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;
171 rc = lu_env_init(&env, LCT_DT_THREAD);
172 spin_lock(&ofd->ofd_inconsistency_lock);
173 thread_set_flags(thread, rc ? SVC_STOPPED : SVC_RUNNING);
174 wake_up_all(&thread->t_ctl_waitq);
175 spin_unlock(&ofd->ofd_inconsistency_lock);
181 GOTO(out_unlocked, rc = -ENOMEM);
183 lrl->lrl_event = LEL_PAIRS_VERIFY_LOCAL;
184 lrl->lrl_active = LFSCK_TYPE_LAYOUT;
186 spin_lock(&ofd->ofd_inconsistency_lock);
188 if (unlikely(!thread_is_running(thread)))
191 while (!list_empty(&ofd->ofd_inconsistency_list)) {
192 oii = list_entry(ofd->ofd_inconsistency_list.next,
193 struct ofd_inconsistency_item,
195 list_del_init(&oii->oii_list);
196 spin_unlock(&ofd->ofd_inconsistency_lock);
197 ofd_inconsistency_verify_one(&env, ofd, oii, lrl);
198 spin_lock(&ofd->ofd_inconsistency_lock);
201 spin_unlock(&ofd->ofd_inconsistency_lock);
202 wait_event_idle(thread->t_ctl_waitq,
203 !list_empty(&ofd->ofd_inconsistency_list) ||
204 !thread_is_running(thread));
205 spin_lock(&ofd->ofd_inconsistency_lock);
208 while (!list_empty(&ofd->ofd_inconsistency_list)) {
209 struct ofd_object *fo;
211 oii = list_entry(ofd->ofd_inconsistency_list.next,
212 struct ofd_inconsistency_item,
214 list_del_init(&oii->oii_list);
216 spin_unlock(&ofd->ofd_inconsistency_lock);
218 ofd_write_lock(&env, fo);
219 fo->ofo_pfid_checking = 0;
220 ofd_write_unlock(&env, fo);
222 ofd_object_put(&env, fo);
224 spin_lock(&ofd->ofd_inconsistency_lock);
232 spin_lock(&ofd->ofd_inconsistency_lock);
234 thread_set_flags(thread, SVC_STOPPED);
235 wake_up_all(&thread->t_ctl_waitq);
236 spin_unlock(&ofd->ofd_inconsistency_lock);
243 * Start parent FID verification thread.
245 * See ofd_inconsistency_verification_main().
247 * \param[in] ofd OFD device
249 * \retval 0 on successful start of thread
250 * \retval negative value on error
252 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
254 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
255 struct task_struct *task;
258 spin_lock(&ofd->ofd_inconsistency_lock);
259 if (unlikely(thread_is_running(thread))) {
260 spin_unlock(&ofd->ofd_inconsistency_lock);
265 thread_set_flags(thread, 0);
266 spin_unlock(&ofd->ofd_inconsistency_lock);
267 task = kthread_run(ofd_inconsistency_verification_main, ofd,
268 "inconsistency_verification");
271 CERROR("%s: cannot start self_repair thread: rc = %d\n",
275 wait_event_idle(thread->t_ctl_waitq,
276 thread_is_running(thread) ||
277 thread_is_stopped(thread));
284 * Stop parent FID verification thread.
286 * \param[in] ofd OFD device
288 * \retval 0 on successful start of thread
289 * \retval -EALREADY if thread is already stopped
291 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
293 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
295 spin_lock(&ofd->ofd_inconsistency_lock);
296 if (thread_is_init(thread) || thread_is_stopped(thread)) {
297 spin_unlock(&ofd->ofd_inconsistency_lock);
302 thread_set_flags(thread, SVC_STOPPING);
303 spin_unlock(&ofd->ofd_inconsistency_lock);
304 wake_up_all(&thread->t_ctl_waitq);
305 wait_event_idle(thread->t_ctl_waitq,
306 thread_is_stopped(thread));
312 * Add new item for parent FID verification.
314 * Prepare new verification item and pass it to the dedicated
315 * verification thread for further processing.
317 * \param[in] env execution environment
318 * \param[in] fo OFD object
319 * \param[in] oa OBDO structure with PFID
321 static void ofd_add_inconsistency_item(const struct lu_env *env,
322 struct ofd_object *fo, struct obdo *oa)
324 struct ofd_device *ofd = ofd_obj2dev(fo);
325 struct ofd_inconsistency_item *oii;
326 struct filter_fid *ff;
333 INIT_LIST_HEAD(&oii->oii_list);
334 lu_object_get(&fo->ofo_obj.do_lu);
337 ff->ff_parent.f_seq = oa->o_parent_seq;
338 ff->ff_parent.f_oid = oa->o_parent_oid;
339 ff->ff_parent.f_stripe_idx = oa->o_stripe_idx;
340 ff->ff_layout = oa->o_layout;
342 spin_lock(&ofd->ofd_inconsistency_lock);
343 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
344 spin_unlock(&ofd->ofd_inconsistency_lock);
350 fo->ofo_pfid_checking = 1;
351 if (list_empty(&ofd->ofd_inconsistency_list))
353 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
354 spin_unlock(&ofd->ofd_inconsistency_lock);
356 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
358 /* XXX: When the found inconsistency exceeds some threshold,
359 * we can trigger the LFSCK to scan part of the system
360 * or the whole system, which depends on how to define
361 * the threshold, a simple way maybe like that: define
362 * the absolute value of how many inconsisteny allowed
363 * to be repaired via self detect/repair mechanism, if
364 * exceeded, then trigger the LFSCK to scan the layout
365 * inconsistency within the whole system. */
369 * Verify parent FID of an object.
371 * Check the parent FID is sane and start extended
372 * verification procedure otherwise.
374 * \param[in] env execution environment
375 * \param[in] fo OFD object
376 * \param[in] oa OBDO structure with PFID
378 * \retval 0 on successful verification
379 * \retval -EINPROGRESS if PFID is being repaired
380 * \retval -EPERM if PFID was verified but still insane
382 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
385 struct lu_fid *pfid = &fo->ofo_ff.ff_parent;
389 if (fid_is_sane(pfid)) {
390 if (likely(oa->o_parent_seq == pfid->f_seq &&
391 oa->o_parent_oid == pfid->f_oid &&
392 oa->o_stripe_idx == pfid->f_stripe_idx))
395 if (fo->ofo_pfid_verified)
399 /* The OST-object may be inconsistent, and we need further verification.
400 * To avoid block the RPC service thread, return -EINPROGRESS to client
401 * and make it retry later. */
402 if (fo->ofo_pfid_checking)
403 RETURN(-EINPROGRESS);
405 rc = ofd_object_ff_load(env, fo);
412 if (likely(oa->o_parent_seq == pfid->f_seq &&
413 oa->o_parent_oid == pfid->f_oid &&
414 oa->o_stripe_idx == pfid->f_stripe_idx))
417 /* Push it to the dedicated thread for further verification. */
418 ofd_add_inconsistency_item(env, fo, oa);
420 RETURN(-EINPROGRESS);
424 * FLR: verify the layout version of object.
426 * \param[in] env execution environment
427 * \param[in] fo OFD object
428 * \param[in] oa OBDO structure with layout version
430 * \retval 0 on successful verification
431 * \retval -EINPROGRESS layout version is in transfer
432 * \retval -ESTALE the layout version on client is stale
434 int ofd_verify_layout_version(const struct lu_env *env,
435 struct ofd_object *fo, const struct obdo *oa)
437 __u32 layout_version;
441 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_OST_SKIP_LV_CHECK)))
444 rc = ofd_object_ff_load(env, fo);
451 layout_version = fo->ofo_ff.ff_layout_version;
452 if (oa->o_layout_version >= layout_version &&
453 oa->o_layout_version <= layout_version + fo->ofo_ff.ff_range)
456 /* normal traffic, decide if to return ESTALE or EINPROGRESS */
457 layout_version &= ~LU_LAYOUT_RESYNC;
459 /* this update is not legitimate */
460 if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) <= layout_version)
461 GOTO(out, rc = -ESTALE);
463 /* layout version may not be transmitted yet */
464 if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) > layout_version)
465 GOTO(out, rc = -EINPROGRESS);
470 CDEBUG(D_INODE, DFID " verify layout version: %u vs. %u/%u, rc: %d\n",
471 PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
472 oa->o_layout_version, fo->ofo_ff.ff_layout_version,
473 fo->ofo_ff.ff_range, rc);
479 * Prepare buffers for read request processing.
481 * This function converts remote buffers from client to local buffers
482 * and prepares the latter.
484 * \param[in] env execution environment
485 * \param[in] exp OBD export of client
486 * \param[in] ofd OFD device
487 * \param[in] fid FID of object
488 * \param[in] la object attributes
489 * \param[in] oa OBDO structure from client
490 * \param[in] niocount number of remote buffers
491 * \param[in] rnb remote buffers
492 * \param[in] nr_local number of local buffers
493 * \param[in] lnb local buffers
494 * \param[in] jobid job ID name
496 * \retval 0 on successful prepare
497 * \retval negative value on error
499 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
500 struct ofd_device *ofd, const struct lu_fid *fid,
501 struct lu_attr *la, struct obdo *oa, int niocount,
502 struct niobuf_remote *rnb, int *nr_local,
503 struct niobuf_local *lnb, char *jobid)
505 struct ofd_object *fo;
506 int i, j, rc, tot_bytes = 0;
507 enum dt_bufs_type dbt = DT_BUFS_TYPE_READ;
508 int maxlnb = *nr_local;
511 LASSERT(env != NULL);
513 fo = ofd_object_find(env, ofd, fid);
518 ofd_read_lock(env, fo);
519 if (!ofd_object_exists(fo))
520 GOTO(unlock, rc = -ENOENT);
522 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
523 rc = ofd_verify_ff(env, fo, oa);
528 if (ptlrpc_connection_is_local(exp->exp_connection))
529 dbt |= DT_BUFS_TYPE_LOCAL;
531 for (*nr_local = 0, i = 0, j = 0; i < niocount; i++) {
533 if (OBD_FAIL_CHECK(OBD_FAIL_OST_2BIG_NIOBUF))
534 rnb[i].rnb_len = 100 * 1024 * 1024;
536 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
537 lnb + j, maxlnb, dbt);
538 if (unlikely(rc < 0))
540 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
541 /* 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;
602 int maxlnb = *nr_local;
605 LASSERT(env != NULL);
606 LASSERT(objcount == 1);
608 if (unlikely(exp->exp_obd->obd_recovering)) {
609 u64 seq = fid_seq(fid);
610 u64 oid = fid_oid(fid);
611 struct ofd_seq *oseq;
613 oseq = ofd_seq_load(env, ofd, seq);
615 CERROR("%s: Can't find FID Sequence %#llx: rc = %d\n",
616 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
617 GOTO(out, rc = -EINVAL);
620 if (oid > ofd_seq_last_oid(oseq)) {
624 mutex_lock(&oseq->os_create_lock);
625 diff = oid - ofd_seq_last_oid(oseq);
627 /* Do sync create if the seq is about to used up */
628 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
629 if (unlikely(oid >= IDIF_MAX_OID - 1))
631 } else if (fid_seq_is_norm(seq)) {
633 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
636 CERROR("%s : invalid o_seq "DOSTID"\n",
637 ofd_name(ofd), POSTID(&oa->o_oi));
638 mutex_unlock(&oseq->os_create_lock);
639 ofd_seq_put(env, oseq);
640 GOTO(out, rc = -EINVAL);
644 u64 next_id = ofd_seq_last_oid(oseq) + 1;
645 int count = ofd_precreate_batch(ofd, diff);
647 rc = ofd_precreate_objects(env, ofd, next_id,
650 mutex_unlock(&oseq->os_create_lock);
651 ofd_seq_put(env, oseq);
658 mutex_unlock(&oseq->os_create_lock);
661 ofd_seq_put(env, oseq);
664 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
665 * space back if possible, we have to do this outside of the lock as
666 * grant preparation may need to sync whole fs thus wait for all the
667 * transactions to complete. */
668 tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
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 if (ptlrpc_connection_is_local(exp->exp_connection))
706 dbt |= DT_BUFS_TYPE_LOCAL;
708 /* parse remote buffers to local buffers and prepare the latter */
709 for (*nr_local = 0, i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
710 if (OBD_FAIL_CHECK(OBD_FAIL_OST_2BIG_NIOBUF))
711 rnb[i].rnb_len += PAGE_SIZE;
712 rc = dt_bufs_get(env, ofd_object_child(fo),
713 rnb + i, lnb + j, maxlnb, dbt);
714 if (unlikely(rc < 0))
716 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
717 /* correct index for local buffers to continue with */
718 for (k = 0; k < rc; k++) {
719 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
720 lnb[j+k].lnb_flags &= ~OBD_BRW_LOCALS;
721 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
722 lnb[j+k].lnb_rc = -ENOSPC;
727 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
728 tot_bytes += rnb[i].rnb_len;
730 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
732 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
733 if (unlikely(rc != 0))
736 ofd_read_unlock(env, fo);
737 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
740 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
741 ofd_read_unlock(env, fo);
742 ofd_object_put(env, fo);
743 /* tgt_grant_prepare_write() was called, so we must commit */
744 tgt_grant_commit(exp, oa->o_grant_used, rc);
746 /* let's still process incoming grant information packed in the oa,
747 * but without enforcing grant since we won't proceed with the write.
748 * Just like a read request actually. */
749 tgt_grant_prepare_read(env, exp, oa);
754 * Prepare bulk IO requests for processing.
756 * This function does initial checks of IO and calls corresponding
757 * functions for read/write processing.
759 * \param[in] env execution environment
760 * \param[in] cmd IO type (read/write)
761 * \param[in] exp OBD export of client
762 * \param[in] oa OBDO structure from request
763 * \param[in] objcount always 1
764 * \param[in] obj object data
765 * \param[in] rnb remote buffers
766 * \param[in] nr_local number of local buffers
767 * \param[in] lnb local buffers
769 * \retval 0 on successful prepare
770 * \retval negative value on error
772 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
773 struct obdo *oa, int objcount, struct obd_ioobj *obj,
774 struct niobuf_remote *rnb, int *nr_local,
775 struct niobuf_local *lnb)
777 struct tgt_session_info *tsi = tgt_ses_info(env);
778 struct ofd_device *ofd = ofd_exp(exp);
779 struct ofd_thread_info *info;
781 const struct lu_fid *fid = &oa->o_oi.oi_fid;
784 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
785 CERROR("%s: bulk has too many pages %d, which exceeds the"
786 "maximum pages per RPC of %d\n",
787 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
791 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
792 info = ofd_info_init(env, exp);
795 info = tsi2ofd_info(tsi);
796 jobid = tsi->tsi_jobid;
801 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
802 struct ofd_seq *oseq;
804 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
806 CERROR("%s: Can not find seq for "DOSTID
807 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
812 if (oseq->os_destroys_in_progress == 0) {
813 /* don't fail lookups for orphan recovery, it causes
814 * later LBUGs when objects still exist during
816 ofd_seq_put(env, oseq);
819 ofd_seq_put(env, oseq);
822 LASSERT(objcount == 1);
823 LASSERT(obj->ioo_bufcnt > 0);
825 if (cmd == OBD_BRW_WRITE) {
826 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
827 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
828 objcount, obj, rnb, nr_local, lnb, jobid);
829 } else if (cmd == OBD_BRW_READ) {
830 tgt_grant_prepare_read(env, exp, oa);
831 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
832 obj->ioo_bufcnt, rnb, nr_local, lnb,
834 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
836 CERROR("%s: wrong cmd %d received!\n",
837 exp->exp_obd->obd_name, cmd);
844 * Drop reference on local buffers for read bulk IO.
846 * This will free all local buffers use by this read request.
848 * \param[in] env execution environment
849 * \param[in] ofd OFD device
850 * \param[in] fid object FID
851 * \param[in] objcount always 1
852 * \param[in] niocount number of local buffers
853 * \param[in] lnb local buffers
855 * \retval 0 on successful execution
856 * \retval negative value on error
859 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
860 const struct lu_fid *fid, int objcount, int niocount,
861 struct niobuf_local *lnb)
863 struct ofd_object *fo;
867 LASSERT(niocount > 0);
869 fo = ofd_object_find(env, ofd, fid);
873 LASSERT(ofd_object_exists(fo));
874 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
876 ofd_read_unlock(env, fo);
877 ofd_object_put(env, fo);
878 /* second put is pair to object_get in ofd_preprw_read */
879 ofd_object_put(env, fo);
885 * Set attributes of object during write bulk IO processing.
887 * Change object attributes and write parent FID into extended
888 * attributes when needed.
890 * \param[in] env execution environment
891 * \param[in] ofd OFD device
892 * \param[in] ofd_obj OFD object
893 * \param[in] la object attributes
896 * \retval 0 on successful attributes update
897 * \retval negative value on error
900 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
901 struct ofd_object *ofd_obj, struct lu_attr *la,
904 struct ofd_thread_info *info = ofd_info(env);
905 struct filter_fid *ff = &info->fti_mds_fid;
906 __u64 valid = la->la_valid;
908 struct dt_object *dt_obj;
916 dt_obj = ofd_object_child(ofd_obj);
917 LASSERT(dt_obj != NULL);
919 la->la_valid &= LA_UID | LA_GID | LA_PROJID;
921 rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
925 if (!la->la_valid && !(oa->o_valid &
926 (OBD_MD_FLFID | OBD_MD_FLOSTLAYOUT | OBD_MD_LAYOUT_VERSION)))
927 /* no attributes to set */
930 th = ofd_trans_create(env, ofd);
932 GOTO(out, rc = PTR_ERR(th));
935 rc = dt_declare_attr_set(env, dt_obj, la, th);
940 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
941 XATTR_NAME_FID, 0, th);
945 /* We don't need a transno for this operation which will be re-executed
946 * anyway when the OST_WRITE (with a transno assigned) is replayed */
947 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
951 ofd_read_lock(env, ofd_obj);
953 /* set uid/gid/projid */
955 rc = dt_attr_set(env, dt_obj, la, th);
957 GOTO(out_unlock, rc);
960 fl = ofd_object_ff_update(env, ofd_obj, oa, ff);
962 GOTO(out_unlock, rc = fl);
964 /* set filter fid EA.
965 * FIXME: it holds read lock of ofd object to modify the XATTR_NAME_FID
966 * while the write lock should be held. However, it should work because
967 * write RPCs only modify ff_{parent,layout} and those information will
968 * be the same from all the write RPCs. The reason that fl is not used
969 * in dt_xattr_set() is to allow this race. */
970 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
971 GOTO(out_unlock, rc);
972 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
973 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
974 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
975 le32_add_cpu(&ff->ff_parent.f_oid, -1);
977 info->fti_buf.lb_buf = ff;
978 info->fti_buf.lb_len = sizeof(*ff);
979 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID, 0, th);
981 filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
983 GOTO(out_unlock, rc);
986 ofd_read_unlock(env, ofd_obj);
988 dt_trans_stop(env, ofd->ofd_osd, th);
990 la->la_valid = valid;
994 struct ofd_soft_sync_callback {
995 struct dt_txn_commit_cb ossc_cb;
996 struct obd_export *ossc_exp;
1000 * Callback function for "soft sync" update.
1002 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
1003 * See ofd_soft_sync_cb_add() below for more details on soft sync.
1005 * \param[in] env execution environment
1006 * \param[in] th transaction handle
1007 * \param[in] cb callback data
1008 * \param[in] err error code
1010 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
1011 struct dt_txn_commit_cb *cb, int err)
1013 struct ofd_soft_sync_callback *ossc;
1015 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
1017 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
1018 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
1020 class_export_cb_put(ossc->ossc_exp);
1025 * Add callback for "soft sync" processing.
1027 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
1028 * flag is set in client buffers. The intention is for this operation to
1029 * commit pages belonging to a client which has "too many" outstanding
1030 * unstable pages in its cache. See LU-2139 for details.
1032 * This function adds callback to be called when commit is done.
1034 * \param[in] th transaction handle
1035 * \param[in] exp OBD export of client
1037 * \retval 0 on successful callback adding
1038 * \retval negative value on error
1040 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
1042 struct ofd_soft_sync_callback *ossc;
1043 struct dt_txn_commit_cb *dcb;
1046 OBD_ALLOC_PTR(ossc);
1050 ossc->ossc_exp = class_export_cb_get(exp);
1052 dcb = &ossc->ossc_cb;
1053 dcb->dcb_func = ofd_cb_soft_sync;
1054 INIT_LIST_HEAD(&dcb->dcb_linkage);
1055 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
1057 rc = dt_trans_cb_add(th, dcb);
1059 class_export_cb_put(exp);
1067 * Commit bulk IO buffers to the storage.
1069 * This function finalizes write IO processing by writing data to the disk.
1070 * That write can be synchronous or asynchronous depending on buffers flags.
1072 * \param[in] env execution environment
1073 * \param[in] exp OBD export of client
1074 * \param[in] ofd OFD device
1075 * \param[in] fid FID of object
1076 * \param[in] la object attributes
1077 * \param[in] ff parent FID of object
1078 * \param[in] objcount always 1
1079 * \param[in] niocount number of local buffers
1080 * \param[in] lnb local buffers
1081 * \param[in] granted grant space consumed for the bulk I/O
1082 * \param[in] old_rc result of processing at this point
1084 * \retval 0 on successful commit
1085 * \retval negative value on error
1088 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1089 struct ofd_device *ofd, const struct lu_fid *fid,
1090 struct lu_attr *la, struct obdo *oa, int objcount,
1091 int niocount, struct niobuf_local *lnb,
1092 unsigned long granted, int old_rc)
1094 struct ofd_thread_info *info = ofd_info(env);
1095 struct filter_export_data *fed = &exp->exp_filter_data;
1096 struct ofd_object *fo;
1097 struct dt_object *o;
1103 bool soft_sync = false;
1104 bool cb_registered = false;
1105 bool fake_write = false;
1109 LASSERT(objcount == 1);
1111 fo = ofd_object_find(env, ofd, fid);
1112 LASSERT(fo != NULL);
1114 o = ofd_object_child(fo);
1118 GOTO(out, rc = old_rc);
1119 if (!ofd_object_exists(fo))
1120 GOTO(out, rc = -ENOENT);
1123 * The first write to each object must set some attributes. It is
1124 * important to set the uid/gid before calling
1125 * dt_declare_write_commit() since quota enforcement is now handled in
1128 rc = ofd_write_attr_set(env, ofd, fo, la, oa);
1132 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1134 /* do fake write, to simulate the write case for performance testing */
1135 if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) {
1136 struct niobuf_local *last = &lnb[niocount - 1];
1137 __u64 file_size = last->lnb_file_offset + last->lnb_len;
1138 __u64 valid = la->la_valid;
1140 la->la_valid = LA_SIZE;
1142 rc = dt_attr_get(env, o, la);
1143 if (rc < 0 && rc != -ENOENT)
1146 if (file_size < la->la_size)
1147 file_size = la->la_size;
1149 /* dirty inode by setting file size */
1150 la->la_valid = valid | LA_SIZE;
1151 la->la_size = file_size;
1157 th = ofd_trans_create(env, ofd);
1159 GOTO(out, rc = PTR_ERR(th));
1161 th->th_sync |= ofd->ofd_sync_journal;
1162 if (th->th_sync == 0) {
1163 for (i = 0; i < niocount; i++) {
1164 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1168 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1173 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1174 GOTO(out_stop, rc = -EINPROGRESS);
1176 if (likely(!fake_write)) {
1177 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1183 /* update [mac]time if needed */
1184 rc = dt_declare_attr_set(env, o, la, th);
1189 rc = ofd_trans_start(env, ofd, fo, th);
1193 ofd_read_lock(env, fo);
1194 if (!ofd_object_exists(fo))
1195 GOTO(out_unlock, rc = -ENOENT);
1197 if (likely(!fake_write)) {
1198 rc = dt_write_commit(env, o, lnb, niocount, th);
1200 GOTO(out_unlock, rc);
1203 /* Don't update timestamps if this write is older than a
1204 * setattr which modifies the timestamps. b=10150 */
1205 if (la->la_valid && tgt_fmd_check(exp, fid, info->fti_xid)) {
1206 rc = dt_attr_set(env, o, la, th);
1208 GOTO(out_unlock, rc);
1211 /* get attr to return */
1212 rc = dt_attr_get(env, o, la);
1215 ofd_read_unlock(env, fo);
1217 /* Force commit to make the just-deleted blocks
1218 * reusable. LU-456 */
1222 /* do this before trans stop in case commit has finished */
1223 if (!th->th_sync && soft_sync && !cb_registered) {
1224 ofd_soft_sync_cb_add(th, exp);
1225 cb_registered = true;
1228 if (rc == 0 && granted > 0) {
1229 if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
1233 rc2 = ofd_trans_stop(env, ofd, th, rc);
1236 if (rc == -ENOSPC && retries++ < 3) {
1237 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1243 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1244 atomic_set(&fed->fed_soft_sync_count, 0);
1245 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1246 ofd->ofd_soft_sync_limit)
1247 dt_commit_async(env, ofd->ofd_osd);
1250 dt_bufs_put(env, o, lnb, niocount);
1251 ofd_object_put(env, fo);
1252 /* second put is pair to object_get in ofd_preprw_write */
1253 ofd_object_put(env, fo);
1255 tgt_grant_commit(exp, granted, old_rc);
1260 * Commit bulk IO to the storage.
1262 * This is companion function to the ofd_preprw(). It finishes bulk IO
1263 * request processing by committing buffers to the storage (WRITE) and/or
1264 * freeing those buffers (read/write). See ofd_commitrw_read() and
1265 * ofd_commitrw_write() for details about each type of IO.
1267 * \param[in] env execution environment
1268 * \param[in] cmd IO type (READ/WRITE)
1269 * \param[in] exp OBD export of client
1270 * \param[in] oa OBDO structure from client
1271 * \param[in] objcount always 1
1272 * \param[in] obj object data
1273 * \param[in] rnb remote buffers
1274 * \param[in] npages number of local buffers
1275 * \param[in] lnb local buffers
1276 * \param[in] old_rc result of processing at this point
1278 * \retval 0 on successful commit
1279 * \retval negative value on error
1281 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1282 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1283 struct niobuf_remote *rnb, int npages,
1284 struct niobuf_local *lnb, int old_rc)
1286 struct ofd_thread_info *info = ofd_info(env);
1287 struct ofd_device *ofd = ofd_exp(exp);
1288 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1292 LASSERT(npages > 0);
1294 if (cmd == OBD_BRW_WRITE) {
1295 struct lu_nodemap *nodemap;
1297 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID |
1298 OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
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,