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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Implementation of cl_io for LOV layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_LOV
40 #include "lov_cl_internal.h"
46 static inline struct lov_io_sub *lov_sub_alloc(struct lov_io *lio, int index)
48 struct lov_io_sub *sub;
50 if (lio->lis_nr_subios == 0) {
51 LASSERT(lio->lis_single_subio_index == -1);
52 sub = &lio->lis_single_subio;
53 lio->lis_single_subio_index = index;
54 memset(sub, 0, sizeof(*sub));
60 INIT_LIST_HEAD(&sub->sub_list);
61 INIT_LIST_HEAD(&sub->sub_linkage);
62 sub->sub_subio_index = index;
68 static inline void lov_sub_free(struct lov_io *lio, struct lov_io_sub *sub)
70 if (sub->sub_subio_index == lio->lis_single_subio_index) {
71 LASSERT(sub == &lio->lis_single_subio);
72 lio->lis_single_subio_index = -1;
78 static void lov_io_sub_fini(const struct lu_env *env, struct lov_io *lio,
79 struct lov_io_sub *sub)
83 cl_io_fini(sub->sub_env, &sub->sub_io);
85 if (sub->sub_env && !IS_ERR(sub->sub_env)) {
86 cl_env_put(sub->sub_env, &sub->sub_refcheck);
93 is_index_within_mirror(struct lov_object *lov, int index, int mirror_index)
95 struct lov_layout_composite *comp = &lov->u.composite;
96 struct lov_mirror_entry *lre = &comp->lo_mirrors[mirror_index];
98 return (index >= lre->lre_start && index <= lre->lre_end);
101 static int lov_io_sub_init(const struct lu_env *env, struct lov_io *lio,
102 struct lov_io_sub *sub)
104 struct lov_object *lov = lio->lis_object;
105 struct cl_io *sub_io;
106 struct cl_object *sub_obj;
107 struct cl_io *io = lio->lis_cl.cis_io;
108 int index = lov_comp_entry(sub->sub_subio_index);
109 int stripe = lov_comp_stripe(sub->sub_subio_index);
111 LASSERT(sub->sub_env == NULL);
114 if (unlikely(!lov_r0(lov, index)->lo_sub ||
115 !lov_r0(lov, index)->lo_sub[stripe]))
118 LASSERTF(ergo(lov_is_flr(lov),
119 is_index_within_mirror(lov, index,
120 lio->lis_mirror_index)),
121 DFID "iot = %d, index = %d, mirror = %d\n",
122 PFID(lu_object_fid(lov2lu(lov))), io->ci_type, index,
123 lio->lis_mirror_index);
125 /* obtain new environment */
126 sub->sub_env = cl_env_get(&sub->sub_refcheck);
127 if (IS_ERR(sub->sub_env)) {
128 result = PTR_ERR(sub->sub_env);
132 sub_obj = lovsub2cl(lov_r0(lov, index)->lo_sub[stripe]);
133 sub_io = &sub->sub_io;
135 sub_io->ci_obj = sub_obj;
136 sub_io->ci_result = 0;
138 sub_io->ci_parent = io;
139 sub_io->ci_lockreq = io->ci_lockreq;
140 sub_io->ci_type = io->ci_type;
141 sub_io->ci_no_srvlock = io->ci_no_srvlock;
142 sub_io->ci_noatime = io->ci_noatime;
143 sub_io->ci_async_readahead = io->ci_async_readahead;
144 sub_io->ci_lock_no_expand = io->ci_lock_no_expand;
145 sub_io->ci_ndelay = io->ci_ndelay;
146 sub_io->ci_layout_version = io->ci_layout_version;
147 sub_io->ci_tried_all_mirrors = io->ci_tried_all_mirrors;
149 result = cl_io_sub_init(sub->sub_env, sub_io, io->ci_type, sub_obj);
152 lov_io_sub_fini(env, lio, sub);
157 struct lov_io_sub *lov_sub_get(const struct lu_env *env,
158 struct lov_io *lio, int index)
160 struct lov_io_sub *sub;
165 list_for_each_entry(sub, &lio->lis_subios, sub_list) {
166 if (sub->sub_subio_index == index) {
173 sub = lov_sub_alloc(lio, index);
175 GOTO(out, rc = -ENOMEM);
177 rc = lov_io_sub_init(env, lio, sub);
179 lov_sub_free(lio, sub);
183 list_add_tail(&sub->sub_list, &lio->lis_subios);
184 lio->lis_nr_subios++;
190 sub->sub_io.ci_noquota = lio->lis_cl.cis_io->ci_noquota;
194 /*****************************************************************************
199 static int lov_io_subio_init(const struct lu_env *env, struct lov_io *lio,
204 LASSERT(lio->lis_object != NULL);
206 INIT_LIST_HEAD(&lio->lis_subios);
207 lio->lis_single_subio_index = -1;
208 lio->lis_nr_subios = 0;
214 * Decide if it will need write intent RPC
216 static int lov_io_mirror_write_intent(struct lov_io *lio,
217 struct lov_object *obj, struct cl_io *io)
219 struct lov_layout_composite *comp = &obj->u.composite;
220 struct lu_extent *ext = &io->ci_write_intent;
221 struct lov_mirror_entry *lre;
222 struct lov_mirror_entry *primary;
223 struct lov_layout_entry *lle;
227 *ext = (typeof(*ext)) { lio->lis_pos, lio->lis_endpos };
228 io->ci_need_write_intent = 0;
230 if (!(io->ci_type == CIT_WRITE || cl_io_is_trunc(io) ||
231 cl_io_is_mkwrite(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 GOTO(out, result = 0);
589 * for truncate, it only needs to instantiate the components
590 * before the truncated size.
592 if (cl_io_is_trunc(io)) {
593 io->ci_write_intent.e_start = 0;
594 /* for writes, e_end is endpos, the location of the file
595 * pointer after the write is completed, so it is not accessed.
596 * For truncate, 'end' is the size, and *is* acccessed.
597 * In other words, writes are [start, end), but truncate is
598 * [start, size], where both are included. So add 1 to the
599 * size when creating the write intent to account for this.
601 io->ci_write_intent.e_end =
602 io->u.ci_setattr.sa_attr.lvb_size + 1;
604 io->ci_write_intent.e_start = lio->lis_pos;
605 io->ci_write_intent.e_end = lio->lis_endpos;
609 lov_foreach_io_layout(index, lio, &io->ci_write_intent) {
610 if (!lsm_entry_inited(obj->lo_lsm, index)) {
611 io->ci_need_write_intent = 1;
616 if (io->ci_need_write_intent && io->ci_designated_mirror > 0) {
618 * REINT_SYNC RPC has already tried to instantiate all of the
619 * components involved, obviously it didn't succeed. Skip this
620 * mirror for now. The server won't be able to figure out
621 * which mirror it should instantiate components
623 CERROR(DFID": trying to instantiate components for designated "
624 "I/O, file state: %d\n",
625 PFID(lu_object_fid(lov2lu(obj))), lov_flr_state(obj));
627 io->ci_need_write_intent = 0;
628 GOTO(out, result = -EIO);
631 if (io->ci_need_write_intent)
632 GOTO(out, result = 1);
640 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
642 struct lov_io *lio = cl2lov_io(env, ios);
643 struct lov_object *lov = cl2lov(ios->cis_obj);
647 LASSERT(list_empty(&lio->lis_active));
649 while (!list_empty(&lio->lis_subios)) {
650 struct lov_io_sub *sub = list_entry(lio->lis_subios.next,
654 list_del_init(&sub->sub_list);
655 lio->lis_nr_subios--;
657 lov_io_sub_fini(env, lio, sub);
658 lov_sub_free(lio, sub);
660 LASSERT(lio->lis_nr_subios == 0);
662 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
663 if (atomic_dec_and_test(&lov->lo_active_ios))
664 wake_up_all(&lov->lo_waitq);
668 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
669 loff_t start, loff_t end)
671 struct cl_io *io = &sub->sub_io;
672 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
673 struct cl_io *parent = lio->lis_cl.cis_io;
674 int index = lov_comp_entry(sub->sub_subio_index);
675 int stripe = lov_comp_stripe(sub->sub_subio_index);
677 switch (io->ci_type) {
679 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
680 io->u.ci_setattr.sa_attr_flags =
681 parent->u.ci_setattr.sa_attr_flags;
682 io->u.ci_setattr.sa_avalid = parent->u.ci_setattr.sa_avalid;
683 io->u.ci_setattr.sa_xvalid = parent->u.ci_setattr.sa_xvalid;
684 io->u.ci_setattr.sa_falloc_mode =
685 parent->u.ci_setattr.sa_falloc_mode;
686 io->u.ci_setattr.sa_stripe_index = stripe;
687 io->u.ci_setattr.sa_parent_fid =
688 parent->u.ci_setattr.sa_parent_fid;
689 /* For SETATTR(fallocate) pass the subtype to lower IO */
690 io->u.ci_setattr.sa_subtype = parent->u.ci_setattr.sa_subtype;
691 if (cl_io_is_fallocate(io)) {
692 io->u.ci_setattr.sa_falloc_offset = start;
693 io->u.ci_setattr.sa_falloc_end = end;
695 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
696 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
698 new_size = lov_size_to_stripe(lsm, index, new_size,
700 io->u.ci_setattr.sa_attr.lvb_size = new_size;
702 lov_lsm2layout(lsm, lsm->lsm_entries[index],
703 &io->u.ci_setattr.sa_layout);
706 case CIT_DATA_VERSION: {
707 io->u.ci_data_version.dv_data_version = 0;
708 io->u.ci_data_version.dv_flags =
709 parent->u.ci_data_version.dv_flags;
713 struct cl_object *obj = parent->ci_obj;
714 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
716 io->u.ci_fault = parent->u.ci_fault;
717 off = lov_size_to_stripe(lsm, index, off, stripe);
718 io->u.ci_fault.ft_index = cl_index(obj, off);
722 io->u.ci_fsync.fi_start = start;
723 io->u.ci_fsync.fi_end = end;
724 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
725 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
730 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
731 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
732 if (cl_io_is_append(parent)) {
733 io->u.ci_wr.wr_append = 1;
735 io->u.ci_rw.crw_pos = start;
736 io->u.ci_rw.crw_count = end - start;
741 io->u.ci_ladvise.li_start = start;
742 io->u.ci_ladvise.li_end = end;
743 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
744 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
745 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
749 io->u.ci_lseek.ls_start = start;
750 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
751 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
761 static loff_t lov_offset_mod(loff_t val, int delta)
763 if (val != OBD_OBJECT_EOF)
768 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
769 struct lov_io_sub *sub, u64 start, u64 end)
773 end = lov_offset_mod(end, 1);
774 lov_io_sub_inherit(sub, lio, start, end);
775 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
777 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
781 list_add_tail(&sub->sub_linkage, &lio->lis_active);
785 static int lov_io_iter_init(const struct lu_env *env,
786 const struct cl_io_slice *ios)
788 struct lov_io *lio = cl2lov_io(env, ios);
789 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
790 struct lov_io_sub *sub;
791 struct lu_extent ext;
797 ext.e_start = lio->lis_pos;
798 ext.e_end = lio->lis_endpos;
800 lov_foreach_io_layout(index, lio, &ext) {
801 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
802 struct lov_layout_raid0 *r0 = &le->lle_raid0;
806 bool tested_trunc_stripe = false;
808 r0->lo_trunc_stripeno = -1;
810 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
811 index, lsm->lsm_entries[index]->lsme_flags);
812 if (!lsm_entry_inited(lsm, index)) {
814 * Read from uninitialized components should return
820 if (lsm_entry_is_foreign(lsm, index))
823 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
824 CERROR("I/O to invalid component: %d, mirror: %d\n",
825 index, lio->lis_mirror_index);
829 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
830 if (!lov_stripe_intersects(lsm, index, stripe,
834 if (unlikely(!r0->lo_sub[stripe])) {
835 if (ios->cis_io->ci_type == CIT_READ ||
836 ios->cis_io->ci_type == CIT_WRITE ||
837 ios->cis_io->ci_type == CIT_FAULT)
843 if (cl_io_is_trunc(ios->cis_io) &&
844 !tested_trunc_stripe) {
848 prev = (stripe == 0) ? r0->lo_nr - 1 :
851 * Only involving previous stripe if the
852 * truncate in this component is at the
853 * beginning of this stripe.
855 tested_trunc_stripe = true;
856 if (ext.e_start < lsm->lsm_entries[index]->
857 lsme_extent.e_start) {
858 /* need previous stripe involvement */
859 r0->lo_trunc_stripeno = prev;
861 tr_start = ext.e_start;
862 tr_start = lov_do_div64(tr_start,
863 stripe_width(lsm, index));
864 /* tr_start %= stripe_swidth */
865 if (tr_start == stripe * lsm->
868 r0->lo_trunc_stripeno = prev;
872 /* if the last stripe is the trunc stripeno */
873 if (r0->lo_trunc_stripeno == stripe)
874 r0->lo_trunc_stripeno = -1;
876 sub = lov_sub_get(env, lio,
877 lov_comp_index(index, stripe));
881 rc = lov_io_add_sub(env, lio, sub, start, end);
888 if (r0->lo_trunc_stripeno != -1) {
889 stripe = r0->lo_trunc_stripeno;
890 if (unlikely(!r0->lo_sub[stripe])) {
891 r0->lo_trunc_stripeno = -1;
894 sub = lov_sub_get(env, lio,
895 lov_comp_index(index, stripe));
900 * the prev sub could be used by another truncate, we'd
901 * skip it. LU-14128 happends when expand truncate +
902 * read get wrong kms.
904 if (!list_empty(&sub->sub_linkage)) {
905 r0->lo_trunc_stripeno = -1;
909 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
911 rc = lov_io_add_sub(env, lio, sub, start, end);
920 static int lov_io_rw_iter_init(const struct lu_env *env,
921 const struct cl_io_slice *ios)
923 struct lov_io *lio = cl2lov_io(env, ios);
924 struct cl_io *io = ios->cis_io;
925 struct lov_stripe_md_entry *lse;
926 loff_t start = io->u.ci_rw.crw_pos;
930 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
933 if (cl_io_is_append(io))
934 RETURN(lov_io_iter_init(env, ios));
936 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
937 if (index < 0) { /* non-existing layout component */
938 if (io->ci_type == CIT_READ) {
940 * TODO: it needs to detect the next component and
941 * then set the next pos
945 RETURN(lov_io_iter_init(env, ios));
951 if (!lov_entry(lio->lis_object, index)->lle_valid &&
952 !io->ci_designated_mirror)
953 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
955 lse = lov_lse(lio->lis_object, index);
957 if (lsme_is_foreign(lse))
960 next = MAX_LFS_FILESIZE;
961 if (lse->lsme_stripe_count > 1) {
962 unsigned long ssize = lse->lsme_stripe_size;
964 lov_do_div64(start, ssize);
965 next = (start + 1) * ssize;
966 if (next <= start * ssize)
967 next = MAX_LFS_FILESIZE;
970 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
971 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
972 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
973 next = min_t(__u64, next, lse->lsme_extent.e_end);
974 next = min_t(loff_t, next, lio->lis_io_endpos);
976 io->ci_continue = next < lio->lis_io_endpos;
977 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
978 lio->lis_pos = io->u.ci_rw.crw_pos;
979 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
981 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
982 (__u64)start, lio->lis_pos, lio->lis_endpos,
983 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
986 * XXX The following call should be optimized: we know, that
987 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
989 RETURN(lov_io_iter_init(env, ios));
992 static int lov_io_setattr_iter_init(const struct lu_env *env,
993 const struct cl_io_slice *ios)
995 struct lov_io *lio = cl2lov_io(env, ios);
996 struct cl_io *io = ios->cis_io;
1000 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1001 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1002 /* no entry found for such offset */
1004 RETURN(io->ci_result = -ENODATA);
1007 RETURN(lov_io_iter_init(env, ios));
1010 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1011 int (*iofunc)(const struct lu_env *, struct cl_io *))
1013 struct cl_io *parent = lio->lis_cl.cis_io;
1014 struct lov_io_sub *sub;
1018 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1019 rc = iofunc(sub->sub_env, &sub->sub_io);
1023 if (parent->ci_result == 0)
1024 parent->ci_result = sub->sub_io.ci_result;
1029 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1032 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1035 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1038 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1041 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1045 * It's possible that lov_io_start() wasn't called against this
1046 * sub-io, either because previous sub-io failed, or upper layer
1049 if (io->ci_state == CIS_IO_GOING)
1052 io->ci_state = CIS_IO_FINISHED;
1056 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1058 cl_io_iter_fini(env, io);
1062 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1064 cl_io_unlock(env, io);
1068 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1072 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1077 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1079 struct lov_io *lio = cl2lov_io(env, ios);
1080 struct cl_io *parent = lio->lis_cl.cis_io;
1081 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1082 struct lov_io_sub *sub;
1085 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1086 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1088 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1090 pdv->dv_data_version += sdv->dv_data_version;
1091 if (pdv->dv_layout_version > sdv->dv_layout_version)
1092 pdv->dv_layout_version = sdv->dv_layout_version;
1094 if (parent->ci_result == 0)
1095 parent->ci_result = sub->sub_io.ci_result;
1101 static void lov_io_iter_fini(const struct lu_env *env,
1102 const struct cl_io_slice *ios)
1104 struct lov_io *lio = cl2lov_io(env, ios);
1108 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1110 while (!list_empty(&lio->lis_active))
1111 list_del_init(lio->lis_active.next);
1115 static void lov_io_unlock(const struct lu_env *env,
1116 const struct cl_io_slice *ios)
1121 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1126 static int lov_io_read_ahead(const struct lu_env *env,
1127 const struct cl_io_slice *ios,
1128 pgoff_t start, struct cl_read_ahead *ra)
1130 struct lov_io *lio = cl2lov_io(env, ios);
1131 struct lov_object *loo = lio->lis_object;
1132 struct cl_object *obj = lov2cl(loo);
1133 struct lov_layout_raid0 *r0;
1134 struct lov_io_sub *sub;
1138 unsigned int pps; /* pages per stripe */
1144 offset = cl_offset(obj, start);
1145 index = lov_io_layout_at(lio, offset);
1146 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1147 lsm_entry_is_foreign(loo->lo_lsm, index))
1150 /* avoid readahead to expand to stale components */
1151 if (!lov_entry(loo, index)->lle_valid)
1154 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1156 r0 = lov_r0(loo, index);
1157 if (unlikely(!r0->lo_sub[stripe]))
1160 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1162 RETURN(PTR_ERR(sub));
1164 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1165 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1166 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1169 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1170 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1176 * Adjust the stripe index by layout of comp. ra->cra_end is the
1177 * maximum page index covered by an underlying DLM lock.
1178 * This function converts cra_end from stripe level to file level, and
1179 * make sure it's not beyond stripe and component boundary.
1182 /* cra_end is stripe level, convert it into file level */
1183 ra_end = ra->cra_end_idx;
1184 if (ra_end != CL_PAGE_EOF)
1185 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1188 /* boundary of current component */
1189 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1190 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1191 ra->cra_end_idx = ra_end - 1;
1193 if (r0->lo_nr == 1) /* single stripe file */
1196 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1198 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1199 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1200 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1201 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1203 /* never exceed the end of the stripe */
1204 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1205 start + pps - start % pps - 1);
1210 * lov implementation of cl_operations::cio_submit() method. It takes a list
1211 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1212 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1215 * Major complication of this function is a need to handle memory cleansing:
1216 * cl_io_submit() is called to write out pages as a part of VM memory
1217 * reclamation, and hence it may not fail due to memory shortages (system
1218 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1219 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1220 * not-memory cleansing context), and in case of memory shortage, these
1221 * pre-allocated resources are used by lov_io_submit() under
1222 * lov_device::ld_mutex mutex.
1224 static int lov_io_submit(const struct lu_env *env,
1225 const struct cl_io_slice *ios,
1226 enum cl_req_type crt, struct cl_2queue *queue)
1228 struct cl_page_list *qin = &queue->c2_qin;
1229 struct lov_io *lio = cl2lov_io(env, ios);
1230 struct lov_io_sub *sub;
1231 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1232 struct cl_page *page;
1233 struct cl_page *tmp;
1238 cl_page_list_init(plist);
1239 while (qin->pl_nr > 0) {
1240 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1242 page = cl_page_list_first(qin);
1243 if (lov_page_is_empty(page)) {
1244 cl_page_list_move(&queue->c2_qout, qin, page);
1247 * it could only be mirror read to get here therefore
1248 * the pages will be transient. We don't care about
1249 * the return code of cl_page_prep() at all.
1251 (void) cl_page_prep(env, ios->cis_io, page, crt);
1252 cl_page_completion(env, page, crt, 0);
1256 cl_2queue_init(cl2q);
1257 cl_page_list_move(&cl2q->c2_qin, qin, page);
1259 index = page->cp_lov_index;
1260 cl_page_list_for_each_safe(page, tmp, qin) {
1261 /* this page is not on this stripe */
1262 if (index != page->cp_lov_index)
1265 cl_page_list_move(&cl2q->c2_qin, qin, page);
1268 sub = lov_sub_get(env, lio, index);
1270 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1276 cl_page_list_splice(&cl2q->c2_qin, plist);
1277 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1278 cl_2queue_fini(env, cl2q);
1284 cl_page_list_splice(plist, qin);
1285 cl_page_list_fini(env, plist);
1290 static int lov_io_commit_async(const struct lu_env *env,
1291 const struct cl_io_slice *ios,
1292 struct cl_page_list *queue, int from, int to,
1295 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1296 struct lov_io *lio = cl2lov_io(env, ios);
1297 struct lov_io_sub *sub;
1298 struct cl_page *page;
1302 if (lio->lis_nr_subios == 1) {
1303 int idx = lio->lis_single_subio_index;
1305 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1307 sub = lov_sub_get(env, lio, idx);
1308 LASSERT(!IS_ERR(sub));
1309 LASSERT(sub == &lio->lis_single_subio);
1310 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1315 cl_page_list_init(plist);
1316 while (queue->pl_nr > 0) {
1320 LASSERT(plist->pl_nr == 0);
1321 page = cl_page_list_first(queue);
1322 LASSERT(!lov_page_is_empty(page));
1324 cl_page_list_move(plist, queue, page);
1326 index = page->cp_lov_index;
1327 while (queue->pl_nr > 0) {
1328 page = cl_page_list_first(queue);
1329 if (index != page->cp_lov_index)
1332 cl_page_list_move(plist, queue, page);
1335 if (queue->pl_nr > 0) /* still has more pages */
1336 stripe_to = PAGE_SIZE;
1338 sub = lov_sub_get(env, lio, index);
1340 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1341 plist, from, stripe_to, cb);
1347 if (plist->pl_nr > 0) /* short write */
1352 if (lov_comp_entry(index) !=
1353 lov_comp_entry(page->cp_lov_index))
1354 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1357 /* for error case, add the page back into the qin list */
1358 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1359 while (plist->pl_nr > 0) {
1360 /* error occurred, add the uncommitted pages back into queue */
1361 page = cl_page_list_last(plist);
1362 cl_page_list_move_head(queue, plist, page);
1368 static int lov_io_fault_start(const struct lu_env *env,
1369 const struct cl_io_slice *ios)
1371 struct cl_fault_io *fio;
1373 struct lov_io_sub *sub;
1377 fio = &ios->cis_io->u.ci_fault;
1378 lio = cl2lov_io(env, ios);
1379 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1380 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1382 RETURN(lov_io_start(env, ios));
1385 static void lov_io_fsync_end(const struct lu_env *env,
1386 const struct cl_io_slice *ios)
1388 struct lov_io *lio = cl2lov_io(env, ios);
1389 struct lov_io_sub *sub;
1390 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1394 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1395 struct cl_io *subio = &sub->sub_io;
1397 lov_io_end_wrapper(sub->sub_env, subio);
1399 if (subio->ci_result == 0)
1400 *written += subio->u.ci_fsync.fi_nr_written;
1405 static void lov_io_lseek_end(const struct lu_env *env,
1406 const struct cl_io_slice *ios)
1408 struct lov_io *lio = cl2lov_io(env, ios);
1409 struct cl_io *io = lio->lis_cl.cis_io;
1410 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1411 struct lov_io_sub *sub;
1412 loff_t offset = -ENXIO;
1414 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1418 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1419 struct cl_io *subio = &sub->sub_io;
1420 int index = lov_comp_entry(sub->sub_subio_index);
1421 int stripe = lov_comp_stripe(sub->sub_subio_index);
1422 loff_t sub_off, lov_off;
1423 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1425 lov_io_end_wrapper(sub->sub_env, subio);
1427 if (io->ci_result == 0)
1428 io->ci_result = sub->sub_io.ci_result;
1433 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1434 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1435 index, stripe, seek_hole ? "HOLE" : "DATA",
1436 subio->u.ci_lseek.ls_start);
1438 /* first subio with positive result is what we need */
1439 sub_off = subio->u.ci_lseek.ls_result;
1440 /* Expected error, offset is out of stripe file size */
1441 if (sub_off == -ENXIO)
1443 /* Any other errors are not expected with ci_result == 0 */
1445 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1447 io->ci_result = sub_off;
1450 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1452 /* the only way to get negatove lov_off here is too big
1453 * result. Return -EOVERFLOW then.
1455 io->ci_result = -EOVERFLOW;
1456 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1457 (u64)lov_off, io->ci_result);
1460 if (lov_off < io->u.ci_lseek.ls_start) {
1461 io->ci_result = -EINVAL;
1462 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1463 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1466 /* resulting offset can be out of component range if stripe
1467 * object is full and its file size was returned as virtual
1468 * hole start. Skip this result, the next component will give
1469 * us correct lseek result but keep possible hole offset in
1470 * case there is no more components ahead
1472 if (lov_off >= comp_end) {
1473 /* must be SEEK_HOLE case */
1474 if (likely(seek_hole)) {
1475 /* save comp end as potential hole offset */
1476 hole_off = max_t(__u64, comp_end, hole_off);
1478 io->ci_result = -EINVAL;
1480 "off %lld >= comp_end %llu: rc = %d\n",
1481 lov_off, comp_end, io->ci_result);
1486 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1487 seek_hole ? "HOLE" : "DATA",
1488 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1489 sub->sub_io.ci_result);
1490 offset = min_t(__u64, offset, lov_off);
1492 /* no result but some component returns hole as component end */
1493 if (seek_hole && offset == -ENXIO && hole_off > 0)
1496 io->u.ci_lseek.ls_result = offset;
1500 static const struct cl_io_operations lov_io_ops = {
1503 .cio_fini = lov_io_fini,
1504 .cio_iter_init = lov_io_rw_iter_init,
1505 .cio_iter_fini = lov_io_iter_fini,
1506 .cio_lock = lov_io_lock,
1507 .cio_unlock = lov_io_unlock,
1508 .cio_start = lov_io_start,
1509 .cio_end = lov_io_end
1512 .cio_fini = lov_io_fini,
1513 .cio_iter_init = lov_io_rw_iter_init,
1514 .cio_iter_fini = lov_io_iter_fini,
1515 .cio_lock = lov_io_lock,
1516 .cio_unlock = lov_io_unlock,
1517 .cio_start = lov_io_start,
1518 .cio_end = lov_io_end
1521 .cio_fini = lov_io_fini,
1522 .cio_iter_init = lov_io_setattr_iter_init,
1523 .cio_iter_fini = lov_io_iter_fini,
1524 .cio_lock = lov_io_lock,
1525 .cio_unlock = lov_io_unlock,
1526 .cio_start = lov_io_start,
1527 .cio_end = lov_io_end
1529 [CIT_DATA_VERSION] = {
1530 .cio_fini = lov_io_fini,
1531 .cio_iter_init = lov_io_iter_init,
1532 .cio_iter_fini = lov_io_iter_fini,
1533 .cio_lock = lov_io_lock,
1534 .cio_unlock = lov_io_unlock,
1535 .cio_start = lov_io_start,
1536 .cio_end = lov_io_data_version_end,
1539 .cio_fini = lov_io_fini,
1540 .cio_iter_init = lov_io_iter_init,
1541 .cio_iter_fini = lov_io_iter_fini,
1542 .cio_lock = lov_io_lock,
1543 .cio_unlock = lov_io_unlock,
1544 .cio_start = lov_io_fault_start,
1545 .cio_end = lov_io_end
1548 .cio_fini = lov_io_fini,
1549 .cio_iter_init = lov_io_iter_init,
1550 .cio_iter_fini = lov_io_iter_fini,
1551 .cio_lock = lov_io_lock,
1552 .cio_unlock = lov_io_unlock,
1553 .cio_start = lov_io_start,
1554 .cio_end = lov_io_fsync_end
1557 .cio_fini = lov_io_fini,
1558 .cio_iter_init = lov_io_iter_init,
1559 .cio_iter_fini = lov_io_iter_fini,
1560 .cio_lock = lov_io_lock,
1561 .cio_unlock = lov_io_unlock,
1562 .cio_start = lov_io_start,
1563 .cio_end = lov_io_end
1566 .cio_fini = lov_io_fini,
1567 .cio_iter_init = lov_io_iter_init,
1568 .cio_iter_fini = lov_io_iter_fini,
1569 .cio_lock = lov_io_lock,
1570 .cio_unlock = lov_io_unlock,
1571 .cio_start = lov_io_start,
1572 .cio_end = lov_io_lseek_end
1575 .cio_fini = lov_io_fini,
1578 .cio_fini = lov_io_fini
1581 .cio_read_ahead = lov_io_read_ahead,
1582 .cio_submit = lov_io_submit,
1583 .cio_commit_async = lov_io_commit_async,
1586 /*****************************************************************************
1588 * Empty lov io operations.
1592 static void lov_empty_io_fini(const struct lu_env *env,
1593 const struct cl_io_slice *ios)
1595 struct lov_object *lov = cl2lov(ios->cis_obj);
1598 if (atomic_dec_and_test(&lov->lo_active_ios))
1599 wake_up_all(&lov->lo_waitq);
1603 static int lov_empty_io_submit(const struct lu_env *env,
1604 const struct cl_io_slice *ios,
1605 enum cl_req_type crt, struct cl_2queue *queue)
1610 static void lov_empty_impossible(const struct lu_env *env,
1611 struct cl_io_slice *ios)
1616 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1619 * An io operation vector for files without stripes.
1621 static const struct cl_io_operations lov_empty_io_ops = {
1624 .cio_fini = lov_empty_io_fini,
1626 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1627 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1628 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1629 .cio_end = LOV_EMPTY_IMPOSSIBLE
1633 .cio_fini = lov_empty_io_fini,
1634 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1635 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1636 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1637 .cio_end = LOV_EMPTY_IMPOSSIBLE
1640 .cio_fini = lov_empty_io_fini,
1641 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1642 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1643 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1644 .cio_end = LOV_EMPTY_IMPOSSIBLE
1647 .cio_fini = lov_empty_io_fini,
1648 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1649 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1650 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1651 .cio_end = LOV_EMPTY_IMPOSSIBLE
1654 .cio_fini = lov_empty_io_fini
1657 .cio_fini = lov_empty_io_fini
1660 .cio_fini = lov_empty_io_fini
1663 .cio_fini = lov_empty_io_fini
1666 .cio_submit = lov_empty_io_submit,
1667 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1670 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1673 struct lov_io *lio = lov_env_io(env);
1674 struct lov_object *lov = cl2lov(obj);
1679 INIT_LIST_HEAD(&lio->lis_active);
1680 result = lov_io_slice_init(lio, lov, io);
1684 result = lov_io_subio_init(env, lio, io);
1686 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1687 atomic_inc(&lov->lo_active_ios);
1691 io->ci_result = result < 0 ? result : 0;
1695 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1698 struct lov_object *lov = cl2lov(obj);
1699 struct lov_io *lio = lov_env_io(env);
1703 lio->lis_object = lov;
1704 switch (io->ci_type) {
1716 case CIT_DATA_VERSION:
1724 CERROR("Page fault on a file without stripes: "DFID"\n",
1725 PFID(lu_object_fid(&obj->co_lu)));
1729 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1730 atomic_inc(&lov->lo_active_ios);
1733 io->ci_result = result < 0 ? result : 0;
1737 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1740 struct lov_object *lov = cl2lov(obj);
1741 struct lov_io *lio = lov_env_io(env);
1745 LASSERT(lov->lo_lsm != NULL);
1746 lio->lis_object = lov;
1748 switch (io->ci_type) {
1750 LASSERTF(0, "invalid type %d\n", io->ci_type);
1751 result = -EOPNOTSUPP;
1757 case CIT_DATA_VERSION:
1762 * the truncate to 0 is managed by MDT:
1763 * - in open, for open O_TRUNC
1764 * - in setattr, for truncate
1767 * the truncate is for size > 0 so triggers a restore,
1768 * also trigger a restore for prealloc/punch
1770 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1771 io->ci_restore_needed = 1;
1780 io->ci_restore_needed = 1;
1786 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1787 atomic_inc(&lov->lo_active_ios);
1790 io->ci_result = result < 0 ? result : 0;
1795 * Return the index in composite:lo_entries by the file offset
1797 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1799 struct lov_object *lov = lio->lis_object;
1800 struct lov_layout_composite *comp = &lov->u.composite;
1801 int start_index = 0;
1802 int end_index = comp->lo_entry_count - 1;
1805 LASSERT(lov->lo_type == LLT_COMP);
1807 /* This is actual file offset so nothing can cover eof. */
1808 if (offset == LUSTRE_EOF)
1811 if (lov_is_flr(lov)) {
1812 struct lov_mirror_entry *lre;
1814 LASSERT(lio->lis_mirror_index >= 0);
1816 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1817 start_index = lre->lre_start;
1818 end_index = lre->lre_end;
1821 for (i = start_index; i <= end_index; i++) {
1822 struct lov_layout_entry *lle = lov_entry(lov, i);
1824 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1826 if ((offset >= lle->lle_extent->e_start &&
1827 offset < lle->lle_extent->e_end) ||
1828 (offset == OBD_OBJECT_EOF &&
1829 lle->lle_extent->e_end == OBD_OBJECT_EOF))