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) ||
231 cl_io_is_fault_writable(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 LASSERT(!primary->lre_stale);
259 lov_foreach_mirror_layout_entry(obj, lle, primary) {
260 LASSERT(lle->lle_valid);
261 if (!lu_extent_is_overlapped(ext, lle->lle_extent))
264 ext->e_start = min(ext->e_start, lle->lle_extent->e_start);
265 ext->e_end = max(ext->e_end, lle->lle_extent->e_end);
269 CERROR(DFID ": cannot find any valid components covering "
270 "file extent "DEXT", mirror: %d\n",
271 PFID(lu_object_fid(lov2lu(obj))), PEXT(ext),
272 primary->lre_mirror_id);
277 lov_foreach_mirror_entry(obj, lre) {
281 lov_foreach_mirror_layout_entry(obj, lle, lre) {
285 if (lu_extent_is_overlapped(ext, lle->lle_extent)) {
292 CDEBUG(D_VFSTRACE, DFID "there are %zd components to be staled to "
293 "modify file extent "DEXT", iot: %d\n",
294 PFID(lu_object_fid(lov2lu(obj))), count, PEXT(ext), io->ci_type);
296 io->ci_need_write_intent = count > 0;
301 static int lov_io_mirror_init(struct lov_io *lio, struct lov_object *obj,
304 struct lov_layout_composite *comp = &obj->u.composite;
310 if (!lov_is_flr(obj)) {
311 /* only locks/pages are manipulated for CIT_MISC op, no
312 * cl_io_loop() will be called, don't check/set mirror info.
314 if (io->ci_type != CIT_MISC) {
315 LASSERT(comp->lo_preferred_mirror == 0);
316 lio->lis_mirror_index = comp->lo_preferred_mirror;
322 /* transfer the layout version for verification */
323 if (io->ci_layout_version == 0)
324 io->ci_layout_version = obj->lo_lsm->lsm_layout_gen;
326 /* find the corresponding mirror for designated mirror IO */
327 if (io->ci_designated_mirror > 0) {
328 struct lov_mirror_entry *entry;
330 LASSERT(!io->ci_ndelay);
332 CDEBUG(D_LAYOUT, "designated I/O mirror state: %d\n",
335 if ((cl_io_is_trunc(io) || io->ci_type == CIT_WRITE) &&
336 (io->ci_layout_version != obj->lo_lsm->lsm_layout_gen)) {
338 * For resync I/O, the ci_layout_version was the layout
339 * version when resync starts. If it doesn't match the
340 * current object layout version, it means the layout
346 io->ci_layout_version |= LU_LAYOUT_RESYNC;
349 lio->lis_mirror_index = -1;
350 lov_foreach_mirror_entry(obj, entry) {
351 if (entry->lre_mirror_id ==
352 io->ci_designated_mirror) {
353 lio->lis_mirror_index = index;
360 RETURN(lio->lis_mirror_index < 0 ? -EINVAL : 0);
363 result = lov_io_mirror_write_intent(lio, obj, io);
367 if (io->ci_need_write_intent) {
368 CDEBUG(D_VFSTRACE, DFID " need write intent for [%llu, %llu)\n",
369 PFID(lu_object_fid(lov2lu(obj))),
370 lio->lis_pos, lio->lis_endpos);
372 if (cl_io_is_trunc(io)) {
374 * for truncate, we uses [size, EOF) to judge whether
375 * a write intent needs to be send, but we need to
376 * restore the write extent to [0, size], in truncate,
377 * the byte in the size position is accessed.
379 io->ci_write_intent.e_start = 0;
380 io->ci_write_intent.e_end =
381 io->u.ci_setattr.sa_attr.lvb_size + 1;
383 /* stop cl_io_init() loop */
387 if (io->ci_ndelay_tried == 0 || /* first time to try */
388 /* reset the mirror index if layout has changed */
389 lio->lis_mirror_layout_gen != obj->lo_lsm->lsm_layout_gen) {
390 lio->lis_mirror_layout_gen = obj->lo_lsm->lsm_layout_gen;
391 index = lio->lis_mirror_index = comp->lo_preferred_mirror;
393 index = lio->lis_mirror_index;
396 /* move mirror index to the next one */
397 index = (index + 1) % comp->lo_mirror_count;
400 for (i = 0; i < comp->lo_mirror_count; i++) {
401 struct lu_extent ext = { .e_start = lio->lis_pos,
402 .e_end = lio->lis_pos + 1 };
403 struct lov_mirror_entry *lre;
404 struct lov_layout_entry *lle;
407 lre = &comp->lo_mirrors[(index + i) % comp->lo_mirror_count];
411 if (lre->lre_foreign)
414 lov_foreach_mirror_layout_entry(obj, lle, lre) {
418 if (lu_extent_is_overlapped(&ext, lle->lle_extent)) {
422 } /* each component of the mirror */
424 index = (index + i) % comp->lo_mirror_count;
429 if (i == comp->lo_mirror_count) {
430 CERROR(DFID": failed to find a component covering "
431 "I/O region at %llu\n",
432 PFID(lu_object_fid(lov2lu(obj))), lio->lis_pos);
434 dump_lsm(D_ERROR, obj->lo_lsm);
439 CDEBUG(D_VFSTRACE, DFID ": flr state: %d, move mirror from %d to %d, "
440 "have retried: %d, mirror count: %d\n",
441 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj),
442 lio->lis_mirror_index, index, io->ci_ndelay_tried,
443 comp->lo_mirror_count);
445 lio->lis_mirror_index = index;
448 * FLR: if all mirrors have been tried once, most likely the network
449 * of this client has been partitioned. We should relinquish CPU for
450 * a while before trying again.
452 if (io->ci_ndelay && io->ci_ndelay_tried > 0 &&
453 (io->ci_ndelay_tried % comp->lo_mirror_count == 0)) {
454 schedule_timeout_interruptible(cfs_time_seconds(1) / 100);
455 if (signal_pending(current))
459 * we'd set ci_tried_all_mirrors to turn off fast mirror
460 * switching for read after we've tried all mirrors several
463 io->ci_tried_all_mirrors = io->ci_ndelay_tried %
464 (comp->lo_mirror_count * 4) == 0;
466 ++io->ci_ndelay_tried;
468 CDEBUG(D_VFSTRACE, "use %sdelayed RPC state for this IO\n",
469 io->ci_ndelay ? "non-" : "");
474 static int lov_io_slice_init(struct lov_io *lio,
475 struct lov_object *obj, struct cl_io *io)
482 lio->lis_object = obj;
484 switch (io->ci_type) {
487 lio->lis_pos = io->u.ci_rw.crw_pos;
488 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
489 lio->lis_io_endpos = lio->lis_endpos;
490 if (cl_io_is_append(io)) {
491 LASSERT(io->ci_type == CIT_WRITE);
494 * If there is LOV EA hole, then we may cannot locate
495 * the current file-tail exactly.
497 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
499 GOTO(out, result = -EIO);
502 lio->lis_endpos = OBD_OBJECT_EOF;
507 if (cl_io_is_fallocate(io)) {
508 lio->lis_pos = io->u.ci_setattr.sa_falloc_offset;
509 lio->lis_endpos = io->u.ci_setattr.sa_falloc_end;
510 } else if (cl_io_is_trunc(io)) {
511 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
512 lio->lis_endpos = OBD_OBJECT_EOF;
515 lio->lis_endpos = OBD_OBJECT_EOF;
519 case CIT_DATA_VERSION:
521 lio->lis_endpos = OBD_OBJECT_EOF;
525 pgoff_t index = io->u.ci_fault.ft_index;
527 lio->lis_pos = cl_offset(io->ci_obj, index);
528 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
533 lio->lis_pos = io->u.ci_fsync.fi_start;
534 lio->lis_endpos = io->u.ci_fsync.fi_end;
539 lio->lis_pos = io->u.ci_ladvise.li_start;
540 lio->lis_endpos = io->u.ci_ladvise.li_end;
545 lio->lis_pos = io->u.ci_lseek.ls_start;
546 lio->lis_endpos = OBD_OBJECT_EOF;
552 lio->lis_endpos = OBD_OBJECT_EOF;
554 if (lov_flr_state(obj) == LCM_FL_RDONLY &&
555 !OBD_FAIL_CHECK(OBD_FAIL_FLR_GLIMPSE_IMMUTABLE))
556 /* SoM is accurate, no need glimpse */
557 GOTO(out, result = 1);
562 lio->lis_endpos = OBD_OBJECT_EOF;
570 * CIT_MISC + ci_ignore_layout can identify the I/O from the OSC layer,
571 * it won't care/access lov layout related info.
573 if (io->ci_ignore_layout && io->ci_type == CIT_MISC)
574 GOTO(out, result = 0);
576 LASSERT(obj->lo_lsm != NULL);
578 result = lov_io_mirror_init(lio, obj, io);
582 /* check if it needs to instantiate layout */
583 if (!(io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io) ||
584 cl_io_is_fallocate(io) ||
585 (cl_io_is_trunc(io) && io->u.ci_setattr.sa_attr.lvb_size > 0)) ||
586 cl_io_is_fault_writable(io))
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);
648 LASSERT(list_empty(&lio->lis_active));
650 while (!list_empty(&lio->lis_subios)) {
651 struct lov_io_sub *sub = list_entry(lio->lis_subios.next,
655 list_del_init(&sub->sub_list);
656 lio->lis_nr_subios--;
658 lov_io_sub_fini(env, lio, sub);
659 lov_sub_free(lio, sub);
661 LASSERT(lio->lis_nr_subios == 0);
663 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
664 if (atomic_dec_and_test(&lov->lo_active_ios))
665 wake_up(&lov->lo_waitq);
669 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
670 loff_t start, loff_t end)
672 struct cl_io *io = &sub->sub_io;
673 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
674 struct cl_io *parent = lio->lis_cl.cis_io;
675 int index = lov_comp_entry(sub->sub_subio_index);
676 int stripe = lov_comp_stripe(sub->sub_subio_index);
678 switch (io->ci_type) {
680 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
681 io->u.ci_setattr.sa_attr_flags =
682 parent->u.ci_setattr.sa_attr_flags;
683 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
684 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
685 io->u.ci_setattr.sa_falloc_mode =
686 parent->u.ci_setattr.sa_falloc_mode;
687 io->u.ci_setattr.sa_stripe_index = stripe;
688 io->u.ci_setattr.sa_parent_fid =
689 parent->u.ci_setattr.sa_parent_fid;
690 /* For SETATTR(fallocate) pass the subtype to lower IO */
691 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
692 if (cl_io_is_fallocate(io)) {
693 io->u.ci_setattr.sa_falloc_offset = start;
694 io->u.ci_setattr.sa_falloc_end = end;
696 if (cl_io_is_trunc(io)) {
697 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
699 new_size = lov_size_to_stripe(lsm, index, new_size,
701 io->u.ci_setattr.sa_attr.lvb_size = new_size;
703 lov_lsm2layout(lsm, lsm->lsm_entries[index],
704 &io->u.ci_setattr.sa_layout);
707 case CIT_DATA_VERSION: {
708 io->u.ci_data_version.dv_data_version = 0;
709 io->u.ci_data_version.dv_flags =
710 parent->u.ci_data_version.dv_flags;
714 struct cl_object *obj = parent->ci_obj;
715 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
717 io->u.ci_fault = parent->u.ci_fault;
718 off = lov_size_to_stripe(lsm, index, off, stripe);
719 io->u.ci_fault.ft_index = cl_index(obj, off);
723 io->u.ci_fsync.fi_start = start;
724 io->u.ci_fsync.fi_end = end;
725 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
726 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
731 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
732 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
733 if (cl_io_is_append(parent)) {
734 io->u.ci_wr.wr_append = 1;
736 io->u.ci_rw.crw_pos = start;
737 io->u.ci_rw.crw_count = end - start;
742 io->u.ci_ladvise.li_start = start;
743 io->u.ci_ladvise.li_end = end;
744 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
745 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
746 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
750 io->u.ci_lseek.ls_start = start;
751 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
752 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
762 static loff_t lov_offset_mod(loff_t val, int delta)
764 if (val != OBD_OBJECT_EOF)
769 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
770 struct lov_io_sub *sub, u64 start, u64 end)
774 end = lov_offset_mod(end, 1);
775 lov_io_sub_inherit(sub, lio, start, end);
776 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
778 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
782 list_add_tail(&sub->sub_linkage, &lio->lis_active);
786 static int lov_io_iter_init(const struct lu_env *env,
787 const struct cl_io_slice *ios)
789 struct lov_io *lio = cl2lov_io(env, ios);
790 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
791 struct lov_io_sub *sub;
792 struct lu_extent ext;
798 ext.e_start = lio->lis_pos;
799 ext.e_end = lio->lis_endpos;
801 lov_foreach_io_layout(index, lio, &ext) {
802 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
803 struct lov_layout_raid0 *r0 = &le->lle_raid0;
807 bool tested_trunc_stripe = false;
809 r0->lo_trunc_stripeno = -1;
811 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
812 index, lsm->lsm_entries[index]->lsme_flags);
813 if (!lsm_entry_inited(lsm, index)) {
815 * Read from uninitialized components should return
821 if (lsm_entry_is_foreign(lsm, index))
824 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
825 CERROR("I/O to invalid component: %d, mirror: %d\n",
826 index, lio->lis_mirror_index);
830 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
831 if (!lov_stripe_intersects(lsm, index, stripe,
835 if (unlikely(!r0->lo_sub[stripe])) {
836 if (ios->cis_io->ci_type == CIT_READ ||
837 ios->cis_io->ci_type == CIT_WRITE ||
838 ios->cis_io->ci_type == CIT_FAULT)
844 if (cl_io_is_trunc(ios->cis_io) &&
845 !tested_trunc_stripe) {
849 prev = (stripe == 0) ? r0->lo_nr - 1 :
852 * Only involving previous stripe if the
853 * truncate in this component is at the
854 * beginning of this stripe.
856 tested_trunc_stripe = true;
857 if (ext.e_start < lsm->lsm_entries[index]->
858 lsme_extent.e_start) {
859 /* need previous stripe involvement */
860 r0->lo_trunc_stripeno = prev;
862 tr_start = ext.e_start;
863 tr_start = lov_do_div64(tr_start,
864 stripe_width(lsm, index));
865 /* tr_start %= stripe_swidth */
866 if (tr_start == stripe * lsm->
869 r0->lo_trunc_stripeno = prev;
873 /* if the last stripe is the trunc stripeno */
874 if (r0->lo_trunc_stripeno == stripe)
875 r0->lo_trunc_stripeno = -1;
877 sub = lov_sub_get(env, lio,
878 lov_comp_index(index, stripe));
882 rc = lov_io_add_sub(env, lio, sub, start, end);
889 if (r0->lo_trunc_stripeno != -1) {
890 stripe = r0->lo_trunc_stripeno;
891 if (unlikely(!r0->lo_sub[stripe])) {
892 r0->lo_trunc_stripeno = -1;
895 sub = lov_sub_get(env, lio,
896 lov_comp_index(index, stripe));
901 * the prev sub could be used by another truncate, we'd
902 * skip it. LU-14128 happends when expand truncate +
903 * read get wrong kms.
905 if (!list_empty(&sub->sub_linkage)) {
906 r0->lo_trunc_stripeno = -1;
910 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
912 rc = lov_io_add_sub(env, lio, sub, start, end);
921 static int lov_io_rw_iter_init(const struct lu_env *env,
922 const struct cl_io_slice *ios)
924 struct lov_io *lio = cl2lov_io(env, ios);
925 struct cl_io *io = ios->cis_io;
926 struct lov_stripe_md_entry *lse;
927 loff_t start = io->u.ci_rw.crw_pos;
931 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
934 if (cl_io_is_append(io))
935 RETURN(lov_io_iter_init(env, ios));
937 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
938 if (index < 0) { /* non-existing layout component */
939 if (io->ci_type == CIT_READ) {
941 * TODO: it needs to detect the next component and
942 * then set the next pos
946 RETURN(lov_io_iter_init(env, ios));
952 if (!lov_entry(lio->lis_object, index)->lle_valid &&
953 !io->ci_designated_mirror)
954 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
956 lse = lov_lse(lio->lis_object, index);
958 if (lsme_is_foreign(lse))
961 next = MAX_LFS_FILESIZE;
962 if (lse->lsme_stripe_count > 1) {
963 unsigned long ssize = lse->lsme_stripe_size;
965 lov_do_div64(start, ssize);
966 next = (start + 1) * ssize;
967 if (next <= start * ssize)
968 next = MAX_LFS_FILESIZE;
971 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
972 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
973 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
974 next = min_t(__u64, next, lse->lsme_extent.e_end);
975 next = min_t(loff_t, next, lio->lis_io_endpos);
977 io->ci_continue = next < lio->lis_io_endpos;
978 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
979 lio->lis_pos = io->u.ci_rw.crw_pos;
980 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
982 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
983 (__u64)start, lio->lis_pos, lio->lis_endpos,
984 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
987 * XXX The following call should be optimized: we know, that
988 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
990 RETURN(lov_io_iter_init(env, ios));
993 static int lov_io_setattr_iter_init(const struct lu_env *env,
994 const struct cl_io_slice *ios)
996 struct lov_io *lio = cl2lov_io(env, ios);
997 struct cl_io *io = ios->cis_io;
1001 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1002 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1003 /* no entry found for such offset */
1005 RETURN(io->ci_result = -ENODATA);
1008 RETURN(lov_io_iter_init(env, ios));
1011 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1012 int (*iofunc)(const struct lu_env *, struct cl_io *))
1014 struct cl_io *parent = lio->lis_cl.cis_io;
1015 struct lov_io_sub *sub;
1019 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1020 rc = iofunc(sub->sub_env, &sub->sub_io);
1024 if (parent->ci_result == 0)
1025 parent->ci_result = sub->sub_io.ci_result;
1030 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1033 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1036 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1039 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1042 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1046 * It's possible that lov_io_start() wasn't called against this
1047 * sub-io, either because previous sub-io failed, or upper layer
1050 if (io->ci_state == CIS_IO_GOING)
1053 io->ci_state = CIS_IO_FINISHED;
1057 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1059 cl_io_iter_fini(env, io);
1063 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1065 cl_io_unlock(env, io);
1069 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1073 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1078 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1080 struct lov_io *lio = cl2lov_io(env, ios);
1081 struct cl_io *parent = lio->lis_cl.cis_io;
1082 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1083 struct lov_io_sub *sub;
1086 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1087 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1089 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1091 pdv->dv_data_version += sdv->dv_data_version;
1092 if (pdv->dv_layout_version > sdv->dv_layout_version)
1093 pdv->dv_layout_version = sdv->dv_layout_version;
1095 if (parent->ci_result == 0)
1096 parent->ci_result = sub->sub_io.ci_result;
1102 static void lov_io_iter_fini(const struct lu_env *env,
1103 const struct cl_io_slice *ios)
1105 struct lov_io *lio = cl2lov_io(env, ios);
1109 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1111 while (!list_empty(&lio->lis_active))
1112 list_del_init(lio->lis_active.next);
1116 static void lov_io_unlock(const struct lu_env *env,
1117 const struct cl_io_slice *ios)
1122 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1127 static int lov_io_read_ahead(const struct lu_env *env,
1128 const struct cl_io_slice *ios,
1129 pgoff_t start, struct cl_read_ahead *ra)
1131 struct lov_io *lio = cl2lov_io(env, ios);
1132 struct lov_object *loo = lio->lis_object;
1133 struct cl_object *obj = lov2cl(loo);
1134 struct lov_layout_raid0 *r0;
1135 struct lov_io_sub *sub;
1139 unsigned int pps; /* pages per stripe */
1145 offset = cl_offset(obj, start);
1146 index = lov_io_layout_at(lio, offset);
1147 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1148 lsm_entry_is_foreign(loo->lo_lsm, index))
1151 /* avoid readahead to expand to stale components */
1152 if (!lov_entry(loo, index)->lle_valid)
1155 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1157 r0 = lov_r0(loo, index);
1158 if (unlikely(!r0->lo_sub[stripe]))
1161 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1163 RETURN(PTR_ERR(sub));
1165 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1166 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1167 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1170 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1171 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1177 * Adjust the stripe index by layout of comp. ra->cra_end is the
1178 * maximum page index covered by an underlying DLM lock.
1179 * This function converts cra_end from stripe level to file level, and
1180 * make sure it's not beyond stripe and component boundary.
1183 /* cra_end is stripe level, convert it into file level */
1184 ra_end = ra->cra_end_idx;
1185 if (ra_end != CL_PAGE_EOF)
1186 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1189 /* boundary of current component */
1190 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1191 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1192 ra->cra_end_idx = ra_end - 1;
1194 if (r0->lo_nr == 1) /* single stripe file */
1197 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1199 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1200 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1201 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1202 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1204 /* never exceed the end of the stripe */
1205 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1206 start + pps - start % pps - 1);
1210 int lov_io_lru_reserve(const struct lu_env *env,
1211 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1213 struct lov_io *lio = cl2lov_io(env, ios);
1214 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1215 struct lov_io_sub *sub;
1216 struct lu_extent ext;
1223 ext.e_end = pos + bytes;
1224 lov_foreach_io_layout(index, lio, &ext) {
1225 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1226 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1231 if (!lsm_entry_inited(lsm, index))
1234 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1235 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1236 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1237 index, lio->lis_mirror_index);
1241 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1242 if (!lov_stripe_intersects(lsm, index, stripe,
1243 &ext, &start, &end))
1246 if (unlikely(!r0->lo_sub[stripe]))
1249 sub = lov_sub_get(env, lio,
1250 lov_comp_index(index, stripe));
1252 return PTR_ERR(sub);
1254 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1265 * lov implementation of cl_operations::cio_submit() method. It takes a list
1266 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1267 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1270 * Major complication of this function is a need to handle memory cleansing:
1271 * cl_io_submit() is called to write out pages as a part of VM memory
1272 * reclamation, and hence it may not fail due to memory shortages (system
1273 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1274 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1275 * not-memory cleansing context), and in case of memory shortage, these
1276 * pre-allocated resources are used by lov_io_submit() under
1277 * lov_device::ld_mutex mutex.
1279 static int lov_io_submit(const struct lu_env *env,
1280 const struct cl_io_slice *ios,
1281 enum cl_req_type crt, struct cl_2queue *queue)
1283 struct cl_page_list *qin = &queue->c2_qin;
1284 struct lov_io *lio = cl2lov_io(env, ios);
1285 struct lov_io_sub *sub;
1286 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1287 struct cl_page *page;
1288 struct cl_page *tmp;
1293 cl_page_list_init(plist);
1294 while (qin->pl_nr > 0) {
1295 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1297 page = cl_page_list_first(qin);
1298 if (lov_page_is_empty(page)) {
1299 cl_page_list_move(&queue->c2_qout, qin, page);
1302 * it could only be mirror read to get here therefore
1303 * the pages will be transient. We don't care about
1304 * the return code of cl_page_prep() at all.
1306 (void) cl_page_prep(env, ios->cis_io, page, crt);
1307 cl_page_completion(env, page, crt, 0);
1311 cl_2queue_init(cl2q);
1312 cl_page_list_move(&cl2q->c2_qin, qin, page);
1314 index = page->cp_lov_index;
1315 cl_page_list_for_each_safe(page, tmp, qin) {
1316 /* this page is not on this stripe */
1317 if (index != page->cp_lov_index)
1320 cl_page_list_move(&cl2q->c2_qin, qin, page);
1323 sub = lov_sub_get(env, lio, index);
1325 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1331 cl_page_list_splice(&cl2q->c2_qin, plist);
1332 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1333 cl_2queue_fini(env, cl2q);
1339 cl_page_list_splice(plist, qin);
1340 cl_page_list_fini(env, plist);
1345 static int lov_io_commit_async(const struct lu_env *env,
1346 const struct cl_io_slice *ios,
1347 struct cl_page_list *queue, int from, int to,
1350 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1351 struct lov_io *lio = cl2lov_io(env, ios);
1352 struct lov_io_sub *sub;
1353 struct cl_page *page;
1357 if (lio->lis_nr_subios == 1) {
1358 int idx = lio->lis_single_subio_index;
1360 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1362 sub = lov_sub_get(env, lio, idx);
1363 LASSERT(!IS_ERR(sub));
1364 LASSERT(sub == &lio->lis_single_subio);
1365 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1370 cl_page_list_init(plist);
1371 while (queue->pl_nr > 0) {
1375 LASSERT(plist->pl_nr == 0);
1376 page = cl_page_list_first(queue);
1377 LASSERT(!lov_page_is_empty(page));
1379 cl_page_list_move(plist, queue, page);
1381 index = page->cp_lov_index;
1382 while (queue->pl_nr > 0) {
1383 page = cl_page_list_first(queue);
1384 if (index != page->cp_lov_index)
1387 cl_page_list_move(plist, queue, page);
1390 if (queue->pl_nr > 0) /* still has more pages */
1391 stripe_to = PAGE_SIZE;
1393 sub = lov_sub_get(env, lio, index);
1395 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1396 plist, from, stripe_to, cb);
1402 if (plist->pl_nr > 0) /* short write */
1407 if (lov_comp_entry(index) !=
1408 lov_comp_entry(page->cp_lov_index))
1409 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1412 /* for error case, add the page back into the qin list */
1413 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1414 while (plist->pl_nr > 0) {
1415 /* error occurred, add the uncommitted pages back into queue */
1416 page = cl_page_list_last(plist);
1417 cl_page_list_move_head(queue, plist, page);
1423 static int lov_io_fault_start(const struct lu_env *env,
1424 const struct cl_io_slice *ios)
1426 struct cl_fault_io *fio;
1428 struct lov_io_sub *sub;
1435 fio = &ios->cis_io->u.ci_fault;
1436 lio = cl2lov_io(env, ios);
1439 * LU-14502: ft_page could be an existing cl_page associated with
1440 * the vmpage covering the fault index, and the page may still
1441 * refer to another mirror of an old IO.
1443 if (lov_is_flr(lio->lis_object)) {
1444 offset = cl_offset(ios->cis_obj, fio->ft_index);
1445 entry = lov_io_layout_at(lio, offset);
1447 CERROR(DFID": page fault index %lu invalid component: "
1449 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1450 fio->ft_index, entry,
1451 lio->lis_mirror_index);
1454 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1457 if (fio->ft_page->cp_lov_index !=
1458 lov_comp_index(entry, stripe)) {
1459 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1460 "at mirror %u comp entry %u stripe %u, "
1461 "been used with comp entry %u stripe %u\n",
1462 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1463 fio->ft_index, lio->lis_mirror_index,
1465 lov_comp_entry(fio->ft_page->cp_lov_index),
1466 lov_comp_stripe(fio->ft_page->cp_lov_index));
1468 fio->ft_page->cp_lov_index =
1469 lov_comp_index(entry, stripe);
1473 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1474 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1476 RETURN(lov_io_start(env, ios));
1479 static int lov_io_setattr_start(const struct lu_env *env,
1480 const struct cl_io_slice *ios)
1482 struct lov_io *lio = cl2lov_io(env, ios);
1483 struct cl_io *parent = ios->cis_io;
1484 struct lov_io_sub *sub;
1485 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1489 if (cl_io_is_fallocate(parent)) {
1490 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1491 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1492 int index = lov_comp_entry(sub->sub_subio_index);
1493 int stripe = lov_comp_stripe(sub->sub_subio_index);
1495 size = lov_size_to_stripe(lsm, index, size, stripe);
1496 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1497 sub->sub_io.u.ci_setattr.sa_avalid =
1498 parent->u.ci_setattr.sa_avalid;
1502 RETURN(lov_io_start(env, ios));
1505 static void lov_io_fsync_end(const struct lu_env *env,
1506 const struct cl_io_slice *ios)
1508 struct lov_io *lio = cl2lov_io(env, ios);
1509 struct lov_io_sub *sub;
1510 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1514 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1515 struct cl_io *subio = &sub->sub_io;
1517 lov_io_end_wrapper(sub->sub_env, subio);
1519 if (subio->ci_result == 0)
1520 *written += subio->u.ci_fsync.fi_nr_written;
1525 static void lov_io_lseek_end(const struct lu_env *env,
1526 const struct cl_io_slice *ios)
1528 struct lov_io *lio = cl2lov_io(env, ios);
1529 struct cl_io *io = lio->lis_cl.cis_io;
1530 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1531 struct lov_io_sub *sub;
1532 loff_t offset = -ENXIO;
1534 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1538 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1539 struct cl_io *subio = &sub->sub_io;
1540 int index = lov_comp_entry(sub->sub_subio_index);
1541 int stripe = lov_comp_stripe(sub->sub_subio_index);
1542 loff_t sub_off, lov_off;
1543 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1545 lov_io_end_wrapper(sub->sub_env, subio);
1547 if (io->ci_result == 0)
1548 io->ci_result = sub->sub_io.ci_result;
1553 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1554 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1555 index, stripe, seek_hole ? "HOLE" : "DATA",
1556 subio->u.ci_lseek.ls_start);
1558 /* first subio with positive result is what we need */
1559 sub_off = subio->u.ci_lseek.ls_result;
1560 /* Expected error, offset is out of stripe file size */
1561 if (sub_off == -ENXIO)
1563 /* Any other errors are not expected with ci_result == 0 */
1565 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1567 io->ci_result = sub_off;
1570 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1572 /* the only way to get negatove lov_off here is too big
1573 * result. Return -EOVERFLOW then.
1575 io->ci_result = -EOVERFLOW;
1576 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1577 (u64)lov_off, io->ci_result);
1580 if (lov_off < io->u.ci_lseek.ls_start) {
1581 io->ci_result = -EINVAL;
1582 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1583 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1586 /* resulting offset can be out of component range if stripe
1587 * object is full and its file size was returned as virtual
1588 * hole start. Skip this result, the next component will give
1589 * us correct lseek result but keep possible hole offset in
1590 * case there is no more components ahead
1592 if (lov_off >= comp_end) {
1593 /* must be SEEK_HOLE case */
1594 if (likely(seek_hole)) {
1595 /* save comp end as potential hole offset */
1596 hole_off = max_t(__u64, comp_end, hole_off);
1598 io->ci_result = -EINVAL;
1600 "off %lld >= comp_end %llu: rc = %d\n",
1601 lov_off, comp_end, io->ci_result);
1606 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1607 seek_hole ? "HOLE" : "DATA",
1608 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1609 sub->sub_io.ci_result);
1610 offset = min_t(__u64, offset, lov_off);
1612 /* no result but some component returns hole as component end */
1613 if (seek_hole && offset == -ENXIO && hole_off > 0)
1616 io->u.ci_lseek.ls_result = offset;
1620 static const struct cl_io_operations lov_io_ops = {
1623 .cio_fini = lov_io_fini,
1624 .cio_iter_init = lov_io_rw_iter_init,
1625 .cio_iter_fini = lov_io_iter_fini,
1626 .cio_lock = lov_io_lock,
1627 .cio_unlock = lov_io_unlock,
1628 .cio_start = lov_io_start,
1629 .cio_end = lov_io_end
1632 .cio_fini = lov_io_fini,
1633 .cio_iter_init = lov_io_rw_iter_init,
1634 .cio_iter_fini = lov_io_iter_fini,
1635 .cio_lock = lov_io_lock,
1636 .cio_unlock = lov_io_unlock,
1637 .cio_start = lov_io_start,
1638 .cio_end = lov_io_end
1641 .cio_fini = lov_io_fini,
1642 .cio_iter_init = lov_io_setattr_iter_init,
1643 .cio_iter_fini = lov_io_iter_fini,
1644 .cio_lock = lov_io_lock,
1645 .cio_unlock = lov_io_unlock,
1646 .cio_start = lov_io_setattr_start,
1647 .cio_end = lov_io_end
1649 [CIT_DATA_VERSION] = {
1650 .cio_fini = lov_io_fini,
1651 .cio_iter_init = lov_io_iter_init,
1652 .cio_iter_fini = lov_io_iter_fini,
1653 .cio_lock = lov_io_lock,
1654 .cio_unlock = lov_io_unlock,
1655 .cio_start = lov_io_start,
1656 .cio_end = lov_io_data_version_end,
1659 .cio_fini = lov_io_fini,
1660 .cio_iter_init = lov_io_iter_init,
1661 .cio_iter_fini = lov_io_iter_fini,
1662 .cio_lock = lov_io_lock,
1663 .cio_unlock = lov_io_unlock,
1664 .cio_start = lov_io_fault_start,
1665 .cio_end = lov_io_end
1668 .cio_fini = lov_io_fini,
1669 .cio_iter_init = lov_io_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_fsync_end
1677 .cio_fini = lov_io_fini,
1678 .cio_iter_init = lov_io_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_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_start,
1692 .cio_end = lov_io_lseek_end
1695 .cio_fini = lov_io_fini,
1698 .cio_fini = lov_io_fini
1701 .cio_read_ahead = lov_io_read_ahead,
1702 .cio_lru_reserve = lov_io_lru_reserve,
1703 .cio_submit = lov_io_submit,
1704 .cio_commit_async = lov_io_commit_async,
1707 /*****************************************************************************
1709 * Empty lov io operations.
1713 static void lov_empty_io_fini(const struct lu_env *env,
1714 const struct cl_io_slice *ios)
1716 struct lov_object *lov = cl2lov(ios->cis_obj);
1719 if (atomic_dec_and_test(&lov->lo_active_ios))
1720 wake_up(&lov->lo_waitq);
1724 static int lov_empty_io_submit(const struct lu_env *env,
1725 const struct cl_io_slice *ios,
1726 enum cl_req_type crt, struct cl_2queue *queue)
1731 static void lov_empty_impossible(const struct lu_env *env,
1732 struct cl_io_slice *ios)
1737 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1740 * An io operation vector for files without stripes.
1742 static const struct cl_io_operations lov_empty_io_ops = {
1745 .cio_fini = lov_empty_io_fini,
1747 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1748 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1749 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1750 .cio_end = LOV_EMPTY_IMPOSSIBLE
1754 .cio_fini = lov_empty_io_fini,
1755 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1756 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1757 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1758 .cio_end = LOV_EMPTY_IMPOSSIBLE
1761 .cio_fini = lov_empty_io_fini,
1762 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1763 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1764 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1765 .cio_end = LOV_EMPTY_IMPOSSIBLE
1768 .cio_fini = lov_empty_io_fini,
1769 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1770 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1771 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1772 .cio_end = LOV_EMPTY_IMPOSSIBLE
1775 .cio_fini = lov_empty_io_fini
1778 .cio_fini = lov_empty_io_fini
1781 .cio_fini = lov_empty_io_fini
1784 .cio_fini = lov_empty_io_fini
1787 .cio_submit = lov_empty_io_submit,
1788 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1791 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1794 struct lov_io *lio = lov_env_io(env);
1795 struct lov_object *lov = cl2lov(obj);
1800 INIT_LIST_HEAD(&lio->lis_active);
1801 result = lov_io_slice_init(lio, lov, io);
1805 result = lov_io_subio_init(env, lio, io);
1807 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1808 atomic_inc(&lov->lo_active_ios);
1812 io->ci_result = result < 0 ? result : 0;
1816 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1819 struct lov_object *lov = cl2lov(obj);
1820 struct lov_io *lio = lov_env_io(env);
1824 lio->lis_object = lov;
1825 switch (io->ci_type) {
1837 case CIT_DATA_VERSION:
1845 CERROR("Page fault on a file without stripes: "DFID"\n",
1846 PFID(lu_object_fid(&obj->co_lu)));
1850 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1851 atomic_inc(&lov->lo_active_ios);
1854 io->ci_result = result < 0 ? result : 0;
1858 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1861 struct lov_object *lov = cl2lov(obj);
1862 struct lov_io *lio = lov_env_io(env);
1866 LASSERT(lov->lo_lsm != NULL);
1867 lio->lis_object = lov;
1869 switch (io->ci_type) {
1871 LASSERTF(0, "invalid type %d\n", io->ci_type);
1872 result = -EOPNOTSUPP;
1878 case CIT_DATA_VERSION:
1883 * the truncate to 0 is managed by MDT:
1884 * - in open, for open O_TRUNC
1885 * - in setattr, for truncate
1888 * the truncate is for size > 0 so triggers a restore,
1889 * also trigger a restore for prealloc/punch
1891 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1892 io->ci_restore_needed = 1;
1901 io->ci_restore_needed = 1;
1907 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1908 atomic_inc(&lov->lo_active_ios);
1911 io->ci_result = result < 0 ? result : 0;
1916 * Return the index in composite:lo_entries by the file offset
1918 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1920 struct lov_object *lov = lio->lis_object;
1921 struct lov_layout_composite *comp = &lov->u.composite;
1922 int start_index = 0;
1923 int end_index = comp->lo_entry_count - 1;
1926 LASSERT(lov->lo_type == LLT_COMP);
1928 /* This is actual file offset so nothing can cover eof. */
1929 if (offset == LUSTRE_EOF)
1932 if (lov_is_flr(lov)) {
1933 struct lov_mirror_entry *lre;
1935 LASSERT(lio->lis_mirror_index >= 0);
1937 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1938 start_index = lre->lre_start;
1939 end_index = lre->lre_end;
1942 for (i = start_index; i <= end_index; i++) {
1943 struct lov_layout_entry *lle = lov_entry(lov, i);
1945 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1947 if ((offset >= lle->lle_extent->e_start &&
1948 offset < lle->lle_extent->e_end) ||
1949 (offset == OBD_OBJECT_EOF &&
1950 lle->lle_extent->e_end == OBD_OBJECT_EOF))