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"
46 struct ofd_inconsistency_item {
47 struct list_head oii_list;
48 struct ofd_object *oii_obj;
49 struct filter_fid oii_ff;
53 * Verify single object for parent FID consistency.
55 * Part of LFSCK processing which checks single object PFID stored in extended
56 * attribute (XATTR) against real FID of MDT parent object received by LFSCK.
57 * This verifies that the OST object is being referenced by only a single MDT
60 * \param[in] env execution environment
61 * \param[in] ofd OFD device
62 * \param[in] oii object-related local data
63 * \param[in] lrl LFSCK request data
65 static void ofd_inconsistency_verify_one(const struct lu_env *env,
66 struct ofd_device *ofd,
67 struct ofd_inconsistency_item *oii,
68 struct lfsck_req_local *lrl)
70 struct ofd_object *fo = oii->oii_obj;
71 struct filter_fid *client_ff = &oii->oii_ff;
72 struct filter_fid *local_ff = &fo->ofo_ff;
75 LASSERT(fo->ofo_pfid_checking);
76 LASSERT(!fo->ofo_pfid_verified);
78 lrl->lrl_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
79 lrl->lrl_ff_client = *client_ff; /* client given PFID. */
80 lrl->lrl_ff_local = *local_ff; /* OST local stored PFID. */
82 rc = lfsck_in_notify_local(env, ofd->ofd_osd, lrl, NULL);
83 ofd_write_lock(env, fo);
84 switch (lrl->lrl_status) {
89 CDEBUG(D_LFSCK, "%s: fail to verify OST local stored "
90 "PFID xattr for "DFID", the client given PFID "
91 DFID", OST local stored PFID "DFID": rc = %d\n",
92 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
93 PFID(&client_ff->ff_parent),
94 PFID(&local_ff->ff_parent), rc);
96 fo->ofo_pfid_verified = 1;
98 case LPVS_INCONSISTENT:
101 ofd->ofd_inconsistency_self_detected++;
103 CDEBUG(D_LFSCK, "%s: fail to verify the client given "
104 "PFID for "DFID", the client given PFID "DFID
105 ", local stored PFID "DFID": rc = %d\n",
106 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
107 PFID(&client_ff->ff_parent),
108 PFID(&local_ff->ff_parent), rc);
110 CDEBUG(D_LFSCK, "%s: both the client given PFID and "
111 "the OST local stored PFID are stale for the "
112 "OST-object "DFID", client given PFID is "DFID
113 ", local stored PFID is "DFID"\n",
114 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
115 PFID(&client_ff->ff_parent),
116 PFID(&local_ff->ff_parent));
118 case LPVS_INCONSISTENT_TOFIX:
119 ofd->ofd_inconsistency_self_detected++;
121 ofd->ofd_inconsistency_self_repaired++;
122 CDEBUG(D_LFSCK, "%s: fixed the staled OST PFID xattr "
123 "for "DFID", with the client given PFID "DFID
124 ", the old stored PFID "DFID"\n",
125 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
126 PFID(&client_ff->ff_parent),
127 PFID(&local_ff->ff_parent));
129 CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
130 "for "DFID", client given PFID "DFID", local "
131 "stored PFID "DFID": rc = %d\n",
132 ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
133 PFID(&client_ff->ff_parent),
134 PFID(&local_ff->ff_parent), rc);
136 local_ff->ff_parent = client_ff->ff_parent;
137 fo->ofo_pfid_verified = 1;
142 fo->ofo_pfid_checking = 0;
143 ofd_write_unlock(env, fo);
145 ofd_object_put(env, fo);
150 * Verification thread to check parent FID consistency.
152 * Kernel thread to check consistency of parent FID for any
153 * new item added for checking by ofd_add_inconsistency_item().
155 * \param[in] args OFD device
157 * \retval 0 on successful thread termination
158 * \retval negative value if thread can't start
160 static int ofd_inconsistency_verification_main(void *args)
163 struct ofd_device *ofd = args;
164 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
165 struct ofd_inconsistency_item *oii;
166 struct lfsck_req_local *lrl = NULL;
167 struct l_wait_info lwi = { 0 };
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 l_wait_event(thread->t_ctl_waitq,
203 !list_empty(&ofd->ofd_inconsistency_list) ||
204 !thread_is_running(thread),
206 spin_lock(&ofd->ofd_inconsistency_lock);
209 while (!list_empty(&ofd->ofd_inconsistency_list)) {
210 struct ofd_object *fo;
212 oii = list_entry(ofd->ofd_inconsistency_list.next,
213 struct ofd_inconsistency_item,
215 list_del_init(&oii->oii_list);
217 spin_unlock(&ofd->ofd_inconsistency_lock);
219 ofd_write_lock(&env, fo);
220 fo->ofo_pfid_checking = 0;
221 ofd_write_unlock(&env, fo);
223 ofd_object_put(&env, fo);
225 spin_lock(&ofd->ofd_inconsistency_lock);
233 spin_lock(&ofd->ofd_inconsistency_lock);
235 thread_set_flags(thread, SVC_STOPPED);
236 wake_up_all(&thread->t_ctl_waitq);
237 spin_unlock(&ofd->ofd_inconsistency_lock);
244 * Start parent FID verification thread.
246 * See ofd_inconsistency_verification_main().
248 * \param[in] ofd OFD device
250 * \retval 0 on successful start of thread
251 * \retval negative value on error
253 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
255 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
256 struct l_wait_info lwi = { 0 };
257 struct task_struct *task;
260 spin_lock(&ofd->ofd_inconsistency_lock);
261 if (unlikely(thread_is_running(thread))) {
262 spin_unlock(&ofd->ofd_inconsistency_lock);
267 thread_set_flags(thread, 0);
268 spin_unlock(&ofd->ofd_inconsistency_lock);
269 task = kthread_run(ofd_inconsistency_verification_main, ofd,
270 "inconsistency_verification");
273 CERROR("%s: cannot start self_repair thread: rc = %d\n",
277 l_wait_event(thread->t_ctl_waitq,
278 thread_is_running(thread) ||
279 thread_is_stopped(thread),
287 * Stop parent FID verification thread.
289 * \param[in] ofd OFD device
291 * \retval 0 on successful start of thread
292 * \retval -EALREADY if thread is already stopped
294 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
296 struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
297 struct l_wait_info lwi = { 0 };
299 spin_lock(&ofd->ofd_inconsistency_lock);
300 if (thread_is_init(thread) || thread_is_stopped(thread)) {
301 spin_unlock(&ofd->ofd_inconsistency_lock);
306 thread_set_flags(thread, SVC_STOPPING);
307 spin_unlock(&ofd->ofd_inconsistency_lock);
308 wake_up_all(&thread->t_ctl_waitq);
309 l_wait_event(thread->t_ctl_waitq,
310 thread_is_stopped(thread),
317 * Add new item for parent FID verification.
319 * Prepare new verification item and pass it to the dedicated
320 * verification thread for further processing.
322 * \param[in] env execution environment
323 * \param[in] fo OFD object
324 * \param[in] oa OBDO structure with PFID
326 static void ofd_add_inconsistency_item(const struct lu_env *env,
327 struct ofd_object *fo, struct obdo *oa)
329 struct ofd_device *ofd = ofd_obj2dev(fo);
330 struct ofd_inconsistency_item *oii;
331 struct filter_fid *ff;
338 INIT_LIST_HEAD(&oii->oii_list);
339 lu_object_get(&fo->ofo_obj.do_lu);
342 ff->ff_parent.f_seq = oa->o_parent_seq;
343 ff->ff_parent.f_oid = oa->o_parent_oid;
344 ff->ff_parent.f_stripe_idx = oa->o_stripe_idx;
345 ff->ff_layout = oa->o_layout;
347 spin_lock(&ofd->ofd_inconsistency_lock);
348 if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
349 spin_unlock(&ofd->ofd_inconsistency_lock);
355 fo->ofo_pfid_checking = 1;
356 if (list_empty(&ofd->ofd_inconsistency_list))
358 list_add_tail(&oii->oii_list, &ofd->ofd_inconsistency_list);
359 spin_unlock(&ofd->ofd_inconsistency_lock);
361 wake_up_all(&ofd->ofd_inconsistency_thread.t_ctl_waitq);
363 /* XXX: When the found inconsistency exceeds some threshold,
364 * we can trigger the LFSCK to scan part of the system
365 * or the whole system, which depends on how to define
366 * the threshold, a simple way maybe like that: define
367 * the absolute value of how many inconsisteny allowed
368 * to be repaired via self detect/repair mechanism, if
369 * exceeded, then trigger the LFSCK to scan the layout
370 * inconsistency within the whole system. */
374 * Verify parent FID of an object.
376 * Check the parent FID is sane and start extended
377 * verification procedure otherwise.
379 * \param[in] env execution environment
380 * \param[in] fo OFD object
381 * \param[in] oa OBDO structure with PFID
383 * \retval 0 on successful verification
384 * \retval -EINPROGRESS if PFID is being repaired
385 * \retval -EPERM if PFID was verified but still insane
387 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
390 struct lu_fid *pfid = &fo->ofo_ff.ff_parent;
394 if (fid_is_sane(pfid)) {
395 if (likely(oa->o_parent_seq == pfid->f_seq &&
396 oa->o_parent_oid == pfid->f_oid &&
397 oa->o_stripe_idx == pfid->f_stripe_idx))
400 if (fo->ofo_pfid_verified)
404 /* The OST-object may be inconsistent, and we need further verification.
405 * To avoid block the RPC service thread, return -EINPROGRESS to client
406 * and make it retry later. */
407 if (fo->ofo_pfid_checking)
408 RETURN(-EINPROGRESS);
410 rc = ofd_object_ff_load(env, fo);
417 if (likely(oa->o_parent_seq == pfid->f_seq &&
418 oa->o_parent_oid == pfid->f_oid &&
419 oa->o_stripe_idx == pfid->f_stripe_idx))
422 /* Push it to the dedicated thread for further verification. */
423 ofd_add_inconsistency_item(env, fo, oa);
425 RETURN(-EINPROGRESS);
429 * Prepare buffers for read request processing.
431 * This function converts remote buffers from client to local buffers
432 * and prepares the latter.
434 * \param[in] env execution environment
435 * \param[in] exp OBD export of client
436 * \param[in] ofd OFD device
437 * \param[in] fid FID of object
438 * \param[in] la object attributes
439 * \param[in] oa OBDO structure from client
440 * \param[in] niocount number of remote buffers
441 * \param[in] rnb remote buffers
442 * \param[in] nr_local number of local buffers
443 * \param[in] lnb local buffers
444 * \param[in] jobid job ID name
446 * \retval 0 on successful prepare
447 * \retval negative value on error
449 static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
450 struct ofd_device *ofd, const struct lu_fid *fid,
451 struct lu_attr *la, struct obdo *oa, int niocount,
452 struct niobuf_remote *rnb, int *nr_local,
453 struct niobuf_local *lnb, char *jobid)
455 struct ofd_object *fo;
456 int i, j, rc, tot_bytes = 0;
459 LASSERT(env != NULL);
461 fo = ofd_object_find(env, ofd, fid);
466 ofd_read_lock(env, fo);
467 if (!ofd_object_exists(fo))
468 GOTO(unlock, rc = -ENOENT);
470 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
471 rc = ofd_verify_ff(env, fo, oa);
477 for (i = 0, j = 0; i < niocount; i++) {
478 rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
480 if (unlikely(rc < 0))
482 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
483 /* correct index for local buffers to continue with */
486 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
487 tot_bytes += rnb[i].rnb_len;
490 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
491 rc = dt_attr_get(env, ofd_object_child(fo), la);
495 rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
499 ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
503 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
505 ofd_read_unlock(env, fo);
506 ofd_object_put(env, fo);
511 * Prepare buffers for write request processing.
513 * This function converts remote buffers from client to local buffers
514 * and prepares the latter. If there is recovery in progress and required
515 * object is missing then it can be re-created before write.
517 * \param[in] env execution environment
518 * \param[in] exp OBD export of client
519 * \param[in] ofd OFD device
520 * \param[in] fid FID of object
521 * \param[in] la object attributes
522 * \param[in] oa OBDO structure from client
523 * \param[in] objcount always 1
524 * \param[in] obj object data
525 * \param[in] rnb remote buffers
526 * \param[in] nr_local number of local buffers
527 * \param[in] lnb local buffers
528 * \param[in] jobid job ID name
530 * \retval 0 on successful prepare
531 * \retval negative value on error
533 static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
534 struct ofd_device *ofd, const struct lu_fid *fid,
535 struct lu_attr *la, struct obdo *oa,
536 int objcount, struct obd_ioobj *obj,
537 struct niobuf_remote *rnb, int *nr_local,
538 struct niobuf_local *lnb, char *jobid)
540 struct ofd_object *fo;
541 int i, j, k, rc = 0, tot_bytes = 0;
544 LASSERT(env != NULL);
545 LASSERT(objcount == 1);
547 if (unlikely(exp->exp_obd->obd_recovering)) {
548 u64 seq = fid_seq(fid);
549 u64 oid = fid_oid(fid);
550 struct ofd_seq *oseq;
552 oseq = ofd_seq_load(env, ofd, seq);
554 CERROR("%s: Can't find FID Sequence %#llx: rc = %d\n",
555 ofd_name(ofd), seq, (int)PTR_ERR(oseq));
556 GOTO(out, rc = -EINVAL);
559 if (oid > ofd_seq_last_oid(oseq)) {
563 mutex_lock(&oseq->os_create_lock);
564 diff = oid - ofd_seq_last_oid(oseq);
566 /* Do sync create if the seq is about to used up */
567 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
568 if (unlikely(oid >= IDIF_MAX_OID - 1))
570 } else if (fid_seq_is_norm(seq)) {
572 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
575 CERROR("%s : invalid o_seq "DOSTID"\n",
576 ofd_name(ofd), POSTID(&oa->o_oi));
577 mutex_unlock(&oseq->os_create_lock);
578 ofd_seq_put(env, oseq);
579 GOTO(out, rc = -EINVAL);
583 u64 next_id = ofd_seq_last_oid(oseq) + 1;
584 int count = ofd_precreate_batch(ofd, diff);
586 rc = ofd_precreate_objects(env, ofd, next_id,
589 mutex_unlock(&oseq->os_create_lock);
590 ofd_seq_put(env, oseq);
597 mutex_unlock(&oseq->os_create_lock);
600 ofd_seq_put(env, oseq);
603 fo = ofd_object_find(env, ofd, fid);
605 GOTO(out, rc = PTR_ERR(fo));
608 ofd_read_lock(env, fo);
609 if (!ofd_object_exists(fo)) {
610 CERROR("%s: BRW to missing obj "DOSTID"\n",
611 exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
612 ofd_read_unlock(env, fo);
613 ofd_object_put(env, fo);
614 GOTO(out, rc = -ENOENT);
617 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
618 rc = ofd_verify_ff(env, fo, oa);
620 ofd_read_unlock(env, fo);
621 ofd_object_put(env, fo);
626 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
627 * space back if possible */
628 tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
630 /* parse remote buffers to local buffers and prepare the latter */
632 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
633 rc = dt_bufs_get(env, ofd_object_child(fo),
634 rnb + i, lnb + j, 1);
635 if (unlikely(rc < 0))
637 LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
638 /* correct index for local buffers to continue with */
639 for (k = 0; k < rc; k++) {
640 lnb[j+k].lnb_flags = rnb[i].rnb_flags;
641 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
642 lnb[j+k].lnb_rc = -ENOSPC;
646 LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
647 tot_bytes += rnb[i].rnb_len;
649 LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
651 rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
652 if (unlikely(rc != 0))
655 ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
658 dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
659 ofd_read_unlock(env, fo);
660 ofd_object_put(env, fo);
661 /* tgt_grant_prepare_write() was called, so we must commit */
662 tgt_grant_commit(exp, oa->o_grant_used, rc);
664 /* let's still process incoming grant information packed in the oa,
665 * but without enforcing grant since we won't proceed with the write.
666 * Just like a read request actually. */
667 tgt_grant_prepare_read(env, exp, oa);
672 * Prepare bulk IO requests for processing.
674 * This function does initial checks of IO and calls corresponding
675 * functions for read/write processing.
677 * \param[in] env execution environment
678 * \param[in] cmd IO type (read/write)
679 * \param[in] exp OBD export of client
680 * \param[in] oa OBDO structure from request
681 * \param[in] objcount always 1
682 * \param[in] obj object data
683 * \param[in] rnb remote buffers
684 * \param[in] nr_local number of local buffers
685 * \param[in] lnb local buffers
687 * \retval 0 on successful prepare
688 * \retval negative value on error
690 int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
691 struct obdo *oa, int objcount, struct obd_ioobj *obj,
692 struct niobuf_remote *rnb, int *nr_local,
693 struct niobuf_local *lnb)
695 struct tgt_session_info *tsi = tgt_ses_info(env);
696 struct ofd_device *ofd = ofd_exp(exp);
697 struct ofd_thread_info *info;
699 const struct lu_fid *fid = &oa->o_oi.oi_fid;
702 if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
703 CERROR("%s: bulk has too many pages %d, which exceeds the"
704 "maximum pages per RPC of %d\n",
705 exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
709 if (tgt_ses_req(tsi) == NULL) { /* echo client case */
710 info = ofd_info_init(env, exp);
713 info = tsi2ofd_info(tsi);
714 jobid = tsi->tsi_jobid;
719 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
720 struct ofd_seq *oseq;
722 oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
724 CERROR("%s: Can not find seq for "DOSTID
725 ": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
730 if (oseq->os_destroys_in_progress == 0) {
731 /* don't fail lookups for orphan recovery, it causes
732 * later LBUGs when objects still exist during
734 ofd_seq_put(env, oseq);
737 ofd_seq_put(env, oseq);
740 LASSERT(objcount == 1);
741 LASSERT(obj->ioo_bufcnt > 0);
743 if (cmd == OBD_BRW_WRITE) {
744 la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
745 rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
746 objcount, obj, rnb, nr_local, lnb, jobid);
747 } else if (cmd == OBD_BRW_READ) {
748 tgt_grant_prepare_read(env, exp, oa);
749 rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
750 obj->ioo_bufcnt, rnb, nr_local, lnb,
752 obdo_from_la(oa, &info->fti_attr, LA_ATIME);
754 CERROR("%s: wrong cmd %d received!\n",
755 exp->exp_obd->obd_name, cmd);
762 * Drop reference on local buffers for read bulk IO.
764 * This will free all local buffers use by this read request.
766 * \param[in] env execution environment
767 * \param[in] ofd OFD device
768 * \param[in] fid object FID
769 * \param[in] objcount always 1
770 * \param[in] niocount number of local buffers
771 * \param[in] lnb local buffers
773 * \retval 0 on successful execution
774 * \retval negative value on error
777 ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
778 const struct lu_fid *fid, int objcount, int niocount,
779 struct niobuf_local *lnb)
781 struct ofd_object *fo;
785 LASSERT(niocount > 0);
787 fo = ofd_object_find(env, ofd, fid);
791 LASSERT(ofd_object_exists(fo));
792 dt_bufs_put(env, ofd_object_child(fo), lnb, niocount);
794 ofd_read_unlock(env, fo);
795 ofd_object_put(env, fo);
796 /* second put is pair to object_get in ofd_preprw_read */
797 ofd_object_put(env, fo);
803 * Set attributes of object during write bulk IO processing.
805 * Change object attributes and write parent FID into extended
806 * attributes when needed.
808 * \param[in] env execution environment
809 * \param[in] ofd OFD device
810 * \param[in] ofd_obj OFD object
811 * \param[in] la object attributes
812 * \param[in] ff parent FID
814 * \retval 0 on successful attributes update
815 * \retval negative value on error
818 ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
819 struct ofd_object *ofd_obj, struct lu_attr *la,
820 struct filter_fid *ff)
822 struct ofd_thread_info *info = ofd_info(env);
823 __u64 valid = la->la_valid;
826 struct dt_object *dt_obj;
833 dt_obj = ofd_object_child(ofd_obj);
834 LASSERT(dt_obj != NULL);
836 la->la_valid &= LA_UID | LA_GID | LA_PROJID;
838 rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
843 rc = ofd_object_ff_load(env, ofd_obj);
850 if (!la->la_valid && !ff_needed)
851 /* no attributes to set */
854 th = ofd_trans_create(env, ofd);
856 GOTO(out, rc = PTR_ERR(th));
859 rc = dt_declare_attr_set(env, dt_obj, la, th);
865 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
866 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
867 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
868 le32_add_cpu(&ff->ff_parent.f_oid, -1);
870 info->fti_buf.lb_buf = ff;
871 info->fti_buf.lb_len = sizeof(*ff);
872 rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
873 XATTR_NAME_FID, 0, th);
878 /* We don't need a transno for this operation which will be re-executed
879 * anyway when the OST_WRITE (with a transno assigned) is replayed */
880 rc = dt_trans_start_local(env, ofd->ofd_osd , th);
884 /* set uid/gid/projid */
886 rc = dt_attr_set(env, dt_obj, la, th);
891 /* set filter fid EA */
893 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
896 rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
899 filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
905 dt_trans_stop(env, ofd->ofd_osd, th);
907 la->la_valid = valid;
911 struct ofd_soft_sync_callback {
912 struct dt_txn_commit_cb ossc_cb;
913 struct obd_export *ossc_exp;
917 * Callback function for "soft sync" update.
919 * Reset fed_soft_sync_count upon committing the "soft_sync" update.
920 * See ofd_soft_sync_cb_add() below for more details on soft sync.
922 * \param[in] env execution environment
923 * \param[in] th transaction handle
924 * \param[in] cb callback data
925 * \param[in] err error code
927 static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
928 struct dt_txn_commit_cb *cb, int err)
930 struct ofd_soft_sync_callback *ossc;
932 ossc = container_of(cb, struct ofd_soft_sync_callback, ossc_cb);
934 CDEBUG(D_INODE, "export %p soft sync count is reset\n", ossc->ossc_exp);
935 atomic_set(&ossc->ossc_exp->exp_filter_data.fed_soft_sync_count, 0);
937 class_export_cb_put(ossc->ossc_exp);
942 * Add callback for "soft sync" processing.
944 * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
945 * flag is set in client buffers. The intention is for this operation to
946 * commit pages belonging to a client which has "too many" outstanding
947 * unstable pages in its cache. See LU-2139 for details.
949 * This function adds callback to be called when commit is done.
951 * \param[in] th transaction handle
952 * \param[in] exp OBD export of client
954 * \retval 0 on successful callback adding
955 * \retval negative value on error
957 static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
959 struct ofd_soft_sync_callback *ossc;
960 struct dt_txn_commit_cb *dcb;
967 ossc->ossc_exp = class_export_cb_get(exp);
969 dcb = &ossc->ossc_cb;
970 dcb->dcb_func = ofd_cb_soft_sync;
971 INIT_LIST_HEAD(&dcb->dcb_linkage);
972 strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
974 rc = dt_trans_cb_add(th, dcb);
976 class_export_cb_put(exp);
984 * Commit bulk IO buffers to the storage.
986 * This function finalizes write IO processing by writing data to the disk.
987 * That write can be synchronous or asynchronous depending on buffers flags.
989 * \param[in] env execution environment
990 * \param[in] exp OBD export of client
991 * \param[in] ofd OFD device
992 * \param[in] fid FID of object
993 * \param[in] la object attributes
994 * \param[in] ff parent FID of object
995 * \param[in] objcount always 1
996 * \param[in] niocount number of local buffers
997 * \param[in] lnb local buffers
998 * \param[in] granted grant space consumed for the bulk I/O
999 * \param[in] old_rc result of processing at this point
1001 * \retval 0 on successful commit
1002 * \retval negative value on error
1005 ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
1006 struct ofd_device *ofd, const struct lu_fid *fid,
1007 struct lu_attr *la, struct filter_fid *ff, int objcount,
1008 int niocount, struct niobuf_local *lnb,
1009 unsigned long granted, int old_rc)
1011 struct filter_export_data *fed = &exp->exp_filter_data;
1012 struct ofd_object *fo;
1013 struct dt_object *o;
1019 bool soft_sync = false;
1020 bool cb_registered = false;
1021 bool fake_write = false;
1025 LASSERT(objcount == 1);
1027 fo = ofd_object_find(env, ofd, fid);
1028 LASSERT(fo != NULL);
1029 LASSERT(ofd_object_exists(fo));
1031 o = ofd_object_child(fo);
1035 GOTO(out, rc = old_rc);
1038 * The first write to each object must set some attributes. It is
1039 * important to set the uid/gid before calling
1040 * dt_declare_write_commit() since quota enforcement is now handled in
1043 rc = ofd_write_attr_set(env, ofd, fo, la, ff);
1047 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
1049 /* do fake write, to simulate the write case for performance testing */
1050 if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) {
1051 struct niobuf_local *last = &lnb[niocount - 1];
1052 __u64 file_size = last->lnb_file_offset + last->lnb_len;
1053 __u64 valid = la->la_valid;
1055 la->la_valid = LA_SIZE;
1057 rc = dt_attr_get(env, o, la);
1058 if (rc < 0 && rc != -ENOENT)
1061 if (file_size < la->la_size)
1062 file_size = la->la_size;
1064 /* dirty inode by setting file size */
1065 la->la_valid = valid | LA_SIZE;
1066 la->la_size = file_size;
1072 th = ofd_trans_create(env, ofd);
1074 GOTO(out, rc = PTR_ERR(th));
1076 th->th_sync |= ofd->ofd_syncjournal;
1077 if (th->th_sync == 0) {
1078 for (i = 0; i < niocount; i++) {
1079 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
1083 if (lnb[i].lnb_flags & OBD_BRW_SOFT_SYNC)
1088 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
1089 GOTO(out_stop, rc = -EINPROGRESS);
1091 if (likely(!fake_write)) {
1092 rc = dt_declare_write_commit(env, o, lnb, niocount, th);
1098 /* update [mac]time if needed */
1099 rc = dt_declare_attr_set(env, o, la, th);
1104 rc = ofd_trans_start(env, ofd, fo, th);
1108 if (likely(!fake_write)) {
1109 rc = dt_write_commit(env, o, lnb, niocount, th);
1115 rc = dt_attr_set(env, o, la, th);
1120 /* get attr to return */
1121 rc = dt_attr_get(env, o, la);
1124 /* Force commit to make the just-deleted blocks
1125 * reusable. LU-456 */
1129 /* do this before trans stop in case commit has finished */
1130 if (!th->th_sync && soft_sync && !cb_registered) {
1131 ofd_soft_sync_cb_add(th, exp);
1132 cb_registered = true;
1135 if (rc == 0 && granted > 0) {
1136 if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
1140 rc2 = ofd_trans_stop(env, ofd, th, rc);
1143 if (rc == -ENOSPC && retries++ < 3) {
1144 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
1150 /* reset fed_soft_sync_count upon non-SOFT_SYNC RPC */
1151 atomic_set(&fed->fed_soft_sync_count, 0);
1152 else if (atomic_inc_return(&fed->fed_soft_sync_count) ==
1153 ofd->ofd_soft_sync_limit)
1154 dt_commit_async(env, ofd->ofd_osd);
1157 dt_bufs_put(env, o, lnb, niocount);
1158 ofd_read_unlock(env, fo);
1159 ofd_object_put(env, fo);
1160 /* second put is pair to object_get in ofd_preprw_write */
1161 ofd_object_put(env, fo);
1163 tgt_grant_commit(exp, granted, old_rc);
1168 * Commit bulk IO to the storage.
1170 * This is companion function to the ofd_preprw(). It finishes bulk IO
1171 * request processing by committing buffers to the storage (WRITE) and/or
1172 * freeing those buffers (read/write). See ofd_commitrw_read() and
1173 * ofd_commitrw_write() for details about each type of IO.
1175 * \param[in] env execution environment
1176 * \param[in] cmd IO type (READ/WRITE)
1177 * \param[in] exp OBD export of client
1178 * \param[in] oa OBDO structure from client
1179 * \param[in] objcount always 1
1180 * \param[in] obj object data
1181 * \param[in] rnb remote buffers
1182 * \param[in] npages number of local buffers
1183 * \param[in] lnb local buffers
1184 * \param[in] old_rc result of processing at this point
1186 * \retval 0 on successful commit
1187 * \retval negative value on error
1189 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1190 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1191 struct niobuf_remote *rnb, int npages,
1192 struct niobuf_local *lnb, int old_rc)
1194 struct ofd_thread_info *info = ofd_info(env);
1195 struct ofd_mod_data *fmd;
1197 struct ofd_device *ofd = ofd_exp(exp);
1198 struct filter_fid *ff = NULL;
1199 const struct lu_fid *fid = &oa->o_oi.oi_fid;
1202 LASSERT(npages > 0);
1204 if (cmd == OBD_BRW_WRITE) {
1205 /* Don't update timestamps if this write is older than a
1206 * setattr which modifies the timestamps. b=10150 */
1208 /* XXX when we start having persistent reservations this needs
1209 * to be changed to ofd_fmd_get() to create the fmd if it
1210 * doesn't already exist so we can store the reservation handle
1212 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID;
1213 fmd = ofd_fmd_find(exp, fid);
1214 if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
1215 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
1217 ofd_fmd_put(exp, fmd);
1218 la_from_obdo(&info->fti_attr, oa, valid);
1220 if (oa->o_valid & OBD_MD_FLFID) {
1221 ff = &info->fti_mds_fid;
1222 ofd_prepare_fidea(ff, oa);
1225 rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
1226 ff, objcount, npages, lnb,
1227 oa->o_grant_used, old_rc);
1229 obdo_from_la(oa, &info->fti_attr,
1230 OFD_VALID_FLAGS | LA_GID | LA_UID |
1233 obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID |
1236 /* don't report overquota flag if we failed before reaching
1238 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
1239 /* return the overquota flags to client */
1240 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
1241 if (oa->o_valid & OBD_MD_FLFLAGS)
1242 oa->o_flags |= OBD_FL_NO_USRQUOTA;
1244 oa->o_flags = OBD_FL_NO_USRQUOTA;
1247 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
1248 if (oa->o_valid & OBD_MD_FLFLAGS)
1249 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
1251 oa->o_flags = OBD_FL_NO_GRPQUOTA;
1253 if (lnb[0].lnb_flags & OBD_BRW_OVER_PRJQUOTA) {
1254 if (oa->o_valid & OBD_MD_FLFLAGS)
1255 oa->o_flags |= OBD_FL_NO_PRJQUOTA;
1257 oa->o_flags = OBD_FL_NO_PRJQUOTA;
1260 oa->o_valid |= OBD_MD_FLFLAGS;
1261 oa->o_valid |= OBD_MD_FLALLQUOTA;
1263 } else if (cmd == OBD_BRW_READ) {
1264 struct ldlm_namespace *ns = ofd->ofd_namespace;
1266 /* If oa != NULL then ofd_preprw_read updated the inode
1267 * atime and we should update the lvb so that other glimpses
1268 * will also get the updated value. bug 5972 */
1269 if (oa && ns && ns->ns_lvbo && ns->ns_lvbo->lvbo_update) {
1270 struct ldlm_resource *rs = NULL;
1272 ost_fid_build_resid(fid, &info->fti_resid);
1273 rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
1276 ldlm_res_lvbo_update(rs, NULL, 1);
1277 ldlm_resource_putref(rs);
1280 rc = ofd_commitrw_read(env, ofd, fid, objcount,