which exceeds the"
770 "maximum pages per RPC of %d\n",
771 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
775 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
776 info = ofd_info_init(env, exp);
779 info = tsi2ofd_info(tsi);
780 jobid = tsi->tsi_jobid;
785 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
786 struct ofd_seq *oseq;
788 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
790 CERROR("%s: Can not find seq for "DOSTID
791 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
796 if (oseq->os_destroys_in_progress == 0) {
797 /* don't fail lookups for orphan recovery, it causes
798 * later LBUGs when objects still exist during
800 ofd_seq_put(env, oseq);
803 ofd_seq_put(env, oseq);
806 LASSERT(objcount == 1);
807 LASSERT(obj->ioo_bufcnt > 0);
809 if (cmd == OBD_BRW_WRITE) {
810 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
811 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
812 objcount, obj, rnb, nr_local, lnb, jobid);
813 } else if (cmd == OBD_BRW_READ) {
814 tgt_grant_prepare_read(env, exp, oa);
815 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
816 obj->ioo_bufcnt, rnb, nr_local, lnb,
818 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
820 CERROR("%s: wrong cmd %d received!\n",
821 exp->exp_obd->obd_name, cmd);
828 * Drop reference on local buffers for read bulk IO.
830 * This will free all local buffers use by this read request.
832 * \param[in] env execution environment
833 * \param[in] ofd OFD device
834 * \param[in] fid object FID
835 * \param[in] objcount always 1
836 * \param[in] niocount number of local buffers
837 * \param[in] lnb local buffers
839 * \retval 0 on successful execution
840 * \retval negative value on error
843 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
844 const struct lu_fid *fid, int objcount, int niocount,
845 struct niobuf_local *lnb)
847 struct ofd_object *fo;
851 LASSERT(niocount > 0);
853 fo = ofd_object_find(env, ofd, fid);
857 LASSERT(ofd_object_exists(fo));
858 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
860 ofd_read_unlock(env, fo);
861 ofd_object_put(env, fo);
862 /* second put is pair to object_get in ofd_preprw_read */
863 ofd_object_put(env, fo);
869 * Set attributes of object during write bulk IO processing.
871 * Change object attributes and write parent FID into extended
872 * attributes when needed.
874 * \param[in] env execution environment
875 * \param[in] ofd OFD device
876 * \param[in] ofd_obj OFD object
877 * \param[in] la object attributes
880 * \retval 0 on successful attributes update
881 * \retval negative value on error
884 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
885 struct ofd_object *ofd_obj, struct lu_attr *la,
888 struct ofd_thread_info *info = ofd_info(env);
889 struct filter_fid *ff = &info->fti_mds_fid;
890 __u64 valid = la->la_valid;
892 struct dt_object *dt_obj;
900 dt_obj = ofd_object_child(ofd_obj);
901 LASSERT(dt_obj != NULL);
903 la->la_valid &= LA_UID | LA_GID | LA_PROJID;
905 rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
909 fl = ofd_object_ff_update(env, ofd_obj, oa, ff);
913 if (!la->la_valid && !fl)
914 /* no attributes to set */
917 th = ofd_trans_create(env, ofd);
919 GOTO(out, rc = PTR_ERR(th));
922 rc = dt_declare_attr_set(env, dt_obj, la, th);
928 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
929 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
930 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
931 le32_add_cpu(&ff->ff_parent.f_oid, -1);
933 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
934 XATTR_NAME_FID, 0, th);
939 /* We don't need a transno for this operation which will be re-executed
940 * anyway when the OST_WRITE (with a transno assigned) is replayed */
941 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
945 /* set uid/gid/projid */
947 rc = dt_attr_set(env, dt_obj, la, th);
952 /* set filter fid EA.
953 * FIXME: it holds read lock of ofd object to modify the XATTR_NAME_FID
954 * while the write lock should be held. However, it should work because
955 * write RPCs only modify ff_{parent,layout} and those information will
956 * be the same from all the write RPCs. The reason that fl is not used
957 * in dt_xattr_set() is to allow this race. */
959 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
962 info->fti_buf.lb_buf = ff;
963 info->fti_buf.lb_len = sizeof(*ff);
964 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
967 filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
973 dt_trans_stop(env, ofd->ofd_osd, th);
975 la->la_valid = valid;
979 struct ofd_soft_sync_callback {
980 struct dt_txn_commit_cb ossc_cb;
981 struct obd_export *ossc_exp;
985 * Callback function for "soft sync" update.
987 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
988 * See ofd_soft_sync_cb_add() below for more details on soft sync.
990 * \param[in] env execution environment
991 * \param[in] th transaction handle
992 * \param[in] cb callback data
993 * \param[in] err error code
995 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
996 struct dt_txn_commit_cb *cb, int err)
998 struct ofd_soft_sync_callback *ossc;
1000 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
1002 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
1003 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
1005 class_export_cb_put(ossc->ossc_exp);
1010 * Add callback for "soft sync" processing.
1012 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
1013 * flag is set in client buffers. The intention is for this operation to
1014 * commit pages belonging to a client which has "too many" outstanding
1015 * unstable pages in its cache. See LU-2139 for details.
1017 * This function adds callback to be called when commit is done.
1019 * \param[in] th transaction handle
1020 * \param[in] exp OBD export of client
1022 * \retval 0 on successful callback adding
1023 * \retval negative value on error
1025 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
1027 struct ofd_soft_sync_callback *ossc;
1028 struct dt_txn_commit_cb *dcb;
1031 OBD_ALLOC_PTR(ossc);
1035 ossc->ossc_exp = class_export_cb_get(exp);
1037 dcb = &ossc->ossc_cb;
1038 dcb->dcb_func = ofd_cb_soft_sync;
1039 INIT_LIST_HEAD(&dcb->dcb_linkage);
1040 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
1042 rc = dt_trans_cb_add(th, dcb);
1044 class_export_cb_put(exp);
1052 * Commit bulk IO buffers to the storage.
1054 * This function finalizes write IO processing by writing data to the disk.
1055 * That write can be synchronous or asynchronous depending on buffers flags.
1057 * \param[in] env execution environment
1058 * \param[in] exp OBD export of client
1059 * \param[in] ofd OFD device
1060 * \param[in] fid FID of object
1061 * \param[in] la object attributes
1062 * \param[in] ff parent FID of object
1063 * \param[in] objcount always 1
1064 * \param[in] niocount number of local buffers
1065 * \param[in] lnb local buffers
1066 * \param[in] granted grant space consumed for the bulk I/O
1067 * \param[in] old_rc result of processing at this point
1069 * \retval 0 on successful commit
1070 * \retval negative value on error
1073 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1074 struct ofd_device *ofd, const struct lu_fid *fid,
1075 struct lu_attr *la, struct obdo *oa, int objcount,
1076 int niocount, struct niobuf_local *lnb,
1077 unsigned long granted, int old_rc)
1079 struct filter_export_data *fed = &exp->exp_filter_data;
1080 struct ofd_object *fo;
1081 struct dt_object *o;
1087 bool soft_sync = false;
1088 bool cb_registered = false;
1089 bool fake_write = false;
1093 LASSERT(objcount == 1);
1095 fo = ofd_object_find(env, ofd, fid);
1096 LASSERT(fo != NULL);
1097 LASSERT(ofd_object_exists(fo));
1099 o = ofd_object_child(fo);
1103 GOTO(out, rc = old_rc);
1106 * The first write to each object must set some attributes. It is
1107 * important to set the uid/gid before calling
1108 * dt_declare_write_commit() since quota enforcement is now handled in
1111 rc = ofd_write_attr_set(env, ofd, fo, la, oa);
1115 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1117 /* do fake write, to simulate the write case for performance testing */
1118 if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) {
1119 struct niobuf_local *last = &lnb[niocount - 1];
1120 __u64 file_size = last->lnb_file_offset + last->lnb_len;
1121 __u64 valid = la->la_valid;
1123 la->la_valid = LA_SIZE;
1125 rc = dt_attr_get(env, o, la);
1126 if (rc < 0 && rc != -ENOENT)
1129 if (file_size < la->la_size)
1130 file_size = la->la_size;
1132 /* dirty inode by setting file size */
1133 la->la_valid = valid | LA_SIZE;
1134 la->la_size = file_size;
1140 th = ofd_trans_create(env, ofd);
1142 GOTO(out, rc = PTR_ERR(th));
1144 th->th_sync |= ofd->ofd_syncjournal;
1145 if (th->th_sync == 0) {
1146 for (i = 0; i < niocount; i++) {
1147 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1151 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1156 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1157 GOTO(out_stop, rc = -EINPROGRESS);
1159 if (likely(!fake_write)) {
1160 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1166 /* update [mac]time if needed */
1167 rc = dt_declare_attr_set(env, o, la, th);
1172 rc = ofd_trans_start(env, ofd, fo, th);
1176 if (likely(!fake_write)) {
1177 rc = dt_write_commit(env, o, lnb, niocount, th);
1183 rc = dt_attr_set(env, o, la, th);
1188 /* get attr to return */
1189 rc = dt_attr_get(env, o, la);
1192 /* Force commit to make the just-deleted blocks
1193 * reusable. LU-456 */
1197 /* do this before trans stop in case commit has finished */
1198 if (!th->th_sync && soft_sync && !cb_registered) {
1199 ofd_soft_sync_cb_add(th, exp);
1200 cb_registered = true;
1203 if (rc == 0 && granted > 0) {
1204 if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
1208 rc2 = ofd_trans_stop(env, ofd, th, rc);
1211 if (rc == -ENOSPC && retries++ < 3) {
1212 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1218 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1219 atomic_set(&fed->fed_soft_sync_count, 0);
1220 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1221 ofd->ofd_soft_sync_limit)
1222 dt_commit_async(env, ofd->ofd_osd);
1225 dt_bufs_put(env, o, lnb, niocount);
1226 ofd_read_unlock(env, fo);
1227 ofd_object_put(env, fo);
1228 /* second put is pair to object_get in ofd_preprw_write */
1229 ofd_object_put(env, fo);
1231 tgt_grant_commit(exp, granted, old_rc);
1236 * Commit bulk IO to the storage.
1238 * This is companion function to the ofd_preprw(). It finishes bulk IO
1239 * request processing by committing buffers to the storage (WRITE) and/or
1240 * freeing those buffers (read/write). See ofd_commitrw_read() and
1241 * ofd_commitrw_write() for details about each type of IO.
1243 * \param[in] env execution environment
1244 * \param[in] cmd IO type (READ/WRITE)
1245 * \param[in] exp OBD export of client
1246 * \param[in] oa OBDO structure from client
1247 * \param[in] objcount always 1
1248 * \param[in] obj object data
1249 * \param[in] rnb remote buffers
1250 * \param[in] npages number of local buffers
1251 * \param[in] lnb local buffers
1252 * \param[in] old_rc result of processing at this point
1254 * \retval 0 on successful commit
1255 * \retval negative value on error
1257 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1258 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1259 struct niobuf_remote *rnb, int npages,
1260 struct niobuf_local *lnb, int old_rc)
1262 struct ofd_thread_info *info = ofd_info(env);
1263 struct ofd_mod_data *fmd;
1265 struct ofd_device *ofd = ofd_exp(exp);
1266 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1269 LASSERT(npages > 0);
1271 if (cmd == OBD_BRW_WRITE) {
1272 struct lu_nodemap *nodemap;
1274 /* Don't update timestamps if this write is older than a
1275 * setattr which modifies the timestamps. b=10150 */
1277 /* XXX when we start having persistent reservations this needs
1278 * to be changed to ofd_fmd_get() to create the fmd if it
1279 * doesn't already exist so we can store the reservation handle
1281 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID;
1282 fmd = ofd_fmd_find(exp, fid);
1283 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1284 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1286 ofd_fmd_put(exp, fmd);
1287 la_from_obdo(&info->fti_attr, oa, valid);
1289 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1290 oa, objcount, npages, lnb,
1291 oa->o_grant_used, old_rc);
1293 obdo_from_la(oa, &info->fti_attr,
1294 OFD_VALID_FLAGS | LA_GID | LA_UID |
1297 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID |
1300 /* don't report overquota flag if we failed before reaching
1302 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1303 /* return the overquota flags to client */
1304 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1305 if (oa->o_valid & OBD_MD_FLFLAGS)
1306 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1308 oa->o_flags = OBD_FL_NO_USRQUOTA;
1311 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1312 if (oa->o_valid & OBD_MD_FLFLAGS)
1313 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1315 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1317 if (lnb[0].lnb_flags & OBD_BRW_OVER_PRJQUOTA) {
1318 if (oa->o_valid & OBD_MD_FLFLAGS)
1319 oa->o_flags |= OBD_FL_NO_PRJQUOTA;
1321 oa->o_flags = OBD_FL_NO_PRJQUOTA;
1324 oa->o_valid |= OBD_MD_FLFLAGS;
1325 oa->o_valid |= OBD_MD_FLALLQUOTA;
1328 /* Convert back to client IDs. LU-9671.
1329 * nodemap_get_from_exp() may fail due to nodemap deactivated,
1330 * server ID will be returned back to client in that case. */
1331 nodemap = nodemap_get_from_exp(exp);
1332 if (nodemap != NULL && !IS_ERR(nodemap)) {
1333 oa->o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1334 NODEMAP_FS_TO_CLIENT,
1336 oa->o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1337 NODEMAP_FS_TO_CLIENT,
1339 nodemap_putref(nodemap);
1341 } else if (cmd == OBD_BRW_READ) {
1342 struct ldlm_namespace *ns = ofd->ofd_namespace;
1344 /* If oa != NULL then ofd_preprw_read updated the inode
1345 * atime and we should update the lvb so that other glimpses
1346 * will also get the updated value. bug 5972 */
1347 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1348 struct ldlm_resource *rs = NULL;
1350 ost_fid_build_resid(fid, &info->fti_resid);
1351 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1354 ldlm_res_lvbo_update(rs, NULL, 1);
1355 ldlm_resource_putref(rs);
1358 rc = ofd_commitrw_read(env, ofd, fid, objcount,