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;
514 int maxlnb = *nr_local;
517 LASSERT(env != NULL);
519 fo = ofd_object_find(env, ofd, fid);
524 ofd_read_lock(env, fo);
525 if (!ofd_object_exists(fo))
526 GOTO(unlock, rc = -ENOENT);
528 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
529 rc = ofd_verify_ff(env, fo, oa);
534 if (ptlrpc_connection_is_local(exp->exp_connection))
535 dbt |= DT_BUFS_TYPE_LOCAL;
537 for (*nr_local = 0, i = 0, j = 0; i < niocount; i++) {
539 if (OBD_FAIL_CHECK(OBD_FAIL_OST_2BIG_NIOBUF))
540 rnb[i].rnb_len = 100 * 1024 * 1024;
542 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
543 lnb + j, maxlnb, dbt);
544 if (unlikely(rc < 0))
546 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
547 /* correct index for local buffers to continue with */
551 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
552 tot_bytes += rnb[i].rnb_len;
555 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
556 rc = dt_attr_get(env, ofd_object_child(fo), la);
560 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
564 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
568 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
570 ofd_read_unlock(env, fo);
571 ofd_object_put(env, fo);
576 * Prepare buffers for write request processing.
578 * This function converts remote buffers from client to local buffers
579 * and prepares the latter. If there is recovery in progress and required
580 * object is missing then it can be re-created before write.
582 * \param[in] env execution environment
583 * \param[in] exp OBD export of client
584 * \param[in] ofd OFD device
585 * \param[in] fid FID of object
586 * \param[in] la object attributes
587 * \param[in] oa OBDO structure from client
588 * \param[in] objcount always 1
589 * \param[in] obj object data
590 * \param[in] rnb remote buffers
591 * \param[in] nr_local number of local buffers
592 * \param[in] lnb local buffers
593 * \param[in] jobid job ID name
595 * \retval 0 on successful prepare
596 * \retval negative value on error
598 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
599 struct ofd_device *ofd, const struct lu_fid *fid,
600 struct lu_attr *la, struct obdo *oa,
601 int objcount, struct obd_ioobj *obj,
602 struct niobuf_remote *rnb, int *nr_local,
603 struct niobuf_local *lnb, char *jobid)
605 struct ofd_object *fo;
606 int i, j, k, rc = 0, tot_bytes = 0;
607 enum dt_bufs_type dbt = DT_BUFS_TYPE_WRITE;
608 int maxlnb = *nr_local;
611 LASSERT(env != NULL);
612 LASSERT(objcount == 1);
614 if (unlikely(exp->exp_obd->obd_recovering)) {
615 u64 seq = fid_seq(fid);
616 u64 oid = fid_oid(fid);
617 struct ofd_seq *oseq;
619 oseq = ofd_seq_load(env, ofd, seq);
621 CERROR("%s: Can't find FID Sequence %#llx: rc = %d\n",
622 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
623 GOTO(out, rc = -EINVAL);
626 if (oid > ofd_seq_last_oid(oseq)) {
630 mutex_lock(&oseq->os_create_lock);
631 diff = oid - ofd_seq_last_oid(oseq);
633 /* Do sync create if the seq is about to used up */
634 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
635 if (unlikely(oid >= IDIF_MAX_OID - 1))
637 } else if (fid_seq_is_norm(seq)) {
639 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
642 CERROR("%s : invalid o_seq "DOSTID"\n",
643 ofd_name(ofd), POSTID(&oa->o_oi));
644 mutex_unlock(&oseq->os_create_lock);
645 ofd_seq_put(env, oseq);
646 GOTO(out, rc = -EINVAL);
650 u64 next_id = ofd_seq_last_oid(oseq) + 1;
651 int count = ofd_precreate_batch(ofd, diff);
653 rc = ofd_precreate_objects(env, ofd, next_id,
656 mutex_unlock(&oseq->os_create_lock);
657 ofd_seq_put(env, oseq);
664 mutex_unlock(&oseq->os_create_lock);
667 ofd_seq_put(env, oseq);
670 fo = ofd_object_find(env, ofd, fid);
672 GOTO(out, rc = PTR_ERR(fo));
675 ofd_read_lock(env, fo);
676 if (!ofd_object_exists(fo)) {
677 CERROR("%s: BRW to missing obj "DOSTID"\n",
678 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
679 ofd_read_unlock(env, fo);
680 ofd_object_put(env, fo);
681 GOTO(out, rc = -ENOENT);
684 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
685 rc = ofd_verify_ff(env, fo, oa);
687 ofd_read_unlock(env, fo);
688 ofd_object_put(env, fo);
693 /* need to verify layout version */
694 if (oa->o_valid & OBD_MD_LAYOUT_VERSION) {
695 rc = ofd_verify_layout_version(env, fo, oa);
697 ofd_read_unlock(env, fo);
698 ofd_object_put(env, fo);
702 oa->o_valid &= ~OBD_MD_LAYOUT_VERSION;
705 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
706 * space back if possible */
707 tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
709 if (ptlrpc_connection_is_local(exp->exp_connection))
710 dbt |= DT_BUFS_TYPE_LOCAL;
712 /* parse remote buffers to local buffers and prepare the latter */
713 for (*nr_local = 0, i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
714 if (OBD_FAIL_CHECK(OBD_FAIL_OST_2BIG_NIOBUF))
715 rnb[i].rnb_len += PAGE_SIZE;
716 rc = dt_bufs_get(env, ofd_object_child(fo),
717 rnb + i, lnb + j, maxlnb, dbt);
718 if (unlikely(rc < 0))
720 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
721 /* correct index for local buffers to continue with */
722 for (k = 0; k < rc; k++) {
723 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
724 lnb[j+k].lnb_flags &= ~OBD_BRW_LOCALS;
725 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
726 lnb[j+k].lnb_rc = -ENOSPC;
731 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
732 tot_bytes += rnb[i].rnb_len;
734 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
736 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
737 if (unlikely(rc != 0))
740 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
743 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
744 ofd_read_unlock(env, fo);
745 ofd_object_put(env, fo);
746 /* tgt_grant_prepare_write() was called, so we must commit */
747 tgt_grant_commit(exp, oa->o_grant_used, rc);
749 /* let's still process incoming grant information packed in the oa,
750 * but without enforcing grant since we won't proceed with the write.
751 * Just like a read request actually. */
752 tgt_grant_prepare_read(env, exp, oa);
757 * Prepare bulk IO requests for processing.
759 * This function does initial checks of IO and calls corresponding
760 * functions for read/write processing.
762 * \param[in] env execution environment
763 * \param[in] cmd IO type (read/write)
764 * \param[in] exp OBD export of client
765 * \param[in] oa OBDO structure from request
766 * \param[in] objcount always 1
767 * \param[in] obj object data
768 * \param[in] rnb remote buffers
769 * \param[in] nr_local number of local buffers
770 * \param[in] lnb local buffers
772 * \retval 0 on successful prepare
773 * \retval negative value on error
775 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
776 struct obdo *oa, int objcount, struct obd_ioobj *obj,
777 struct niobuf_remote *rnb, int *nr_local,
778 struct niobuf_local *lnb)
780 struct tgt_session_info *tsi = tgt_ses_info(env);
781 struct ofd_device *ofd = ofd_exp(exp);
782 struct ofd_thread_info *info;
784 const struct lu_fid *fid = &oa->o_oi.oi_fid;
787 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
788 CERROR("%s: bulk has too many pages %d, which exceeds the"
789 "maximum pages per RPC of %d\n",
790 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
794 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
795 info = ofd_info_init(env, exp);
798 info = tsi2ofd_info(tsi);
799 jobid = tsi->tsi_jobid;
804 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
805 struct ofd_seq *oseq;
807 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
809 CERROR("%s: Can not find seq for "DOSTID
810 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
815 if (oseq->os_destroys_in_progress == 0) {
816 /* don't fail lookups for orphan recovery, it causes
817 * later LBUGs when objects still exist during
819 ofd_seq_put(env, oseq);
822 ofd_seq_put(env, oseq);
825 LASSERT(objcount == 1);
826 LASSERT(obj->ioo_bufcnt > 0);
828 if (cmd == OBD_BRW_WRITE) {
829 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
830 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
831 objcount, obj, rnb, nr_local, lnb, jobid);
832 } else if (cmd == OBD_BRW_READ) {
833 tgt_grant_prepare_read(env, exp, oa);
834 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
835 obj->ioo_bufcnt, rnb, nr_local, lnb,
837 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
839 CERROR("%s: wrong cmd %d received!\n",
840 exp->exp_obd->obd_name, cmd);
847 * Drop reference on local buffers for read bulk IO.
849 * This will free all local buffers use by this read request.
851 * \param[in] env execution environment
852 * \param[in] ofd OFD device
853 * \param[in] fid object FID
854 * \param[in] objcount always 1
855 * \param[in] niocount number of local buffers
856 * \param[in] lnb local buffers
858 * \retval 0 on successful execution
859 * \retval negative value on error
862 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
863 const struct lu_fid *fid, int objcount, int niocount,
864 struct niobuf_local *lnb)
866 struct ofd_object *fo;
870 LASSERT(niocount > 0);
872 fo = ofd_object_find(env, ofd, fid);
876 LASSERT(ofd_object_exists(fo));
877 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
879 ofd_read_unlock(env, fo);
880 ofd_object_put(env, fo);
881 /* second put is pair to object_get in ofd_preprw_read */
882 ofd_object_put(env, fo);
888 * Set attributes of object during write bulk IO processing.
890 * Change object attributes and write parent FID into extended
891 * attributes when needed.
893 * \param[in] env execution environment
894 * \param[in] ofd OFD device
895 * \param[in] ofd_obj OFD object
896 * \param[in] la object attributes
899 * \retval 0 on successful attributes update
900 * \retval negative value on error
903 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
904 struct ofd_object *ofd_obj, struct lu_attr *la,
907 struct ofd_thread_info *info = ofd_info(env);
908 struct filter_fid *ff = &info->fti_mds_fid;
909 __u64 valid = la->la_valid;
911 struct dt_object *dt_obj;
919 dt_obj = ofd_object_child(ofd_obj);
920 LASSERT(dt_obj != NULL);
922 la->la_valid &= LA_UID | LA_GID | LA_PROJID;
924 rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
928 fl = ofd_object_ff_update(env, ofd_obj, oa, ff);
932 if (!la->la_valid && !fl)
933 /* no attributes to set */
936 th = ofd_trans_create(env, ofd);
938 GOTO(out, rc = PTR_ERR(th));
941 rc = dt_declare_attr_set(env, dt_obj, la, th);
947 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
948 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
949 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
950 le32_add_cpu(&ff->ff_parent.f_oid, -1);
952 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
953 XATTR_NAME_FID, 0, th);
958 /* We don't need a transno for this operation which will be re-executed
959 * anyway when the OST_WRITE (with a transno assigned) is replayed */
960 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
964 /* set uid/gid/projid */
966 rc = dt_attr_set(env, dt_obj, la, th);
971 /* set filter fid EA.
972 * FIXME: it holds read lock of ofd object to modify the XATTR_NAME_FID
973 * while the write lock should be held. However, it should work because
974 * write RPCs only modify ff_{parent,layout} and those information will
975 * be the same from all the write RPCs. The reason that fl is not used
976 * in dt_xattr_set() is to allow this race. */
978 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
981 info->fti_buf.lb_buf = ff;
982 info->fti_buf.lb_len = sizeof(*ff);
983 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
986 filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
992 dt_trans_stop(env, ofd->ofd_osd, th);
994 la->la_valid = valid;
998 struct ofd_soft_sync_callback {
999 struct dt_txn_commit_cb ossc_cb;
1000 struct obd_export *ossc_exp;
1004 * Callback function for "soft sync" update.
1006 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
1007 * See ofd_soft_sync_cb_add() below for more details on soft sync.
1009 * \param[in] env execution environment
1010 * \param[in] th transaction handle
1011 * \param[in] cb callback data
1012 * \param[in] err error code
1014 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
1015 struct dt_txn_commit_cb *cb, int err)
1017 struct ofd_soft_sync_callback *ossc;
1019 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
1021 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
1022 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
1024 class_export_cb_put(ossc->ossc_exp);
1029 * Add callback for "soft sync" processing.
1031 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
1032 * flag is set in client buffers. The intention is for this operation to
1033 * commit pages belonging to a client which has "too many" outstanding
1034 * unstable pages in its cache. See LU-2139 for details.
1036 * This function adds callback to be called when commit is done.
1038 * \param[in] th transaction handle
1039 * \param[in] exp OBD export of client
1041 * \retval 0 on successful callback adding
1042 * \retval negative value on error
1044 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
1046 struct ofd_soft_sync_callback *ossc;
1047 struct dt_txn_commit_cb *dcb;
1050 OBD_ALLOC_PTR(ossc);
1054 ossc->ossc_exp = class_export_cb_get(exp);
1056 dcb = &ossc->ossc_cb;
1057 dcb->dcb_func = ofd_cb_soft_sync;
1058 INIT_LIST_HEAD(&dcb->dcb_linkage);
1059 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
1061 rc = dt_trans_cb_add(th, dcb);
1063 class_export_cb_put(exp);
1071 * Commit bulk IO buffers to the storage.
1073 * This function finalizes write IO processing by writing data to the disk.
1074 * That write can be synchronous or asynchronous depending on buffers flags.
1076 * \param[in] env execution environment
1077 * \param[in] exp OBD export of client
1078 * \param[in] ofd OFD device
1079 * \param[in] fid FID of object
1080 * \param[in] la object attributes
1081 * \param[in] ff parent FID of object
1082 * \param[in] objcount always 1
1083 * \param[in] niocount number of local buffers
1084 * \param[in] lnb local buffers
1085 * \param[in] granted grant space consumed for the bulk I/O
1086 * \param[in] old_rc result of processing at this point
1088 * \retval 0 on successful commit
1089 * \retval negative value on error
1092 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1093 struct ofd_device *ofd, const struct lu_fid *fid,
1094 struct lu_attr *la, struct obdo *oa, int objcount,
1095 int niocount, struct niobuf_local *lnb,
1096 unsigned long granted, int old_rc)
1098 struct filter_export_data *fed = &exp->exp_filter_data;
1099 struct ofd_object *fo;
1100 struct dt_object *o;
1106 bool soft_sync = false;
1107 bool cb_registered = false;
1108 bool fake_write = false;
1112 LASSERT(objcount == 1);
1114 fo = ofd_object_find(env, ofd, fid);
1115 LASSERT(fo != NULL);
1116 LASSERT(ofd_object_exists(fo));
1118 o = ofd_object_child(fo);
1122 GOTO(out, rc = old_rc);
1125 * The first write to each object must set some attributes. It is
1126 * important to set the uid/gid before calling
1127 * dt_declare_write_commit() since quota enforcement is now handled in
1130 rc = ofd_write_attr_set(env, ofd, fo, la, oa);
1134 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1136 /* do fake write, to simulate the write case for performance testing */
1137 if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) {
1138 struct niobuf_local *last = &lnb[niocount - 1];
1139 __u64 file_size = last->lnb_file_offset + last->lnb_len;
1140 __u64 valid = la->la_valid;
1142 la->la_valid = LA_SIZE;
1144 rc = dt_attr_get(env, o, la);
1145 if (rc < 0 && rc != -ENOENT)
1148 if (file_size < la->la_size)
1149 file_size = la->la_size;
1151 /* dirty inode by setting file size */
1152 la->la_valid = valid | LA_SIZE;
1153 la->la_size = file_size;
1159 th = ofd_trans_create(env, ofd);
1161 GOTO(out, rc = PTR_ERR(th));
1163 th->th_sync |= ofd->ofd_sync_journal;
1164 if (th->th_sync == 0) {
1165 for (i = 0; i < niocount; i++) {
1166 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1170 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1175 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1176 GOTO(out_stop, rc = -EINPROGRESS);
1178 if (likely(!fake_write)) {
1179 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1185 /* update [mac]time if needed */
1186 rc = dt_declare_attr_set(env, o, la, th);
1191 rc = ofd_trans_start(env, ofd, fo, th);
1195 if (likely(!fake_write)) {
1196 rc = dt_write_commit(env, o, lnb, niocount, th);
1202 rc = dt_attr_set(env, o, la, th);
1207 /* get attr to return */
1208 rc = dt_attr_get(env, o, la);
1211 /* Force commit to make the just-deleted blocks
1212 * reusable. LU-456 */
1216 /* do this before trans stop in case commit has finished */
1217 if (!th->th_sync && soft_sync && !cb_registered) {
1218 ofd_soft_sync_cb_add(th, exp);
1219 cb_registered = true;
1222 if (rc == 0 && granted > 0) {
1223 if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
1227 rc2 = ofd_trans_stop(env, ofd, th, rc);
1230 if (rc == -ENOSPC && retries++ < 3) {
1231 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1237 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1238 atomic_set(&fed->fed_soft_sync_count, 0);
1239 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1240 ofd->ofd_soft_sync_limit)
1241 dt_commit_async(env, ofd->ofd_osd);
1244 dt_bufs_put(env, o, lnb, niocount);
1245 ofd_read_unlock(env, fo);
1246 ofd_object_put(env, fo);
1247 /* second put is pair to object_get in ofd_preprw_write */
1248 ofd_object_put(env, fo);
1250 tgt_grant_commit(exp, granted, old_rc);
1255 * Commit bulk IO to the storage.
1257 * This is companion function to the ofd_preprw(). It finishes bulk IO
1258 * request processing by committing buffers to the storage (WRITE) and/or
1259 * freeing those buffers (read/write). See ofd_commitrw_read() and
1260 * ofd_commitrw_write() for details about each type of IO.
1262 * \param[in] env execution environment
1263 * \param[in] cmd IO type (READ/WRITE)
1264 * \param[in] exp OBD export of client
1265 * \param[in] oa OBDO structure from client
1266 * \param[in] objcount always 1
1267 * \param[in] obj object data
1268 * \param[in] rnb remote buffers
1269 * \param[in] npages number of local buffers
1270 * \param[in] lnb local buffers
1271 * \param[in] old_rc result of processing at this point
1273 * \retval 0 on successful commit
1274 * \retval negative value on error
1276 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1277 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1278 struct niobuf_remote *rnb, int npages,
1279 struct niobuf_local *lnb, int old_rc)
1281 struct ofd_thread_info *info = ofd_info(env);
1282 struct ofd_device *ofd = ofd_exp(exp);
1283 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1287 LASSERT(npages > 0);
1289 if (cmd == OBD_BRW_WRITE) {
1290 struct lu_nodemap *nodemap;
1292 /* Don't update timestamps if this write is older than a
1293 * setattr which modifies the timestamps. b=10150 */
1294 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID;
1295 if (tgt_fmd_check(exp, fid, info->fti_xid))
1296 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1299 la_from_obdo(&info->fti_attr, oa, valid);
1301 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1302 oa, objcount, npages, lnb,
1303 oa->o_grant_used, old_rc);
1305 obdo_from_la(oa, &info->fti_attr,
1306 OFD_VALID_FLAGS | LA_GID | LA_UID |
1309 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID |
1312 /* don't report overquota flag if we failed before reaching
1314 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1315 /* return the overquota flags to client */
1316 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1317 if (oa->o_valid & OBD_MD_FLFLAGS)
1318 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1320 oa->o_flags = OBD_FL_NO_USRQUOTA;
1323 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1324 if (oa->o_valid & OBD_MD_FLFLAGS)
1325 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1327 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1329 if (lnb[0].lnb_flags & OBD_BRW_OVER_PRJQUOTA) {
1330 if (oa->o_valid & OBD_MD_FLFLAGS)
1331 oa->o_flags |= OBD_FL_NO_PRJQUOTA;
1333 oa->o_flags = OBD_FL_NO_PRJQUOTA;
1336 oa->o_valid |= OBD_MD_FLFLAGS;
1337 oa->o_valid |= OBD_MD_FLALLQUOTA;
1340 /* Convert back to client IDs. LU-9671.
1341 * nodemap_get_from_exp() may fail due to nodemap deactivated,
1342 * server ID will be returned back to client in that case. */
1343 nodemap = nodemap_get_from_exp(exp);
1344 if (nodemap != NULL && !IS_ERR(nodemap)) {
1345 oa->o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1346 NODEMAP_FS_TO_CLIENT,
1348 oa->o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1349 NODEMAP_FS_TO_CLIENT,
1351 nodemap_putref(nodemap);
1353 } else if (cmd == OBD_BRW_READ) {
1354 struct ldlm_namespace *ns = ofd->ofd_namespace;
1356 /* If oa != NULL then ofd_preprw_read updated the inode
1357 * atime and we should update the lvb so that other glimpses
1358 * will also get the updated value. bug 5972 */
1359 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1360 struct ldlm_resource *rs = NULL;
1362 ost_fid_build_resid(fid, &info->fti_resid);
1363 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1366 ldlm_res_lvbo_update(rs, NULL, 1);
1367 ldlm_resource_putref(rs);
1370 rc = ofd_commitrw_read(env, ofd, fid, objcount,