4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * Implementation of cl_io for LOV layer.
33 * Author: Nikita Danilov <nikita.danilov@sun.com>
34 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
37 #define DEBUG_SUBSYSTEM S_LOV
39 #include "lov_cl_internal.h"
45 static inline struct lov_io_sub *lov_sub_alloc(struct lov_io *lio, int index)
47 struct lov_io_sub *sub;
49 if (lio->lis_nr_subios == 0) {
50 LASSERT(lio->lis_single_subio_index == -1);
51 sub = &lio->lis_single_subio;
52 lio->lis_single_subio_index = index;
53 memset(sub, 0, sizeof(*sub));
59 INIT_LIST_HEAD(&sub->sub_list);
60 INIT_LIST_HEAD(&sub->sub_linkage);
61 sub->sub_subio_index = index;
67 static inline void lov_sub_free(struct lov_io *lio, struct lov_io_sub *sub)
69 if (sub->sub_subio_index == lio->lis_single_subio_index) {
70 LASSERT(sub == &lio->lis_single_subio);
71 lio->lis_single_subio_index = -1;
77 static void lov_io_sub_fini(const struct lu_env *env, struct lov_io *lio,
78 struct lov_io_sub *sub)
82 cl_io_fini(sub->sub_env, &sub->sub_io);
84 if (sub->sub_env && !IS_ERR(sub->sub_env)) {
85 cl_env_put(sub->sub_env, &sub->sub_refcheck);
92 is_index_within_mirror(struct lov_object *lov, int index, int mirror_index)
94 struct lov_layout_composite *comp = &lov->u.composite;
95 struct lov_mirror_entry *lre = &comp->lo_mirrors[mirror_index];
97 return (index >= lre->lre_start && index <= lre->lre_end);
100 static int lov_io_sub_init(const struct lu_env *env, struct lov_io *lio,
101 struct lov_io_sub *sub)
103 struct lov_object *lov = lio->lis_object;
104 struct cl_io *sub_io;
105 struct cl_object *sub_obj;
106 struct cl_io *io = lio->lis_cl.cis_io;
107 int index = lov_comp_entry(sub->sub_subio_index);
108 int stripe = lov_comp_stripe(sub->sub_subio_index);
110 LASSERT(sub->sub_env == NULL);
113 if (unlikely(!lov_r0(lov, index)->lo_sub ||
114 !lov_r0(lov, index)->lo_sub[stripe]))
117 LASSERTF(ergo(lov_is_flr(lov),
118 is_index_within_mirror(lov, index,
119 lio->lis_mirror_index)),
120 DFID "iot = %d, index = %d, mirror = %d\n",
121 PFID(lu_object_fid(lov2lu(lov))), io->ci_type, index,
122 lio->lis_mirror_index);
124 /* obtain new environment */
125 sub->sub_env = cl_env_get(&sub->sub_refcheck);
126 if (IS_ERR(sub->sub_env)) {
127 result = PTR_ERR(sub->sub_env);
131 sub_obj = lovsub2cl(lov_r0(lov, index)->lo_sub[stripe]);
132 sub_io = &sub->sub_io;
134 sub_io->ci_obj = sub_obj;
135 sub_io->ci_result = 0;
137 sub_io->ci_parent = io;
138 sub_io->ci_lockreq = io->ci_lockreq;
139 sub_io->ci_type = io->ci_type;
140 sub_io->ci_no_srvlock = io->ci_no_srvlock;
141 sub_io->ci_noatime = io->ci_noatime;
142 sub_io->ci_async_readahead = io->ci_async_readahead;
143 sub_io->ci_lock_no_expand = io->ci_lock_no_expand;
144 sub_io->ci_ndelay = io->ci_ndelay;
145 sub_io->ci_layout_version = io->ci_layout_version;
146 sub_io->ci_tried_all_mirrors = io->ci_tried_all_mirrors;
148 result = cl_io_sub_init(sub->sub_env, sub_io, io->ci_type, sub_obj);
151 lov_io_sub_fini(env, lio, sub);
156 struct lov_io_sub *lov_sub_get(const struct lu_env *env,
157 struct lov_io *lio, int index)
159 struct lov_io_sub *sub;
164 list_for_each_entry(sub, &lio->lis_subios, sub_list) {
165 if (sub->sub_subio_index == index) {
172 sub = lov_sub_alloc(lio, index);
174 GOTO(out, rc = -ENOMEM);
176 rc = lov_io_sub_init(env, lio, sub);
178 lov_sub_free(lio, sub);
182 list_add_tail(&sub->sub_list, &lio->lis_subios);
183 lio->lis_nr_subios++;
189 sub->sub_io.ci_noquota = lio->lis_cl.cis_io->ci_noquota;
193 /*****************************************************************************
198 static int lov_io_subio_init(const struct lu_env *env, struct lov_io *lio,
203 LASSERT(lio->lis_object != NULL);
205 INIT_LIST_HEAD(&lio->lis_subios);
206 lio->lis_single_subio_index = -1;
207 lio->lis_nr_subios = 0;
213 * Decide if it will need write intent RPC
215 static int lov_io_mirror_write_intent(struct lov_io *lio,
216 struct lov_object *obj, struct cl_io *io)
218 struct lov_layout_composite *comp = &obj->u.composite;
219 struct lu_extent *ext = &io->ci_write_intent;
220 struct lov_mirror_entry *lre;
221 struct lov_mirror_entry *primary;
222 struct lov_layout_entry *lle;
226 *ext = (typeof(*ext)) { lio->lis_pos, lio->lis_endpos };
227 io->ci_need_write_intent = 0;
229 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
230 cl_io_is_fallocate(io) || cl_io_is_trunc(io)))
234 * FLR: check if it needs to send a write intent RPC to server.
235 * Writing to sync_pending file needs write intent RPC to change
236 * the file state back to write_pending, so that the layout version
237 * can be increased when the state changes to sync_pending at a later
238 * time. Otherwise there exists a chance that an evicted client may
239 * dirty the file data while resync client is working on it.
240 * Designated I/O is allowed for resync workload.
242 if (lov_flr_state(obj) == LCM_FL_RDONLY ||
243 (lov_flr_state(obj) == LCM_FL_SYNC_PENDING &&
244 io->ci_designated_mirror == 0)) {
245 io->ci_need_write_intent = 1;
249 LASSERT((lov_flr_state(obj) == LCM_FL_WRITE_PENDING));
250 LASSERT(comp->lo_preferred_mirror >= 0);
253 * need to iterate all components to see if there are
254 * multiple components covering the writing component
256 primary = &comp->lo_mirrors[comp->lo_preferred_mirror];
257 LASSERT(!primary->lre_stale);
258 lov_foreach_mirror_layout_entry(obj, lle, primary) {
259 LASSERT(lle->lle_valid);
260 if (!lu_extent_is_overlapped(ext, lle->lle_extent))
263 ext->e_start = min(ext->e_start, lle->lle_extent->e_start);
264 ext->e_end = max(ext->e_end, lle->lle_extent->e_end);
268 CERROR(DFID ": cannot find any valid components covering "
269 "file extent "DEXT", mirror: %d\n",
270 PFID(lu_object_fid(lov2lu(obj))), PEXT(ext),
271 primary->lre_mirror_id);
276 lov_foreach_mirror_entry(obj, lre) {
280 lov_foreach_mirror_layout_entry(obj, lle, lre) {
284 if (lu_extent_is_overlapped(ext, lle->lle_extent)) {
291 CDEBUG(D_VFSTRACE, DFID "there are %zd components to be staled to "
292 "modify file extent "DEXT", iot: %d\n",
293 PFID(lu_object_fid(lov2lu(obj))), count, PEXT(ext), io->ci_type);
295 io->ci_need_write_intent = count > 0;
300 static int lov_io_mirror_init(struct lov_io *lio, struct lov_object *obj,
303 struct lov_layout_composite *comp = &obj->u.composite;
309 if (!lov_is_flr(obj)) {
310 /* only locks/pages are manipulated for CIT_MISC op, no
311 * cl_io_loop() will be called, don't check/set mirror info.
313 if (io->ci_type != CIT_MISC) {
314 LASSERT(comp->lo_preferred_mirror == 0);
315 lio->lis_mirror_index = comp->lo_preferred_mirror;
321 /* transfer the layout version for verification */
322 if (io->ci_layout_version == 0)
323 io->ci_layout_version = obj->lo_lsm->lsm_layout_gen;
325 /* find the corresponding mirror for designated mirror IO */
326 if (io->ci_designated_mirror > 0) {
327 struct lov_mirror_entry *entry;
329 LASSERT(!io->ci_ndelay);
331 CDEBUG(D_LAYOUT, "designated I/O mirror state: %d\n",
334 if ((cl_io_is_trunc(io) || io->ci_type == CIT_WRITE ||
335 cl_io_is_fallocate(io)) &&
336 (io->ci_layout_version != obj->lo_lsm->lsm_layout_gen)) {
338 * For resync I/O, the ci_layout_version was the layout
339 * version when resync starts. If it doesn't match the
340 * current object layout version, it means the layout
346 io->ci_layout_version |= LU_LAYOUT_RESYNC;
349 lio->lis_mirror_index = -1;
350 lov_foreach_mirror_entry(obj, entry) {
351 if (entry->lre_mirror_id ==
352 io->ci_designated_mirror) {
353 lio->lis_mirror_index = index;
360 RETURN(lio->lis_mirror_index < 0 ? -EINVAL : 0);
363 result = lov_io_mirror_write_intent(lio, obj, io);
367 if (io->ci_need_write_intent) {
368 CDEBUG(D_VFSTRACE, DFID " need write intent for [%llu, %llu)\n",
369 PFID(lu_object_fid(lov2lu(obj))),
370 lio->lis_pos, lio->lis_endpos);
372 if (cl_io_is_trunc(io)) {
374 * for truncate, we uses [size, EOF) to judge whether
375 * a write intent needs to be send, but we need to
376 * restore the write extent to [0, size], in truncate,
377 * the byte in the size position is accessed.
379 io->ci_write_intent.e_start = 0;
380 io->ci_write_intent.e_end =
381 io->u.ci_setattr.sa_attr.lvb_size + 1;
383 /* stop cl_io_init() loop */
387 if (io->ci_ndelay_tried == 0 || /* first time to try */
388 /* reset the mirror index if layout has changed */
389 lio->lis_mirror_layout_gen != obj->lo_lsm->lsm_layout_gen) {
390 lio->lis_mirror_layout_gen = obj->lo_lsm->lsm_layout_gen;
391 index = lio->lis_mirror_index = comp->lo_preferred_mirror;
393 index = lio->lis_mirror_index;
396 /* move mirror index to the next one */
397 index = (index + 1) % comp->lo_mirror_count;
400 for (i = 0; i < comp->lo_mirror_count; i++) {
401 struct lu_extent ext = { .e_start = lio->lis_pos,
402 .e_end = lio->lis_pos + 1 };
403 struct lov_mirror_entry *lre;
404 struct lov_layout_entry *lle;
407 lre = &comp->lo_mirrors[(index + i) % comp->lo_mirror_count];
411 if (lre->lre_foreign)
414 lov_foreach_mirror_layout_entry(obj, lle, lre) {
418 if (lu_extent_is_overlapped(&ext, lle->lle_extent)) {
422 } /* each component of the mirror */
424 index = (index + i) % comp->lo_mirror_count;
429 if (i == comp->lo_mirror_count) {
430 CERROR(DFID": failed to find a component covering "
431 "I/O region at %llu\n",
432 PFID(lu_object_fid(lov2lu(obj))), lio->lis_pos);
434 dump_lsm(D_ERROR, obj->lo_lsm);
439 CDEBUG(D_VFSTRACE, DFID ": flr state: %d, move mirror from %d to %d, "
440 "have retried: %d, mirror count: %d\n",
441 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj),
442 lio->lis_mirror_index, index, io->ci_ndelay_tried,
443 comp->lo_mirror_count);
445 lio->lis_mirror_index = index;
448 * FLR: if all mirrors have been tried once, most likely the network
449 * of this client has been partitioned. We should relinquish CPU for
450 * a while before trying again.
452 if (io->ci_ndelay && io->ci_ndelay_tried > 0 &&
453 (io->ci_ndelay_tried % comp->lo_mirror_count == 0)) {
454 schedule_timeout_interruptible(cfs_time_seconds(1) / 100);
455 if (signal_pending(current))
459 * we'd set ci_tried_all_mirrors to turn off fast mirror
460 * switching for read after we've tried all mirrors several
463 io->ci_tried_all_mirrors = io->ci_ndelay_tried %
464 (comp->lo_mirror_count * 4) == 0;
466 ++io->ci_ndelay_tried;
468 CDEBUG(D_VFSTRACE, "use %sdelayed RPC state for this IO\n",
469 io->ci_ndelay ? "non-" : "");
474 static int lov_io_slice_init(struct lov_io *lio,
475 struct lov_object *obj, struct cl_io *io)
482 lio->lis_object = obj;
483 lio->lis_cached_entry = LIS_CACHE_ENTRY_NONE;
485 switch (io->ci_type) {
488 lio->lis_pos = io->u.ci_rw.crw_pos;
489 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
490 lio->lis_io_endpos = lio->lis_endpos;
491 if (cl_io_is_append(io)) {
492 LASSERT(io->ci_type == CIT_WRITE);
495 * If there is LOV EA hole, then we may cannot locate
496 * the current file-tail exactly.
498 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
500 GOTO(out, result = -EIO);
503 lio->lis_endpos = OBD_OBJECT_EOF;
508 if (cl_io_is_fallocate(io)) {
509 lio->lis_pos = io->u.ci_setattr.sa_falloc_offset;
510 lio->lis_endpos = io->u.ci_setattr.sa_falloc_end;
511 } else if (cl_io_is_trunc(io)) {
512 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
513 lio->lis_endpos = OBD_OBJECT_EOF;
516 lio->lis_endpos = OBD_OBJECT_EOF;
520 case CIT_DATA_VERSION:
522 lio->lis_endpos = OBD_OBJECT_EOF;
526 pgoff_t index = io->u.ci_fault.ft_index;
528 lio->lis_pos = cl_offset(io->ci_obj, index);
529 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
534 lio->lis_pos = io->u.ci_fsync.fi_start;
535 lio->lis_endpos = io->u.ci_fsync.fi_end;
540 lio->lis_pos = io->u.ci_ladvise.li_start;
541 lio->lis_endpos = io->u.ci_ladvise.li_end;
546 lio->lis_pos = io->u.ci_lseek.ls_start;
547 lio->lis_endpos = OBD_OBJECT_EOF;
553 lio->lis_endpos = OBD_OBJECT_EOF;
555 if (lov_flr_state(obj) == LCM_FL_RDONLY &&
556 !OBD_FAIL_CHECK(OBD_FAIL_FLR_GLIMPSE_IMMUTABLE))
557 /* SoM is accurate, no need glimpse */
558 GOTO(out, result = 1);
563 lio->lis_endpos = OBD_OBJECT_EOF;
571 * CIT_MISC + ci_ignore_layout can identify the I/O from the OSC layer,
572 * it won't care/access lov layout related info.
574 if (io->ci_ignore_layout && io->ci_type == CIT_MISC)
575 GOTO(out, result = 0);
577 LASSERT(obj->lo_lsm != NULL);
579 result = lov_io_mirror_init(lio, obj, io);
583 /* check if it needs to instantiate layout */
584 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
585 cl_io_is_fallocate(io) ||
586 (cl_io_is_trunc(io) && io->u.ci_setattr.sa_attr.lvb_size > 0)))
587 GOTO(out, result = 0);
590 * for truncate, it only needs to instantiate the components
591 * before the truncated size.
593 if (cl_io_is_trunc(io)) {
594 io->ci_write_intent.e_start = 0;
595 /* for writes, e_end is endpos, the location of the file
596 * pointer after the write is completed, so it is not accessed.
597 * For truncate, 'end' is the size, and *is* acccessed.
598 * In other words, writes are [start, end), but truncate is
599 * [start, size], where both are included. So add 1 to the
600 * size when creating the write intent to account for this.
602 io->ci_write_intent.e_end =
603 io->u.ci_setattr.sa_attr.lvb_size + 1;
605 io->ci_write_intent.e_start = lio->lis_pos;
606 io->ci_write_intent.e_end = lio->lis_endpos;
610 lov_foreach_io_layout(index, lio, &io->ci_write_intent) {
611 if (!lsm_entry_inited(obj->lo_lsm, index)) {
612 io->ci_need_write_intent = 1;
617 if (io->ci_need_write_intent && io->ci_designated_mirror > 0) {
619 * REINT_SYNC RPC has already tried to instantiate all of the
620 * components involved, obviously it didn't succeed. Skip this
621 * mirror for now. The server won't be able to figure out
622 * which mirror it should instantiate components
624 CERROR(DFID": trying to instantiate components for designated "
625 "I/O, file state: %d\n",
626 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj));
628 io->ci_need_write_intent = 0;
629 GOTO(out, result = -EIO);
632 if (io->ci_need_write_intent)
633 GOTO(out, result = 1);
641 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
643 struct lov_io *lio = cl2lov_io(env, ios);
644 struct lov_object *lov = cl2lov(ios->cis_obj);
645 struct lov_io_sub *sub;
648 LASSERT(list_empty(&lio->lis_active));
650 while ((sub = list_first_entry_or_null(&lio->lis_subios,
652 sub_list)) != NULL) {
653 list_del_init(&sub->sub_list);
654 lio->lis_nr_subios--;
656 lov_io_sub_fini(env, lio, sub);
657 lov_sub_free(lio, sub);
659 LASSERT(lio->lis_nr_subios == 0);
661 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
662 if (atomic_dec_and_test(&lov->lo_active_ios))
663 wake_up(&lov->lo_waitq);
667 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
668 loff_t start, loff_t end)
670 struct cl_io *io = &sub->sub_io;
671 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
672 struct cl_io *parent = lio->lis_cl.cis_io;
673 int index = lov_comp_entry(sub->sub_subio_index);
674 int stripe = lov_comp_stripe(sub->sub_subio_index);
676 switch (io->ci_type) {
678 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
679 io->u.ci_setattr.sa_attr_flags =
680 parent->u.ci_setattr.sa_attr_flags;
681 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
682 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
683 io->u.ci_setattr.sa_falloc_mode =
684 parent->u.ci_setattr.sa_falloc_mode;
685 io->u.ci_setattr.sa_stripe_index = stripe;
686 io->u.ci_setattr.sa_parent_fid =
687 parent->u.ci_setattr.sa_parent_fid;
688 /* For SETATTR(fallocate) pass the subtype to lower IO */
689 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
690 if (cl_io_is_fallocate(io)) {
691 io->u.ci_setattr.sa_falloc_offset = start;
692 io->u.ci_setattr.sa_falloc_end = end;
693 io->u.ci_setattr.sa_falloc_uid =
694 parent->u.ci_setattr.sa_falloc_uid;
695 io->u.ci_setattr.sa_falloc_gid =
696 parent->u.ci_setattr.sa_falloc_gid;
697 io->u.ci_setattr.sa_falloc_projid =
698 parent->u.ci_setattr.sa_falloc_projid;
700 if (cl_io_is_trunc(io)) {
701 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
703 new_size = lov_size_to_stripe(lsm, index, new_size,
705 io->u.ci_setattr.sa_attr.lvb_size = new_size;
707 lov_lsm2layout(lsm, lsm->lsm_entries[index],
708 &io->u.ci_setattr.sa_layout);
711 case CIT_DATA_VERSION: {
712 io->u.ci_data_version.dv_data_version = 0;
713 io->u.ci_data_version.dv_flags =
714 parent->u.ci_data_version.dv_flags;
718 struct cl_object *obj = parent->ci_obj;
719 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
721 io->u.ci_fault = parent->u.ci_fault;
722 off = lov_size_to_stripe(lsm, index, off, stripe);
723 io->u.ci_fault.ft_index = cl_index(obj, off);
727 io->u.ci_fsync.fi_start = start;
728 io->u.ci_fsync.fi_end = end;
729 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
730 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
735 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
736 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
737 if (cl_io_is_append(parent)) {
738 io->u.ci_wr.wr_append = 1;
740 io->u.ci_rw.crw_pos = start;
741 io->u.ci_rw.crw_count = end - start;
746 io->u.ci_ladvise.li_start = start;
747 io->u.ci_ladvise.li_end = end;
748 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
749 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
750 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
754 io->u.ci_lseek.ls_start = start;
755 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
756 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
766 static loff_t lov_offset_mod(loff_t val, int delta)
768 if (val != OBD_OBJECT_EOF)
773 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
774 struct lov_io_sub *sub, u64 start, u64 end)
778 end = lov_offset_mod(end, 1);
779 lov_io_sub_inherit(sub, lio, start, end);
780 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
782 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
786 list_add_tail(&sub->sub_linkage, &lio->lis_active);
790 static int lov_io_iter_init(const struct lu_env *env,
791 const struct cl_io_slice *ios)
793 struct lov_io *lio = cl2lov_io(env, ios);
794 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
795 bool is_trunc = cl_io_is_trunc(ios->cis_io);
796 struct lov_io_sub *sub;
797 struct lu_extent ext;
803 ext.e_start = lio->lis_pos;
804 ext.e_end = lio->lis_endpos;
807 OBD_ALLOC_PTR_ARRAY(lio->lis_trunc_stripe_index,
808 lio->lis_object->u.composite.lo_entry_count);
809 if (lio->lis_trunc_stripe_index == NULL)
813 lov_foreach_io_layout(index, lio, &ext) {
814 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
815 struct lov_layout_raid0 *r0 = &le->lle_raid0;
819 bool tested_trunc_stripe = false;
822 lio->lis_trunc_stripe_index[index] = -1;
824 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
825 index, lsm->lsm_entries[index]->lsme_flags);
826 if (!lsm_entry_inited(lsm, index)) {
828 * Read from uninitialized components should return
834 if (lsm_entry_is_foreign(lsm, index))
837 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
838 CERROR("I/O to invalid component: %d, mirror: %d\n",
839 index, lio->lis_mirror_index);
843 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
844 if (!lov_stripe_intersects(lsm, index, stripe,
848 if (unlikely(!r0->lo_sub[stripe])) {
849 if (ios->cis_io->ci_type == CIT_READ ||
850 ios->cis_io->ci_type == CIT_WRITE ||
851 ios->cis_io->ci_type == CIT_FAULT)
857 if (is_trunc && !tested_trunc_stripe) {
861 prev = (stripe == 0) ? r0->lo_nr - 1 :
864 * Only involving previous stripe if the
865 * truncate in this component is at the
866 * beginning of this stripe.
868 tested_trunc_stripe = true;
869 if (ext.e_start < lsm->lsm_entries[index]->
870 lsme_extent.e_start) {
871 /* need previous stripe involvement */
872 lio->lis_trunc_stripe_index[index] = prev;
874 tr_start = ext.e_start;
875 tr_start = lov_do_div64(tr_start,
876 stripe_width(lsm, index));
877 /* tr_start %= stripe_swidth */
878 if (tr_start == stripe * lsm->
881 lio->lis_trunc_stripe_index[index] = prev;
885 /* if the last stripe is the trunc stripeno */
886 if (is_trunc && lio->lis_trunc_stripe_index[index] == stripe)
887 lio->lis_trunc_stripe_index[index] = -1;
889 sub = lov_sub_get(env, lio,
890 lov_comp_index(index, stripe));
894 rc = lov_io_add_sub(env, lio, sub, start, end);
901 if (is_trunc && lio->lis_trunc_stripe_index[index] != -1) {
902 stripe = lio->lis_trunc_stripe_index[index];
903 if (unlikely(!r0->lo_sub[stripe])) {
904 lio->lis_trunc_stripe_index[index] = -1;
907 sub = lov_sub_get(env, lio,
908 lov_comp_index(index, stripe));
913 * the prev sub could be used by another truncate, we'd
914 * skip it. LU-14128 happends when expand truncate +
915 * read get wrong kms.
917 if (!list_empty(&sub->sub_linkage)) {
918 lio->lis_trunc_stripe_index[index] = -1;
922 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
924 rc = lov_io_add_sub(env, lio, sub, start, end);
933 static int lov_io_rw_iter_init(const struct lu_env *env,
934 const struct cl_io_slice *ios)
936 struct lov_io *lio = cl2lov_io(env, ios);
937 struct cl_io *io = ios->cis_io;
938 struct lov_stripe_md_entry *lse;
939 loff_t start = io->u.ci_rw.crw_pos;
943 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
946 if (cl_io_is_append(io))
947 RETURN(lov_io_iter_init(env, ios));
949 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
950 if (index < 0) { /* non-existing layout component */
951 if (io->ci_type == CIT_READ) {
953 * TODO: it needs to detect the next component and
954 * then set the next pos
958 RETURN(lov_io_iter_init(env, ios));
964 if (!lov_entry(lio->lis_object, index)->lle_valid &&
965 !io->ci_designated_mirror)
966 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
968 lse = lov_lse(lio->lis_object, index);
970 if (lsme_is_foreign(lse))
973 next = MAX_LFS_FILESIZE;
974 if (lse->lsme_stripe_count > 1) {
975 unsigned long ssize = lse->lsme_stripe_size;
977 lov_do_div64(start, ssize);
978 next = (start + 1) * ssize;
979 if (next <= start * ssize)
980 next = MAX_LFS_FILESIZE;
983 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
984 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
985 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
986 next = min_t(__u64, next, lse->lsme_extent.e_end);
987 next = min_t(loff_t, next, lio->lis_io_endpos);
989 io->ci_continue = next < lio->lis_io_endpos;
990 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
991 lio->lis_pos = io->u.ci_rw.crw_pos;
992 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
994 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
995 (__u64)start, lio->lis_pos, lio->lis_endpos,
996 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
999 * XXX The following call should be optimized: we know, that
1000 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
1002 RETURN(lov_io_iter_init(env, ios));
1005 static int lov_io_setattr_iter_init(const struct lu_env *env,
1006 const struct cl_io_slice *ios)
1008 struct lov_io *lio = cl2lov_io(env, ios);
1009 struct cl_io *io = ios->cis_io;
1013 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1014 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1015 /* no entry found for such offset */
1017 RETURN(io->ci_result = -ENODATA);
1020 RETURN(lov_io_iter_init(env, ios));
1023 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1024 int (*iofunc)(const struct lu_env *, struct cl_io *))
1026 struct cl_io *parent = lio->lis_cl.cis_io;
1027 struct lov_io_sub *sub;
1031 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1032 rc = iofunc(sub->sub_env, &sub->sub_io);
1036 if (parent->ci_result == 0)
1037 parent->ci_result = sub->sub_io.ci_result;
1042 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1045 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1048 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1051 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1054 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1058 * It's possible that lov_io_start() wasn't called against this
1059 * sub-io, either because previous sub-io failed, or upper layer
1062 if (io->ci_state == CIS_IO_GOING)
1065 io->ci_state = CIS_IO_FINISHED;
1069 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1071 cl_io_iter_fini(env, io);
1075 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1077 cl_io_unlock(env, io);
1081 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1085 /* Before ending each i/o, we must set lis_cached_entry to tell the
1086 * next i/o not to use stale cached lis information.
1088 cl2lov_io(env, ios)->lis_cached_entry = LIS_CACHE_ENTRY_NONE;
1090 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1095 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1097 struct lov_io *lio = cl2lov_io(env, ios);
1098 struct cl_io *parent = lio->lis_cl.cis_io;
1099 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1100 struct lov_io_sub *sub;
1103 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1104 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1106 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1108 pdv->dv_data_version += sdv->dv_data_version;
1109 if (pdv->dv_layout_version > sdv->dv_layout_version)
1110 pdv->dv_layout_version = sdv->dv_layout_version;
1112 if (parent->ci_result == 0)
1113 parent->ci_result = sub->sub_io.ci_result;
1119 static void lov_io_iter_fini(const struct lu_env *env,
1120 const struct cl_io_slice *ios)
1122 struct lov_io *lio = cl2lov_io(env, ios);
1127 if (lio->lis_trunc_stripe_index != NULL)
1128 OBD_FREE_PTR_ARRAY(lio->lis_trunc_stripe_index,
1129 lio->lis_object->u.composite.lo_entry_count);
1130 lio->lis_trunc_stripe_index = NULL;
1132 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1134 while (!list_empty(&lio->lis_active))
1135 list_del_init(lio->lis_active.next);
1139 static void lov_io_unlock(const struct lu_env *env,
1140 const struct cl_io_slice *ios)
1145 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1150 static int lov_io_read_ahead(const struct lu_env *env,
1151 const struct cl_io_slice *ios,
1152 pgoff_t start, struct cl_read_ahead *ra)
1154 struct lov_io *lio = cl2lov_io(env, ios);
1155 struct lov_object *loo = lio->lis_object;
1156 struct cl_object *obj = lov2cl(loo);
1157 struct lov_layout_raid0 *r0;
1158 struct lov_io_sub *sub;
1162 unsigned int pps; /* pages per stripe */
1168 offset = cl_offset(obj, start);
1169 index = lov_io_layout_at(lio, offset);
1170 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1171 lsm_entry_is_foreign(loo->lo_lsm, index))
1174 /* avoid readahead to expand to stale components */
1175 if (!lov_entry(loo, index)->lle_valid)
1178 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1180 r0 = lov_r0(loo, index);
1181 if (unlikely(!r0->lo_sub[stripe]))
1184 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1186 RETURN(PTR_ERR(sub));
1188 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1189 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1190 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1193 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1194 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1200 * Adjust the stripe index by layout of comp. ra->cra_end is the
1201 * maximum page index covered by an underlying DLM lock.
1202 * This function converts cra_end from stripe level to file level, and
1203 * make sure it's not beyond stripe and component boundary.
1206 /* cra_end is stripe level, convert it into file level */
1207 ra_end = ra->cra_end_idx;
1208 if (ra_end != CL_PAGE_EOF)
1209 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1212 /* boundary of current component */
1213 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1214 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1215 ra->cra_end_idx = ra_end - 1;
1217 if (r0->lo_nr == 1) /* single stripe file */
1220 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1222 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1223 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1224 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1225 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1227 /* never exceed the end of the stripe */
1228 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1229 start + pps - start % pps - 1);
1233 int lov_io_lru_reserve(const struct lu_env *env,
1234 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1236 struct lov_io *lio = cl2lov_io(env, ios);
1237 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1238 struct lov_io_sub *sub;
1239 struct lu_extent ext;
1246 ext.e_end = pos + bytes;
1247 lov_foreach_io_layout(index, lio, &ext) {
1248 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1249 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1254 if (!lsm_entry_inited(lsm, index))
1257 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1258 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1259 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1260 index, lio->lis_mirror_index);
1264 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1265 if (!lov_stripe_intersects(lsm, index, stripe,
1266 &ext, &start, &end))
1269 if (unlikely(!r0->lo_sub[stripe]))
1272 sub = lov_sub_get(env, lio,
1273 lov_comp_index(index, stripe));
1275 return PTR_ERR(sub);
1277 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1288 * lov implementation of cl_operations::cio_submit() method. It takes a list
1289 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1290 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1293 * Major complication of this function is a need to handle memory cleansing:
1294 * cl_io_submit() is called to write out pages as a part of VM memory
1295 * reclamation, and hence it may not fail due to memory shortages (system
1296 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1297 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1298 * not-memory cleansing context), and in case of memory shortage, these
1299 * pre-allocated resources are used by lov_io_submit() under
1300 * lov_device::ld_mutex mutex.
1302 static int lov_io_submit(const struct lu_env *env,
1303 const struct cl_io_slice *ios,
1304 enum cl_req_type crt, struct cl_2queue *queue)
1306 struct cl_page_list *qin = &queue->c2_qin;
1307 struct lov_io *lio = cl2lov_io(env, ios);
1308 struct lov_io_sub *sub;
1309 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1310 struct cl_page *page = cl_page_list_first(qin);
1311 struct cl_page *tmp;
1317 if (page->cp_type == CPT_TRANSIENT)
1320 cl_page_list_init(plist);
1321 while (qin->pl_nr > 0) {
1322 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1324 page = cl_page_list_first(qin);
1325 if (lov_page_is_empty(page)) {
1326 cl_page_list_move(&queue->c2_qout, qin, page);
1329 * it could only be mirror read to get here therefore
1330 * the pages will be transient. We don't care about
1331 * the return code of cl_page_prep() at all.
1333 (void) cl_page_prep(env, ios->cis_io, page, crt);
1334 cl_page_completion(env, page, crt, 0);
1338 cl_2queue_init(cl2q);
1339 cl_page_list_move(&cl2q->c2_qin, qin, page);
1341 index = page->cp_lov_index;
1342 /* DIO is already split by stripe */
1344 cl_page_list_for_each_safe(page, tmp, qin) {
1345 /* this page is not on this stripe */
1346 if (index != page->cp_lov_index)
1349 cl_page_list_move(&cl2q->c2_qin, qin, page);
1352 cl_page_list_splice(qin, &cl2q->c2_qin);
1355 sub = lov_sub_get(env, lio, index);
1357 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1363 cl_page_list_splice(&cl2q->c2_qin, plist);
1364 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1365 cl_2queue_fini(env, cl2q);
1371 cl_page_list_splice(plist, qin);
1372 cl_page_list_fini(env, plist);
1377 static int lov_io_commit_async(const struct lu_env *env,
1378 const struct cl_io_slice *ios,
1379 struct cl_page_list *queue, int from, int to,
1382 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1383 struct lov_io *lio = cl2lov_io(env, ios);
1384 struct lov_io_sub *sub;
1385 struct cl_page *page;
1389 if (lio->lis_nr_subios == 1) {
1390 int idx = lio->lis_single_subio_index;
1392 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1394 sub = lov_sub_get(env, lio, idx);
1395 LASSERT(!IS_ERR(sub));
1396 LASSERT(sub == &lio->lis_single_subio);
1397 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1402 cl_page_list_init(plist);
1403 while (queue->pl_nr > 0) {
1407 LASSERT(plist->pl_nr == 0);
1408 page = cl_page_list_first(queue);
1409 LASSERT(!lov_page_is_empty(page));
1411 cl_page_list_move(plist, queue, page);
1413 index = page->cp_lov_index;
1414 while (queue->pl_nr > 0) {
1415 page = cl_page_list_first(queue);
1416 if (index != page->cp_lov_index)
1419 cl_page_list_move(plist, queue, page);
1422 if (queue->pl_nr > 0) /* still has more pages */
1423 stripe_to = PAGE_SIZE;
1425 sub = lov_sub_get(env, lio, index);
1427 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1428 plist, from, stripe_to, cb);
1434 if (plist->pl_nr > 0) /* short write */
1439 if (lov_comp_entry(index) !=
1440 lov_comp_entry(page->cp_lov_index))
1441 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1444 /* for error case, add the page back into the qin list */
1445 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1446 while (plist->pl_nr > 0) {
1447 /* error occurred, add the uncommitted pages back into queue */
1448 page = cl_page_list_last(plist);
1449 cl_page_list_move_head(queue, plist, page);
1455 static int lov_io_fault_start(const struct lu_env *env,
1456 const struct cl_io_slice *ios)
1458 struct cl_fault_io *fio;
1460 struct lov_io_sub *sub;
1467 fio = &ios->cis_io->u.ci_fault;
1468 lio = cl2lov_io(env, ios);
1471 * LU-14502: ft_page could be an existing cl_page associated with
1472 * the vmpage covering the fault index, and the page may still
1473 * refer to another mirror of an old IO.
1475 if (lov_is_flr(lio->lis_object)) {
1476 offset = cl_offset(ios->cis_obj, fio->ft_index);
1477 entry = lov_io_layout_at(lio, offset);
1479 CERROR(DFID": page fault index %lu invalid component: "
1481 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1482 fio->ft_index, entry,
1483 lio->lis_mirror_index);
1486 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1489 if (fio->ft_page->cp_lov_index !=
1490 lov_comp_index(entry, stripe)) {
1491 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1492 "at mirror %u comp entry %u stripe %u, "
1493 "been used with comp entry %u stripe %u\n",
1494 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1495 fio->ft_index, lio->lis_mirror_index,
1497 lov_comp_entry(fio->ft_page->cp_lov_index),
1498 lov_comp_stripe(fio->ft_page->cp_lov_index));
1500 fio->ft_page->cp_lov_index =
1501 lov_comp_index(entry, stripe);
1505 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1506 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1508 RETURN(lov_io_start(env, ios));
1511 static int lov_io_setattr_start(const struct lu_env *env,
1512 const struct cl_io_slice *ios)
1514 struct lov_io *lio = cl2lov_io(env, ios);
1515 struct cl_io *parent = ios->cis_io;
1516 struct lov_io_sub *sub;
1517 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1521 if (cl_io_is_fallocate(parent)) {
1522 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1523 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1524 int index = lov_comp_entry(sub->sub_subio_index);
1525 int stripe = lov_comp_stripe(sub->sub_subio_index);
1527 size = lov_size_to_stripe(lsm, index, size, stripe);
1528 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1529 sub->sub_io.u.ci_setattr.sa_avalid =
1530 parent->u.ci_setattr.sa_avalid;
1534 RETURN(lov_io_start(env, ios));
1537 static void lov_io_fsync_end(const struct lu_env *env,
1538 const struct cl_io_slice *ios)
1540 struct lov_io *lio = cl2lov_io(env, ios);
1541 struct lov_io_sub *sub;
1542 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1546 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1547 struct cl_io *subio = &sub->sub_io;
1549 lov_io_end_wrapper(sub->sub_env, subio);
1551 if (subio->ci_result == 0)
1552 *written += subio->u.ci_fsync.fi_nr_written;
1557 static void lov_io_lseek_end(const struct lu_env *env,
1558 const struct cl_io_slice *ios)
1560 struct lov_io *lio = cl2lov_io(env, ios);
1561 struct cl_io *io = lio->lis_cl.cis_io;
1562 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1563 struct lov_io_sub *sub;
1564 loff_t offset = -ENXIO;
1566 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1570 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1571 struct cl_io *subio = &sub->sub_io;
1572 int index = lov_comp_entry(sub->sub_subio_index);
1573 int stripe = lov_comp_stripe(sub->sub_subio_index);
1574 loff_t sub_off, lov_off;
1575 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1577 lov_io_end_wrapper(sub->sub_env, subio);
1579 if (io->ci_result == 0)
1580 io->ci_result = sub->sub_io.ci_result;
1585 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1586 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1587 index, stripe, seek_hole ? "HOLE" : "DATA",
1588 subio->u.ci_lseek.ls_start);
1590 /* first subio with positive result is what we need */
1591 sub_off = subio->u.ci_lseek.ls_result;
1592 /* Expected error, offset is out of stripe file size */
1593 if (sub_off == -ENXIO)
1595 /* Any other errors are not expected with ci_result == 0 */
1597 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1599 io->ci_result = sub_off;
1602 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1604 /* the only way to get negatove lov_off here is too big
1605 * result. Return -EOVERFLOW then.
1607 io->ci_result = -EOVERFLOW;
1608 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1609 (u64)lov_off, io->ci_result);
1612 if (lov_off < io->u.ci_lseek.ls_start) {
1613 io->ci_result = -EINVAL;
1614 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1615 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1618 /* resulting offset can be out of component range if stripe
1619 * object is full and its file size was returned as virtual
1620 * hole start. Skip this result, the next component will give
1621 * us correct lseek result but keep possible hole offset in
1622 * case there is no more components ahead
1624 if (lov_off >= comp_end) {
1625 /* must be SEEK_HOLE case */
1626 if (likely(seek_hole)) {
1627 /* save comp end as potential hole offset */
1628 hole_off = max_t(__u64, comp_end, hole_off);
1630 io->ci_result = -EINVAL;
1632 "off %lld >= comp_end %llu: rc = %d\n",
1633 lov_off, comp_end, io->ci_result);
1638 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1639 seek_hole ? "HOLE" : "DATA",
1640 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1641 sub->sub_io.ci_result);
1642 offset = min_t(__u64, offset, lov_off);
1644 /* no result but some component returns hole as component end */
1645 if (seek_hole && offset == -ENXIO && hole_off > 0)
1648 io->u.ci_lseek.ls_result = offset;
1652 static const struct cl_io_operations lov_io_ops = {
1655 .cio_fini = lov_io_fini,
1656 .cio_iter_init = lov_io_rw_iter_init,
1657 .cio_iter_fini = lov_io_iter_fini,
1658 .cio_lock = lov_io_lock,
1659 .cio_unlock = lov_io_unlock,
1660 .cio_start = lov_io_start,
1661 .cio_end = lov_io_end
1664 .cio_fini = lov_io_fini,
1665 .cio_iter_init = lov_io_rw_iter_init,
1666 .cio_iter_fini = lov_io_iter_fini,
1667 .cio_lock = lov_io_lock,
1668 .cio_unlock = lov_io_unlock,
1669 .cio_start = lov_io_start,
1670 .cio_end = lov_io_end
1673 .cio_fini = lov_io_fini,
1674 .cio_iter_init = lov_io_setattr_iter_init,
1675 .cio_iter_fini = lov_io_iter_fini,
1676 .cio_lock = lov_io_lock,
1677 .cio_unlock = lov_io_unlock,
1678 .cio_start = lov_io_setattr_start,
1679 .cio_end = lov_io_end
1681 [CIT_DATA_VERSION] = {
1682 .cio_fini = lov_io_fini,
1683 .cio_iter_init = lov_io_iter_init,
1684 .cio_iter_fini = lov_io_iter_fini,
1685 .cio_lock = lov_io_lock,
1686 .cio_unlock = lov_io_unlock,
1687 .cio_start = lov_io_start,
1688 .cio_end = lov_io_data_version_end,
1691 .cio_fini = lov_io_fini,
1692 .cio_iter_init = lov_io_iter_init,
1693 .cio_iter_fini = lov_io_iter_fini,
1694 .cio_lock = lov_io_lock,
1695 .cio_unlock = lov_io_unlock,
1696 .cio_start = lov_io_fault_start,
1697 .cio_end = lov_io_end
1700 .cio_fini = lov_io_fini,
1701 .cio_iter_init = lov_io_iter_init,
1702 .cio_iter_fini = lov_io_iter_fini,
1703 .cio_lock = lov_io_lock,
1704 .cio_unlock = lov_io_unlock,
1705 .cio_start = lov_io_start,
1706 .cio_end = lov_io_fsync_end
1709 .cio_fini = lov_io_fini,
1710 .cio_iter_init = lov_io_iter_init,
1711 .cio_iter_fini = lov_io_iter_fini,
1712 .cio_lock = lov_io_lock,
1713 .cio_unlock = lov_io_unlock,
1714 .cio_start = lov_io_start,
1715 .cio_end = lov_io_end
1718 .cio_fini = lov_io_fini,
1719 .cio_iter_init = lov_io_iter_init,
1720 .cio_iter_fini = lov_io_iter_fini,
1721 .cio_lock = lov_io_lock,
1722 .cio_unlock = lov_io_unlock,
1723 .cio_start = lov_io_start,
1724 .cio_end = lov_io_lseek_end
1727 .cio_fini = lov_io_fini,
1730 .cio_fini = lov_io_fini
1733 .cio_read_ahead = lov_io_read_ahead,
1734 .cio_lru_reserve = lov_io_lru_reserve,
1735 .cio_submit = lov_io_submit,
1736 .cio_commit_async = lov_io_commit_async,
1739 /*****************************************************************************
1741 * Empty lov io operations.
1745 static void lov_empty_io_fini(const struct lu_env *env,
1746 const struct cl_io_slice *ios)
1748 struct lov_object *lov = cl2lov(ios->cis_obj);
1751 if (atomic_dec_and_test(&lov->lo_active_ios))
1752 wake_up(&lov->lo_waitq);
1756 static int lov_empty_io_submit(const struct lu_env *env,
1757 const struct cl_io_slice *ios,
1758 enum cl_req_type crt, struct cl_2queue *queue)
1763 static void lov_empty_impossible(const struct lu_env *env,
1764 struct cl_io_slice *ios)
1769 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1772 * An io operation vector for files without stripes.
1774 static const struct cl_io_operations lov_empty_io_ops = {
1777 .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
1786 .cio_fini = lov_empty_io_fini,
1787 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1788 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1789 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1790 .cio_end = LOV_EMPTY_IMPOSSIBLE
1793 .cio_fini = lov_empty_io_fini,
1794 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1795 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1796 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1797 .cio_end = LOV_EMPTY_IMPOSSIBLE
1800 .cio_fini = lov_empty_io_fini,
1801 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1802 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1803 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1804 .cio_end = LOV_EMPTY_IMPOSSIBLE
1807 .cio_fini = lov_empty_io_fini
1810 .cio_fini = lov_empty_io_fini
1813 .cio_fini = lov_empty_io_fini
1816 .cio_fini = lov_empty_io_fini
1819 .cio_submit = lov_empty_io_submit,
1820 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1823 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1826 struct lov_io *lio = lov_env_io(env);
1827 struct lov_object *lov = cl2lov(obj);
1832 INIT_LIST_HEAD(&lio->lis_active);
1833 result = lov_io_slice_init(lio, lov, io);
1837 result = lov_io_subio_init(env, lio, io);
1839 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1840 atomic_inc(&lov->lo_active_ios);
1844 io->ci_result = result < 0 ? result : 0;
1848 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1851 struct lov_object *lov = cl2lov(obj);
1852 struct lov_io *lio = lov_env_io(env);
1856 lio->lis_object = lov;
1857 switch (io->ci_type) {
1869 case CIT_DATA_VERSION:
1877 CERROR("Page fault on a file without stripes: "DFID"\n",
1878 PFID(lu_object_fid(&obj->co_lu)));
1882 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1883 atomic_inc(&lov->lo_active_ios);
1886 io->ci_result = result < 0 ? result : 0;
1890 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1893 struct lov_object *lov = cl2lov(obj);
1894 struct lov_io *lio = lov_env_io(env);
1898 LASSERT(lov->lo_lsm != NULL);
1899 lio->lis_object = lov;
1901 switch (io->ci_type) {
1903 LASSERTF(0, "invalid type %d\n", io->ci_type);
1904 result = -EOPNOTSUPP;
1910 case CIT_DATA_VERSION:
1915 * the truncate to 0 is managed by MDT:
1916 * - in open, for open O_TRUNC
1917 * - in setattr, for truncate
1920 * the truncate is for size > 0 so triggers a restore,
1921 * also trigger a restore for prealloc/punch
1923 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1924 io->ci_restore_needed = 1;
1933 io->ci_restore_needed = 1;
1939 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1940 atomic_inc(&lov->lo_active_ios);
1943 io->ci_result = result < 0 ? result : 0;
1948 * Return the index in composite:lo_entries by the file offset
1950 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1952 struct lov_object *lov = lio->lis_object;
1953 struct lov_layout_composite *comp = &lov->u.composite;
1954 int start_index = 0;
1955 int end_index = comp->lo_entry_count - 1;
1958 LASSERT(lov->lo_type == LLT_COMP);
1960 /* This is actual file offset so nothing can cover eof. */
1961 if (offset == LUSTRE_EOF)
1964 if (lov_is_flr(lov)) {
1965 struct lov_mirror_entry *lre;
1967 LASSERT(lio->lis_mirror_index >= 0);
1969 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1970 start_index = lre->lre_start;
1971 end_index = lre->lre_end;
1974 for (i = start_index; i <= end_index; i++) {
1975 struct lov_layout_entry *lle = lov_entry(lov, i);
1977 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1979 if ((offset >= lle->lle_extent->e_start &&
1980 offset < lle->lle_extent->e_end) ||
1981 (offset == OBD_OBJECT_EOF &&
1982 lle->lle_extent->e_end == OBD_OBJECT_EOF))