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 lu_object *lobj = lov2lu(obj);
219 struct lov_layout_composite *comp = &obj->u.composite;
220 struct lu_extent *ext = &io->ci_write_intent;
221 struct lov_mirror_entry *lre;
222 struct lov_mirror_entry *primary;
223 struct lov_layout_entry *lle;
227 *ext = (typeof(*ext)) { lio->lis_pos, lio->lis_endpos };
228 io->ci_need_write_intent = 0;
230 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
231 cl_io_is_fallocate(io) || cl_io_is_trunc(io)))
235 * FLR: check if it needs to send a write intent RPC to server.
236 * Writing to sync_pending file needs write intent RPC to change
237 * the file state back to write_pending, so that the layout version
238 * can be increased when the state changes to sync_pending at a later
239 * time. Otherwise there exists a chance that an evicted client may
240 * dirty the file data while resync client is working on it.
241 * Designated I/O is allowed for resync workload.
243 if (lov_flr_state(obj) == LCM_FL_RDONLY ||
244 (lov_flr_state(obj) == LCM_FL_SYNC_PENDING &&
245 io->ci_designated_mirror == 0)) {
246 io->ci_need_write_intent = 1;
250 LASSERT((lov_flr_state(obj) == LCM_FL_WRITE_PENDING));
251 LASSERT(comp->lo_preferred_mirror >= 0);
254 * need to iterate all components to see if there are
255 * multiple components covering the writing component
257 primary = &comp->lo_mirrors[comp->lo_preferred_mirror];
258 if (primary->lre_stale || !primary->lre_valid) {
260 * new server could pick a primary mirror which old client
261 * does not recognize, and old client would mark it as
264 CERROR(DFID ": cannot find known valid non-stale mirror, "
265 "could be new server picked a mirror which this client "
266 "does not recognize.\n",
267 PFID(lu_object_fid(lobj)));
271 lov_foreach_mirror_layout_entry(obj, lle, primary) {
272 LASSERT(lle->lle_valid);
273 if (!lu_extent_is_overlapped(ext, lle->lle_extent))
276 ext->e_start = min(ext->e_start, lle->lle_extent->e_start);
277 ext->e_end = max(ext->e_end, lle->lle_extent->e_end);
281 CERROR(DFID ": cannot find any valid components covering "
282 "file extent "DEXT", mirror: %d\n",
283 PFID(lu_object_fid(lobj)), PEXT(ext),
284 primary->lre_mirror_id);
289 lov_foreach_mirror_entry(obj, lre) {
293 lov_foreach_mirror_layout_entry(obj, lle, lre) {
297 if (lu_extent_is_overlapped(ext, lle->lle_extent)) {
304 CDEBUG(D_VFSTRACE, DFID "there are %zd components to be staled to "
305 "modify file extent "DEXT", iot: %d\n",
306 PFID(lu_object_fid(lobj)), count, PEXT(ext), io->ci_type);
308 io->ci_need_write_intent = count > 0;
313 static int lov_io_mirror_init(struct lov_io *lio, struct lov_object *obj,
316 struct lov_layout_composite *comp = &obj->u.composite;
322 if (!lov_is_flr(obj)) {
323 /* only locks/pages are manipulated for CIT_MISC op, no
324 * cl_io_loop() will be called, don't check/set mirror info.
326 if (io->ci_type != CIT_MISC) {
327 LASSERT(comp->lo_preferred_mirror == 0);
328 lio->lis_mirror_index = comp->lo_preferred_mirror;
334 /* transfer the layout version for verification */
335 if (io->ci_layout_version == 0)
336 io->ci_layout_version = obj->lo_lsm->lsm_layout_gen;
338 /* find the corresponding mirror for designated mirror IO */
339 if (io->ci_designated_mirror > 0) {
340 struct lov_mirror_entry *entry;
342 LASSERT(!io->ci_ndelay);
344 CDEBUG(D_LAYOUT, "designated I/O mirror state: %d\n",
347 if ((cl_io_is_trunc(io) || io->ci_type == CIT_WRITE ||
348 cl_io_is_fallocate(io)) &&
349 (io->ci_layout_version != obj->lo_lsm->lsm_layout_gen)) {
351 * For resync I/O, the ci_layout_version was the layout
352 * version when resync starts. If it doesn't match the
353 * current object layout version, it means the layout
359 io->ci_layout_version |= LU_LAYOUT_RESYNC;
362 lio->lis_mirror_index = -1;
363 lov_foreach_mirror_entry(obj, entry) {
364 if (entry->lre_mirror_id ==
365 io->ci_designated_mirror) {
366 lio->lis_mirror_index = index;
373 RETURN(lio->lis_mirror_index < 0 ? -EINVAL : 0);
376 result = lov_io_mirror_write_intent(lio, obj, io);
380 if (io->ci_need_write_intent) {
381 CDEBUG(D_VFSTRACE, DFID " need write intent for [%llu, %llu)\n",
382 PFID(lu_object_fid(lov2lu(obj))),
383 lio->lis_pos, lio->lis_endpos);
385 if (cl_io_is_trunc(io)) {
387 * for truncate, we uses [size, EOF) to judge whether
388 * a write intent needs to be send, but we need to
389 * restore the write extent to [0, size], in truncate,
390 * the byte in the size position is accessed.
392 io->ci_write_intent.e_start = 0;
393 io->ci_write_intent.e_end =
394 io->u.ci_setattr.sa_attr.lvb_size + 1;
396 /* stop cl_io_init() loop */
400 if (io->ci_ndelay_tried == 0 || /* first time to try */
401 /* reset the mirror index if layout has changed */
402 lio->lis_mirror_layout_gen != obj->lo_lsm->lsm_layout_gen) {
403 lio->lis_mirror_layout_gen = obj->lo_lsm->lsm_layout_gen;
404 index = lio->lis_mirror_index = comp->lo_preferred_mirror;
406 index = lio->lis_mirror_index;
409 /* move mirror index to the next one */
410 index = (index + 1) % comp->lo_mirror_count;
413 for (i = 0; i < comp->lo_mirror_count; i++) {
414 struct lu_extent ext = { .e_start = lio->lis_pos,
415 .e_end = lio->lis_pos + 1 };
416 struct lov_mirror_entry *lre;
417 struct lov_layout_entry *lle;
420 lre = &comp->lo_mirrors[(index + i) % comp->lo_mirror_count];
424 if (lre->lre_foreign)
427 lov_foreach_mirror_layout_entry(obj, lle, lre) {
431 if (lu_extent_is_overlapped(&ext, lle->lle_extent)) {
435 } /* each component of the mirror */
437 index = (index + i) % comp->lo_mirror_count;
442 if (i == comp->lo_mirror_count) {
443 CERROR(DFID": failed to find a component covering "
444 "I/O region at %llu\n",
445 PFID(lu_object_fid(lov2lu(obj))), lio->lis_pos);
447 dump_lsm(D_ERROR, obj->lo_lsm);
452 CDEBUG(D_VFSTRACE, DFID ": flr state: %d, move mirror from %d to %d, "
453 "have retried: %d, mirror count: %d\n",
454 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj),
455 lio->lis_mirror_index, index, io->ci_ndelay_tried,
456 comp->lo_mirror_count);
458 lio->lis_mirror_index = index;
461 * FLR: if all mirrors have been tried once, most likely the network
462 * of this client has been partitioned. We should relinquish CPU for
463 * a while before trying again.
465 if (io->ci_ndelay && io->ci_ndelay_tried > 0 &&
466 (io->ci_ndelay_tried % comp->lo_mirror_count == 0)) {
467 schedule_timeout_interruptible(cfs_time_seconds(1) / 100);
468 if (signal_pending(current))
472 * we'd set ci_tried_all_mirrors to turn off fast mirror
473 * switching for read after we've tried all mirrors several
476 io->ci_tried_all_mirrors = io->ci_ndelay_tried %
477 (comp->lo_mirror_count * 4) == 0;
479 ++io->ci_ndelay_tried;
481 CDEBUG(D_VFSTRACE, "use %sdelayed RPC state for this IO\n",
482 io->ci_ndelay ? "non-" : "");
487 static int lov_io_slice_init(struct lov_io *lio,
488 struct lov_object *obj, struct cl_io *io)
495 lio->lis_object = obj;
496 lio->lis_cached_entry = LIS_CACHE_ENTRY_NONE;
498 switch (io->ci_type) {
501 lio->lis_pos = io->u.ci_rw.crw_pos;
502 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_bytes;
503 lio->lis_io_endpos = lio->lis_endpos;
504 if (cl_io_is_append(io)) {
505 LASSERT(io->ci_type == CIT_WRITE);
508 * If there is LOV EA hole, then we may cannot locate
509 * the current file-tail exactly.
511 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
513 GOTO(out, result = -EIO);
516 lio->lis_endpos = OBD_OBJECT_EOF;
521 if (cl_io_is_fallocate(io)) {
522 lio->lis_pos = io->u.ci_setattr.sa_falloc_offset;
523 lio->lis_endpos = io->u.ci_setattr.sa_falloc_end;
524 } else if (cl_io_is_trunc(io)) {
525 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
526 lio->lis_endpos = OBD_OBJECT_EOF;
529 lio->lis_endpos = OBD_OBJECT_EOF;
533 case CIT_DATA_VERSION:
535 lio->lis_endpos = OBD_OBJECT_EOF;
539 pgoff_t index = io->u.ci_fault.ft_index;
541 lio->lis_pos = index << PAGE_SHIFT;
542 lio->lis_endpos = (index + 1) << PAGE_SHIFT;
547 lio->lis_pos = io->u.ci_fsync.fi_start;
548 lio->lis_endpos = io->u.ci_fsync.fi_end;
553 lio->lis_pos = io->u.ci_ladvise.li_start;
554 lio->lis_endpos = io->u.ci_ladvise.li_end;
559 lio->lis_pos = io->u.ci_lseek.ls_start;
560 lio->lis_endpos = OBD_OBJECT_EOF;
566 lio->lis_endpos = OBD_OBJECT_EOF;
571 lio->lis_endpos = OBD_OBJECT_EOF;
579 * CIT_MISC + ci_ignore_layout can identify the I/O from the OSC layer,
580 * it won't care/access lov layout related info.
582 if (io->ci_ignore_layout && io->ci_type == CIT_MISC)
583 GOTO(out, result = 0);
585 LASSERT(obj->lo_lsm != NULL);
587 result = lov_io_mirror_init(lio, obj, io);
591 /* check if it needs to instantiate layout */
592 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
593 cl_io_is_fallocate(io) ||
594 (cl_io_is_trunc(io) && io->u.ci_setattr.sa_attr.lvb_size > 0)))
595 GOTO(out, result = 0);
598 * for truncate, it only needs to instantiate the components
599 * before the truncated size.
601 if (cl_io_is_trunc(io)) {
602 io->ci_write_intent.e_start = 0;
603 /* for writes, e_end is endpos, the location of the file
604 * pointer after the write is completed, so it is not accessed.
605 * For truncate, 'end' is the size, and *is* acccessed.
606 * In other words, writes are [start, end), but truncate is
607 * [start, size], where both are included. So add 1 to the
608 * size when creating the write intent to account for this.
610 io->ci_write_intent.e_end =
611 io->u.ci_setattr.sa_attr.lvb_size + 1;
613 io->ci_write_intent.e_start = lio->lis_pos;
614 io->ci_write_intent.e_end = lio->lis_endpos;
618 lov_foreach_io_layout(index, lio, &io->ci_write_intent) {
619 if (!lsm_entry_inited(obj->lo_lsm, index)) {
620 io->ci_need_write_intent = 1;
625 if (io->ci_need_write_intent && io->ci_designated_mirror > 0) {
627 * REINT_SYNC RPC has already tried to instantiate all of the
628 * components involved, obviously it didn't succeed. Skip this
629 * mirror for now. The server won't be able to figure out
630 * which mirror it should instantiate components
632 CERROR(DFID": trying to instantiate components for designated "
633 "I/O, file state: %d\n",
634 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj));
636 io->ci_need_write_intent = 0;
637 GOTO(out, result = -EIO);
640 if (io->ci_need_write_intent)
641 GOTO(out, result = 1);
649 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
651 struct lov_io *lio = cl2lov_io(env, ios);
652 struct lov_object *lov = cl2lov(ios->cis_obj);
653 struct lov_io_sub *sub;
656 LASSERT(list_empty(&lio->lis_active));
658 while ((sub = list_first_entry_or_null(&lio->lis_subios,
660 sub_list)) != NULL) {
661 list_del_init(&sub->sub_list);
662 lio->lis_nr_subios--;
664 lov_io_sub_fini(env, lio, sub);
665 lov_sub_free(lio, sub);
667 LASSERT(lio->lis_nr_subios == 0);
669 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
670 if (atomic_dec_and_test(&lov->lo_active_ios))
671 wake_up(&lov->lo_waitq);
675 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
676 loff_t start, loff_t end)
678 struct cl_io *io = &sub->sub_io;
679 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
680 struct cl_io *parent = lio->lis_cl.cis_io;
681 int index = lov_comp_entry(sub->sub_subio_index);
682 int stripe = lov_comp_stripe(sub->sub_subio_index);
684 switch (io->ci_type) {
686 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
687 io->u.ci_setattr.sa_attr_flags =
688 parent->u.ci_setattr.sa_attr_flags;
689 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
690 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
691 io->u.ci_setattr.sa_falloc_mode =
692 parent->u.ci_setattr.sa_falloc_mode;
693 io->u.ci_setattr.sa_stripe_index = stripe;
694 io->u.ci_setattr.sa_parent_fid =
695 parent->u.ci_setattr.sa_parent_fid;
696 /* For SETATTR(fallocate) pass the subtype to lower IO */
697 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
698 if (cl_io_is_fallocate(io)) {
699 io->u.ci_setattr.sa_falloc_offset = start;
700 io->u.ci_setattr.sa_falloc_end = end;
701 io->u.ci_setattr.sa_falloc_uid =
702 parent->u.ci_setattr.sa_falloc_uid;
703 io->u.ci_setattr.sa_falloc_gid =
704 parent->u.ci_setattr.sa_falloc_gid;
705 io->u.ci_setattr.sa_falloc_projid =
706 parent->u.ci_setattr.sa_falloc_projid;
708 if (cl_io_is_trunc(io)) {
709 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
711 new_size = lov_size_to_stripe(lsm, index, new_size,
713 io->u.ci_setattr.sa_attr.lvb_size = new_size;
715 lov_lsm2layout(lsm, lsm->lsm_entries[index],
716 &io->u.ci_setattr.sa_layout);
719 case CIT_DATA_VERSION: {
720 io->u.ci_data_version.dv_data_version = 0;
721 io->u.ci_data_version.dv_flags =
722 parent->u.ci_data_version.dv_flags;
726 loff_t off = parent->u.ci_fault.ft_index << PAGE_SHIFT;
728 io->u.ci_fault = parent->u.ci_fault;
729 off = lov_size_to_stripe(lsm, index, off, stripe);
730 io->u.ci_fault.ft_index = off >> PAGE_SHIFT;
734 io->u.ci_fsync.fi_start = start;
735 io->u.ci_fsync.fi_end = end;
736 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
737 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
742 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
743 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
744 if (cl_io_is_append(parent)) {
745 io->u.ci_wr.wr_append = 1;
747 io->u.ci_rw.crw_pos = start;
748 io->u.ci_rw.crw_bytes = end - start;
753 io->u.ci_ladvise.li_start = start;
754 io->u.ci_ladvise.li_end = end;
755 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
756 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
757 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
761 io->u.ci_lseek.ls_start = start;
762 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
763 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
773 static loff_t lov_offset_mod(loff_t val, int delta)
775 if (val != OBD_OBJECT_EOF)
780 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
781 struct lov_io_sub *sub, u64 start, u64 end)
785 end = lov_offset_mod(end, 1);
786 lov_io_sub_inherit(sub, lio, start, end);
787 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
789 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
793 list_add_tail(&sub->sub_linkage, &lio->lis_active);
797 static int lov_io_iter_init(const struct lu_env *env,
798 const struct cl_io_slice *ios)
800 struct lov_io *lio = cl2lov_io(env, ios);
801 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
802 bool is_trunc = cl_io_is_trunc(ios->cis_io);
803 struct lov_io_sub *sub;
804 struct lu_extent ext;
810 ext.e_start = lio->lis_pos;
811 ext.e_end = lio->lis_endpos;
814 OBD_ALLOC_PTR_ARRAY(lio->lis_trunc_stripe_index,
815 lio->lis_object->u.composite.lo_entry_count);
816 if (lio->lis_trunc_stripe_index == NULL)
820 lov_foreach_io_layout(index, lio, &ext) {
821 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
822 struct lov_layout_raid0 *r0 = &le->lle_raid0;
826 bool tested_trunc_stripe = false;
829 lio->lis_trunc_stripe_index[index] = -1;
831 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
832 index, lsm->lsm_entries[index]->lsme_flags);
833 if (!lsm_entry_inited(lsm, index)) {
835 * Read from uninitialized components should return
841 if (lsm_entry_is_foreign(lsm, index))
844 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
845 CERROR("I/O to invalid component: %d, mirror: %d\n",
846 index, lio->lis_mirror_index);
850 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
851 if (!lov_stripe_intersects(lsm, index, stripe,
855 if (unlikely(!r0->lo_sub[stripe])) {
856 if (ios->cis_io->ci_type == CIT_READ ||
857 ios->cis_io->ci_type == CIT_WRITE ||
858 ios->cis_io->ci_type == CIT_FAULT)
864 if (is_trunc && !tested_trunc_stripe) {
868 prev = (stripe == 0) ? r0->lo_nr - 1 :
871 * Only involving previous stripe if the
872 * truncate in this component is at the
873 * beginning of this stripe.
875 tested_trunc_stripe = true;
876 if (ext.e_start < lsm->lsm_entries[index]->
877 lsme_extent.e_start) {
878 /* need previous stripe involvement */
879 lio->lis_trunc_stripe_index[index] = prev;
881 tr_start = ext.e_start;
882 tr_start = lov_do_div64(tr_start,
883 stripe_width(lsm, index));
884 /* tr_start %= stripe_swidth */
885 if (tr_start == stripe * lsm->
888 lio->lis_trunc_stripe_index[index] = prev;
892 /* if the last stripe is the trunc stripeno */
893 if (is_trunc && lio->lis_trunc_stripe_index[index] == stripe)
894 lio->lis_trunc_stripe_index[index] = -1;
896 sub = lov_sub_get(env, lio,
897 lov_comp_index(index, stripe));
901 rc = lov_io_add_sub(env, lio, sub, start, end);
908 if (is_trunc && lio->lis_trunc_stripe_index[index] != -1) {
909 stripe = lio->lis_trunc_stripe_index[index];
910 if (unlikely(!r0->lo_sub[stripe])) {
911 lio->lis_trunc_stripe_index[index] = -1;
914 sub = lov_sub_get(env, lio,
915 lov_comp_index(index, stripe));
920 * the prev sub could be used by another truncate, we'd
921 * skip it. LU-14128 happends when expand truncate +
922 * read get wrong kms.
924 if (!list_empty(&sub->sub_linkage)) {
925 lio->lis_trunc_stripe_index[index] = -1;
929 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
931 rc = lov_io_add_sub(env, lio, sub, start, end);
940 static int lov_io_rw_iter_init(const struct lu_env *env,
941 const struct cl_io_slice *ios)
943 struct lov_io *lio = cl2lov_io(env, ios);
944 struct cl_io *io = ios->cis_io;
945 struct lov_stripe_md_entry *lse;
946 loff_t start = io->u.ci_rw.crw_pos;
950 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
953 if (cl_io_is_append(io))
954 RETURN(lov_io_iter_init(env, ios));
956 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
957 if (index < 0) { /* non-existing layout component */
958 if (io->ci_type == CIT_READ) {
960 * TODO: it needs to detect the next component and
961 * then set the next pos
965 RETURN(lov_io_iter_init(env, ios));
971 if (!lov_entry(lio->lis_object, index)->lle_valid &&
972 !io->ci_designated_mirror)
973 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
975 lse = lov_lse(lio->lis_object, index);
977 if (lsme_is_foreign(lse))
980 next = MAX_LFS_FILESIZE;
981 if (lse->lsme_stripe_count > 1) {
982 unsigned long ssize = lse->lsme_stripe_size;
984 lov_do_div64(start, ssize);
985 next = (start + 1) * ssize;
986 if (next <= start * ssize)
987 next = MAX_LFS_FILESIZE;
990 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
991 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
992 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
993 next = min_t(__u64, next, lse->lsme_extent.e_end);
994 next = min_t(loff_t, next, lio->lis_io_endpos);
996 io->ci_continue = next < lio->lis_io_endpos;
997 io->u.ci_rw.crw_bytes = next - io->u.ci_rw.crw_pos;
998 lio->lis_pos = io->u.ci_rw.crw_pos;
999 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_bytes;
1001 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
1002 (__u64)start, lio->lis_pos, lio->lis_endpos,
1003 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_bytes);
1006 * XXX The following call should be optimized: we know, that
1007 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
1009 RETURN(lov_io_iter_init(env, ios));
1012 static int lov_io_setattr_iter_init(const struct lu_env *env,
1013 const struct cl_io_slice *ios)
1015 struct lov_io *lio = cl2lov_io(env, ios);
1016 struct cl_io *io = ios->cis_io;
1020 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1021 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1022 /* no entry found for such offset */
1024 RETURN(io->ci_result = -ENODATA);
1027 RETURN(lov_io_iter_init(env, ios));
1030 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1031 int (*iofunc)(const struct lu_env *, struct cl_io *))
1033 struct cl_io *parent = lio->lis_cl.cis_io;
1034 struct lov_io_sub *sub;
1038 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1039 rc = iofunc(sub->sub_env, &sub->sub_io);
1042 * fsync race with truncate, we'd continue to other
1043 * OST object's fsync to potentially discard
1044 * caching pages (osc_cache_writeback_range).
1046 if (rc == -ENOENT && parent->ci_type == CIT_FSYNC)
1051 if (parent->ci_result == 0)
1052 parent->ci_result = sub->sub_io.ci_result;
1057 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1060 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1063 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1066 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1069 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1073 * It's possible that lov_io_start() wasn't called against this
1074 * sub-io, either because previous sub-io failed, or upper layer
1077 if (io->ci_state == CIS_IO_GOING)
1080 io->ci_state = CIS_IO_FINISHED;
1084 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1086 cl_io_iter_fini(env, io);
1090 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1092 cl_io_unlock(env, io);
1096 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1100 /* Before ending each i/o, we must set lis_cached_entry to tell the
1101 * next i/o not to use stale cached lis information.
1103 cl2lov_io(env, ios)->lis_cached_entry = LIS_CACHE_ENTRY_NONE;
1105 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1110 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1112 struct lov_io *lio = cl2lov_io(env, ios);
1113 struct cl_io *parent = lio->lis_cl.cis_io;
1114 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1115 struct lov_io_sub *sub;
1118 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1119 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1121 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1123 pdv->dv_data_version += sdv->dv_data_version;
1124 if (pdv->dv_layout_version > sdv->dv_layout_version)
1125 pdv->dv_layout_version = sdv->dv_layout_version;
1127 if (parent->ci_result == 0)
1128 parent->ci_result = sub->sub_io.ci_result;
1134 static void lov_io_iter_fini(const struct lu_env *env,
1135 const struct cl_io_slice *ios)
1137 struct lov_io *lio = cl2lov_io(env, ios);
1142 if (lio->lis_trunc_stripe_index != NULL)
1143 OBD_FREE_PTR_ARRAY(lio->lis_trunc_stripe_index,
1144 lio->lis_object->u.composite.lo_entry_count);
1145 lio->lis_trunc_stripe_index = NULL;
1147 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1149 while (!list_empty(&lio->lis_active))
1150 list_del_init(lio->lis_active.next);
1154 static void lov_io_unlock(const struct lu_env *env,
1155 const struct cl_io_slice *ios)
1160 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1165 static int lov_io_read_ahead(const struct lu_env *env,
1166 const struct cl_io_slice *ios,
1167 pgoff_t start, struct cl_read_ahead *ra)
1169 struct lov_io *lio = cl2lov_io(env, ios);
1170 struct lov_object *loo = lio->lis_object;
1171 struct lov_layout_raid0 *r0;
1172 struct lov_io_sub *sub;
1176 unsigned int pps; /* pages per stripe */
1182 offset = start << PAGE_SHIFT;
1183 index = lov_io_layout_at(lio, offset);
1184 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1185 lsm_entry_is_foreign(loo->lo_lsm, index))
1188 /* avoid readahead to expand to stale components */
1189 if (!lov_entry(loo, index)->lle_valid)
1192 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1194 r0 = lov_r0(loo, index);
1195 if (unlikely(!r0->lo_sub[stripe]))
1198 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1200 RETURN(PTR_ERR(sub));
1202 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1203 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1204 suboff >> PAGE_SHIFT, ra);
1206 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1207 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1213 * Adjust the stripe index by layout of comp. ra->cra_end is the
1214 * maximum page index covered by an underlying DLM lock.
1215 * This function converts cra_end from stripe level to file level, and
1216 * make sure it's not beyond stripe and component boundary.
1219 /* cra_end is stripe level, convert it into file level */
1220 ra_end = ra->cra_end_idx;
1221 if (ra_end != CL_PAGE_EOF)
1222 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1225 /* boundary of current component */
1226 ra_end = lov_io_extent(lio, index)->e_end >> PAGE_SHIFT;
1227 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1228 ra->cra_end_idx = ra_end - 1;
1230 if (r0->lo_nr == 1) /* single stripe file */
1233 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1235 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1236 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1237 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1238 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1240 /* never exceed the end of the stripe */
1241 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1242 start + pps - start % pps - 1);
1246 int lov_io_lru_reserve(const struct lu_env *env,
1247 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1249 struct lov_io *lio = cl2lov_io(env, ios);
1250 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1251 struct lov_io_sub *sub;
1252 struct lu_extent ext;
1259 ext.e_end = pos + bytes;
1260 lov_foreach_io_layout(index, lio, &ext) {
1261 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1262 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1267 if (!lsm_entry_inited(lsm, index))
1270 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1271 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1272 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1273 index, lio->lis_mirror_index);
1277 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1278 if (!lov_stripe_intersects(lsm, index, stripe,
1279 &ext, &start, &end))
1282 if (unlikely(!r0->lo_sub[stripe]))
1285 sub = lov_sub_get(env, lio,
1286 lov_comp_index(index, stripe));
1288 return PTR_ERR(sub);
1290 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1301 * lov implementation of cl_operations::cio_submit() method. It takes a list
1302 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1303 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1306 * Major complication of this function is a need to handle memory cleansing:
1307 * cl_io_submit() is called to write out pages as a part of VM memory
1308 * reclamation, and hence it may not fail due to memory shortages (system
1309 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1310 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1311 * not-memory cleansing context), and in case of memory shortage, these
1312 * pre-allocated resources are used by lov_io_submit() under
1313 * lov_device::ld_mutex mutex.
1315 static int lov_io_submit(const struct lu_env *env,
1316 const struct cl_io_slice *ios,
1317 enum cl_req_type crt, struct cl_2queue *queue)
1319 struct cl_page_list *qin = &queue->c2_qin;
1320 struct lov_io *lio = cl2lov_io(env, ios);
1321 struct lov_io_sub *sub;
1322 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1323 struct cl_page *page = cl_page_list_first(qin);
1324 struct cl_page *tmp;
1330 if (page->cp_type == CPT_TRANSIENT)
1333 cl_page_list_init(plist);
1334 while (qin->pl_nr > 0) {
1335 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1337 page = cl_page_list_first(qin);
1338 if (lov_page_is_empty(page)) {
1339 cl_page_list_move(&queue->c2_qout, qin, page);
1342 * it could only be mirror read to get here therefore
1343 * the pages will be transient. We don't care about
1344 * the return code of cl_page_prep() at all.
1346 (void) cl_page_prep(env, ios->cis_io, page, crt);
1347 cl_page_completion(env, page, crt, 0);
1351 cl_2queue_init(cl2q);
1352 cl_page_list_move(&cl2q->c2_qin, qin, page);
1354 index = page->cp_lov_index;
1355 /* DIO is already split by stripe */
1357 cl_page_list_for_each_safe(page, tmp, qin) {
1358 /* this page is not on this stripe */
1359 if (index != page->cp_lov_index)
1362 cl_page_list_move(&cl2q->c2_qin, qin, page);
1365 cl_page_list_splice(qin, &cl2q->c2_qin);
1368 sub = lov_sub_get(env, lio, index);
1370 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1376 cl_page_list_splice(&cl2q->c2_qin, plist);
1377 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1378 cl_2queue_fini(env, cl2q);
1384 cl_page_list_splice(plist, qin);
1385 cl_page_list_fini(env, plist);
1390 static int lov_io_commit_async(const struct lu_env *env,
1391 const struct cl_io_slice *ios,
1392 struct cl_page_list *queue, int from, int to,
1395 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1396 struct lov_io *lio = cl2lov_io(env, ios);
1397 struct lov_io_sub *sub;
1398 struct cl_page *page;
1402 if (lio->lis_nr_subios == 1) {
1403 int idx = lio->lis_single_subio_index;
1405 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1407 sub = lov_sub_get(env, lio, idx);
1408 LASSERT(!IS_ERR(sub));
1409 LASSERT(sub == &lio->lis_single_subio);
1410 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1415 cl_page_list_init(plist);
1416 while (queue->pl_nr > 0) {
1420 LASSERT(plist->pl_nr == 0);
1421 page = cl_page_list_first(queue);
1422 LASSERT(!lov_page_is_empty(page));
1424 cl_page_list_move(plist, queue, page);
1426 index = page->cp_lov_index;
1427 while (queue->pl_nr > 0) {
1428 page = cl_page_list_first(queue);
1429 if (index != page->cp_lov_index)
1432 cl_page_list_move(plist, queue, page);
1435 if (queue->pl_nr > 0) /* still has more pages */
1436 stripe_to = PAGE_SIZE;
1438 sub = lov_sub_get(env, lio, index);
1440 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1441 plist, from, stripe_to, cb);
1447 if (plist->pl_nr > 0) /* short write */
1452 if (lov_comp_entry(index) !=
1453 lov_comp_entry(page->cp_lov_index))
1454 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1457 /* for error case, add the page back into the qin list */
1458 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1459 while (plist->pl_nr > 0) {
1460 /* error occurred, add the uncommitted pages back into queue */
1461 page = cl_page_list_last(plist);
1462 cl_page_list_move_head(queue, plist, page);
1468 static int lov_io_fault_start(const struct lu_env *env,
1469 const struct cl_io_slice *ios)
1471 struct cl_fault_io *fio;
1473 struct lov_io_sub *sub;
1480 fio = &ios->cis_io->u.ci_fault;
1481 lio = cl2lov_io(env, ios);
1484 * LU-14502: ft_page could be an existing cl_page associated with
1485 * the vmpage covering the fault index, and the page may still
1486 * refer to another mirror of an old IO.
1488 if (lov_is_flr(lio->lis_object)) {
1489 offset = fio->ft_index << PAGE_SHIFT;;
1490 entry = lov_io_layout_at(lio, offset);
1492 CERROR(DFID": page fault index %lu invalid component: "
1494 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1495 fio->ft_index, entry,
1496 lio->lis_mirror_index);
1499 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1502 if (fio->ft_page->cp_lov_index !=
1503 lov_comp_index(entry, stripe)) {
1504 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1505 "at mirror %u comp entry %u stripe %u, "
1506 "been used with comp entry %u stripe %u\n",
1507 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1508 fio->ft_index, lio->lis_mirror_index,
1510 lov_comp_entry(fio->ft_page->cp_lov_index),
1511 lov_comp_stripe(fio->ft_page->cp_lov_index));
1513 fio->ft_page->cp_lov_index =
1514 lov_comp_index(entry, stripe);
1518 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1519 sub->sub_io.u.ci_fault.ft_bytes = fio->ft_bytes;
1521 RETURN(lov_io_start(env, ios));
1524 static int lov_io_setattr_start(const struct lu_env *env,
1525 const struct cl_io_slice *ios)
1527 struct lov_io *lio = cl2lov_io(env, ios);
1528 struct cl_io *parent = ios->cis_io;
1529 struct lov_io_sub *sub;
1530 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1534 if (cl_io_is_fallocate(parent)) {
1535 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1536 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1537 int index = lov_comp_entry(sub->sub_subio_index);
1538 int stripe = lov_comp_stripe(sub->sub_subio_index);
1540 size = lov_size_to_stripe(lsm, index, size, stripe);
1541 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1542 sub->sub_io.u.ci_setattr.sa_avalid =
1543 parent->u.ci_setattr.sa_avalid;
1547 RETURN(lov_io_start(env, ios));
1550 static void lov_io_fsync_end(const struct lu_env *env,
1551 const struct cl_io_slice *ios)
1553 struct lov_io *lio = cl2lov_io(env, ios);
1554 struct lov_io_sub *sub;
1555 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1559 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1560 struct cl_io *subio = &sub->sub_io;
1562 lov_io_end_wrapper(sub->sub_env, subio);
1564 if (subio->ci_result == 0)
1565 *written += subio->u.ci_fsync.fi_nr_written;
1570 static void lov_io_lseek_end(const struct lu_env *env,
1571 const struct cl_io_slice *ios)
1573 struct lov_io *lio = cl2lov_io(env, ios);
1574 struct cl_io *io = lio->lis_cl.cis_io;
1575 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1576 struct lov_io_sub *sub;
1577 loff_t offset = -ENXIO;
1579 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1583 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1584 struct cl_io *subio = &sub->sub_io;
1585 int index = lov_comp_entry(sub->sub_subio_index);
1586 int stripe = lov_comp_stripe(sub->sub_subio_index);
1587 loff_t sub_off, lov_off;
1588 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1590 lov_io_end_wrapper(sub->sub_env, subio);
1592 if (io->ci_result == 0)
1593 io->ci_result = sub->sub_io.ci_result;
1598 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1599 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1600 index, stripe, seek_hole ? "HOLE" : "DATA",
1601 subio->u.ci_lseek.ls_start);
1603 /* first subio with positive result is what we need */
1604 sub_off = subio->u.ci_lseek.ls_result;
1605 /* Expected error, offset is out of stripe file size */
1606 if (sub_off == -ENXIO)
1608 /* Any other errors are not expected with ci_result == 0 */
1610 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1612 io->ci_result = sub_off;
1615 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1617 /* the only way to get negatove lov_off here is too big
1618 * result. Return -EOVERFLOW then.
1620 io->ci_result = -EOVERFLOW;
1621 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1622 (u64)lov_off, io->ci_result);
1625 if (lov_off < io->u.ci_lseek.ls_start) {
1626 io->ci_result = -EINVAL;
1627 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1628 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1631 /* resulting offset can be out of component range if stripe
1632 * object is full and its file size was returned as virtual
1633 * hole start. Skip this result, the next component will give
1634 * us correct lseek result but keep possible hole offset in
1635 * case there is no more components ahead
1637 if (lov_off >= comp_end) {
1638 /* must be SEEK_HOLE case */
1639 if (likely(seek_hole)) {
1640 /* save comp end as potential hole offset */
1641 hole_off = max_t(__u64, comp_end, hole_off);
1643 io->ci_result = -EINVAL;
1645 "off %lld >= comp_end %llu: rc = %d\n",
1646 lov_off, comp_end, io->ci_result);
1651 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1652 seek_hole ? "HOLE" : "DATA",
1653 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1654 sub->sub_io.ci_result);
1655 offset = min_t(__u64, offset, lov_off);
1657 /* no result but some component returns hole as component end */
1658 if (seek_hole && offset == -ENXIO && hole_off > 0)
1661 io->u.ci_lseek.ls_result = offset;
1665 static const struct cl_io_operations lov_io_ops = {
1668 .cio_fini = lov_io_fini,
1669 .cio_iter_init = lov_io_rw_iter_init,
1670 .cio_iter_fini = lov_io_iter_fini,
1671 .cio_lock = lov_io_lock,
1672 .cio_unlock = lov_io_unlock,
1673 .cio_start = lov_io_start,
1674 .cio_end = lov_io_end
1677 .cio_fini = lov_io_fini,
1678 .cio_iter_init = lov_io_rw_iter_init,
1679 .cio_iter_fini = lov_io_iter_fini,
1680 .cio_lock = lov_io_lock,
1681 .cio_unlock = lov_io_unlock,
1682 .cio_start = lov_io_start,
1683 .cio_end = lov_io_end
1686 .cio_fini = lov_io_fini,
1687 .cio_iter_init = lov_io_setattr_iter_init,
1688 .cio_iter_fini = lov_io_iter_fini,
1689 .cio_lock = lov_io_lock,
1690 .cio_unlock = lov_io_unlock,
1691 .cio_start = lov_io_setattr_start,
1692 .cio_end = lov_io_end
1694 [CIT_DATA_VERSION] = {
1695 .cio_fini = lov_io_fini,
1696 .cio_iter_init = lov_io_iter_init,
1697 .cio_iter_fini = lov_io_iter_fini,
1698 .cio_lock = lov_io_lock,
1699 .cio_unlock = lov_io_unlock,
1700 .cio_start = lov_io_start,
1701 .cio_end = lov_io_data_version_end,
1704 .cio_fini = lov_io_fini,
1705 .cio_iter_init = lov_io_iter_init,
1706 .cio_iter_fini = lov_io_iter_fini,
1707 .cio_lock = lov_io_lock,
1708 .cio_unlock = lov_io_unlock,
1709 .cio_start = lov_io_fault_start,
1710 .cio_end = lov_io_end
1713 .cio_fini = lov_io_fini,
1714 .cio_iter_init = lov_io_iter_init,
1715 .cio_iter_fini = lov_io_iter_fini,
1716 .cio_lock = lov_io_lock,
1717 .cio_unlock = lov_io_unlock,
1718 .cio_start = lov_io_start,
1719 .cio_end = lov_io_fsync_end
1722 .cio_fini = lov_io_fini,
1723 .cio_iter_init = lov_io_iter_init,
1724 .cio_iter_fini = lov_io_iter_fini,
1725 .cio_lock = lov_io_lock,
1726 .cio_unlock = lov_io_unlock,
1727 .cio_start = lov_io_start,
1728 .cio_end = lov_io_end
1731 .cio_fini = lov_io_fini,
1732 .cio_iter_init = lov_io_iter_init,
1733 .cio_iter_fini = lov_io_iter_fini,
1734 .cio_lock = lov_io_lock,
1735 .cio_unlock = lov_io_unlock,
1736 .cio_start = lov_io_start,
1737 .cio_end = lov_io_lseek_end
1740 .cio_fini = lov_io_fini,
1743 .cio_fini = lov_io_fini
1746 .cio_read_ahead = lov_io_read_ahead,
1747 .cio_lru_reserve = lov_io_lru_reserve,
1748 .cio_submit = lov_io_submit,
1749 .cio_commit_async = lov_io_commit_async,
1752 /*****************************************************************************
1754 * Empty lov io operations.
1758 static void lov_empty_io_fini(const struct lu_env *env,
1759 const struct cl_io_slice *ios)
1761 struct lov_object *lov = cl2lov(ios->cis_obj);
1764 if (atomic_dec_and_test(&lov->lo_active_ios))
1765 wake_up(&lov->lo_waitq);
1769 static int lov_empty_io_submit(const struct lu_env *env,
1770 const struct cl_io_slice *ios,
1771 enum cl_req_type crt, struct cl_2queue *queue)
1776 static void lov_empty_impossible(const struct lu_env *env,
1777 struct cl_io_slice *ios)
1782 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1785 * An io operation vector for files without stripes.
1787 static const struct cl_io_operations lov_empty_io_ops = {
1790 .cio_fini = lov_empty_io_fini,
1792 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1793 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1794 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1795 .cio_end = LOV_EMPTY_IMPOSSIBLE
1799 .cio_fini = lov_empty_io_fini,
1800 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1801 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1802 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1803 .cio_end = LOV_EMPTY_IMPOSSIBLE
1806 .cio_fini = lov_empty_io_fini,
1807 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1808 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1809 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1810 .cio_end = LOV_EMPTY_IMPOSSIBLE
1813 .cio_fini = lov_empty_io_fini,
1814 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1815 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1816 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1817 .cio_end = LOV_EMPTY_IMPOSSIBLE
1820 .cio_fini = lov_empty_io_fini
1823 .cio_fini = lov_empty_io_fini
1826 .cio_fini = lov_empty_io_fini
1829 .cio_fini = lov_empty_io_fini
1832 .cio_submit = lov_empty_io_submit,
1833 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1836 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1839 struct lov_io *lio = lov_env_io(env);
1840 struct lov_object *lov = cl2lov(obj);
1845 INIT_LIST_HEAD(&lio->lis_active);
1846 result = lov_io_slice_init(lio, lov, io);
1850 result = lov_io_subio_init(env, lio, io);
1852 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1853 atomic_inc(&lov->lo_active_ios);
1857 io->ci_result = result < 0 ? result : 0;
1861 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1864 struct lov_object *lov = cl2lov(obj);
1865 struct lov_io *lio = lov_env_io(env);
1869 lio->lis_object = lov;
1870 switch (io->ci_type) {
1882 case CIT_DATA_VERSION:
1890 CERROR("Page fault on a file without stripes: "DFID"\n",
1891 PFID(lu_object_fid(&obj->co_lu)));
1895 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1896 atomic_inc(&lov->lo_active_ios);
1899 io->ci_result = result < 0 ? result : 0;
1903 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1906 struct lov_object *lov = cl2lov(obj);
1907 struct lov_io *lio = lov_env_io(env);
1911 LASSERT(lov->lo_lsm != NULL);
1912 lio->lis_object = lov;
1914 switch (io->ci_type) {
1916 LASSERTF(0, "invalid type %d\n", io->ci_type);
1917 result = -EOPNOTSUPP;
1923 case CIT_DATA_VERSION:
1928 * the truncate to 0 is managed by MDT:
1929 * - in open, for open O_TRUNC
1930 * - in setattr, for truncate
1933 * the truncate is for size > 0 so triggers a restore,
1934 * also trigger a restore for prealloc/punch
1936 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1937 io->ci_restore_needed = 1;
1946 io->ci_restore_needed = 1;
1952 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1953 atomic_inc(&lov->lo_active_ios);
1956 io->ci_result = result < 0 ? result : 0;
1961 * Return the index in composite:lo_entries by the file offset
1963 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1965 struct lov_object *lov = lio->lis_object;
1966 struct lov_layout_composite *comp = &lov->u.composite;
1967 int start_index = 0;
1968 int end_index = comp->lo_entry_count - 1;
1971 LASSERT(lov->lo_type == LLT_COMP);
1973 /* This is actual file offset so nothing can cover eof. */
1974 if (offset == LUSTRE_EOF)
1977 if (lov_is_flr(lov)) {
1978 struct lov_mirror_entry *lre;
1980 LASSERT(lio->lis_mirror_index >= 0);
1982 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1983 start_index = lre->lre_start;
1984 end_index = lre->lre_end;
1987 for (i = start_index; i <= end_index; i++) {
1988 struct lov_layout_entry *lle = lov_entry(lov, i);
1990 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1992 if ((offset >= lle->lle_extent->e_start &&
1993 offset < lle->lle_extent->e_end) ||
1994 (offset == OBD_OBJECT_EOF &&
1995 lle->lle_extent->e_end == OBD_OBJECT_EOF))