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 cl_io_is_fallocate(io)) &&
336 (io->ci_layout_version != obj->lo_lsm->lsm_layout_gen)) {
338 * For resync I/O, the ci_layout_version was the layout
339 * version when resync starts. If it doesn't match the
340 * current object layout version, it means the layout
346 io->ci_layout_version |= LU_LAYOUT_RESYNC;
349 lio->lis_mirror_index = -1;
350 lov_foreach_mirror_entry(obj, entry) {
351 if (entry->lre_mirror_id ==
352 io->ci_designated_mirror) {
353 lio->lis_mirror_index = index;
360 RETURN(lio->lis_mirror_index < 0 ? -EINVAL : 0);
363 result = lov_io_mirror_write_intent(lio, obj, io);
367 if (io->ci_need_write_intent) {
368 CDEBUG(D_VFSTRACE, DFID " need write intent for [%llu, %llu)\n",
369 PFID(lu_object_fid(lov2lu(obj))),
370 lio->lis_pos, lio->lis_endpos);
372 if (cl_io_is_trunc(io)) {
374 * for truncate, we uses [size, EOF) to judge whether
375 * a write intent needs to be send, but we need to
376 * restore the write extent to [0, size], in truncate,
377 * the byte in the size position is accessed.
379 io->ci_write_intent.e_start = 0;
380 io->ci_write_intent.e_end =
381 io->u.ci_setattr.sa_attr.lvb_size + 1;
383 /* stop cl_io_init() loop */
387 if (io->ci_ndelay_tried == 0 || /* first time to try */
388 /* reset the mirror index if layout has changed */
389 lio->lis_mirror_layout_gen != obj->lo_lsm->lsm_layout_gen) {
390 lio->lis_mirror_layout_gen = obj->lo_lsm->lsm_layout_gen;
391 index = lio->lis_mirror_index = comp->lo_preferred_mirror;
393 index = lio->lis_mirror_index;
396 /* move mirror index to the next one */
397 index = (index + 1) % comp->lo_mirror_count;
400 for (i = 0; i < comp->lo_mirror_count; i++) {
401 struct lu_extent ext = { .e_start = lio->lis_pos,
402 .e_end = lio->lis_pos + 1 };
403 struct lov_mirror_entry *lre;
404 struct lov_layout_entry *lle;
407 lre = &comp->lo_mirrors[(index + i) % comp->lo_mirror_count];
411 if (lre->lre_foreign)
414 lov_foreach_mirror_layout_entry(obj, lle, lre) {
418 if (lu_extent_is_overlapped(&ext, lle->lle_extent)) {
422 } /* each component of the mirror */
424 index = (index + i) % comp->lo_mirror_count;
429 if (i == comp->lo_mirror_count) {
430 CERROR(DFID": failed to find a component covering "
431 "I/O region at %llu\n",
432 PFID(lu_object_fid(lov2lu(obj))), lio->lis_pos);
434 dump_lsm(D_ERROR, obj->lo_lsm);
439 CDEBUG(D_VFSTRACE, DFID ": flr state: %d, move mirror from %d to %d, "
440 "have retried: %d, mirror count: %d\n",
441 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj),
442 lio->lis_mirror_index, index, io->ci_ndelay_tried,
443 comp->lo_mirror_count);
445 lio->lis_mirror_index = index;
448 * FLR: if all mirrors have been tried once, most likely the network
449 * of this client has been partitioned. We should relinquish CPU for
450 * a while before trying again.
452 if (io->ci_ndelay && io->ci_ndelay_tried > 0 &&
453 (io->ci_ndelay_tried % comp->lo_mirror_count == 0)) {
454 schedule_timeout_interruptible(cfs_time_seconds(1) / 100);
455 if (signal_pending(current))
459 * we'd set ci_tried_all_mirrors to turn off fast mirror
460 * switching for read after we've tried all mirrors several
463 io->ci_tried_all_mirrors = io->ci_ndelay_tried %
464 (comp->lo_mirror_count * 4) == 0;
466 ++io->ci_ndelay_tried;
468 CDEBUG(D_VFSTRACE, "use %sdelayed RPC state for this IO\n",
469 io->ci_ndelay ? "non-" : "");
474 static int lov_io_slice_init(struct lov_io *lio,
475 struct lov_object *obj, struct cl_io *io)
482 lio->lis_object = obj;
483 lio->lis_cached_entry = LIS_CACHE_ENTRY_NONE;
485 switch (io->ci_type) {
488 lio->lis_pos = io->u.ci_rw.crw_pos;
489 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
490 lio->lis_io_endpos = lio->lis_endpos;
491 if (cl_io_is_append(io)) {
492 LASSERT(io->ci_type == CIT_WRITE);
495 * If there is LOV EA hole, then we may cannot locate
496 * the current file-tail exactly.
498 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
500 GOTO(out, result = -EIO);
503 lio->lis_endpos = OBD_OBJECT_EOF;
508 if (cl_io_is_fallocate(io)) {
509 lio->lis_pos = io->u.ci_setattr.sa_falloc_offset;
510 lio->lis_endpos = io->u.ci_setattr.sa_falloc_end;
511 } else if (cl_io_is_trunc(io)) {
512 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
513 lio->lis_endpos = OBD_OBJECT_EOF;
516 lio->lis_endpos = OBD_OBJECT_EOF;
520 case CIT_DATA_VERSION:
522 lio->lis_endpos = OBD_OBJECT_EOF;
526 pgoff_t index = io->u.ci_fault.ft_index;
528 lio->lis_pos = cl_offset(io->ci_obj, index);
529 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
534 lio->lis_pos = io->u.ci_fsync.fi_start;
535 lio->lis_endpos = io->u.ci_fsync.fi_end;
540 lio->lis_pos = io->u.ci_ladvise.li_start;
541 lio->lis_endpos = io->u.ci_ladvise.li_end;
546 lio->lis_pos = io->u.ci_lseek.ls_start;
547 lio->lis_endpos = OBD_OBJECT_EOF;
553 lio->lis_endpos = OBD_OBJECT_EOF;
555 if (lov_flr_state(obj) == LCM_FL_RDONLY &&
556 !OBD_FAIL_CHECK(OBD_FAIL_FLR_GLIMPSE_IMMUTABLE))
557 /* SoM is accurate, no need glimpse */
558 GOTO(out, result = 1);
563 lio->lis_endpos = OBD_OBJECT_EOF;
571 * CIT_MISC + ci_ignore_layout can identify the I/O from the OSC layer,
572 * it won't care/access lov layout related info.
574 if (io->ci_ignore_layout && io->ci_type == CIT_MISC)
575 GOTO(out, result = 0);
577 LASSERT(obj->lo_lsm != NULL);
579 result = lov_io_mirror_init(lio, obj, io);
583 /* check if it needs to instantiate layout */
584 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
585 cl_io_is_fallocate(io) ||
586 (cl_io_is_trunc(io) && io->u.ci_setattr.sa_attr.lvb_size > 0)))
587 GOTO(out, result = 0);
590 * for truncate, it only needs to instantiate the components
591 * before the truncated size.
593 if (cl_io_is_trunc(io)) {
594 io->ci_write_intent.e_start = 0;
595 /* for writes, e_end is endpos, the location of the file
596 * pointer after the write is completed, so it is not accessed.
597 * For truncate, 'end' is the size, and *is* acccessed.
598 * In other words, writes are [start, end), but truncate is
599 * [start, size], where both are included. So add 1 to the
600 * size when creating the write intent to account for this.
602 io->ci_write_intent.e_end =
603 io->u.ci_setattr.sa_attr.lvb_size + 1;
605 io->ci_write_intent.e_start = lio->lis_pos;
606 io->ci_write_intent.e_end = lio->lis_endpos;
610 lov_foreach_io_layout(index, lio, &io->ci_write_intent) {
611 if (!lsm_entry_inited(obj->lo_lsm, index)) {
612 io->ci_need_write_intent = 1;
617 if (io->ci_need_write_intent && io->ci_designated_mirror > 0) {
619 * REINT_SYNC RPC has already tried to instantiate all of the
620 * components involved, obviously it didn't succeed. Skip this
621 * mirror for now. The server won't be able to figure out
622 * which mirror it should instantiate components
624 CERROR(DFID": trying to instantiate components for designated "
625 "I/O, file state: %d\n",
626 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj));
628 io->ci_need_write_intent = 0;
629 GOTO(out, result = -EIO);
632 if (io->ci_need_write_intent)
633 GOTO(out, result = 1);
641 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
643 struct lov_io *lio = cl2lov_io(env, ios);
644 struct lov_object *lov = cl2lov(ios->cis_obj);
645 struct lov_io_sub *sub;
648 LASSERT(list_empty(&lio->lis_active));
650 while ((sub = list_first_entry_or_null(&lio->lis_subios,
652 sub_list)) != NULL) {
653 list_del_init(&sub->sub_list);
654 lio->lis_nr_subios--;
656 lov_io_sub_fini(env, lio, sub);
657 lov_sub_free(lio, sub);
659 LASSERT(lio->lis_nr_subios == 0);
661 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
662 if (atomic_dec_and_test(&lov->lo_active_ios))
663 wake_up(&lov->lo_waitq);
667 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
668 loff_t start, loff_t end)
670 struct cl_io *io = &sub->sub_io;
671 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
672 struct cl_io *parent = lio->lis_cl.cis_io;
673 int index = lov_comp_entry(sub->sub_subio_index);
674 int stripe = lov_comp_stripe(sub->sub_subio_index);
676 switch (io->ci_type) {
678 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
679 io->u.ci_setattr.sa_attr_flags =
680 parent->u.ci_setattr.sa_attr_flags;
681 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
682 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
683 io->u.ci_setattr.sa_falloc_mode =
684 parent->u.ci_setattr.sa_falloc_mode;
685 io->u.ci_setattr.sa_stripe_index = stripe;
686 io->u.ci_setattr.sa_parent_fid =
687 parent->u.ci_setattr.sa_parent_fid;
688 /* For SETATTR(fallocate) pass the subtype to lower IO */
689 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
690 if (cl_io_is_fallocate(io)) {
691 io->u.ci_setattr.sa_falloc_offset = start;
692 io->u.ci_setattr.sa_falloc_end = end;
693 io->u.ci_setattr.sa_falloc_uid =
694 parent->u.ci_setattr.sa_falloc_uid;
695 io->u.ci_setattr.sa_falloc_gid =
696 parent->u.ci_setattr.sa_falloc_gid;
698 if (cl_io_is_trunc(io)) {
699 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
701 new_size = lov_size_to_stripe(lsm, index, new_size,
703 io->u.ci_setattr.sa_attr.lvb_size = new_size;
705 lov_lsm2layout(lsm, lsm->lsm_entries[index],
706 &io->u.ci_setattr.sa_layout);
709 case CIT_DATA_VERSION: {
710 io->u.ci_data_version.dv_data_version = 0;
711 io->u.ci_data_version.dv_flags =
712 parent->u.ci_data_version.dv_flags;
716 struct cl_object *obj = parent->ci_obj;
717 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
719 io->u.ci_fault = parent->u.ci_fault;
720 off = lov_size_to_stripe(lsm, index, off, stripe);
721 io->u.ci_fault.ft_index = cl_index(obj, off);
725 io->u.ci_fsync.fi_start = start;
726 io->u.ci_fsync.fi_end = end;
727 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
728 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
733 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
734 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
735 if (cl_io_is_append(parent)) {
736 io->u.ci_wr.wr_append = 1;
738 io->u.ci_rw.crw_pos = start;
739 io->u.ci_rw.crw_count = end - start;
744 io->u.ci_ladvise.li_start = start;
745 io->u.ci_ladvise.li_end = end;
746 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
747 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
748 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
752 io->u.ci_lseek.ls_start = start;
753 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
754 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
764 static loff_t lov_offset_mod(loff_t val, int delta)
766 if (val != OBD_OBJECT_EOF)
771 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
772 struct lov_io_sub *sub, u64 start, u64 end)
776 end = lov_offset_mod(end, 1);
777 lov_io_sub_inherit(sub, lio, start, end);
778 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
780 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
784 list_add_tail(&sub->sub_linkage, &lio->lis_active);
788 static int lov_io_iter_init(const struct lu_env *env,
789 const struct cl_io_slice *ios)
791 struct lov_io *lio = cl2lov_io(env, ios);
792 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
793 struct lov_io_sub *sub;
794 struct lu_extent ext;
800 ext.e_start = lio->lis_pos;
801 ext.e_end = lio->lis_endpos;
803 lov_foreach_io_layout(index, lio, &ext) {
804 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
805 struct lov_layout_raid0 *r0 = &le->lle_raid0;
809 bool tested_trunc_stripe = false;
811 r0->lo_trunc_stripeno = -1;
813 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
814 index, lsm->lsm_entries[index]->lsme_flags);
815 if (!lsm_entry_inited(lsm, index)) {
817 * Read from uninitialized components should return
823 if (lsm_entry_is_foreign(lsm, index))
826 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
827 CERROR("I/O to invalid component: %d, mirror: %d\n",
828 index, lio->lis_mirror_index);
832 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
833 if (!lov_stripe_intersects(lsm, index, stripe,
837 if (unlikely(!r0->lo_sub[stripe])) {
838 if (ios->cis_io->ci_type == CIT_READ ||
839 ios->cis_io->ci_type == CIT_WRITE ||
840 ios->cis_io->ci_type == CIT_FAULT)
846 if (cl_io_is_trunc(ios->cis_io) &&
847 !tested_trunc_stripe) {
851 prev = (stripe == 0) ? r0->lo_nr - 1 :
854 * Only involving previous stripe if the
855 * truncate in this component is at the
856 * beginning of this stripe.
858 tested_trunc_stripe = true;
859 if (ext.e_start < lsm->lsm_entries[index]->
860 lsme_extent.e_start) {
861 /* need previous stripe involvement */
862 r0->lo_trunc_stripeno = prev;
864 tr_start = ext.e_start;
865 tr_start = lov_do_div64(tr_start,
866 stripe_width(lsm, index));
867 /* tr_start %= stripe_swidth */
868 if (tr_start == stripe * lsm->
871 r0->lo_trunc_stripeno = prev;
875 /* if the last stripe is the trunc stripeno */
876 if (r0->lo_trunc_stripeno == stripe)
877 r0->lo_trunc_stripeno = -1;
879 sub = lov_sub_get(env, lio,
880 lov_comp_index(index, stripe));
884 rc = lov_io_add_sub(env, lio, sub, start, end);
891 if (r0->lo_trunc_stripeno != -1) {
892 stripe = r0->lo_trunc_stripeno;
893 if (unlikely(!r0->lo_sub[stripe])) {
894 r0->lo_trunc_stripeno = -1;
897 sub = lov_sub_get(env, lio,
898 lov_comp_index(index, stripe));
903 * the prev sub could be used by another truncate, we'd
904 * skip it. LU-14128 happends when expand truncate +
905 * read get wrong kms.
907 if (!list_empty(&sub->sub_linkage)) {
908 r0->lo_trunc_stripeno = -1;
912 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
914 rc = lov_io_add_sub(env, lio, sub, start, end);
923 static int lov_io_rw_iter_init(const struct lu_env *env,
924 const struct cl_io_slice *ios)
926 struct lov_io *lio = cl2lov_io(env, ios);
927 struct cl_io *io = ios->cis_io;
928 struct lov_stripe_md_entry *lse;
929 loff_t start = io->u.ci_rw.crw_pos;
933 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
936 if (cl_io_is_append(io))
937 RETURN(lov_io_iter_init(env, ios));
939 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
940 if (index < 0) { /* non-existing layout component */
941 if (io->ci_type == CIT_READ) {
943 * TODO: it needs to detect the next component and
944 * then set the next pos
948 RETURN(lov_io_iter_init(env, ios));
954 if (!lov_entry(lio->lis_object, index)->lle_valid &&
955 !io->ci_designated_mirror)
956 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
958 lse = lov_lse(lio->lis_object, index);
960 if (lsme_is_foreign(lse))
963 next = MAX_LFS_FILESIZE;
964 if (lse->lsme_stripe_count > 1) {
965 unsigned long ssize = lse->lsme_stripe_size;
967 lov_do_div64(start, ssize);
968 next = (start + 1) * ssize;
969 if (next <= start * ssize)
970 next = MAX_LFS_FILESIZE;
973 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
974 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
975 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
976 next = min_t(__u64, next, lse->lsme_extent.e_end);
977 next = min_t(loff_t, next, lio->lis_io_endpos);
979 io->ci_continue = next < lio->lis_io_endpos;
980 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
981 lio->lis_pos = io->u.ci_rw.crw_pos;
982 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
984 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
985 (__u64)start, lio->lis_pos, lio->lis_endpos,
986 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
989 * XXX The following call should be optimized: we know, that
990 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
992 RETURN(lov_io_iter_init(env, ios));
995 static int lov_io_setattr_iter_init(const struct lu_env *env,
996 const struct cl_io_slice *ios)
998 struct lov_io *lio = cl2lov_io(env, ios);
999 struct cl_io *io = ios->cis_io;
1003 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1004 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1005 /* no entry found for such offset */
1007 RETURN(io->ci_result = -ENODATA);
1010 RETURN(lov_io_iter_init(env, ios));
1013 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1014 int (*iofunc)(const struct lu_env *, struct cl_io *))
1016 struct cl_io *parent = lio->lis_cl.cis_io;
1017 struct lov_io_sub *sub;
1021 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1022 rc = iofunc(sub->sub_env, &sub->sub_io);
1026 if (parent->ci_result == 0)
1027 parent->ci_result = sub->sub_io.ci_result;
1032 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1035 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1038 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1041 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1044 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1048 * It's possible that lov_io_start() wasn't called against this
1049 * sub-io, either because previous sub-io failed, or upper layer
1052 if (io->ci_state == CIS_IO_GOING)
1055 io->ci_state = CIS_IO_FINISHED;
1059 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1061 cl_io_iter_fini(env, io);
1065 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1067 cl_io_unlock(env, io);
1071 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1075 /* Before ending each i/o, we must set lis_cached_entry to tell the
1076 * next i/o not to use stale cached lis information.
1078 cl2lov_io(env, ios)->lis_cached_entry = LIS_CACHE_ENTRY_NONE;
1080 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1085 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1087 struct lov_io *lio = cl2lov_io(env, ios);
1088 struct cl_io *parent = lio->lis_cl.cis_io;
1089 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1090 struct lov_io_sub *sub;
1093 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1094 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1096 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1098 pdv->dv_data_version += sdv->dv_data_version;
1099 if (pdv->dv_layout_version > sdv->dv_layout_version)
1100 pdv->dv_layout_version = sdv->dv_layout_version;
1102 if (parent->ci_result == 0)
1103 parent->ci_result = sub->sub_io.ci_result;
1109 static void lov_io_iter_fini(const struct lu_env *env,
1110 const struct cl_io_slice *ios)
1112 struct lov_io *lio = cl2lov_io(env, ios);
1116 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1118 while (!list_empty(&lio->lis_active))
1119 list_del_init(lio->lis_active.next);
1123 static void lov_io_unlock(const struct lu_env *env,
1124 const struct cl_io_slice *ios)
1129 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1134 static int lov_io_read_ahead(const struct lu_env *env,
1135 const struct cl_io_slice *ios,
1136 pgoff_t start, struct cl_read_ahead *ra)
1138 struct lov_io *lio = cl2lov_io(env, ios);
1139 struct lov_object *loo = lio->lis_object;
1140 struct cl_object *obj = lov2cl(loo);
1141 struct lov_layout_raid0 *r0;
1142 struct lov_io_sub *sub;
1146 unsigned int pps; /* pages per stripe */
1152 offset = cl_offset(obj, start);
1153 index = lov_io_layout_at(lio, offset);
1154 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1155 lsm_entry_is_foreign(loo->lo_lsm, index))
1158 /* avoid readahead to expand to stale components */
1159 if (!lov_entry(loo, index)->lle_valid)
1162 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1164 r0 = lov_r0(loo, index);
1165 if (unlikely(!r0->lo_sub[stripe]))
1168 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1170 RETURN(PTR_ERR(sub));
1172 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1173 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1174 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1177 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1178 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1184 * Adjust the stripe index by layout of comp. ra->cra_end is the
1185 * maximum page index covered by an underlying DLM lock.
1186 * This function converts cra_end from stripe level to file level, and
1187 * make sure it's not beyond stripe and component boundary.
1190 /* cra_end is stripe level, convert it into file level */
1191 ra_end = ra->cra_end_idx;
1192 if (ra_end != CL_PAGE_EOF)
1193 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1196 /* boundary of current component */
1197 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1198 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1199 ra->cra_end_idx = ra_end - 1;
1201 if (r0->lo_nr == 1) /* single stripe file */
1204 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1206 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1207 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1208 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1209 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1211 /* never exceed the end of the stripe */
1212 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1213 start + pps - start % pps - 1);
1217 int lov_io_lru_reserve(const struct lu_env *env,
1218 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1220 struct lov_io *lio = cl2lov_io(env, ios);
1221 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1222 struct lov_io_sub *sub;
1223 struct lu_extent ext;
1230 ext.e_end = pos + bytes;
1231 lov_foreach_io_layout(index, lio, &ext) {
1232 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1233 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1238 if (!lsm_entry_inited(lsm, index))
1241 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1242 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1243 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1244 index, lio->lis_mirror_index);
1248 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1249 if (!lov_stripe_intersects(lsm, index, stripe,
1250 &ext, &start, &end))
1253 if (unlikely(!r0->lo_sub[stripe]))
1256 sub = lov_sub_get(env, lio,
1257 lov_comp_index(index, stripe));
1259 return PTR_ERR(sub);
1261 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1272 * lov implementation of cl_operations::cio_submit() method. It takes a list
1273 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1274 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1277 * Major complication of this function is a need to handle memory cleansing:
1278 * cl_io_submit() is called to write out pages as a part of VM memory
1279 * reclamation, and hence it may not fail due to memory shortages (system
1280 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1281 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1282 * not-memory cleansing context), and in case of memory shortage, these
1283 * pre-allocated resources are used by lov_io_submit() under
1284 * lov_device::ld_mutex mutex.
1286 static int lov_io_submit(const struct lu_env *env,
1287 const struct cl_io_slice *ios,
1288 enum cl_req_type crt, struct cl_2queue *queue)
1290 struct cl_page_list *qin = &queue->c2_qin;
1291 struct lov_io *lio = cl2lov_io(env, ios);
1292 struct lov_io_sub *sub;
1293 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1294 struct cl_page *page = cl_page_list_first(qin);
1295 struct cl_page *tmp;
1301 if (page->cp_type == CPT_TRANSIENT)
1304 cl_page_list_init(plist);
1305 while (qin->pl_nr > 0) {
1306 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1308 page = cl_page_list_first(qin);
1309 if (lov_page_is_empty(page)) {
1310 cl_page_list_move(&queue->c2_qout, qin, page);
1313 * it could only be mirror read to get here therefore
1314 * the pages will be transient. We don't care about
1315 * the return code of cl_page_prep() at all.
1317 (void) cl_page_prep(env, ios->cis_io, page, crt);
1318 cl_page_completion(env, page, crt, 0);
1322 cl_2queue_init(cl2q);
1323 cl_page_list_move(&cl2q->c2_qin, qin, page);
1325 index = page->cp_lov_index;
1326 /* DIO is already split by stripe */
1328 cl_page_list_for_each_safe(page, tmp, qin) {
1329 /* this page is not on this stripe */
1330 if (index != page->cp_lov_index)
1333 cl_page_list_move(&cl2q->c2_qin, qin, page);
1336 cl_page_list_splice(qin, &cl2q->c2_qin);
1339 sub = lov_sub_get(env, lio, index);
1341 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1347 cl_page_list_splice(&cl2q->c2_qin, plist);
1348 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1349 cl_2queue_fini(env, cl2q);
1355 cl_page_list_splice(plist, qin);
1356 cl_page_list_fini(env, plist);
1361 static int lov_io_commit_async(const struct lu_env *env,
1362 const struct cl_io_slice *ios,
1363 struct cl_page_list *queue, int from, int to,
1366 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1367 struct lov_io *lio = cl2lov_io(env, ios);
1368 struct lov_io_sub *sub;
1369 struct cl_page *page;
1373 if (lio->lis_nr_subios == 1) {
1374 int idx = lio->lis_single_subio_index;
1376 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1378 sub = lov_sub_get(env, lio, idx);
1379 LASSERT(!IS_ERR(sub));
1380 LASSERT(sub == &lio->lis_single_subio);
1381 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1386 cl_page_list_init(plist);
1387 while (queue->pl_nr > 0) {
1391 LASSERT(plist->pl_nr == 0);
1392 page = cl_page_list_first(queue);
1393 LASSERT(!lov_page_is_empty(page));
1395 cl_page_list_move(plist, queue, page);
1397 index = page->cp_lov_index;
1398 while (queue->pl_nr > 0) {
1399 page = cl_page_list_first(queue);
1400 if (index != page->cp_lov_index)
1403 cl_page_list_move(plist, queue, page);
1406 if (queue->pl_nr > 0) /* still has more pages */
1407 stripe_to = PAGE_SIZE;
1409 sub = lov_sub_get(env, lio, index);
1411 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1412 plist, from, stripe_to, cb);
1418 if (plist->pl_nr > 0) /* short write */
1423 if (lov_comp_entry(index) !=
1424 lov_comp_entry(page->cp_lov_index))
1425 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1428 /* for error case, add the page back into the qin list */
1429 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1430 while (plist->pl_nr > 0) {
1431 /* error occurred, add the uncommitted pages back into queue */
1432 page = cl_page_list_last(plist);
1433 cl_page_list_move_head(queue, plist, page);
1439 static int lov_io_fault_start(const struct lu_env *env,
1440 const struct cl_io_slice *ios)
1442 struct cl_fault_io *fio;
1444 struct lov_io_sub *sub;
1451 fio = &ios->cis_io->u.ci_fault;
1452 lio = cl2lov_io(env, ios);
1455 * LU-14502: ft_page could be an existing cl_page associated with
1456 * the vmpage covering the fault index, and the page may still
1457 * refer to another mirror of an old IO.
1459 if (lov_is_flr(lio->lis_object)) {
1460 offset = cl_offset(ios->cis_obj, fio->ft_index);
1461 entry = lov_io_layout_at(lio, offset);
1463 CERROR(DFID": page fault index %lu invalid component: "
1465 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1466 fio->ft_index, entry,
1467 lio->lis_mirror_index);
1470 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1473 if (fio->ft_page->cp_lov_index !=
1474 lov_comp_index(entry, stripe)) {
1475 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1476 "at mirror %u comp entry %u stripe %u, "
1477 "been used with comp entry %u stripe %u\n",
1478 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1479 fio->ft_index, lio->lis_mirror_index,
1481 lov_comp_entry(fio->ft_page->cp_lov_index),
1482 lov_comp_stripe(fio->ft_page->cp_lov_index));
1484 fio->ft_page->cp_lov_index =
1485 lov_comp_index(entry, stripe);
1489 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1490 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1492 RETURN(lov_io_start(env, ios));
1495 static int lov_io_setattr_start(const struct lu_env *env,
1496 const struct cl_io_slice *ios)
1498 struct lov_io *lio = cl2lov_io(env, ios);
1499 struct cl_io *parent = ios->cis_io;
1500 struct lov_io_sub *sub;
1501 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1505 if (cl_io_is_fallocate(parent)) {
1506 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1507 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1508 int index = lov_comp_entry(sub->sub_subio_index);
1509 int stripe = lov_comp_stripe(sub->sub_subio_index);
1511 size = lov_size_to_stripe(lsm, index, size, stripe);
1512 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1513 sub->sub_io.u.ci_setattr.sa_avalid =
1514 parent->u.ci_setattr.sa_avalid;
1518 RETURN(lov_io_start(env, ios));
1521 static void lov_io_fsync_end(const struct lu_env *env,
1522 const struct cl_io_slice *ios)
1524 struct lov_io *lio = cl2lov_io(env, ios);
1525 struct lov_io_sub *sub;
1526 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1530 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1531 struct cl_io *subio = &sub->sub_io;
1533 lov_io_end_wrapper(sub->sub_env, subio);
1535 if (subio->ci_result == 0)
1536 *written += subio->u.ci_fsync.fi_nr_written;
1541 static void lov_io_lseek_end(const struct lu_env *env,
1542 const struct cl_io_slice *ios)
1544 struct lov_io *lio = cl2lov_io(env, ios);
1545 struct cl_io *io = lio->lis_cl.cis_io;
1546 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1547 struct lov_io_sub *sub;
1548 loff_t offset = -ENXIO;
1550 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1554 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1555 struct cl_io *subio = &sub->sub_io;
1556 int index = lov_comp_entry(sub->sub_subio_index);
1557 int stripe = lov_comp_stripe(sub->sub_subio_index);
1558 loff_t sub_off, lov_off;
1559 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1561 lov_io_end_wrapper(sub->sub_env, subio);
1563 if (io->ci_result == 0)
1564 io->ci_result = sub->sub_io.ci_result;
1569 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1570 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1571 index, stripe, seek_hole ? "HOLE" : "DATA",
1572 subio->u.ci_lseek.ls_start);
1574 /* first subio with positive result is what we need */
1575 sub_off = subio->u.ci_lseek.ls_result;
1576 /* Expected error, offset is out of stripe file size */
1577 if (sub_off == -ENXIO)
1579 /* Any other errors are not expected with ci_result == 0 */
1581 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1583 io->ci_result = sub_off;
1586 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1588 /* the only way to get negatove lov_off here is too big
1589 * result. Return -EOVERFLOW then.
1591 io->ci_result = -EOVERFLOW;
1592 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1593 (u64)lov_off, io->ci_result);
1596 if (lov_off < io->u.ci_lseek.ls_start) {
1597 io->ci_result = -EINVAL;
1598 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1599 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1602 /* resulting offset can be out of component range if stripe
1603 * object is full and its file size was returned as virtual
1604 * hole start. Skip this result, the next component will give
1605 * us correct lseek result but keep possible hole offset in
1606 * case there is no more components ahead
1608 if (lov_off >= comp_end) {
1609 /* must be SEEK_HOLE case */
1610 if (likely(seek_hole)) {
1611 /* save comp end as potential hole offset */
1612 hole_off = max_t(__u64, comp_end, hole_off);
1614 io->ci_result = -EINVAL;
1616 "off %lld >= comp_end %llu: rc = %d\n",
1617 lov_off, comp_end, io->ci_result);
1622 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1623 seek_hole ? "HOLE" : "DATA",
1624 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1625 sub->sub_io.ci_result);
1626 offset = min_t(__u64, offset, lov_off);
1628 /* no result but some component returns hole as component end */
1629 if (seek_hole && offset == -ENXIO && hole_off > 0)
1632 io->u.ci_lseek.ls_result = offset;
1636 static const struct cl_io_operations lov_io_ops = {
1639 .cio_fini = lov_io_fini,
1640 .cio_iter_init = lov_io_rw_iter_init,
1641 .cio_iter_fini = lov_io_iter_fini,
1642 .cio_lock = lov_io_lock,
1643 .cio_unlock = lov_io_unlock,
1644 .cio_start = lov_io_start,
1645 .cio_end = lov_io_end
1648 .cio_fini = lov_io_fini,
1649 .cio_iter_init = lov_io_rw_iter_init,
1650 .cio_iter_fini = lov_io_iter_fini,
1651 .cio_lock = lov_io_lock,
1652 .cio_unlock = lov_io_unlock,
1653 .cio_start = lov_io_start,
1654 .cio_end = lov_io_end
1657 .cio_fini = lov_io_fini,
1658 .cio_iter_init = lov_io_setattr_iter_init,
1659 .cio_iter_fini = lov_io_iter_fini,
1660 .cio_lock = lov_io_lock,
1661 .cio_unlock = lov_io_unlock,
1662 .cio_start = lov_io_setattr_start,
1663 .cio_end = lov_io_end
1665 [CIT_DATA_VERSION] = {
1666 .cio_fini = lov_io_fini,
1667 .cio_iter_init = lov_io_iter_init,
1668 .cio_iter_fini = lov_io_iter_fini,
1669 .cio_lock = lov_io_lock,
1670 .cio_unlock = lov_io_unlock,
1671 .cio_start = lov_io_start,
1672 .cio_end = lov_io_data_version_end,
1675 .cio_fini = lov_io_fini,
1676 .cio_iter_init = lov_io_iter_init,
1677 .cio_iter_fini = lov_io_iter_fini,
1678 .cio_lock = lov_io_lock,
1679 .cio_unlock = lov_io_unlock,
1680 .cio_start = lov_io_fault_start,
1681 .cio_end = lov_io_end
1684 .cio_fini = lov_io_fini,
1685 .cio_iter_init = lov_io_iter_init,
1686 .cio_iter_fini = lov_io_iter_fini,
1687 .cio_lock = lov_io_lock,
1688 .cio_unlock = lov_io_unlock,
1689 .cio_start = lov_io_start,
1690 .cio_end = lov_io_fsync_end
1693 .cio_fini = lov_io_fini,
1694 .cio_iter_init = lov_io_iter_init,
1695 .cio_iter_fini = lov_io_iter_fini,
1696 .cio_lock = lov_io_lock,
1697 .cio_unlock = lov_io_unlock,
1698 .cio_start = lov_io_start,
1699 .cio_end = lov_io_end
1702 .cio_fini = lov_io_fini,
1703 .cio_iter_init = lov_io_iter_init,
1704 .cio_iter_fini = lov_io_iter_fini,
1705 .cio_lock = lov_io_lock,
1706 .cio_unlock = lov_io_unlock,
1707 .cio_start = lov_io_start,
1708 .cio_end = lov_io_lseek_end
1711 .cio_fini = lov_io_fini,
1714 .cio_fini = lov_io_fini
1717 .cio_read_ahead = lov_io_read_ahead,
1718 .cio_lru_reserve = lov_io_lru_reserve,
1719 .cio_submit = lov_io_submit,
1720 .cio_commit_async = lov_io_commit_async,
1723 /*****************************************************************************
1725 * Empty lov io operations.
1729 static void lov_empty_io_fini(const struct lu_env *env,
1730 const struct cl_io_slice *ios)
1732 struct lov_object *lov = cl2lov(ios->cis_obj);
1735 if (atomic_dec_and_test(&lov->lo_active_ios))
1736 wake_up(&lov->lo_waitq);
1740 static int lov_empty_io_submit(const struct lu_env *env,
1741 const struct cl_io_slice *ios,
1742 enum cl_req_type crt, struct cl_2queue *queue)
1747 static void lov_empty_impossible(const struct lu_env *env,
1748 struct cl_io_slice *ios)
1753 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1756 * An io operation vector for files without stripes.
1758 static const struct cl_io_operations lov_empty_io_ops = {
1761 .cio_fini = lov_empty_io_fini,
1763 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1764 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1765 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1766 .cio_end = LOV_EMPTY_IMPOSSIBLE
1770 .cio_fini = lov_empty_io_fini,
1771 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1772 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1773 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1774 .cio_end = LOV_EMPTY_IMPOSSIBLE
1777 .cio_fini = lov_empty_io_fini,
1778 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1779 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1780 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1781 .cio_end = LOV_EMPTY_IMPOSSIBLE
1784 .cio_fini = lov_empty_io_fini,
1785 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1786 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1787 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1788 .cio_end = LOV_EMPTY_IMPOSSIBLE
1791 .cio_fini = lov_empty_io_fini
1794 .cio_fini = lov_empty_io_fini
1797 .cio_fini = lov_empty_io_fini
1800 .cio_fini = lov_empty_io_fini
1803 .cio_submit = lov_empty_io_submit,
1804 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1807 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1810 struct lov_io *lio = lov_env_io(env);
1811 struct lov_object *lov = cl2lov(obj);
1816 INIT_LIST_HEAD(&lio->lis_active);
1817 result = lov_io_slice_init(lio, lov, io);
1821 result = lov_io_subio_init(env, lio, io);
1823 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1824 atomic_inc(&lov->lo_active_ios);
1828 io->ci_result = result < 0 ? result : 0;
1832 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1835 struct lov_object *lov = cl2lov(obj);
1836 struct lov_io *lio = lov_env_io(env);
1840 lio->lis_object = lov;
1841 switch (io->ci_type) {
1853 case CIT_DATA_VERSION:
1861 CERROR("Page fault on a file without stripes: "DFID"\n",
1862 PFID(lu_object_fid(&obj->co_lu)));
1866 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1867 atomic_inc(&lov->lo_active_ios);
1870 io->ci_result = result < 0 ? result : 0;
1874 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1877 struct lov_object *lov = cl2lov(obj);
1878 struct lov_io *lio = lov_env_io(env);
1882 LASSERT(lov->lo_lsm != NULL);
1883 lio->lis_object = lov;
1885 switch (io->ci_type) {
1887 LASSERTF(0, "invalid type %d\n", io->ci_type);
1888 result = -EOPNOTSUPP;
1894 case CIT_DATA_VERSION:
1899 * the truncate to 0 is managed by MDT:
1900 * - in open, for open O_TRUNC
1901 * - in setattr, for truncate
1904 * the truncate is for size > 0 so triggers a restore,
1905 * also trigger a restore for prealloc/punch
1907 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1908 io->ci_restore_needed = 1;
1917 io->ci_restore_needed = 1;
1923 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1924 atomic_inc(&lov->lo_active_ios);
1927 io->ci_result = result < 0 ? result : 0;
1932 * Return the index in composite:lo_entries by the file offset
1934 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1936 struct lov_object *lov = lio->lis_object;
1937 struct lov_layout_composite *comp = &lov->u.composite;
1938 int start_index = 0;
1939 int end_index = comp->lo_entry_count - 1;
1942 LASSERT(lov->lo_type == LLT_COMP);
1944 /* This is actual file offset so nothing can cover eof. */
1945 if (offset == LUSTRE_EOF)
1948 if (lov_is_flr(lov)) {
1949 struct lov_mirror_entry *lre;
1951 LASSERT(lio->lis_mirror_index >= 0);
1953 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1954 start_index = lre->lre_start;
1955 end_index = lre->lre_end;
1958 for (i = start_index; i <= end_index; i++) {
1959 struct lov_layout_entry *lle = lov_entry(lov, i);
1961 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1963 if ((offset >= lle->lle_extent->e_start &&
1964 offset < lle->lle_extent->e_end) ||
1965 (offset == OBD_OBJECT_EOF &&
1966 lle->lle_extent->e_end == OBD_OBJECT_EOF))