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) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * Implementation of cl_io for LOV layer.
33 * Author: Nikita Danilov <nikita.danilov@sun.com>
34 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
37 #define DEBUG_SUBSYSTEM S_LOV
39 #include "lov_cl_internal.h"
45 static inline struct lov_io_sub *lov_sub_alloc(struct lov_io *lio, int index)
47 struct lov_io_sub *sub;
49 if (lio->lis_nr_subios == 0) {
50 LASSERT(lio->lis_single_subio_index == -1);
51 sub = &lio->lis_single_subio;
52 lio->lis_single_subio_index = index;
53 memset(sub, 0, sizeof(*sub));
59 INIT_LIST_HEAD(&sub->sub_list);
60 INIT_LIST_HEAD(&sub->sub_linkage);
61 sub->sub_subio_index = index;
67 static inline void lov_sub_free(struct lov_io *lio, struct lov_io_sub *sub)
69 if (sub->sub_subio_index == lio->lis_single_subio_index) {
70 LASSERT(sub == &lio->lis_single_subio);
71 lio->lis_single_subio_index = -1;
77 static void lov_io_sub_fini(const struct lu_env *env, struct lov_io *lio,
78 struct lov_io_sub *sub)
82 cl_io_fini(sub->sub_env, &sub->sub_io);
84 if (sub->sub_env && !IS_ERR(sub->sub_env)) {
85 cl_env_put(sub->sub_env, &sub->sub_refcheck);
92 is_index_within_mirror(struct lov_object *lov, int index, int mirror_index)
94 struct lov_layout_composite *comp = &lov->u.composite;
95 struct lov_mirror_entry *lre = &comp->lo_mirrors[mirror_index];
97 return (index >= lre->lre_start && index <= lre->lre_end);
100 static int lov_io_sub_init(const struct lu_env *env, struct lov_io *lio,
101 struct lov_io_sub *sub)
103 struct lov_object *lov = lio->lis_object;
104 struct cl_io *sub_io;
105 struct cl_object *sub_obj;
106 struct cl_io *io = lio->lis_cl.cis_io;
107 int index = lov_comp_entry(sub->sub_subio_index);
108 int stripe = lov_comp_stripe(sub->sub_subio_index);
110 LASSERT(sub->sub_env == NULL);
113 if (unlikely(!lov_r0(lov, index)->lo_sub ||
114 !lov_r0(lov, index)->lo_sub[stripe]))
117 LASSERTF(ergo(lov_is_flr(lov),
118 is_index_within_mirror(lov, index,
119 lio->lis_mirror_index)),
120 DFID "iot = %d, index = %d, mirror = %d\n",
121 PFID(lu_object_fid(lov2lu(lov))), io->ci_type, index,
122 lio->lis_mirror_index);
124 /* obtain new environment */
125 sub->sub_env = cl_env_get(&sub->sub_refcheck);
126 if (IS_ERR(sub->sub_env)) {
127 result = PTR_ERR(sub->sub_env);
131 sub_obj = lovsub2cl(lov_r0(lov, index)->lo_sub[stripe]);
132 sub_io = &sub->sub_io;
134 sub_io->ci_obj = sub_obj;
135 sub_io->ci_result = 0;
137 sub_io->ci_parent = io;
138 sub_io->ci_lockreq = io->ci_lockreq;
139 sub_io->ci_type = io->ci_type;
140 sub_io->ci_no_srvlock = io->ci_no_srvlock;
141 sub_io->ci_noatime = io->ci_noatime;
142 sub_io->ci_async_readahead = io->ci_async_readahead;
143 sub_io->ci_lock_no_expand = io->ci_lock_no_expand;
144 sub_io->ci_ndelay = io->ci_ndelay;
145 sub_io->ci_layout_version = io->ci_layout_version;
146 sub_io->ci_tried_all_mirrors = io->ci_tried_all_mirrors;
148 result = cl_io_sub_init(sub->sub_env, sub_io, io->ci_type, sub_obj);
151 lov_io_sub_fini(env, lio, sub);
156 struct lov_io_sub *lov_sub_get(const struct lu_env *env,
157 struct lov_io *lio, int index)
159 struct lov_io_sub *sub;
164 list_for_each_entry(sub, &lio->lis_subios, sub_list) {
165 if (sub->sub_subio_index == index) {
172 sub = lov_sub_alloc(lio, index);
174 GOTO(out, rc = -ENOMEM);
176 rc = lov_io_sub_init(env, lio, sub);
178 lov_sub_free(lio, sub);
182 list_add_tail(&sub->sub_list, &lio->lis_subios);
183 lio->lis_nr_subios++;
189 sub->sub_io.ci_noquota = lio->lis_cl.cis_io->ci_noquota;
193 /*****************************************************************************
198 static int lov_io_subio_init(const struct lu_env *env, struct lov_io *lio,
203 LASSERT(lio->lis_object != NULL);
205 INIT_LIST_HEAD(&lio->lis_subios);
206 lio->lis_single_subio_index = -1;
207 lio->lis_nr_subios = 0;
213 * Decide if it will need write intent RPC
215 static int lov_io_mirror_write_intent(struct lov_io *lio,
216 struct lov_object *obj, struct cl_io *io)
218 struct lov_layout_composite *comp = &obj->u.composite;
219 struct lu_extent *ext = &io->ci_write_intent;
220 struct lov_mirror_entry *lre;
221 struct lov_mirror_entry *primary;
222 struct lov_layout_entry *lle;
226 *ext = (typeof(*ext)) { lio->lis_pos, lio->lis_endpos };
227 io->ci_need_write_intent = 0;
229 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
230 cl_io_is_fallocate(io) || cl_io_is_trunc(io)))
234 * FLR: check if it needs to send a write intent RPC to server.
235 * Writing to sync_pending file needs write intent RPC to change
236 * the file state back to write_pending, so that the layout version
237 * can be increased when the state changes to sync_pending at a later
238 * time. Otherwise there exists a chance that an evicted client may
239 * dirty the file data while resync client is working on it.
240 * Designated I/O is allowed for resync workload.
242 if (lov_flr_state(obj) == LCM_FL_RDONLY ||
243 (lov_flr_state(obj) == LCM_FL_SYNC_PENDING &&
244 io->ci_designated_mirror == 0)) {
245 io->ci_need_write_intent = 1;
249 LASSERT((lov_flr_state(obj) == LCM_FL_WRITE_PENDING));
250 LASSERT(comp->lo_preferred_mirror >= 0);
253 * need to iterate all components to see if there are
254 * multiple components covering the writing component
256 primary = &comp->lo_mirrors[comp->lo_preferred_mirror];
257 LASSERT(!primary->lre_stale);
258 lov_foreach_mirror_layout_entry(obj, lle, primary) {
259 LASSERT(lle->lle_valid);
260 if (!lu_extent_is_overlapped(ext, lle->lle_extent))
263 ext->e_start = min(ext->e_start, lle->lle_extent->e_start);
264 ext->e_end = max(ext->e_end, lle->lle_extent->e_end);
268 CERROR(DFID ": cannot find any valid components covering "
269 "file extent "DEXT", mirror: %d\n",
270 PFID(lu_object_fid(lov2lu(obj))), PEXT(ext),
271 primary->lre_mirror_id);
276 lov_foreach_mirror_entry(obj, lre) {
280 lov_foreach_mirror_layout_entry(obj, lle, lre) {
284 if (lu_extent_is_overlapped(ext, lle->lle_extent)) {
291 CDEBUG(D_VFSTRACE, DFID "there are %zd components to be staled to "
292 "modify file extent "DEXT", iot: %d\n",
293 PFID(lu_object_fid(lov2lu(obj))), count, PEXT(ext), io->ci_type);
295 io->ci_need_write_intent = count > 0;
300 static int lov_io_mirror_init(struct lov_io *lio, struct lov_object *obj,
303 struct lov_layout_composite *comp = &obj->u.composite;
309 if (!lov_is_flr(obj)) {
310 /* only locks/pages are manipulated for CIT_MISC op, no
311 * cl_io_loop() will be called, don't check/set mirror info.
313 if (io->ci_type != CIT_MISC) {
314 LASSERT(comp->lo_preferred_mirror == 0);
315 lio->lis_mirror_index = comp->lo_preferred_mirror;
321 /* transfer the layout version for verification */
322 if (io->ci_layout_version == 0)
323 io->ci_layout_version = obj->lo_lsm->lsm_layout_gen;
325 /* find the corresponding mirror for designated mirror IO */
326 if (io->ci_designated_mirror > 0) {
327 struct lov_mirror_entry *entry;
329 LASSERT(!io->ci_ndelay);
331 CDEBUG(D_LAYOUT, "designated I/O mirror state: %d\n",
334 if ((cl_io_is_trunc(io) || io->ci_type == CIT_WRITE) &&
335 (io->ci_layout_version != obj->lo_lsm->lsm_layout_gen)) {
337 * For resync I/O, the ci_layout_version was the layout
338 * version when resync starts. If it doesn't match the
339 * current object layout version, it means the layout
345 io->ci_layout_version |= LU_LAYOUT_RESYNC;
348 lio->lis_mirror_index = -1;
349 lov_foreach_mirror_entry(obj, entry) {
350 if (entry->lre_mirror_id ==
351 io->ci_designated_mirror) {
352 lio->lis_mirror_index = index;
359 RETURN(lio->lis_mirror_index < 0 ? -EINVAL : 0);
362 result = lov_io_mirror_write_intent(lio, obj, io);
366 if (io->ci_need_write_intent) {
367 CDEBUG(D_VFSTRACE, DFID " need write intent for [%llu, %llu)\n",
368 PFID(lu_object_fid(lov2lu(obj))),
369 lio->lis_pos, lio->lis_endpos);
371 if (cl_io_is_trunc(io)) {
373 * for truncate, we uses [size, EOF) to judge whether
374 * a write intent needs to be send, but we need to
375 * restore the write extent to [0, size], in truncate,
376 * the byte in the size position is accessed.
378 io->ci_write_intent.e_start = 0;
379 io->ci_write_intent.e_end =
380 io->u.ci_setattr.sa_attr.lvb_size + 1;
382 /* stop cl_io_init() loop */
386 if (io->ci_ndelay_tried == 0 || /* first time to try */
387 /* reset the mirror index if layout has changed */
388 lio->lis_mirror_layout_gen != obj->lo_lsm->lsm_layout_gen) {
389 lio->lis_mirror_layout_gen = obj->lo_lsm->lsm_layout_gen;
390 index = lio->lis_mirror_index = comp->lo_preferred_mirror;
392 index = lio->lis_mirror_index;
395 /* move mirror index to the next one */
396 index = (index + 1) % comp->lo_mirror_count;
399 for (i = 0; i < comp->lo_mirror_count; i++) {
400 struct lu_extent ext = { .e_start = lio->lis_pos,
401 .e_end = lio->lis_pos + 1 };
402 struct lov_mirror_entry *lre;
403 struct lov_layout_entry *lle;
406 lre = &comp->lo_mirrors[(index + i) % comp->lo_mirror_count];
410 if (lre->lre_foreign)
413 lov_foreach_mirror_layout_entry(obj, lle, lre) {
417 if (lu_extent_is_overlapped(&ext, lle->lle_extent)) {
421 } /* each component of the mirror */
423 index = (index + i) % comp->lo_mirror_count;
428 if (i == comp->lo_mirror_count) {
429 CERROR(DFID": failed to find a component covering "
430 "I/O region at %llu\n",
431 PFID(lu_object_fid(lov2lu(obj))), lio->lis_pos);
433 dump_lsm(D_ERROR, obj->lo_lsm);
438 CDEBUG(D_VFSTRACE, DFID ": flr state: %d, move mirror from %d to %d, "
439 "have retried: %d, mirror count: %d\n",
440 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj),
441 lio->lis_mirror_index, index, io->ci_ndelay_tried,
442 comp->lo_mirror_count);
444 lio->lis_mirror_index = index;
447 * FLR: if all mirrors have been tried once, most likely the network
448 * of this client has been partitioned. We should relinquish CPU for
449 * a while before trying again.
451 if (io->ci_ndelay && io->ci_ndelay_tried > 0 &&
452 (io->ci_ndelay_tried % comp->lo_mirror_count == 0)) {
453 schedule_timeout_interruptible(cfs_time_seconds(1) / 100);
454 if (signal_pending(current))
458 * we'd set ci_tried_all_mirrors to turn off fast mirror
459 * switching for read after we've tried all mirrors several
462 io->ci_tried_all_mirrors = io->ci_ndelay_tried %
463 (comp->lo_mirror_count * 4) == 0;
465 ++io->ci_ndelay_tried;
467 CDEBUG(D_VFSTRACE, "use %sdelayed RPC state for this IO\n",
468 io->ci_ndelay ? "non-" : "");
473 static int lov_io_slice_init(struct lov_io *lio,
474 struct lov_object *obj, struct cl_io *io)
481 lio->lis_object = obj;
483 switch (io->ci_type) {
486 lio->lis_pos = io->u.ci_rw.crw_pos;
487 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
488 lio->lis_io_endpos = lio->lis_endpos;
489 if (cl_io_is_append(io)) {
490 LASSERT(io->ci_type == CIT_WRITE);
493 * If there is LOV EA hole, then we may cannot locate
494 * the current file-tail exactly.
496 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
498 GOTO(out, result = -EIO);
501 lio->lis_endpos = OBD_OBJECT_EOF;
506 if (cl_io_is_fallocate(io)) {
507 lio->lis_pos = io->u.ci_setattr.sa_falloc_offset;
508 lio->lis_endpos = io->u.ci_setattr.sa_falloc_end;
509 } else if (cl_io_is_trunc(io)) {
510 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
511 lio->lis_endpos = OBD_OBJECT_EOF;
514 lio->lis_endpos = OBD_OBJECT_EOF;
518 case CIT_DATA_VERSION:
520 lio->lis_endpos = OBD_OBJECT_EOF;
524 pgoff_t index = io->u.ci_fault.ft_index;
526 lio->lis_pos = cl_offset(io->ci_obj, index);
527 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
532 lio->lis_pos = io->u.ci_fsync.fi_start;
533 lio->lis_endpos = io->u.ci_fsync.fi_end;
538 lio->lis_pos = io->u.ci_ladvise.li_start;
539 lio->lis_endpos = io->u.ci_ladvise.li_end;
544 lio->lis_pos = io->u.ci_lseek.ls_start;
545 lio->lis_endpos = OBD_OBJECT_EOF;
551 lio->lis_endpos = OBD_OBJECT_EOF;
553 if (lov_flr_state(obj) == LCM_FL_RDONLY &&
554 !OBD_FAIL_CHECK(OBD_FAIL_FLR_GLIMPSE_IMMUTABLE))
555 /* SoM is accurate, no need glimpse */
556 GOTO(out, result = 1);
561 lio->lis_endpos = OBD_OBJECT_EOF;
569 * CIT_MISC + ci_ignore_layout can identify the I/O from the OSC layer,
570 * it won't care/access lov layout related info.
572 if (io->ci_ignore_layout && io->ci_type == CIT_MISC)
573 GOTO(out, result = 0);
575 LASSERT(obj->lo_lsm != NULL);
577 result = lov_io_mirror_init(lio, obj, io);
581 /* check if it needs to instantiate layout */
582 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
583 cl_io_is_fallocate(io) ||
584 (cl_io_is_trunc(io) && io->u.ci_setattr.sa_attr.lvb_size > 0)))
585 GOTO(out, result = 0);
588 * for truncate, it only needs to instantiate the components
589 * before the truncated size.
591 if (cl_io_is_trunc(io)) {
592 io->ci_write_intent.e_start = 0;
593 /* for writes, e_end is endpos, the location of the file
594 * pointer after the write is completed, so it is not accessed.
595 * For truncate, 'end' is the size, and *is* acccessed.
596 * In other words, writes are [start, end), but truncate is
597 * [start, size], where both are included. So add 1 to the
598 * size when creating the write intent to account for this.
600 io->ci_write_intent.e_end =
601 io->u.ci_setattr.sa_attr.lvb_size + 1;
603 io->ci_write_intent.e_start = lio->lis_pos;
604 io->ci_write_intent.e_end = lio->lis_endpos;
608 lov_foreach_io_layout(index, lio, &io->ci_write_intent) {
609 if (!lsm_entry_inited(obj->lo_lsm, index)) {
610 io->ci_need_write_intent = 1;
615 if (io->ci_need_write_intent && io->ci_designated_mirror > 0) {
617 * REINT_SYNC RPC has already tried to instantiate all of the
618 * components involved, obviously it didn't succeed. Skip this
619 * mirror for now. The server won't be able to figure out
620 * which mirror it should instantiate components
622 CERROR(DFID": trying to instantiate components for designated "
623 "I/O, file state: %d\n",
624 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj));
626 io->ci_need_write_intent = 0;
627 GOTO(out, result = -EIO);
630 if (io->ci_need_write_intent)
631 GOTO(out, result = 1);
639 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
641 struct lov_io *lio = cl2lov_io(env, ios);
642 struct lov_object *lov = cl2lov(ios->cis_obj);
643 struct lov_io_sub *sub;
646 LASSERT(list_empty(&lio->lis_active));
648 while ((sub = list_first_entry_or_null(&lio->lis_subios,
650 sub_list)) != NULL) {
651 list_del_init(&sub->sub_list);
652 lio->lis_nr_subios--;
654 lov_io_sub_fini(env, lio, sub);
655 lov_sub_free(lio, sub);
657 LASSERT(lio->lis_nr_subios == 0);
659 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
660 if (atomic_dec_and_test(&lov->lo_active_ios))
661 wake_up(&lov->lo_waitq);
665 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
666 loff_t start, loff_t end)
668 struct cl_io *io = &sub->sub_io;
669 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
670 struct cl_io *parent = lio->lis_cl.cis_io;
671 int index = lov_comp_entry(sub->sub_subio_index);
672 int stripe = lov_comp_stripe(sub->sub_subio_index);
674 switch (io->ci_type) {
676 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
677 io->u.ci_setattr.sa_attr_flags =
678 parent->u.ci_setattr.sa_attr_flags;
679 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
680 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
681 io->u.ci_setattr.sa_falloc_mode =
682 parent->u.ci_setattr.sa_falloc_mode;
683 io->u.ci_setattr.sa_stripe_index = stripe;
684 io->u.ci_setattr.sa_parent_fid =
685 parent->u.ci_setattr.sa_parent_fid;
686 /* For SETATTR(fallocate) pass the subtype to lower IO */
687 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
688 if (cl_io_is_fallocate(io)) {
689 io->u.ci_setattr.sa_falloc_offset = start;
690 io->u.ci_setattr.sa_falloc_end = end;
692 if (cl_io_is_trunc(io)) {
693 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
695 new_size = lov_size_to_stripe(lsm, index, new_size,
697 io->u.ci_setattr.sa_attr.lvb_size = new_size;
699 lov_lsm2layout(lsm, lsm->lsm_entries[index],
700 &io->u.ci_setattr.sa_layout);
703 case CIT_DATA_VERSION: {
704 io->u.ci_data_version.dv_data_version = 0;
705 io->u.ci_data_version.dv_flags =
706 parent->u.ci_data_version.dv_flags;
710 struct cl_object *obj = parent->ci_obj;
711 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
713 io->u.ci_fault = parent->u.ci_fault;
714 off = lov_size_to_stripe(lsm, index, off, stripe);
715 io->u.ci_fault.ft_index = cl_index(obj, off);
719 io->u.ci_fsync.fi_start = start;
720 io->u.ci_fsync.fi_end = end;
721 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
722 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
727 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
728 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
729 if (cl_io_is_append(parent)) {
730 io->u.ci_wr.wr_append = 1;
732 io->u.ci_rw.crw_pos = start;
733 io->u.ci_rw.crw_count = end - start;
738 io->u.ci_ladvise.li_start = start;
739 io->u.ci_ladvise.li_end = end;
740 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
741 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
742 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
746 io->u.ci_lseek.ls_start = start;
747 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
748 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
758 static loff_t lov_offset_mod(loff_t val, int delta)
760 if (val != OBD_OBJECT_EOF)
765 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
766 struct lov_io_sub *sub, u64 start, u64 end)
770 end = lov_offset_mod(end, 1);
771 lov_io_sub_inherit(sub, lio, start, end);
772 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
774 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
778 list_add_tail(&sub->sub_linkage, &lio->lis_active);
782 static int lov_io_iter_init(const struct lu_env *env,
783 const struct cl_io_slice *ios)
785 struct lov_io *lio = cl2lov_io(env, ios);
786 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
787 struct lov_io_sub *sub;
788 struct lu_extent ext;
794 ext.e_start = lio->lis_pos;
795 ext.e_end = lio->lis_endpos;
797 lov_foreach_io_layout(index, lio, &ext) {
798 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
799 struct lov_layout_raid0 *r0 = &le->lle_raid0;
803 bool tested_trunc_stripe = false;
805 r0->lo_trunc_stripeno = -1;
807 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
808 index, lsm->lsm_entries[index]->lsme_flags);
809 if (!lsm_entry_inited(lsm, index)) {
811 * Read from uninitialized components should return
817 if (lsm_entry_is_foreign(lsm, index))
820 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
821 CERROR("I/O to invalid component: %d, mirror: %d\n",
822 index, lio->lis_mirror_index);
826 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
827 if (!lov_stripe_intersects(lsm, index, stripe,
831 if (unlikely(!r0->lo_sub[stripe])) {
832 if (ios->cis_io->ci_type == CIT_READ ||
833 ios->cis_io->ci_type == CIT_WRITE ||
834 ios->cis_io->ci_type == CIT_FAULT)
840 if (cl_io_is_trunc(ios->cis_io) &&
841 !tested_trunc_stripe) {
845 prev = (stripe == 0) ? r0->lo_nr - 1 :
848 * Only involving previous stripe if the
849 * truncate in this component is at the
850 * beginning of this stripe.
852 tested_trunc_stripe = true;
853 if (ext.e_start < lsm->lsm_entries[index]->
854 lsme_extent.e_start) {
855 /* need previous stripe involvement */
856 r0->lo_trunc_stripeno = prev;
858 tr_start = ext.e_start;
859 tr_start = lov_do_div64(tr_start,
860 stripe_width(lsm, index));
861 /* tr_start %= stripe_swidth */
862 if (tr_start == stripe * lsm->
865 r0->lo_trunc_stripeno = prev;
869 /* if the last stripe is the trunc stripeno */
870 if (r0->lo_trunc_stripeno == stripe)
871 r0->lo_trunc_stripeno = -1;
873 sub = lov_sub_get(env, lio,
874 lov_comp_index(index, stripe));
878 rc = lov_io_add_sub(env, lio, sub, start, end);
885 if (r0->lo_trunc_stripeno != -1) {
886 stripe = r0->lo_trunc_stripeno;
887 if (unlikely(!r0->lo_sub[stripe])) {
888 r0->lo_trunc_stripeno = -1;
891 sub = lov_sub_get(env, lio,
892 lov_comp_index(index, stripe));
897 * the prev sub could be used by another truncate, we'd
898 * skip it. LU-14128 happends when expand truncate +
899 * read get wrong kms.
901 if (!list_empty(&sub->sub_linkage)) {
902 r0->lo_trunc_stripeno = -1;
906 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
908 rc = lov_io_add_sub(env, lio, sub, start, end);
917 static int lov_io_rw_iter_init(const struct lu_env *env,
918 const struct cl_io_slice *ios)
920 struct lov_io *lio = cl2lov_io(env, ios);
921 struct cl_io *io = ios->cis_io;
922 struct lov_stripe_md_entry *lse;
923 loff_t start = io->u.ci_rw.crw_pos;
927 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
930 if (cl_io_is_append(io))
931 RETURN(lov_io_iter_init(env, ios));
933 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
934 if (index < 0) { /* non-existing layout component */
935 if (io->ci_type == CIT_READ) {
937 * TODO: it needs to detect the next component and
938 * then set the next pos
942 RETURN(lov_io_iter_init(env, ios));
948 if (!lov_entry(lio->lis_object, index)->lle_valid &&
949 !io->ci_designated_mirror)
950 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
952 lse = lov_lse(lio->lis_object, index);
954 if (lsme_is_foreign(lse))
957 next = MAX_LFS_FILESIZE;
958 if (lse->lsme_stripe_count > 1) {
959 unsigned long ssize = lse->lsme_stripe_size;
961 lov_do_div64(start, ssize);
962 next = (start + 1) * ssize;
963 if (next <= start * ssize)
964 next = MAX_LFS_FILESIZE;
967 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
968 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
969 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
970 next = min_t(__u64, next, lse->lsme_extent.e_end);
971 next = min_t(loff_t, next, lio->lis_io_endpos);
973 io->ci_continue = next < lio->lis_io_endpos;
974 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
975 lio->lis_pos = io->u.ci_rw.crw_pos;
976 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
978 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
979 (__u64)start, lio->lis_pos, lio->lis_endpos,
980 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
983 * XXX The following call should be optimized: we know, that
984 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
986 RETURN(lov_io_iter_init(env, ios));
989 static int lov_io_setattr_iter_init(const struct lu_env *env,
990 const struct cl_io_slice *ios)
992 struct lov_io *lio = cl2lov_io(env, ios);
993 struct cl_io *io = ios->cis_io;
997 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
998 index = lov_io_layout_at(lio, lio->lis_pos - 1);
999 /* no entry found for such offset */
1001 RETURN(io->ci_result = -ENODATA);
1004 RETURN(lov_io_iter_init(env, ios));
1007 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1008 int (*iofunc)(const struct lu_env *, struct cl_io *))
1010 struct cl_io *parent = lio->lis_cl.cis_io;
1011 struct lov_io_sub *sub;
1015 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1016 rc = iofunc(sub->sub_env, &sub->sub_io);
1020 if (parent->ci_result == 0)
1021 parent->ci_result = sub->sub_io.ci_result;
1026 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1029 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1032 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1035 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1038 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1042 * It's possible that lov_io_start() wasn't called against this
1043 * sub-io, either because previous sub-io failed, or upper layer
1046 if (io->ci_state == CIS_IO_GOING)
1049 io->ci_state = CIS_IO_FINISHED;
1053 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1055 cl_io_iter_fini(env, io);
1059 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1061 cl_io_unlock(env, io);
1065 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1069 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1074 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1076 struct lov_io *lio = cl2lov_io(env, ios);
1077 struct cl_io *parent = lio->lis_cl.cis_io;
1078 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1079 struct lov_io_sub *sub;
1082 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1083 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1085 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1087 pdv->dv_data_version += sdv->dv_data_version;
1088 if (pdv->dv_layout_version > sdv->dv_layout_version)
1089 pdv->dv_layout_version = sdv->dv_layout_version;
1091 if (parent->ci_result == 0)
1092 parent->ci_result = sub->sub_io.ci_result;
1098 static void lov_io_iter_fini(const struct lu_env *env,
1099 const struct cl_io_slice *ios)
1101 struct lov_io *lio = cl2lov_io(env, ios);
1105 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1107 while (!list_empty(&lio->lis_active))
1108 list_del_init(lio->lis_active.next);
1112 static void lov_io_unlock(const struct lu_env *env,
1113 const struct cl_io_slice *ios)
1118 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1123 static int lov_io_read_ahead(const struct lu_env *env,
1124 const struct cl_io_slice *ios,
1125 pgoff_t start, struct cl_read_ahead *ra)
1127 struct lov_io *lio = cl2lov_io(env, ios);
1128 struct lov_object *loo = lio->lis_object;
1129 struct cl_object *obj = lov2cl(loo);
1130 struct lov_layout_raid0 *r0;
1131 struct lov_io_sub *sub;
1135 unsigned int pps; /* pages per stripe */
1141 offset = cl_offset(obj, start);
1142 index = lov_io_layout_at(lio, offset);
1143 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1144 lsm_entry_is_foreign(loo->lo_lsm, index))
1147 /* avoid readahead to expand to stale components */
1148 if (!lov_entry(loo, index)->lle_valid)
1151 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1153 r0 = lov_r0(loo, index);
1154 if (unlikely(!r0->lo_sub[stripe]))
1157 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1159 RETURN(PTR_ERR(sub));
1161 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1162 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1163 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1166 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1167 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1173 * Adjust the stripe index by layout of comp. ra->cra_end is the
1174 * maximum page index covered by an underlying DLM lock.
1175 * This function converts cra_end from stripe level to file level, and
1176 * make sure it's not beyond stripe and component boundary.
1179 /* cra_end is stripe level, convert it into file level */
1180 ra_end = ra->cra_end_idx;
1181 if (ra_end != CL_PAGE_EOF)
1182 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1185 /* boundary of current component */
1186 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1187 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1188 ra->cra_end_idx = ra_end - 1;
1190 if (r0->lo_nr == 1) /* single stripe file */
1193 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1195 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1196 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1197 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1198 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1200 /* never exceed the end of the stripe */
1201 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1202 start + pps - start % pps - 1);
1206 int lov_io_lru_reserve(const struct lu_env *env,
1207 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1209 struct lov_io *lio = cl2lov_io(env, ios);
1210 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1211 struct lov_io_sub *sub;
1212 struct lu_extent ext;
1219 ext.e_end = pos + bytes;
1220 lov_foreach_io_layout(index, lio, &ext) {
1221 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1222 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1227 if (!lsm_entry_inited(lsm, index))
1230 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1231 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1232 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1233 index, lio->lis_mirror_index);
1237 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1238 if (!lov_stripe_intersects(lsm, index, stripe,
1239 &ext, &start, &end))
1242 if (unlikely(!r0->lo_sub[stripe]))
1245 sub = lov_sub_get(env, lio,
1246 lov_comp_index(index, stripe));
1248 return PTR_ERR(sub);
1250 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1261 * lov implementation of cl_operations::cio_submit() method. It takes a list
1262 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1263 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1266 * Major complication of this function is a need to handle memory cleansing:
1267 * cl_io_submit() is called to write out pages as a part of VM memory
1268 * reclamation, and hence it may not fail due to memory shortages (system
1269 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1270 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1271 * not-memory cleansing context), and in case of memory shortage, these
1272 * pre-allocated resources are used by lov_io_submit() under
1273 * lov_device::ld_mutex mutex.
1275 static int lov_io_submit(const struct lu_env *env,
1276 const struct cl_io_slice *ios,
1277 enum cl_req_type crt, struct cl_2queue *queue)
1279 struct cl_page_list *qin = &queue->c2_qin;
1280 struct lov_io *lio = cl2lov_io(env, ios);
1281 struct lov_io_sub *sub;
1282 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1283 struct cl_page *page = cl_page_list_first(qin);
1284 struct cl_page *tmp;
1290 if (page->cp_type == CPT_TRANSIENT)
1293 cl_page_list_init(plist);
1294 while (qin->pl_nr > 0) {
1295 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1297 page = cl_page_list_first(qin);
1298 if (lov_page_is_empty(page)) {
1299 cl_page_list_move(&queue->c2_qout, qin, page);
1302 * it could only be mirror read to get here therefore
1303 * the pages will be transient. We don't care about
1304 * the return code of cl_page_prep() at all.
1306 (void) cl_page_prep(env, ios->cis_io, page, crt);
1307 cl_page_completion(env, page, crt, 0);
1311 cl_2queue_init(cl2q);
1312 cl_page_list_move(&cl2q->c2_qin, qin, page);
1314 index = page->cp_lov_index;
1315 /* DIO is already split by stripe */
1317 cl_page_list_for_each_safe(page, tmp, qin) {
1318 /* this page is not on this stripe */
1319 if (index != page->cp_lov_index)
1322 cl_page_list_move(&cl2q->c2_qin, qin, page);
1325 cl_page_list_splice(qin, &cl2q->c2_qin);
1328 sub = lov_sub_get(env, lio, index);
1330 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1336 cl_page_list_splice(&cl2q->c2_qin, plist);
1337 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1338 cl_2queue_fini(env, cl2q);
1344 cl_page_list_splice(plist, qin);
1345 cl_page_list_fini(env, plist);
1350 static int lov_io_commit_async(const struct lu_env *env,
1351 const struct cl_io_slice *ios,
1352 struct cl_page_list *queue, int from, int to,
1355 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1356 struct lov_io *lio = cl2lov_io(env, ios);
1357 struct lov_io_sub *sub;
1358 struct cl_page *page;
1362 if (lio->lis_nr_subios == 1) {
1363 int idx = lio->lis_single_subio_index;
1365 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1367 sub = lov_sub_get(env, lio, idx);
1368 LASSERT(!IS_ERR(sub));
1369 LASSERT(sub == &lio->lis_single_subio);
1370 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1375 cl_page_list_init(plist);
1376 while (queue->pl_nr > 0) {
1380 LASSERT(plist->pl_nr == 0);
1381 page = cl_page_list_first(queue);
1382 LASSERT(!lov_page_is_empty(page));
1384 cl_page_list_move(plist, queue, page);
1386 index = page->cp_lov_index;
1387 while (queue->pl_nr > 0) {
1388 page = cl_page_list_first(queue);
1389 if (index != page->cp_lov_index)
1392 cl_page_list_move(plist, queue, page);
1395 if (queue->pl_nr > 0) /* still has more pages */
1396 stripe_to = PAGE_SIZE;
1398 sub = lov_sub_get(env, lio, index);
1400 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1401 plist, from, stripe_to, cb);
1407 if (plist->pl_nr > 0) /* short write */
1412 if (lov_comp_entry(index) !=
1413 lov_comp_entry(page->cp_lov_index))
1414 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1417 /* for error case, add the page back into the qin list */
1418 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1419 while (plist->pl_nr > 0) {
1420 /* error occurred, add the uncommitted pages back into queue */
1421 page = cl_page_list_last(plist);
1422 cl_page_list_move_head(queue, plist, page);
1428 static int lov_io_fault_start(const struct lu_env *env,
1429 const struct cl_io_slice *ios)
1431 struct cl_fault_io *fio;
1433 struct lov_io_sub *sub;
1440 fio = &ios->cis_io->u.ci_fault;
1441 lio = cl2lov_io(env, ios);
1444 * LU-14502: ft_page could be an existing cl_page associated with
1445 * the vmpage covering the fault index, and the page may still
1446 * refer to another mirror of an old IO.
1448 if (lov_is_flr(lio->lis_object)) {
1449 offset = cl_offset(ios->cis_obj, fio->ft_index);
1450 entry = lov_io_layout_at(lio, offset);
1452 CERROR(DFID": page fault index %lu invalid component: "
1454 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1455 fio->ft_index, entry,
1456 lio->lis_mirror_index);
1459 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1462 if (fio->ft_page->cp_lov_index !=
1463 lov_comp_index(entry, stripe)) {
1464 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1465 "at mirror %u comp entry %u stripe %u, "
1466 "been used with comp entry %u stripe %u\n",
1467 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1468 fio->ft_index, lio->lis_mirror_index,
1470 lov_comp_entry(fio->ft_page->cp_lov_index),
1471 lov_comp_stripe(fio->ft_page->cp_lov_index));
1473 fio->ft_page->cp_lov_index =
1474 lov_comp_index(entry, stripe);
1478 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1479 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1481 RETURN(lov_io_start(env, ios));
1484 static int lov_io_setattr_start(const struct lu_env *env,
1485 const struct cl_io_slice *ios)
1487 struct lov_io *lio = cl2lov_io(env, ios);
1488 struct cl_io *parent = ios->cis_io;
1489 struct lov_io_sub *sub;
1490 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1494 if (cl_io_is_fallocate(parent)) {
1495 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1496 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1497 int index = lov_comp_entry(sub->sub_subio_index);
1498 int stripe = lov_comp_stripe(sub->sub_subio_index);
1500 size = lov_size_to_stripe(lsm, index, size, stripe);
1501 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1502 sub->sub_io.u.ci_setattr.sa_avalid =
1503 parent->u.ci_setattr.sa_avalid;
1507 RETURN(lov_io_start(env, ios));
1510 static void lov_io_fsync_end(const struct lu_env *env,
1511 const struct cl_io_slice *ios)
1513 struct lov_io *lio = cl2lov_io(env, ios);
1514 struct lov_io_sub *sub;
1515 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1519 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1520 struct cl_io *subio = &sub->sub_io;
1522 lov_io_end_wrapper(sub->sub_env, subio);
1524 if (subio->ci_result == 0)
1525 *written += subio->u.ci_fsync.fi_nr_written;
1530 static void lov_io_lseek_end(const struct lu_env *env,
1531 const struct cl_io_slice *ios)
1533 struct lov_io *lio = cl2lov_io(env, ios);
1534 struct cl_io *io = lio->lis_cl.cis_io;
1535 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1536 struct lov_io_sub *sub;
1537 loff_t offset = -ENXIO;
1539 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1543 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1544 struct cl_io *subio = &sub->sub_io;
1545 int index = lov_comp_entry(sub->sub_subio_index);
1546 int stripe = lov_comp_stripe(sub->sub_subio_index);
1547 loff_t sub_off, lov_off;
1548 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1550 lov_io_end_wrapper(sub->sub_env, subio);
1552 if (io->ci_result == 0)
1553 io->ci_result = sub->sub_io.ci_result;
1558 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1559 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1560 index, stripe, seek_hole ? "HOLE" : "DATA",
1561 subio->u.ci_lseek.ls_start);
1563 /* first subio with positive result is what we need */
1564 sub_off = subio->u.ci_lseek.ls_result;
1565 /* Expected error, offset is out of stripe file size */
1566 if (sub_off == -ENXIO)
1568 /* Any other errors are not expected with ci_result == 0 */
1570 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1572 io->ci_result = sub_off;
1575 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1577 /* the only way to get negatove lov_off here is too big
1578 * result. Return -EOVERFLOW then.
1580 io->ci_result = -EOVERFLOW;
1581 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1582 (u64)lov_off, io->ci_result);
1585 if (lov_off < io->u.ci_lseek.ls_start) {
1586 io->ci_result = -EINVAL;
1587 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1588 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1591 /* resulting offset can be out of component range if stripe
1592 * object is full and its file size was returned as virtual
1593 * hole start. Skip this result, the next component will give
1594 * us correct lseek result but keep possible hole offset in
1595 * case there is no more components ahead
1597 if (lov_off >= comp_end) {
1598 /* must be SEEK_HOLE case */
1599 if (likely(seek_hole)) {
1600 /* save comp end as potential hole offset */
1601 hole_off = max_t(__u64, comp_end, hole_off);
1603 io->ci_result = -EINVAL;
1605 "off %lld >= comp_end %llu: rc = %d\n",
1606 lov_off, comp_end, io->ci_result);
1611 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1612 seek_hole ? "HOLE" : "DATA",
1613 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1614 sub->sub_io.ci_result);
1615 offset = min_t(__u64, offset, lov_off);
1617 /* no result but some component returns hole as component end */
1618 if (seek_hole && offset == -ENXIO && hole_off > 0)
1621 io->u.ci_lseek.ls_result = offset;
1625 static const struct cl_io_operations lov_io_ops = {
1628 .cio_fini = lov_io_fini,
1629 .cio_iter_init = lov_io_rw_iter_init,
1630 .cio_iter_fini = lov_io_iter_fini,
1631 .cio_lock = lov_io_lock,
1632 .cio_unlock = lov_io_unlock,
1633 .cio_start = lov_io_start,
1634 .cio_end = lov_io_end
1637 .cio_fini = lov_io_fini,
1638 .cio_iter_init = lov_io_rw_iter_init,
1639 .cio_iter_fini = lov_io_iter_fini,
1640 .cio_lock = lov_io_lock,
1641 .cio_unlock = lov_io_unlock,
1642 .cio_start = lov_io_start,
1643 .cio_end = lov_io_end
1646 .cio_fini = lov_io_fini,
1647 .cio_iter_init = lov_io_setattr_iter_init,
1648 .cio_iter_fini = lov_io_iter_fini,
1649 .cio_lock = lov_io_lock,
1650 .cio_unlock = lov_io_unlock,
1651 .cio_start = lov_io_setattr_start,
1652 .cio_end = lov_io_end
1654 [CIT_DATA_VERSION] = {
1655 .cio_fini = lov_io_fini,
1656 .cio_iter_init = lov_io_iter_init,
1657 .cio_iter_fini = lov_io_iter_fini,
1658 .cio_lock = lov_io_lock,
1659 .cio_unlock = lov_io_unlock,
1660 .cio_start = lov_io_start,
1661 .cio_end = lov_io_data_version_end,
1664 .cio_fini = lov_io_fini,
1665 .cio_iter_init = lov_io_iter_init,
1666 .cio_iter_fini = lov_io_iter_fini,
1667 .cio_lock = lov_io_lock,
1668 .cio_unlock = lov_io_unlock,
1669 .cio_start = lov_io_fault_start,
1670 .cio_end = lov_io_end
1673 .cio_fini = lov_io_fini,
1674 .cio_iter_init = lov_io_iter_init,
1675 .cio_iter_fini = lov_io_iter_fini,
1676 .cio_lock = lov_io_lock,
1677 .cio_unlock = lov_io_unlock,
1678 .cio_start = lov_io_start,
1679 .cio_end = lov_io_fsync_end
1682 .cio_fini = lov_io_fini,
1683 .cio_iter_init = lov_io_iter_init,
1684 .cio_iter_fini = lov_io_iter_fini,
1685 .cio_lock = lov_io_lock,
1686 .cio_unlock = lov_io_unlock,
1687 .cio_start = lov_io_start,
1688 .cio_end = lov_io_end
1691 .cio_fini = lov_io_fini,
1692 .cio_iter_init = lov_io_iter_init,
1693 .cio_iter_fini = lov_io_iter_fini,
1694 .cio_lock = lov_io_lock,
1695 .cio_unlock = lov_io_unlock,
1696 .cio_start = lov_io_start,
1697 .cio_end = lov_io_lseek_end
1700 .cio_fini = lov_io_fini,
1703 .cio_fini = lov_io_fini
1706 .cio_read_ahead = lov_io_read_ahead,
1707 .cio_lru_reserve = lov_io_lru_reserve,
1708 .cio_submit = lov_io_submit,
1709 .cio_commit_async = lov_io_commit_async,
1712 /*****************************************************************************
1714 * Empty lov io operations.
1718 static void lov_empty_io_fini(const struct lu_env *env,
1719 const struct cl_io_slice *ios)
1721 struct lov_object *lov = cl2lov(ios->cis_obj);
1724 if (atomic_dec_and_test(&lov->lo_active_ios))
1725 wake_up(&lov->lo_waitq);
1729 static int lov_empty_io_submit(const struct lu_env *env,
1730 const struct cl_io_slice *ios,
1731 enum cl_req_type crt, struct cl_2queue *queue)
1736 static void lov_empty_impossible(const struct lu_env *env,
1737 struct cl_io_slice *ios)
1742 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1745 * An io operation vector for files without stripes.
1747 static const struct cl_io_operations lov_empty_io_ops = {
1750 .cio_fini = lov_empty_io_fini,
1752 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1753 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1754 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1755 .cio_end = LOV_EMPTY_IMPOSSIBLE
1759 .cio_fini = lov_empty_io_fini,
1760 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1761 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1762 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1763 .cio_end = LOV_EMPTY_IMPOSSIBLE
1766 .cio_fini = lov_empty_io_fini,
1767 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1768 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1769 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1770 .cio_end = LOV_EMPTY_IMPOSSIBLE
1773 .cio_fini = lov_empty_io_fini,
1774 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1775 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1776 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1777 .cio_end = LOV_EMPTY_IMPOSSIBLE
1780 .cio_fini = lov_empty_io_fini
1783 .cio_fini = lov_empty_io_fini
1786 .cio_fini = lov_empty_io_fini
1789 .cio_fini = lov_empty_io_fini
1792 .cio_submit = lov_empty_io_submit,
1793 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1796 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1799 struct lov_io *lio = lov_env_io(env);
1800 struct lov_object *lov = cl2lov(obj);
1805 INIT_LIST_HEAD(&lio->lis_active);
1806 result = lov_io_slice_init(lio, lov, io);
1810 result = lov_io_subio_init(env, lio, io);
1812 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1813 atomic_inc(&lov->lo_active_ios);
1817 io->ci_result = result < 0 ? result : 0;
1821 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1824 struct lov_object *lov = cl2lov(obj);
1825 struct lov_io *lio = lov_env_io(env);
1829 lio->lis_object = lov;
1830 switch (io->ci_type) {
1842 case CIT_DATA_VERSION:
1850 CERROR("Page fault on a file without stripes: "DFID"\n",
1851 PFID(lu_object_fid(&obj->co_lu)));
1855 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1856 atomic_inc(&lov->lo_active_ios);
1859 io->ci_result = result < 0 ? result : 0;
1863 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1866 struct lov_object *lov = cl2lov(obj);
1867 struct lov_io *lio = lov_env_io(env);
1871 LASSERT(lov->lo_lsm != NULL);
1872 lio->lis_object = lov;
1874 switch (io->ci_type) {
1876 LASSERTF(0, "invalid type %d\n", io->ci_type);
1877 result = -EOPNOTSUPP;
1883 case CIT_DATA_VERSION:
1888 * the truncate to 0 is managed by MDT:
1889 * - in open, for open O_TRUNC
1890 * - in setattr, for truncate
1893 * the truncate is for size > 0 so triggers a restore,
1894 * also trigger a restore for prealloc/punch
1896 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1897 io->ci_restore_needed = 1;
1906 io->ci_restore_needed = 1;
1912 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1913 atomic_inc(&lov->lo_active_ios);
1916 io->ci_result = result < 0 ? result : 0;
1921 * Return the index in composite:lo_entries by the file offset
1923 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1925 struct lov_object *lov = lio->lis_object;
1926 struct lov_layout_composite *comp = &lov->u.composite;
1927 int start_index = 0;
1928 int end_index = comp->lo_entry_count - 1;
1931 LASSERT(lov->lo_type == LLT_COMP);
1933 /* This is actual file offset so nothing can cover eof. */
1934 if (offset == LUSTRE_EOF)
1937 if (lov_is_flr(lov)) {
1938 struct lov_mirror_entry *lre;
1940 LASSERT(lio->lis_mirror_index >= 0);
1942 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1943 start_index = lre->lre_start;
1944 end_index = lre->lre_end;
1947 for (i = start_index; i <= end_index; i++) {
1948 struct lov_layout_entry *lle = lov_entry(lov, i);
1950 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1952 if ((offset >= lle->lle_extent->e_start &&
1953 offset < lle->lle_extent->e_end) ||
1954 (offset == OBD_OBJECT_EOF &&
1955 lle->lle_extent->e_end == OBD_OBJECT_EOF))