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;
484 switch (io->ci_type) {
487 lio->lis_pos = io->u.ci_rw.crw_pos;
488 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
489 lio->lis_io_endpos = lio->lis_endpos;
490 if (cl_io_is_append(io)) {
491 LASSERT(io->ci_type == CIT_WRITE);
494 * If there is LOV EA hole, then we may cannot locate
495 * the current file-tail exactly.
497 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
499 GOTO(out, result = -EIO);
502 lio->lis_endpos = OBD_OBJECT_EOF;
507 if (cl_io_is_fallocate(io)) {
508 lio->lis_pos = io->u.ci_setattr.sa_falloc_offset;
509 lio->lis_endpos = io->u.ci_setattr.sa_falloc_end;
510 } else if (cl_io_is_trunc(io)) {
511 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
512 lio->lis_endpos = OBD_OBJECT_EOF;
515 lio->lis_endpos = OBD_OBJECT_EOF;
519 case CIT_DATA_VERSION:
521 lio->lis_endpos = OBD_OBJECT_EOF;
525 pgoff_t index = io->u.ci_fault.ft_index;
527 lio->lis_pos = cl_offset(io->ci_obj, index);
528 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
533 lio->lis_pos = io->u.ci_fsync.fi_start;
534 lio->lis_endpos = io->u.ci_fsync.fi_end;
539 lio->lis_pos = io->u.ci_ladvise.li_start;
540 lio->lis_endpos = io->u.ci_ladvise.li_end;
545 lio->lis_pos = io->u.ci_lseek.ls_start;
546 lio->lis_endpos = OBD_OBJECT_EOF;
552 lio->lis_endpos = OBD_OBJECT_EOF;
554 if (lov_flr_state(obj) == LCM_FL_RDONLY &&
555 !OBD_FAIL_CHECK(OBD_FAIL_FLR_GLIMPSE_IMMUTABLE))
556 /* SoM is accurate, no need glimpse */
557 GOTO(out, result = 1);
562 lio->lis_endpos = OBD_OBJECT_EOF;
570 * CIT_MISC + ci_ignore_layout can identify the I/O from the OSC layer,
571 * it won't care/access lov layout related info.
573 if (io->ci_ignore_layout && io->ci_type == CIT_MISC)
574 GOTO(out, result = 0);
576 LASSERT(obj->lo_lsm != NULL);
578 result = lov_io_mirror_init(lio, obj, io);
582 /* check if it needs to instantiate layout */
583 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
584 cl_io_is_fallocate(io) ||
585 (cl_io_is_trunc(io) && io->u.ci_setattr.sa_attr.lvb_size > 0)))
586 GOTO(out, result = 0);
589 * for truncate, it only needs to instantiate the components
590 * before the truncated size.
592 if (cl_io_is_trunc(io)) {
593 io->ci_write_intent.e_start = 0;
594 /* for writes, e_end is endpos, the location of the file
595 * pointer after the write is completed, so it is not accessed.
596 * For truncate, 'end' is the size, and *is* acccessed.
597 * In other words, writes are [start, end), but truncate is
598 * [start, size], where both are included. So add 1 to the
599 * size when creating the write intent to account for this.
601 io->ci_write_intent.e_end =
602 io->u.ci_setattr.sa_attr.lvb_size + 1;
604 io->ci_write_intent.e_start = lio->lis_pos;
605 io->ci_write_intent.e_end = lio->lis_endpos;
609 lov_foreach_io_layout(index, lio, &io->ci_write_intent) {
610 if (!lsm_entry_inited(obj->lo_lsm, index)) {
611 io->ci_need_write_intent = 1;
616 if (io->ci_need_write_intent && io->ci_designated_mirror > 0) {
618 * REINT_SYNC RPC has already tried to instantiate all of the
619 * components involved, obviously it didn't succeed. Skip this
620 * mirror for now. The server won't be able to figure out
621 * which mirror it should instantiate components
623 CERROR(DFID": trying to instantiate components for designated "
624 "I/O, file state: %d\n",
625 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj));
627 io->ci_need_write_intent = 0;
628 GOTO(out, result = -EIO);
631 if (io->ci_need_write_intent)
632 GOTO(out, result = 1);
640 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
642 struct lov_io *lio = cl2lov_io(env, ios);
643 struct lov_object *lov = cl2lov(ios->cis_obj);
644 struct lov_io_sub *sub;
647 LASSERT(list_empty(&lio->lis_active));
649 while ((sub = list_first_entry_or_null(&lio->lis_subios,
651 sub_list)) != NULL) {
652 list_del_init(&sub->sub_list);
653 lio->lis_nr_subios--;
655 lov_io_sub_fini(env, lio, sub);
656 lov_sub_free(lio, sub);
658 LASSERT(lio->lis_nr_subios == 0);
660 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
661 if (atomic_dec_and_test(&lov->lo_active_ios))
662 wake_up(&lov->lo_waitq);
666 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
667 loff_t start, loff_t end)
669 struct cl_io *io = &sub->sub_io;
670 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
671 struct cl_io *parent = lio->lis_cl.cis_io;
672 int index = lov_comp_entry(sub->sub_subio_index);
673 int stripe = lov_comp_stripe(sub->sub_subio_index);
675 switch (io->ci_type) {
677 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
678 io->u.ci_setattr.sa_attr_flags =
679 parent->u.ci_setattr.sa_attr_flags;
680 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
681 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
682 io->u.ci_setattr.sa_falloc_mode =
683 parent->u.ci_setattr.sa_falloc_mode;
684 io->u.ci_setattr.sa_stripe_index = stripe;
685 io->u.ci_setattr.sa_parent_fid =
686 parent->u.ci_setattr.sa_parent_fid;
687 /* For SETATTR(fallocate) pass the subtype to lower IO */
688 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
689 if (cl_io_is_fallocate(io)) {
690 io->u.ci_setattr.sa_falloc_offset = start;
691 io->u.ci_setattr.sa_falloc_end = end;
692 io->u.ci_setattr.sa_falloc_uid =
693 parent->u.ci_setattr.sa_falloc_uid;
694 io->u.ci_setattr.sa_falloc_gid =
695 parent->u.ci_setattr.sa_falloc_gid;
697 if (cl_io_is_trunc(io)) {
698 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
700 new_size = lov_size_to_stripe(lsm, index, new_size,
702 io->u.ci_setattr.sa_attr.lvb_size = new_size;
704 lov_lsm2layout(lsm, lsm->lsm_entries[index],
705 &io->u.ci_setattr.sa_layout);
708 case CIT_DATA_VERSION: {
709 io->u.ci_data_version.dv_data_version = 0;
710 io->u.ci_data_version.dv_flags =
711 parent->u.ci_data_version.dv_flags;
715 struct cl_object *obj = parent->ci_obj;
716 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
718 io->u.ci_fault = parent->u.ci_fault;
719 off = lov_size_to_stripe(lsm, index, off, stripe);
720 io->u.ci_fault.ft_index = cl_index(obj, off);
724 io->u.ci_fsync.fi_start = start;
725 io->u.ci_fsync.fi_end = end;
726 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
727 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
732 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
733 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
734 if (cl_io_is_append(parent)) {
735 io->u.ci_wr.wr_append = 1;
737 io->u.ci_rw.crw_pos = start;
738 io->u.ci_rw.crw_count = end - start;
743 io->u.ci_ladvise.li_start = start;
744 io->u.ci_ladvise.li_end = end;
745 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
746 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
747 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
751 io->u.ci_lseek.ls_start = start;
752 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
753 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
763 static loff_t lov_offset_mod(loff_t val, int delta)
765 if (val != OBD_OBJECT_EOF)
770 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
771 struct lov_io_sub *sub, u64 start, u64 end)
775 end = lov_offset_mod(end, 1);
776 lov_io_sub_inherit(sub, lio, start, end);
777 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
779 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
783 list_add_tail(&sub->sub_linkage, &lio->lis_active);
787 static int lov_io_iter_init(const struct lu_env *env,
788 const struct cl_io_slice *ios)
790 struct lov_io *lio = cl2lov_io(env, ios);
791 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
792 struct lov_io_sub *sub;
793 struct lu_extent ext;
799 ext.e_start = lio->lis_pos;
800 ext.e_end = lio->lis_endpos;
802 lov_foreach_io_layout(index, lio, &ext) {
803 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
804 struct lov_layout_raid0 *r0 = &le->lle_raid0;
808 bool tested_trunc_stripe = false;
810 r0->lo_trunc_stripeno = -1;
812 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
813 index, lsm->lsm_entries[index]->lsme_flags);
814 if (!lsm_entry_inited(lsm, index)) {
816 * Read from uninitialized components should return
822 if (lsm_entry_is_foreign(lsm, index))
825 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
826 CERROR("I/O to invalid component: %d, mirror: %d\n",
827 index, lio->lis_mirror_index);
831 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
832 if (!lov_stripe_intersects(lsm, index, stripe,
836 if (unlikely(!r0->lo_sub[stripe])) {
837 if (ios->cis_io->ci_type == CIT_READ ||
838 ios->cis_io->ci_type == CIT_WRITE ||
839 ios->cis_io->ci_type == CIT_FAULT)
845 if (cl_io_is_trunc(ios->cis_io) &&
846 !tested_trunc_stripe) {
850 prev = (stripe == 0) ? r0->lo_nr - 1 :
853 * Only involving previous stripe if the
854 * truncate in this component is at the
855 * beginning of this stripe.
857 tested_trunc_stripe = true;
858 if (ext.e_start < lsm->lsm_entries[index]->
859 lsme_extent.e_start) {
860 /* need previous stripe involvement */
861 r0->lo_trunc_stripeno = prev;
863 tr_start = ext.e_start;
864 tr_start = lov_do_div64(tr_start,
865 stripe_width(lsm, index));
866 /* tr_start %= stripe_swidth */
867 if (tr_start == stripe * lsm->
870 r0->lo_trunc_stripeno = prev;
874 /* if the last stripe is the trunc stripeno */
875 if (r0->lo_trunc_stripeno == stripe)
876 r0->lo_trunc_stripeno = -1;
878 sub = lov_sub_get(env, lio,
879 lov_comp_index(index, stripe));
883 rc = lov_io_add_sub(env, lio, sub, start, end);
890 if (r0->lo_trunc_stripeno != -1) {
891 stripe = r0->lo_trunc_stripeno;
892 if (unlikely(!r0->lo_sub[stripe])) {
893 r0->lo_trunc_stripeno = -1;
896 sub = lov_sub_get(env, lio,
897 lov_comp_index(index, stripe));
902 * the prev sub could be used by another truncate, we'd
903 * skip it. LU-14128 happends when expand truncate +
904 * read get wrong kms.
906 if (!list_empty(&sub->sub_linkage)) {
907 r0->lo_trunc_stripeno = -1;
911 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
913 rc = lov_io_add_sub(env, lio, sub, start, end);
922 static int lov_io_rw_iter_init(const struct lu_env *env,
923 const struct cl_io_slice *ios)
925 struct lov_io *lio = cl2lov_io(env, ios);
926 struct cl_io *io = ios->cis_io;
927 struct lov_stripe_md_entry *lse;
928 loff_t start = io->u.ci_rw.crw_pos;
932 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
935 if (cl_io_is_append(io))
936 RETURN(lov_io_iter_init(env, ios));
938 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
939 if (index < 0) { /* non-existing layout component */
940 if (io->ci_type == CIT_READ) {
942 * TODO: it needs to detect the next component and
943 * then set the next pos
947 RETURN(lov_io_iter_init(env, ios));
953 if (!lov_entry(lio->lis_object, index)->lle_valid &&
954 !io->ci_designated_mirror)
955 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
957 lse = lov_lse(lio->lis_object, index);
959 if (lsme_is_foreign(lse))
962 next = MAX_LFS_FILESIZE;
963 if (lse->lsme_stripe_count > 1) {
964 unsigned long ssize = lse->lsme_stripe_size;
966 lov_do_div64(start, ssize);
967 next = (start + 1) * ssize;
968 if (next <= start * ssize)
969 next = MAX_LFS_FILESIZE;
972 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
973 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
974 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
975 next = min_t(__u64, next, lse->lsme_extent.e_end);
976 next = min_t(loff_t, next, lio->lis_io_endpos);
978 io->ci_continue = next < lio->lis_io_endpos;
979 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
980 lio->lis_pos = io->u.ci_rw.crw_pos;
981 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
983 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
984 (__u64)start, lio->lis_pos, lio->lis_endpos,
985 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
988 * XXX The following call should be optimized: we know, that
989 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
991 RETURN(lov_io_iter_init(env, ios));
994 static int lov_io_setattr_iter_init(const struct lu_env *env,
995 const struct cl_io_slice *ios)
997 struct lov_io *lio = cl2lov_io(env, ios);
998 struct cl_io *io = ios->cis_io;
1002 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1003 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1004 /* no entry found for such offset */
1006 RETURN(io->ci_result = -ENODATA);
1009 RETURN(lov_io_iter_init(env, ios));
1012 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1013 int (*iofunc)(const struct lu_env *, struct cl_io *))
1015 struct cl_io *parent = lio->lis_cl.cis_io;
1016 struct lov_io_sub *sub;
1020 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1021 rc = iofunc(sub->sub_env, &sub->sub_io);
1025 if (parent->ci_result == 0)
1026 parent->ci_result = sub->sub_io.ci_result;
1031 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1034 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1037 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1040 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1043 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1047 * It's possible that lov_io_start() wasn't called against this
1048 * sub-io, either because previous sub-io failed, or upper layer
1051 if (io->ci_state == CIS_IO_GOING)
1054 io->ci_state = CIS_IO_FINISHED;
1058 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1060 cl_io_iter_fini(env, io);
1064 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1066 cl_io_unlock(env, io);
1070 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1074 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1079 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1081 struct lov_io *lio = cl2lov_io(env, ios);
1082 struct cl_io *parent = lio->lis_cl.cis_io;
1083 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1084 struct lov_io_sub *sub;
1087 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1088 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1090 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1092 pdv->dv_data_version += sdv->dv_data_version;
1093 if (pdv->dv_layout_version > sdv->dv_layout_version)
1094 pdv->dv_layout_version = sdv->dv_layout_version;
1096 if (parent->ci_result == 0)
1097 parent->ci_result = sub->sub_io.ci_result;
1103 static void lov_io_iter_fini(const struct lu_env *env,
1104 const struct cl_io_slice *ios)
1106 struct lov_io *lio = cl2lov_io(env, ios);
1110 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1112 while (!list_empty(&lio->lis_active))
1113 list_del_init(lio->lis_active.next);
1117 static void lov_io_unlock(const struct lu_env *env,
1118 const struct cl_io_slice *ios)
1123 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1128 static int lov_io_read_ahead(const struct lu_env *env,
1129 const struct cl_io_slice *ios,
1130 pgoff_t start, struct cl_read_ahead *ra)
1132 struct lov_io *lio = cl2lov_io(env, ios);
1133 struct lov_object *loo = lio->lis_object;
1134 struct cl_object *obj = lov2cl(loo);
1135 struct lov_layout_raid0 *r0;
1136 struct lov_io_sub *sub;
1140 unsigned int pps; /* pages per stripe */
1146 offset = cl_offset(obj, start);
1147 index = lov_io_layout_at(lio, offset);
1148 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1149 lsm_entry_is_foreign(loo->lo_lsm, index))
1152 /* avoid readahead to expand to stale components */
1153 if (!lov_entry(loo, index)->lle_valid)
1156 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1158 r0 = lov_r0(loo, index);
1159 if (unlikely(!r0->lo_sub[stripe]))
1162 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1164 RETURN(PTR_ERR(sub));
1166 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1167 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1168 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1171 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1172 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1178 * Adjust the stripe index by layout of comp. ra->cra_end is the
1179 * maximum page index covered by an underlying DLM lock.
1180 * This function converts cra_end from stripe level to file level, and
1181 * make sure it's not beyond stripe and component boundary.
1184 /* cra_end is stripe level, convert it into file level */
1185 ra_end = ra->cra_end_idx;
1186 if (ra_end != CL_PAGE_EOF)
1187 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1190 /* boundary of current component */
1191 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1192 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1193 ra->cra_end_idx = ra_end - 1;
1195 if (r0->lo_nr == 1) /* single stripe file */
1198 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1200 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1201 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1202 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1203 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1205 /* never exceed the end of the stripe */
1206 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1207 start + pps - start % pps - 1);
1211 int lov_io_lru_reserve(const struct lu_env *env,
1212 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1214 struct lov_io *lio = cl2lov_io(env, ios);
1215 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1216 struct lov_io_sub *sub;
1217 struct lu_extent ext;
1224 ext.e_end = pos + bytes;
1225 lov_foreach_io_layout(index, lio, &ext) {
1226 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1227 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1232 if (!lsm_entry_inited(lsm, index))
1235 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1236 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1237 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1238 index, lio->lis_mirror_index);
1242 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1243 if (!lov_stripe_intersects(lsm, index, stripe,
1244 &ext, &start, &end))
1247 if (unlikely(!r0->lo_sub[stripe]))
1250 sub = lov_sub_get(env, lio,
1251 lov_comp_index(index, stripe));
1253 return PTR_ERR(sub);
1255 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1266 * lov implementation of cl_operations::cio_submit() method. It takes a list
1267 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1268 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1271 * Major complication of this function is a need to handle memory cleansing:
1272 * cl_io_submit() is called to write out pages as a part of VM memory
1273 * reclamation, and hence it may not fail due to memory shortages (system
1274 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1275 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1276 * not-memory cleansing context), and in case of memory shortage, these
1277 * pre-allocated resources are used by lov_io_submit() under
1278 * lov_device::ld_mutex mutex.
1280 static int lov_io_submit(const struct lu_env *env,
1281 const struct cl_io_slice *ios,
1282 enum cl_req_type crt, struct cl_2queue *queue)
1284 struct cl_page_list *qin = &queue->c2_qin;
1285 struct lov_io *lio = cl2lov_io(env, ios);
1286 struct lov_io_sub *sub;
1287 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1288 struct cl_page *page = cl_page_list_first(qin);
1289 struct cl_page *tmp;
1295 if (page->cp_type == CPT_TRANSIENT)
1298 cl_page_list_init(plist);
1299 while (qin->pl_nr > 0) {
1300 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1302 page = cl_page_list_first(qin);
1303 if (lov_page_is_empty(page)) {
1304 cl_page_list_move(&queue->c2_qout, qin, page);
1307 * it could only be mirror read to get here therefore
1308 * the pages will be transient. We don't care about
1309 * the return code of cl_page_prep() at all.
1311 (void) cl_page_prep(env, ios->cis_io, page, crt);
1312 cl_page_completion(env, page, crt, 0);
1316 cl_2queue_init(cl2q);
1317 cl_page_list_move(&cl2q->c2_qin, qin, page);
1319 index = page->cp_lov_index;
1320 /* DIO is already split by stripe */
1322 cl_page_list_for_each_safe(page, tmp, qin) {
1323 /* this page is not on this stripe */
1324 if (index != page->cp_lov_index)
1327 cl_page_list_move(&cl2q->c2_qin, qin, page);
1330 cl_page_list_splice(qin, &cl2q->c2_qin);
1333 sub = lov_sub_get(env, lio, index);
1335 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1341 cl_page_list_splice(&cl2q->c2_qin, plist);
1342 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1343 cl_2queue_fini(env, cl2q);
1349 cl_page_list_splice(plist, qin);
1350 cl_page_list_fini(env, plist);
1355 static int lov_io_commit_async(const struct lu_env *env,
1356 const struct cl_io_slice *ios,
1357 struct cl_page_list *queue, int from, int to,
1360 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1361 struct lov_io *lio = cl2lov_io(env, ios);
1362 struct lov_io_sub *sub;
1363 struct cl_page *page;
1367 if (lio->lis_nr_subios == 1) {
1368 int idx = lio->lis_single_subio_index;
1370 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1372 sub = lov_sub_get(env, lio, idx);
1373 LASSERT(!IS_ERR(sub));
1374 LASSERT(sub == &lio->lis_single_subio);
1375 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1380 cl_page_list_init(plist);
1381 while (queue->pl_nr > 0) {
1385 LASSERT(plist->pl_nr == 0);
1386 page = cl_page_list_first(queue);
1387 LASSERT(!lov_page_is_empty(page));
1389 cl_page_list_move(plist, queue, page);
1391 index = page->cp_lov_index;
1392 while (queue->pl_nr > 0) {
1393 page = cl_page_list_first(queue);
1394 if (index != page->cp_lov_index)
1397 cl_page_list_move(plist, queue, page);
1400 if (queue->pl_nr > 0) /* still has more pages */
1401 stripe_to = PAGE_SIZE;
1403 sub = lov_sub_get(env, lio, index);
1405 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1406 plist, from, stripe_to, cb);
1412 if (plist->pl_nr > 0) /* short write */
1417 if (lov_comp_entry(index) !=
1418 lov_comp_entry(page->cp_lov_index))
1419 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1422 /* for error case, add the page back into the qin list */
1423 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1424 while (plist->pl_nr > 0) {
1425 /* error occurred, add the uncommitted pages back into queue */
1426 page = cl_page_list_last(plist);
1427 cl_page_list_move_head(queue, plist, page);
1433 static int lov_io_fault_start(const struct lu_env *env,
1434 const struct cl_io_slice *ios)
1436 struct cl_fault_io *fio;
1438 struct lov_io_sub *sub;
1445 fio = &ios->cis_io->u.ci_fault;
1446 lio = cl2lov_io(env, ios);
1449 * LU-14502: ft_page could be an existing cl_page associated with
1450 * the vmpage covering the fault index, and the page may still
1451 * refer to another mirror of an old IO.
1453 if (lov_is_flr(lio->lis_object)) {
1454 offset = cl_offset(ios->cis_obj, fio->ft_index);
1455 entry = lov_io_layout_at(lio, offset);
1457 CERROR(DFID": page fault index %lu invalid component: "
1459 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1460 fio->ft_index, entry,
1461 lio->lis_mirror_index);
1464 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1467 if (fio->ft_page->cp_lov_index !=
1468 lov_comp_index(entry, stripe)) {
1469 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1470 "at mirror %u comp entry %u stripe %u, "
1471 "been used with comp entry %u stripe %u\n",
1472 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1473 fio->ft_index, lio->lis_mirror_index,
1475 lov_comp_entry(fio->ft_page->cp_lov_index),
1476 lov_comp_stripe(fio->ft_page->cp_lov_index));
1478 fio->ft_page->cp_lov_index =
1479 lov_comp_index(entry, stripe);
1483 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1484 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1486 RETURN(lov_io_start(env, ios));
1489 static int lov_io_setattr_start(const struct lu_env *env,
1490 const struct cl_io_slice *ios)
1492 struct lov_io *lio = cl2lov_io(env, ios);
1493 struct cl_io *parent = ios->cis_io;
1494 struct lov_io_sub *sub;
1495 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1499 if (cl_io_is_fallocate(parent)) {
1500 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1501 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1502 int index = lov_comp_entry(sub->sub_subio_index);
1503 int stripe = lov_comp_stripe(sub->sub_subio_index);
1505 size = lov_size_to_stripe(lsm, index, size, stripe);
1506 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1507 sub->sub_io.u.ci_setattr.sa_avalid =
1508 parent->u.ci_setattr.sa_avalid;
1512 RETURN(lov_io_start(env, ios));
1515 static void lov_io_fsync_end(const struct lu_env *env,
1516 const struct cl_io_slice *ios)
1518 struct lov_io *lio = cl2lov_io(env, ios);
1519 struct lov_io_sub *sub;
1520 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1524 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1525 struct cl_io *subio = &sub->sub_io;
1527 lov_io_end_wrapper(sub->sub_env, subio);
1529 if (subio->ci_result == 0)
1530 *written += subio->u.ci_fsync.fi_nr_written;
1535 static void lov_io_lseek_end(const struct lu_env *env,
1536 const struct cl_io_slice *ios)
1538 struct lov_io *lio = cl2lov_io(env, ios);
1539 struct cl_io *io = lio->lis_cl.cis_io;
1540 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1541 struct lov_io_sub *sub;
1542 loff_t offset = -ENXIO;
1544 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1548 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1549 struct cl_io *subio = &sub->sub_io;
1550 int index = lov_comp_entry(sub->sub_subio_index);
1551 int stripe = lov_comp_stripe(sub->sub_subio_index);
1552 loff_t sub_off, lov_off;
1553 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1555 lov_io_end_wrapper(sub->sub_env, subio);
1557 if (io->ci_result == 0)
1558 io->ci_result = sub->sub_io.ci_result;
1563 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1564 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1565 index, stripe, seek_hole ? "HOLE" : "DATA",
1566 subio->u.ci_lseek.ls_start);
1568 /* first subio with positive result is what we need */
1569 sub_off = subio->u.ci_lseek.ls_result;
1570 /* Expected error, offset is out of stripe file size */
1571 if (sub_off == -ENXIO)
1573 /* Any other errors are not expected with ci_result == 0 */
1575 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1577 io->ci_result = sub_off;
1580 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1582 /* the only way to get negatove lov_off here is too big
1583 * result. Return -EOVERFLOW then.
1585 io->ci_result = -EOVERFLOW;
1586 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1587 (u64)lov_off, io->ci_result);
1590 if (lov_off < io->u.ci_lseek.ls_start) {
1591 io->ci_result = -EINVAL;
1592 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1593 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1596 /* resulting offset can be out of component range if stripe
1597 * object is full and its file size was returned as virtual
1598 * hole start. Skip this result, the next component will give
1599 * us correct lseek result but keep possible hole offset in
1600 * case there is no more components ahead
1602 if (lov_off >= comp_end) {
1603 /* must be SEEK_HOLE case */
1604 if (likely(seek_hole)) {
1605 /* save comp end as potential hole offset */
1606 hole_off = max_t(__u64, comp_end, hole_off);
1608 io->ci_result = -EINVAL;
1610 "off %lld >= comp_end %llu: rc = %d\n",
1611 lov_off, comp_end, io->ci_result);
1616 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1617 seek_hole ? "HOLE" : "DATA",
1618 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1619 sub->sub_io.ci_result);
1620 offset = min_t(__u64, offset, lov_off);
1622 /* no result but some component returns hole as component end */
1623 if (seek_hole && offset == -ENXIO && hole_off > 0)
1626 io->u.ci_lseek.ls_result = offset;
1630 static const struct cl_io_operations lov_io_ops = {
1633 .cio_fini = lov_io_fini,
1634 .cio_iter_init = lov_io_rw_iter_init,
1635 .cio_iter_fini = lov_io_iter_fini,
1636 .cio_lock = lov_io_lock,
1637 .cio_unlock = lov_io_unlock,
1638 .cio_start = lov_io_start,
1639 .cio_end = lov_io_end
1642 .cio_fini = lov_io_fini,
1643 .cio_iter_init = lov_io_rw_iter_init,
1644 .cio_iter_fini = lov_io_iter_fini,
1645 .cio_lock = lov_io_lock,
1646 .cio_unlock = lov_io_unlock,
1647 .cio_start = lov_io_start,
1648 .cio_end = lov_io_end
1651 .cio_fini = lov_io_fini,
1652 .cio_iter_init = lov_io_setattr_iter_init,
1653 .cio_iter_fini = lov_io_iter_fini,
1654 .cio_lock = lov_io_lock,
1655 .cio_unlock = lov_io_unlock,
1656 .cio_start = lov_io_setattr_start,
1657 .cio_end = lov_io_end
1659 [CIT_DATA_VERSION] = {
1660 .cio_fini = lov_io_fini,
1661 .cio_iter_init = lov_io_iter_init,
1662 .cio_iter_fini = lov_io_iter_fini,
1663 .cio_lock = lov_io_lock,
1664 .cio_unlock = lov_io_unlock,
1665 .cio_start = lov_io_start,
1666 .cio_end = lov_io_data_version_end,
1669 .cio_fini = lov_io_fini,
1670 .cio_iter_init = lov_io_iter_init,
1671 .cio_iter_fini = lov_io_iter_fini,
1672 .cio_lock = lov_io_lock,
1673 .cio_unlock = lov_io_unlock,
1674 .cio_start = lov_io_fault_start,
1675 .cio_end = lov_io_end
1678 .cio_fini = lov_io_fini,
1679 .cio_iter_init = lov_io_iter_init,
1680 .cio_iter_fini = lov_io_iter_fini,
1681 .cio_lock = lov_io_lock,
1682 .cio_unlock = lov_io_unlock,
1683 .cio_start = lov_io_start,
1684 .cio_end = lov_io_fsync_end
1687 .cio_fini = lov_io_fini,
1688 .cio_iter_init = lov_io_iter_init,
1689 .cio_iter_fini = lov_io_iter_fini,
1690 .cio_lock = lov_io_lock,
1691 .cio_unlock = lov_io_unlock,
1692 .cio_start = lov_io_start,
1693 .cio_end = lov_io_end
1696 .cio_fini = lov_io_fini,
1697 .cio_iter_init = lov_io_iter_init,
1698 .cio_iter_fini = lov_io_iter_fini,
1699 .cio_lock = lov_io_lock,
1700 .cio_unlock = lov_io_unlock,
1701 .cio_start = lov_io_start,
1702 .cio_end = lov_io_lseek_end
1705 .cio_fini = lov_io_fini,
1708 .cio_fini = lov_io_fini
1711 .cio_read_ahead = lov_io_read_ahead,
1712 .cio_lru_reserve = lov_io_lru_reserve,
1713 .cio_submit = lov_io_submit,
1714 .cio_commit_async = lov_io_commit_async,
1717 /*****************************************************************************
1719 * Empty lov io operations.
1723 static void lov_empty_io_fini(const struct lu_env *env,
1724 const struct cl_io_slice *ios)
1726 struct lov_object *lov = cl2lov(ios->cis_obj);
1729 if (atomic_dec_and_test(&lov->lo_active_ios))
1730 wake_up(&lov->lo_waitq);
1734 static int lov_empty_io_submit(const struct lu_env *env,
1735 const struct cl_io_slice *ios,
1736 enum cl_req_type crt, struct cl_2queue *queue)
1741 static void lov_empty_impossible(const struct lu_env *env,
1742 struct cl_io_slice *ios)
1747 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1750 * An io operation vector for files without stripes.
1752 static const struct cl_io_operations lov_empty_io_ops = {
1755 .cio_fini = lov_empty_io_fini,
1757 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1758 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1759 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1760 .cio_end = LOV_EMPTY_IMPOSSIBLE
1764 .cio_fini = lov_empty_io_fini,
1765 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1766 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1767 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1768 .cio_end = LOV_EMPTY_IMPOSSIBLE
1771 .cio_fini = lov_empty_io_fini,
1772 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1773 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1774 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1775 .cio_end = LOV_EMPTY_IMPOSSIBLE
1778 .cio_fini = lov_empty_io_fini,
1779 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1780 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1781 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1782 .cio_end = LOV_EMPTY_IMPOSSIBLE
1785 .cio_fini = lov_empty_io_fini
1788 .cio_fini = lov_empty_io_fini
1791 .cio_fini = lov_empty_io_fini
1794 .cio_fini = lov_empty_io_fini
1797 .cio_submit = lov_empty_io_submit,
1798 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1801 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1804 struct lov_io *lio = lov_env_io(env);
1805 struct lov_object *lov = cl2lov(obj);
1810 INIT_LIST_HEAD(&lio->lis_active);
1811 result = lov_io_slice_init(lio, lov, io);
1815 result = lov_io_subio_init(env, lio, io);
1817 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1818 atomic_inc(&lov->lo_active_ios);
1822 io->ci_result = result < 0 ? result : 0;
1826 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1829 struct lov_object *lov = cl2lov(obj);
1830 struct lov_io *lio = lov_env_io(env);
1834 lio->lis_object = lov;
1835 switch (io->ci_type) {
1847 case CIT_DATA_VERSION:
1855 CERROR("Page fault on a file without stripes: "DFID"\n",
1856 PFID(lu_object_fid(&obj->co_lu)));
1860 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1861 atomic_inc(&lov->lo_active_ios);
1864 io->ci_result = result < 0 ? result : 0;
1868 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1871 struct lov_object *lov = cl2lov(obj);
1872 struct lov_io *lio = lov_env_io(env);
1876 LASSERT(lov->lo_lsm != NULL);
1877 lio->lis_object = lov;
1879 switch (io->ci_type) {
1881 LASSERTF(0, "invalid type %d\n", io->ci_type);
1882 result = -EOPNOTSUPP;
1888 case CIT_DATA_VERSION:
1893 * the truncate to 0 is managed by MDT:
1894 * - in open, for open O_TRUNC
1895 * - in setattr, for truncate
1898 * the truncate is for size > 0 so triggers a restore,
1899 * also trigger a restore for prealloc/punch
1901 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1902 io->ci_restore_needed = 1;
1911 io->ci_restore_needed = 1;
1917 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1918 atomic_inc(&lov->lo_active_ios);
1921 io->ci_result = result < 0 ? result : 0;
1926 * Return the index in composite:lo_entries by the file offset
1928 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1930 struct lov_object *lov = lio->lis_object;
1931 struct lov_layout_composite *comp = &lov->u.composite;
1932 int start_index = 0;
1933 int end_index = comp->lo_entry_count - 1;
1936 LASSERT(lov->lo_type == LLT_COMP);
1938 /* This is actual file offset so nothing can cover eof. */
1939 if (offset == LUSTRE_EOF)
1942 if (lov_is_flr(lov)) {
1943 struct lov_mirror_entry *lre;
1945 LASSERT(lio->lis_mirror_index >= 0);
1947 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1948 start_index = lre->lre_start;
1949 end_index = lre->lre_end;
1952 for (i = start_index; i <= end_index; i++) {
1953 struct lov_layout_entry *lle = lov_entry(lov, i);
1955 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1957 if ((offset >= lle->lle_extent->e_start &&
1958 offset < lle->lle_extent->e_end) ||
1959 (offset == OBD_OBJECT_EOF &&
1960 lle->lle_extent->e_end == OBD_OBJECT_EOF))