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);
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;
694 if (cl_io_is_trunc(io)) {
695 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
697 new_size = lov_size_to_stripe(lsm, index, new_size,
699 io->u.ci_setattr.sa_attr.lvb_size = new_size;
701 lov_lsm2layout(lsm, lsm->lsm_entries[index],
702 &io->u.ci_setattr.sa_layout);
705 case CIT_DATA_VERSION: {
706 io->u.ci_data_version.dv_data_version = 0;
707 io->u.ci_data_version.dv_flags =
708 parent->u.ci_data_version.dv_flags;
712 struct cl_object *obj = parent->ci_obj;
713 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
715 io->u.ci_fault = parent->u.ci_fault;
716 off = lov_size_to_stripe(lsm, index, off, stripe);
717 io->u.ci_fault.ft_index = cl_index(obj, off);
721 io->u.ci_fsync.fi_start = start;
722 io->u.ci_fsync.fi_end = end;
723 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
724 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
729 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
730 io->ci_tried_all_mirrors = parent->ci_tried_all_mirrors;
731 if (cl_io_is_append(parent)) {
732 io->u.ci_wr.wr_append = 1;
734 io->u.ci_rw.crw_pos = start;
735 io->u.ci_rw.crw_count = end - start;
740 io->u.ci_ladvise.li_start = start;
741 io->u.ci_ladvise.li_end = end;
742 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
743 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
744 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
748 io->u.ci_lseek.ls_start = start;
749 io->u.ci_lseek.ls_whence = parent->u.ci_lseek.ls_whence;
750 io->u.ci_lseek.ls_result = parent->u.ci_lseek.ls_result;
760 static loff_t lov_offset_mod(loff_t val, int delta)
762 if (val != OBD_OBJECT_EOF)
767 static int lov_io_add_sub(const struct lu_env *env, struct lov_io *lio,
768 struct lov_io_sub *sub, u64 start, u64 end)
772 end = lov_offset_mod(end, 1);
773 lov_io_sub_inherit(sub, lio, start, end);
774 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
776 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
780 list_add_tail(&sub->sub_linkage, &lio->lis_active);
784 static int lov_io_iter_init(const struct lu_env *env,
785 const struct cl_io_slice *ios)
787 struct lov_io *lio = cl2lov_io(env, ios);
788 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
789 struct lov_io_sub *sub;
790 struct lu_extent ext;
796 ext.e_start = lio->lis_pos;
797 ext.e_end = lio->lis_endpos;
799 lov_foreach_io_layout(index, lio, &ext) {
800 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
801 struct lov_layout_raid0 *r0 = &le->lle_raid0;
805 bool tested_trunc_stripe = false;
807 r0->lo_trunc_stripeno = -1;
809 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
810 index, lsm->lsm_entries[index]->lsme_flags);
811 if (!lsm_entry_inited(lsm, index)) {
813 * Read from uninitialized components should return
819 if (lsm_entry_is_foreign(lsm, index))
822 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
823 CERROR("I/O to invalid component: %d, mirror: %d\n",
824 index, lio->lis_mirror_index);
828 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
829 if (!lov_stripe_intersects(lsm, index, stripe,
833 if (unlikely(!r0->lo_sub[stripe])) {
834 if (ios->cis_io->ci_type == CIT_READ ||
835 ios->cis_io->ci_type == CIT_WRITE ||
836 ios->cis_io->ci_type == CIT_FAULT)
842 if (cl_io_is_trunc(ios->cis_io) &&
843 !tested_trunc_stripe) {
847 prev = (stripe == 0) ? r0->lo_nr - 1 :
850 * Only involving previous stripe if the
851 * truncate in this component is at the
852 * beginning of this stripe.
854 tested_trunc_stripe = true;
855 if (ext.e_start < lsm->lsm_entries[index]->
856 lsme_extent.e_start) {
857 /* need previous stripe involvement */
858 r0->lo_trunc_stripeno = prev;
860 tr_start = ext.e_start;
861 tr_start = lov_do_div64(tr_start,
862 stripe_width(lsm, index));
863 /* tr_start %= stripe_swidth */
864 if (tr_start == stripe * lsm->
867 r0->lo_trunc_stripeno = prev;
871 /* if the last stripe is the trunc stripeno */
872 if (r0->lo_trunc_stripeno == stripe)
873 r0->lo_trunc_stripeno = -1;
875 sub = lov_sub_get(env, lio,
876 lov_comp_index(index, stripe));
880 rc = lov_io_add_sub(env, lio, sub, start, end);
887 if (r0->lo_trunc_stripeno != -1) {
888 stripe = r0->lo_trunc_stripeno;
889 if (unlikely(!r0->lo_sub[stripe])) {
890 r0->lo_trunc_stripeno = -1;
893 sub = lov_sub_get(env, lio,
894 lov_comp_index(index, stripe));
899 * the prev sub could be used by another truncate, we'd
900 * skip it. LU-14128 happends when expand truncate +
901 * read get wrong kms.
903 if (!list_empty(&sub->sub_linkage)) {
904 r0->lo_trunc_stripeno = -1;
908 (void)lov_stripe_intersects(lsm, index, stripe, &ext,
910 rc = lov_io_add_sub(env, lio, sub, start, end);
919 static int lov_io_rw_iter_init(const struct lu_env *env,
920 const struct cl_io_slice *ios)
922 struct lov_io *lio = cl2lov_io(env, ios);
923 struct cl_io *io = ios->cis_io;
924 struct lov_stripe_md_entry *lse;
925 loff_t start = io->u.ci_rw.crw_pos;
929 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
932 if (cl_io_is_append(io))
933 RETURN(lov_io_iter_init(env, ios));
935 index = lov_io_layout_at(lio, io->u.ci_rw.crw_pos);
936 if (index < 0) { /* non-existing layout component */
937 if (io->ci_type == CIT_READ) {
939 * TODO: it needs to detect the next component and
940 * then set the next pos
944 RETURN(lov_io_iter_init(env, ios));
950 if (!lov_entry(lio->lis_object, index)->lle_valid &&
951 !io->ci_designated_mirror)
952 RETURN(io->ci_type == CIT_READ ? -EAGAIN : -EIO);
954 lse = lov_lse(lio->lis_object, index);
956 if (lsme_is_foreign(lse))
959 next = MAX_LFS_FILESIZE;
960 if (lse->lsme_stripe_count > 1) {
961 unsigned long ssize = lse->lsme_stripe_size;
963 lov_do_div64(start, ssize);
964 next = (start + 1) * ssize;
965 if (next <= start * ssize)
966 next = MAX_LFS_FILESIZE;
969 LASSERTF(io->u.ci_rw.crw_pos >= lse->lsme_extent.e_start,
970 "pos %lld, [%lld, %lld)\n", io->u.ci_rw.crw_pos,
971 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
972 next = min_t(__u64, next, lse->lsme_extent.e_end);
973 next = min_t(loff_t, next, lio->lis_io_endpos);
975 io->ci_continue = next < lio->lis_io_endpos;
976 io->u.ci_rw.crw_count = next - io->u.ci_rw.crw_pos;
977 lio->lis_pos = io->u.ci_rw.crw_pos;
978 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
980 "stripe: %llu chunk: [%llu, %llu) %llu, %zd\n",
981 (__u64)start, lio->lis_pos, lio->lis_endpos,
982 (__u64)lio->lis_io_endpos, io->u.ci_rw.crw_count);
985 * XXX The following call should be optimized: we know, that
986 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
988 RETURN(lov_io_iter_init(env, ios));
991 static int lov_io_setattr_iter_init(const struct lu_env *env,
992 const struct cl_io_slice *ios)
994 struct lov_io *lio = cl2lov_io(env, ios);
995 struct cl_io *io = ios->cis_io;
999 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
1000 index = lov_io_layout_at(lio, lio->lis_pos - 1);
1001 /* no entry found for such offset */
1003 RETURN(io->ci_result = -ENODATA);
1006 RETURN(lov_io_iter_init(env, ios));
1009 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
1010 int (*iofunc)(const struct lu_env *, struct cl_io *))
1012 struct cl_io *parent = lio->lis_cl.cis_io;
1013 struct lov_io_sub *sub;
1017 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1018 rc = iofunc(sub->sub_env, &sub->sub_io);
1022 if (parent->ci_result == 0)
1023 parent->ci_result = sub->sub_io.ci_result;
1028 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
1031 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
1034 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
1037 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
1040 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
1044 * It's possible that lov_io_start() wasn't called against this
1045 * sub-io, either because previous sub-io failed, or upper layer
1048 if (io->ci_state == CIS_IO_GOING)
1051 io->ci_state = CIS_IO_FINISHED;
1055 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
1057 cl_io_iter_fini(env, io);
1061 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
1063 cl_io_unlock(env, io);
1067 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
1071 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
1076 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
1078 struct lov_io *lio = cl2lov_io(env, ios);
1079 struct cl_io *parent = lio->lis_cl.cis_io;
1080 struct cl_data_version_io *pdv = &parent->u.ci_data_version;
1081 struct lov_io_sub *sub;
1084 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1085 struct cl_data_version_io *sdv = &sub->sub_io.u.ci_data_version;
1087 lov_io_end_wrapper(sub->sub_env, &sub->sub_io);
1089 pdv->dv_data_version += sdv->dv_data_version;
1090 if (pdv->dv_layout_version > sdv->dv_layout_version)
1091 pdv->dv_layout_version = sdv->dv_layout_version;
1093 if (parent->ci_result == 0)
1094 parent->ci_result = sub->sub_io.ci_result;
1100 static void lov_io_iter_fini(const struct lu_env *env,
1101 const struct cl_io_slice *ios)
1103 struct lov_io *lio = cl2lov_io(env, ios);
1107 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
1109 while (!list_empty(&lio->lis_active))
1110 list_del_init(lio->lis_active.next);
1114 static void lov_io_unlock(const struct lu_env *env,
1115 const struct cl_io_slice *ios)
1120 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
1125 static int lov_io_read_ahead(const struct lu_env *env,
1126 const struct cl_io_slice *ios,
1127 pgoff_t start, struct cl_read_ahead *ra)
1129 struct lov_io *lio = cl2lov_io(env, ios);
1130 struct lov_object *loo = lio->lis_object;
1131 struct cl_object *obj = lov2cl(loo);
1132 struct lov_layout_raid0 *r0;
1133 struct lov_io_sub *sub;
1137 unsigned int pps; /* pages per stripe */
1143 offset = cl_offset(obj, start);
1144 index = lov_io_layout_at(lio, offset);
1145 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index) ||
1146 lsm_entry_is_foreign(loo->lo_lsm, index))
1149 /* avoid readahead to expand to stale components */
1150 if (!lov_entry(loo, index)->lle_valid)
1153 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
1155 r0 = lov_r0(loo, index);
1156 if (unlikely(!r0->lo_sub[stripe]))
1159 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
1161 RETURN(PTR_ERR(sub));
1163 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
1164 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
1165 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
1168 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
1169 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx,
1175 * Adjust the stripe index by layout of comp. ra->cra_end is the
1176 * maximum page index covered by an underlying DLM lock.
1177 * This function converts cra_end from stripe level to file level, and
1178 * make sure it's not beyond stripe and component boundary.
1181 /* cra_end is stripe level, convert it into file level */
1182 ra_end = ra->cra_end_idx;
1183 if (ra_end != CL_PAGE_EOF)
1184 ra->cra_end_idx = lov_stripe_pgoff(loo->lo_lsm, index,
1187 /* boundary of current component */
1188 ra_end = cl_index(obj, (loff_t)lov_io_extent(lio, index)->e_end);
1189 if (ra_end != CL_PAGE_EOF && ra->cra_end_idx >= ra_end)
1190 ra->cra_end_idx = ra_end - 1;
1192 if (r0->lo_nr == 1) /* single stripe file */
1195 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
1197 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %d, "
1198 "stripe_size = %u, stripe no = %u, start index = %lu\n",
1199 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end_idx, pps, index,
1200 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
1202 /* never exceed the end of the stripe */
1203 ra->cra_end_idx = min_t(pgoff_t, ra->cra_end_idx,
1204 start + pps - start % pps - 1);
1208 int lov_io_lru_reserve(const struct lu_env *env,
1209 const struct cl_io_slice *ios, loff_t pos, size_t bytes)
1211 struct lov_io *lio = cl2lov_io(env, ios);
1212 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1213 struct lov_io_sub *sub;
1214 struct lu_extent ext;
1221 ext.e_end = pos + bytes;
1222 lov_foreach_io_layout(index, lio, &ext) {
1223 struct lov_layout_entry *le = lov_entry(lio->lis_object, index);
1224 struct lov_layout_raid0 *r0 = &le->lle_raid0;
1229 if (!lsm_entry_inited(lsm, index))
1232 if (!le->lle_valid && !ios->cis_io->ci_designated_mirror) {
1233 CERROR(DFID": I/O to invalid component: %d, mirror: %d\n",
1234 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1235 index, lio->lis_mirror_index);
1239 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
1240 if (!lov_stripe_intersects(lsm, index, stripe,
1241 &ext, &start, &end))
1244 if (unlikely(!r0->lo_sub[stripe]))
1247 sub = lov_sub_get(env, lio,
1248 lov_comp_index(index, stripe));
1250 return PTR_ERR(sub);
1252 rc = cl_io_lru_reserve(sub->sub_env, &sub->sub_io, start,
1263 * lov implementation of cl_operations::cio_submit() method. It takes a list
1264 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
1265 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
1268 * Major complication of this function is a need to handle memory cleansing:
1269 * cl_io_submit() is called to write out pages as a part of VM memory
1270 * reclamation, and hence it may not fail due to memory shortages (system
1271 * dead-locks otherwise). To deal with this, some resources (sub-lists,
1272 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
1273 * not-memory cleansing context), and in case of memory shortage, these
1274 * pre-allocated resources are used by lov_io_submit() under
1275 * lov_device::ld_mutex mutex.
1277 static int lov_io_submit(const struct lu_env *env,
1278 const struct cl_io_slice *ios,
1279 enum cl_req_type crt, struct cl_2queue *queue)
1281 struct cl_page_list *qin = &queue->c2_qin;
1282 struct lov_io *lio = cl2lov_io(env, ios);
1283 struct lov_io_sub *sub;
1284 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1285 struct cl_page *page = cl_page_list_first(qin);
1286 struct cl_page *tmp;
1292 if (page->cp_type == CPT_TRANSIENT)
1295 cl_page_list_init(plist);
1296 while (qin->pl_nr > 0) {
1297 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
1299 page = cl_page_list_first(qin);
1300 if (lov_page_is_empty(page)) {
1301 cl_page_list_move(&queue->c2_qout, qin, page);
1304 * it could only be mirror read to get here therefore
1305 * the pages will be transient. We don't care about
1306 * the return code of cl_page_prep() at all.
1308 (void) cl_page_prep(env, ios->cis_io, page, crt);
1309 cl_page_completion(env, page, crt, 0);
1313 cl_2queue_init(cl2q);
1314 cl_page_list_move(&cl2q->c2_qin, qin, page);
1316 index = page->cp_lov_index;
1317 /* DIO is already split by stripe */
1319 cl_page_list_for_each_safe(page, tmp, qin) {
1320 /* this page is not on this stripe */
1321 if (index != page->cp_lov_index)
1324 cl_page_list_move(&cl2q->c2_qin, qin, page);
1327 cl_page_list_splice(qin, &cl2q->c2_qin);
1330 sub = lov_sub_get(env, lio, index);
1332 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
1338 cl_page_list_splice(&cl2q->c2_qin, plist);
1339 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
1340 cl_2queue_fini(env, cl2q);
1346 cl_page_list_splice(plist, qin);
1347 cl_page_list_fini(env, plist);
1352 static int lov_io_commit_async(const struct lu_env *env,
1353 const struct cl_io_slice *ios,
1354 struct cl_page_list *queue, int from, int to,
1357 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
1358 struct lov_io *lio = cl2lov_io(env, ios);
1359 struct lov_io_sub *sub;
1360 struct cl_page *page;
1364 if (lio->lis_nr_subios == 1) {
1365 int idx = lio->lis_single_subio_index;
1367 LASSERT(!lov_page_is_empty(cl_page_list_first(queue)));
1369 sub = lov_sub_get(env, lio, idx);
1370 LASSERT(!IS_ERR(sub));
1371 LASSERT(sub == &lio->lis_single_subio);
1372 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
1377 cl_page_list_init(plist);
1378 while (queue->pl_nr > 0) {
1382 LASSERT(plist->pl_nr == 0);
1383 page = cl_page_list_first(queue);
1384 LASSERT(!lov_page_is_empty(page));
1386 cl_page_list_move(plist, queue, page);
1388 index = page->cp_lov_index;
1389 while (queue->pl_nr > 0) {
1390 page = cl_page_list_first(queue);
1391 if (index != page->cp_lov_index)
1394 cl_page_list_move(plist, queue, page);
1397 if (queue->pl_nr > 0) /* still has more pages */
1398 stripe_to = PAGE_SIZE;
1400 sub = lov_sub_get(env, lio, index);
1402 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
1403 plist, from, stripe_to, cb);
1409 if (plist->pl_nr > 0) /* short write */
1414 if (lov_comp_entry(index) !=
1415 lov_comp_entry(page->cp_lov_index))
1416 cl_io_extent_release(sub->sub_env, &sub->sub_io);
1419 /* for error case, add the page back into the qin list */
1420 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
1421 while (plist->pl_nr > 0) {
1422 /* error occurred, add the uncommitted pages back into queue */
1423 page = cl_page_list_last(plist);
1424 cl_page_list_move_head(queue, plist, page);
1430 static int lov_io_fault_start(const struct lu_env *env,
1431 const struct cl_io_slice *ios)
1433 struct cl_fault_io *fio;
1435 struct lov_io_sub *sub;
1442 fio = &ios->cis_io->u.ci_fault;
1443 lio = cl2lov_io(env, ios);
1446 * LU-14502: ft_page could be an existing cl_page associated with
1447 * the vmpage covering the fault index, and the page may still
1448 * refer to another mirror of an old IO.
1450 if (lov_is_flr(lio->lis_object)) {
1451 offset = cl_offset(ios->cis_obj, fio->ft_index);
1452 entry = lov_io_layout_at(lio, offset);
1454 CERROR(DFID": page fault index %lu invalid component: "
1456 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1457 fio->ft_index, entry,
1458 lio->lis_mirror_index);
1461 stripe = lov_stripe_number(lio->lis_object->lo_lsm,
1464 if (fio->ft_page->cp_lov_index !=
1465 lov_comp_index(entry, stripe)) {
1466 CDEBUG(D_INFO, DFID": page fault at index %lu, "
1467 "at mirror %u comp entry %u stripe %u, "
1468 "been used with comp entry %u stripe %u\n",
1469 PFID(lu_object_fid(&ios->cis_obj->co_lu)),
1470 fio->ft_index, lio->lis_mirror_index,
1472 lov_comp_entry(fio->ft_page->cp_lov_index),
1473 lov_comp_stripe(fio->ft_page->cp_lov_index));
1475 fio->ft_page->cp_lov_index =
1476 lov_comp_index(entry, stripe);
1480 sub = lov_sub_get(env, lio, fio->ft_page->cp_lov_index);
1481 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
1483 RETURN(lov_io_start(env, ios));
1486 static int lov_io_setattr_start(const struct lu_env *env,
1487 const struct cl_io_slice *ios)
1489 struct lov_io *lio = cl2lov_io(env, ios);
1490 struct cl_io *parent = ios->cis_io;
1491 struct lov_io_sub *sub;
1492 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1496 if (cl_io_is_fallocate(parent)) {
1497 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1498 loff_t size = parent->u.ci_setattr.sa_attr.lvb_size;
1499 int index = lov_comp_entry(sub->sub_subio_index);
1500 int stripe = lov_comp_stripe(sub->sub_subio_index);
1502 size = lov_size_to_stripe(lsm, index, size, stripe);
1503 sub->sub_io.u.ci_setattr.sa_attr.lvb_size = size;
1504 sub->sub_io.u.ci_setattr.sa_avalid =
1505 parent->u.ci_setattr.sa_avalid;
1509 RETURN(lov_io_start(env, ios));
1512 static void lov_io_fsync_end(const struct lu_env *env,
1513 const struct cl_io_slice *ios)
1515 struct lov_io *lio = cl2lov_io(env, ios);
1516 struct lov_io_sub *sub;
1517 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
1521 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1522 struct cl_io *subio = &sub->sub_io;
1524 lov_io_end_wrapper(sub->sub_env, subio);
1526 if (subio->ci_result == 0)
1527 *written += subio->u.ci_fsync.fi_nr_written;
1532 static void lov_io_lseek_end(const struct lu_env *env,
1533 const struct cl_io_slice *ios)
1535 struct lov_io *lio = cl2lov_io(env, ios);
1536 struct cl_io *io = lio->lis_cl.cis_io;
1537 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
1538 struct lov_io_sub *sub;
1539 loff_t offset = -ENXIO;
1541 bool seek_hole = io->u.ci_lseek.ls_whence == SEEK_HOLE;
1545 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
1546 struct cl_io *subio = &sub->sub_io;
1547 int index = lov_comp_entry(sub->sub_subio_index);
1548 int stripe = lov_comp_stripe(sub->sub_subio_index);
1549 loff_t sub_off, lov_off;
1550 __u64 comp_end = lsm->lsm_entries[index]->lsme_extent.e_end;
1552 lov_io_end_wrapper(sub->sub_env, subio);
1554 if (io->ci_result == 0)
1555 io->ci_result = sub->sub_io.ci_result;
1560 CDEBUG(D_INFO, DFID": entry %x stripe %u: SEEK_%s from %lld\n",
1561 PFID(lu_object_fid(lov2lu(lio->lis_object))),
1562 index, stripe, seek_hole ? "HOLE" : "DATA",
1563 subio->u.ci_lseek.ls_start);
1565 /* first subio with positive result is what we need */
1566 sub_off = subio->u.ci_lseek.ls_result;
1567 /* Expected error, offset is out of stripe file size */
1568 if (sub_off == -ENXIO)
1570 /* Any other errors are not expected with ci_result == 0 */
1572 CDEBUG(D_INFO, "unexpected error: rc = %lld\n",
1574 io->ci_result = sub_off;
1577 lov_off = lov_stripe_size(lsm, index, sub_off + 1, stripe) - 1;
1579 /* the only way to get negatove lov_off here is too big
1580 * result. Return -EOVERFLOW then.
1582 io->ci_result = -EOVERFLOW;
1583 CDEBUG(D_INFO, "offset %llu is too big: rc = %d\n",
1584 (u64)lov_off, io->ci_result);
1587 if (lov_off < io->u.ci_lseek.ls_start) {
1588 io->ci_result = -EINVAL;
1589 CDEBUG(D_INFO, "offset %lld < start %lld: rc = %d\n",
1590 sub_off, io->u.ci_lseek.ls_start, io->ci_result);
1593 /* resulting offset can be out of component range if stripe
1594 * object is full and its file size was returned as virtual
1595 * hole start. Skip this result, the next component will give
1596 * us correct lseek result but keep possible hole offset in
1597 * case there is no more components ahead
1599 if (lov_off >= comp_end) {
1600 /* must be SEEK_HOLE case */
1601 if (likely(seek_hole)) {
1602 /* save comp end as potential hole offset */
1603 hole_off = max_t(__u64, comp_end, hole_off);
1605 io->ci_result = -EINVAL;
1607 "off %lld >= comp_end %llu: rc = %d\n",
1608 lov_off, comp_end, io->ci_result);
1613 CDEBUG(D_INFO, "SEEK_%s: %lld->%lld/%lld: rc = %d\n",
1614 seek_hole ? "HOLE" : "DATA",
1615 subio->u.ci_lseek.ls_start, sub_off, lov_off,
1616 sub->sub_io.ci_result);
1617 offset = min_t(__u64, offset, lov_off);
1619 /* no result but some component returns hole as component end */
1620 if (seek_hole && offset == -ENXIO && hole_off > 0)
1623 io->u.ci_lseek.ls_result = offset;
1627 static const struct cl_io_operations lov_io_ops = {
1630 .cio_fini = lov_io_fini,
1631 .cio_iter_init = lov_io_rw_iter_init,
1632 .cio_iter_fini = lov_io_iter_fini,
1633 .cio_lock = lov_io_lock,
1634 .cio_unlock = lov_io_unlock,
1635 .cio_start = lov_io_start,
1636 .cio_end = lov_io_end
1639 .cio_fini = lov_io_fini,
1640 .cio_iter_init = lov_io_rw_iter_init,
1641 .cio_iter_fini = lov_io_iter_fini,
1642 .cio_lock = lov_io_lock,
1643 .cio_unlock = lov_io_unlock,
1644 .cio_start = lov_io_start,
1645 .cio_end = lov_io_end
1648 .cio_fini = lov_io_fini,
1649 .cio_iter_init = lov_io_setattr_iter_init,
1650 .cio_iter_fini = lov_io_iter_fini,
1651 .cio_lock = lov_io_lock,
1652 .cio_unlock = lov_io_unlock,
1653 .cio_start = lov_io_setattr_start,
1654 .cio_end = lov_io_end
1656 [CIT_DATA_VERSION] = {
1657 .cio_fini = lov_io_fini,
1658 .cio_iter_init = lov_io_iter_init,
1659 .cio_iter_fini = lov_io_iter_fini,
1660 .cio_lock = lov_io_lock,
1661 .cio_unlock = lov_io_unlock,
1662 .cio_start = lov_io_start,
1663 .cio_end = lov_io_data_version_end,
1666 .cio_fini = lov_io_fini,
1667 .cio_iter_init = lov_io_iter_init,
1668 .cio_iter_fini = lov_io_iter_fini,
1669 .cio_lock = lov_io_lock,
1670 .cio_unlock = lov_io_unlock,
1671 .cio_start = lov_io_fault_start,
1672 .cio_end = lov_io_end
1675 .cio_fini = lov_io_fini,
1676 .cio_iter_init = lov_io_iter_init,
1677 .cio_iter_fini = lov_io_iter_fini,
1678 .cio_lock = lov_io_lock,
1679 .cio_unlock = lov_io_unlock,
1680 .cio_start = lov_io_start,
1681 .cio_end = lov_io_fsync_end
1684 .cio_fini = lov_io_fini,
1685 .cio_iter_init = lov_io_iter_init,
1686 .cio_iter_fini = lov_io_iter_fini,
1687 .cio_lock = lov_io_lock,
1688 .cio_unlock = lov_io_unlock,
1689 .cio_start = lov_io_start,
1690 .cio_end = lov_io_end
1693 .cio_fini = lov_io_fini,
1694 .cio_iter_init = lov_io_iter_init,
1695 .cio_iter_fini = lov_io_iter_fini,
1696 .cio_lock = lov_io_lock,
1697 .cio_unlock = lov_io_unlock,
1698 .cio_start = lov_io_start,
1699 .cio_end = lov_io_lseek_end
1702 .cio_fini = lov_io_fini,
1705 .cio_fini = lov_io_fini
1708 .cio_read_ahead = lov_io_read_ahead,
1709 .cio_lru_reserve = lov_io_lru_reserve,
1710 .cio_submit = lov_io_submit,
1711 .cio_commit_async = lov_io_commit_async,
1714 /*****************************************************************************
1716 * Empty lov io operations.
1720 static void lov_empty_io_fini(const struct lu_env *env,
1721 const struct cl_io_slice *ios)
1723 struct lov_object *lov = cl2lov(ios->cis_obj);
1726 if (atomic_dec_and_test(&lov->lo_active_ios))
1727 wake_up(&lov->lo_waitq);
1731 static int lov_empty_io_submit(const struct lu_env *env,
1732 const struct cl_io_slice *ios,
1733 enum cl_req_type crt, struct cl_2queue *queue)
1738 static void lov_empty_impossible(const struct lu_env *env,
1739 struct cl_io_slice *ios)
1744 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1747 * An io operation vector for files without stripes.
1749 static const struct cl_io_operations lov_empty_io_ops = {
1752 .cio_fini = lov_empty_io_fini,
1754 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1755 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1756 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1757 .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,
1776 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1777 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1778 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1779 .cio_end = LOV_EMPTY_IMPOSSIBLE
1782 .cio_fini = lov_empty_io_fini
1785 .cio_fini = lov_empty_io_fini
1788 .cio_fini = lov_empty_io_fini
1791 .cio_fini = lov_empty_io_fini
1794 .cio_submit = lov_empty_io_submit,
1795 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1798 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1801 struct lov_io *lio = lov_env_io(env);
1802 struct lov_object *lov = cl2lov(obj);
1807 INIT_LIST_HEAD(&lio->lis_active);
1808 result = lov_io_slice_init(lio, lov, io);
1812 result = lov_io_subio_init(env, lio, io);
1814 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1815 atomic_inc(&lov->lo_active_ios);
1819 io->ci_result = result < 0 ? result : 0;
1823 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1826 struct lov_object *lov = cl2lov(obj);
1827 struct lov_io *lio = lov_env_io(env);
1831 lio->lis_object = lov;
1832 switch (io->ci_type) {
1844 case CIT_DATA_VERSION:
1852 CERROR("Page fault on a file without stripes: "DFID"\n",
1853 PFID(lu_object_fid(&obj->co_lu)));
1857 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1858 atomic_inc(&lov->lo_active_ios);
1861 io->ci_result = result < 0 ? result : 0;
1865 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1868 struct lov_object *lov = cl2lov(obj);
1869 struct lov_io *lio = lov_env_io(env);
1873 LASSERT(lov->lo_lsm != NULL);
1874 lio->lis_object = lov;
1876 switch (io->ci_type) {
1878 LASSERTF(0, "invalid type %d\n", io->ci_type);
1879 result = -EOPNOTSUPP;
1885 case CIT_DATA_VERSION:
1890 * the truncate to 0 is managed by MDT:
1891 * - in open, for open O_TRUNC
1892 * - in setattr, for truncate
1895 * the truncate is for size > 0 so triggers a restore,
1896 * also trigger a restore for prealloc/punch
1898 if (cl_io_is_trunc(io) || cl_io_is_fallocate(io)) {
1899 io->ci_restore_needed = 1;
1908 io->ci_restore_needed = 1;
1914 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1915 atomic_inc(&lov->lo_active_ios);
1918 io->ci_result = result < 0 ? result : 0;
1923 * Return the index in composite:lo_entries by the file offset
1925 int lov_io_layout_at(struct lov_io *lio, __u64 offset)
1927 struct lov_object *lov = lio->lis_object;
1928 struct lov_layout_composite *comp = &lov->u.composite;
1929 int start_index = 0;
1930 int end_index = comp->lo_entry_count - 1;
1933 LASSERT(lov->lo_type == LLT_COMP);
1935 /* This is actual file offset so nothing can cover eof. */
1936 if (offset == LUSTRE_EOF)
1939 if (lov_is_flr(lov)) {
1940 struct lov_mirror_entry *lre;
1942 LASSERT(lio->lis_mirror_index >= 0);
1944 lre = &comp->lo_mirrors[lio->lis_mirror_index];
1945 start_index = lre->lre_start;
1946 end_index = lre->lre_end;
1949 for (i = start_index; i <= end_index; i++) {
1950 struct lov_layout_entry *lle = lov_entry(lov, i);
1952 LASSERT(!lsme_is_foreign(lle->lle_lsme));
1954 if ((offset >= lle->lle_extent->e_start &&
1955 offset < lle->lle_extent->e_end) ||
1956 (offset == OBD_OBJECT_EOF &&
1957 lle->lle_extent->e_end == OBD_OBJECT_EOF))