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, 2016, 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 != NULL && !IS_ERR(sub->sub_env)) {
86 cl_env_put(sub->sub_env, &sub->sub_refcheck);
92 static int lov_io_sub_init(const struct lu_env *env, struct lov_io *lio,
93 struct lov_io_sub *sub)
95 struct lov_object *lov = lio->lis_object;
97 struct cl_object *sub_obj;
98 struct cl_io *io = lio->lis_cl.cis_io;
99 int index = lov_comp_entry(sub->sub_subio_index);
100 int stripe = lov_comp_stripe(sub->sub_subio_index);
102 LASSERT(sub->sub_env == NULL);
105 if (unlikely(!lov_r0(lov, index)->lo_sub ||
106 !lov_r0(lov, index)->lo_sub[stripe]))
109 /* obtain new environment */
110 sub->sub_env = cl_env_get(&sub->sub_refcheck);
111 if (IS_ERR(sub->sub_env))
112 result = PTR_ERR(sub->sub_env);
114 sub_obj = lovsub2cl(lov_r0(lov, index)->lo_sub[stripe]);
115 sub_io = &sub->sub_io;
117 sub_io->ci_obj = sub_obj;
118 sub_io->ci_result = 0;
120 sub_io->ci_parent = io;
121 sub_io->ci_lockreq = io->ci_lockreq;
122 sub_io->ci_type = io->ci_type;
123 sub_io->ci_no_srvlock = io->ci_no_srvlock;
124 sub_io->ci_noatime = io->ci_noatime;
125 sub_io->ci_pio = io->ci_pio;
126 sub_io->ci_lock_no_expand = io->ci_lock_no_expand;
128 result = cl_io_sub_init(sub->sub_env, sub_io, io->ci_type, sub_obj);
131 lov_io_sub_fini(env, lio, sub);
136 struct lov_io_sub *lov_sub_get(const struct lu_env *env,
137 struct lov_io *lio, int index)
139 struct lov_io_sub *sub;
144 list_for_each_entry(sub, &lio->lis_subios, sub_list) {
145 if (sub->sub_subio_index == index) {
152 sub = lov_sub_alloc(lio, index);
154 GOTO(out, rc = -ENOMEM);
156 rc = lov_io_sub_init(env, lio, sub);
158 lov_sub_free(lio, sub);
162 list_add_tail(&sub->sub_list, &lio->lis_subios);
163 lio->lis_nr_subios++;
171 /*****************************************************************************
177 int lov_page_index(const struct cl_page *page)
179 const struct cl_page_slice *slice;
182 slice = cl_page_at(page, &lov_device_type);
183 LASSERT(slice != NULL);
184 LASSERT(slice->cpl_obj != NULL);
186 RETURN(cl2lov_page(slice)->lps_index);
189 static int lov_io_subio_init(const struct lu_env *env, struct lov_io *lio,
194 LASSERT(lio->lis_object != NULL);
196 INIT_LIST_HEAD(&lio->lis_subios);
197 lio->lis_single_subio_index = -1;
198 lio->lis_nr_subios = 0;
203 static int lov_io_slice_init(struct lov_io *lio,
204 struct lov_object *obj, struct cl_io *io)
209 lio->lis_object = obj;
211 LASSERT(obj->lo_lsm != NULL);
213 switch (io->ci_type) {
216 lio->lis_pos = io->u.ci_rw.rw_range.cir_pos;
217 lio->lis_endpos = lio->lis_pos + io->u.ci_rw.rw_range.cir_count;
218 lio->lis_io_endpos = lio->lis_endpos;
219 if (cl_io_is_append(io)) {
220 LASSERT(io->ci_type == CIT_WRITE);
222 /* If there is LOV EA hole, then we may cannot locate
223 * the current file-tail exactly. */
224 if (unlikely(obj->lo_lsm->lsm_entries[0]->lsme_pattern &
229 lio->lis_endpos = OBD_OBJECT_EOF;
234 if (cl_io_is_trunc(io))
235 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
238 lio->lis_endpos = OBD_OBJECT_EOF;
241 case CIT_DATA_VERSION:
243 lio->lis_endpos = OBD_OBJECT_EOF;
247 pgoff_t index = io->u.ci_fault.ft_index;
248 lio->lis_pos = cl_offset(io->ci_obj, index);
249 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
254 lio->lis_pos = io->u.ci_fsync.fi_start;
255 lio->lis_endpos = io->u.ci_fsync.fi_end;
260 lio->lis_pos = io->u.ci_ladvise.li_start;
261 lio->lis_endpos = io->u.ci_ladvise.li_end;
268 lio->lis_endpos = OBD_OBJECT_EOF;
278 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
280 struct lov_io *lio = cl2lov_io(env, ios);
281 struct lov_object *lov = cl2lov(ios->cis_obj);
285 LASSERT(list_empty(&lio->lis_active));
287 while (!list_empty(&lio->lis_subios)) {
288 struct lov_io_sub *sub = list_entry(lio->lis_subios.next,
292 list_del_init(&sub->sub_list);
293 lio->lis_nr_subios--;
295 lov_io_sub_fini(env, lio, sub);
296 lov_sub_free(lio, sub);
298 LASSERT(lio->lis_nr_subios == 0);
300 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
301 if (atomic_dec_and_test(&lov->lo_active_ios))
302 wake_up_all(&lov->lo_waitq);
306 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
307 loff_t start, loff_t end)
309 struct cl_io *io = &sub->sub_io;
310 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
311 struct cl_io *parent = lio->lis_cl.cis_io;
312 int index = lov_comp_entry(sub->sub_subio_index);
313 int stripe = lov_comp_stripe(sub->sub_subio_index);
315 io->ci_pio = parent->ci_pio;
316 switch (io->ci_type) {
318 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
319 io->u.ci_setattr.sa_attr_flags =
320 parent->u.ci_setattr.sa_attr_flags;
321 io->u.ci_setattr.sa_valid = parent->u.ci_setattr.sa_valid;
322 io->u.ci_setattr.sa_stripe_index = stripe;
323 io->u.ci_setattr.sa_parent_fid =
324 parent->u.ci_setattr.sa_parent_fid;
325 if (cl_io_is_trunc(io)) {
326 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
328 new_size = lov_size_to_stripe(lsm, index, new_size,
330 io->u.ci_setattr.sa_attr.lvb_size = new_size;
332 lov_lsm2layout(lsm, lsm->lsm_entries[index],
333 &io->u.ci_setattr.sa_layout);
336 case CIT_DATA_VERSION: {
337 io->u.ci_data_version.dv_data_version = 0;
338 io->u.ci_data_version.dv_flags =
339 parent->u.ci_data_version.dv_flags;
343 struct cl_object *obj = parent->ci_obj;
344 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
346 io->u.ci_fault = parent->u.ci_fault;
347 off = lov_size_to_stripe(lsm, index, off, stripe);
348 io->u.ci_fault.ft_index = cl_index(obj, off);
352 io->u.ci_fsync.fi_start = start;
353 io->u.ci_fsync.fi_end = end;
354 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
355 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
360 io->u.ci_rw.rw_ptask = parent->u.ci_rw.rw_ptask;
361 io->u.ci_rw.rw_iter = parent->u.ci_rw.rw_iter;
362 io->u.ci_rw.rw_iocb = parent->u.ci_rw.rw_iocb;
363 io->u.ci_rw.rw_file = parent->u.ci_rw.rw_file;
364 io->u.ci_rw.rw_sync = parent->u.ci_rw.rw_sync;
365 if (cl_io_is_append(parent)) {
366 io->u.ci_rw.rw_append = 1;
368 io->u.ci_rw.rw_range.cir_pos = start;
369 io->u.ci_rw.rw_range.cir_count = end - start;
374 io->u.ci_ladvise.li_start = start;
375 io->u.ci_ladvise.li_end = end;
376 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
377 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
378 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
388 static loff_t lov_offset_mod(loff_t val, int delta)
390 if (val != OBD_OBJECT_EOF)
395 static int lov_io_iter_init(const struct lu_env *env,
396 const struct cl_io_slice *ios)
398 struct cl_io *io = ios->cis_io;
399 struct lov_io *lio = cl2lov_io(env, ios);
400 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
401 struct lov_io_sub *sub;
402 struct lov_layout_entry *le;
403 struct lu_extent ext;
409 ext.e_start = lio->lis_pos;
410 ext.e_end = lio->lis_endpos;
413 lov_foreach_layout_entry(lio->lis_object, le) {
414 struct lov_layout_raid0 *r0 = &le->lle_raid0;
420 if (!lu_extent_is_overlapped(&ext, &le->lle_extent))
423 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
424 index - 1, lsm->lsm_entries[index - 1]->lsme_flags);
425 if (!lsm_entry_inited(lsm, index - 1)) {
426 /* truncate IO will trigger write intent as well, and
427 * it's handled in lov_io_setattr_iter_init() */
428 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io)) {
429 io->ci_need_write_intent = 1;
430 /* execute it in main thread */
436 /* Read from uninitialized components should return
437 * zero filled pages. */
441 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
442 if (!lov_stripe_intersects(lsm, index - 1, stripe,
446 if (unlikely(r0->lo_sub[stripe] == NULL)) {
447 if (ios->cis_io->ci_type == CIT_READ ||
448 ios->cis_io->ci_type == CIT_WRITE ||
449 ios->cis_io->ci_type == CIT_FAULT)
455 end = lov_offset_mod(end, 1);
456 sub = lov_sub_get(env, lio,
457 lov_comp_index(index - 1, stripe));
463 lov_io_sub_inherit(sub, lio, start, end);
464 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
466 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
471 "shrink stripe: {%d, %d} range: [%llu, %llu)\n",
472 index, stripe, start, end);
474 list_add_tail(&sub->sub_linkage, &lio->lis_active);
482 static int lov_io_rw_iter_init(const struct lu_env *env,
483 const struct cl_io_slice *ios)
485 struct cl_io *io = ios->cis_io;
486 struct lov_io *lio = cl2lov_io(env, ios);
487 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
488 struct lov_stripe_md_entry *lse;
489 struct cl_io_range *range = &io->u.ci_rw.rw_range;
490 loff_t start = range->cir_pos;
494 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
497 if (cl_io_is_append(io))
498 RETURN(lov_io_iter_init(env, ios));
500 index = lov_lsm_entry(lsm, range->cir_pos);
501 if (index < 0) { /* non-existing layout component */
502 if (io->ci_type == CIT_READ) {
503 /* TODO: it needs to detect the next component and
504 * then set the next pos */
506 /* execute it in main thread */
509 RETURN(lov_io_iter_init(env, ios));
515 lse = lov_lse(lio->lis_object, index);
517 next = MAX_LFS_FILESIZE;
518 if (lse->lsme_stripe_count > 1) {
519 unsigned long ssize = lse->lsme_stripe_size;
521 lov_do_div64(start, ssize);
522 next = (start + 1) * ssize;
523 if (next <= start * ssize)
524 next = MAX_LFS_FILESIZE;
527 LASSERTF(range->cir_pos >= lse->lsme_extent.e_start,
528 "pos %lld, [%lld, %lld)\n", range->cir_pos,
529 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
530 next = min_t(__u64, next, lse->lsme_extent.e_end);
531 next = min_t(loff_t, next, lio->lis_io_endpos);
533 io->ci_continue = next < lio->lis_io_endpos;
534 range->cir_count = next - range->cir_pos;
535 lio->lis_pos = range->cir_pos;
536 lio->lis_endpos = range->cir_pos + range->cir_count;
538 "stripe: {%d, %llu} range: [%llu, %llu) end: %llu, count: %zd\n",
539 index, start, lio->lis_pos, lio->lis_endpos,
540 lio->lis_io_endpos, range->cir_count);
542 if (!io->ci_continue) {
543 /* the last piece of IO, execute it in main thread */
548 /* it only splits IO here for parallel IO,
549 * there will be no actual IO going to occur,
550 * so it doesn't need to invoke lov_io_iter_init()
551 * to initialize sub IOs. */
552 if (!lsm_entry_inited(lsm, index)) {
553 io->ci_need_write_intent = 1;
560 * XXX The following call should be optimized: we know, that
561 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
563 RETURN(lov_io_iter_init(env, ios));
566 static int lov_io_setattr_iter_init(const struct lu_env *env,
567 const struct cl_io_slice *ios)
569 struct lov_io *lio = cl2lov_io(env, ios);
570 struct cl_io *io = ios->cis_io;
571 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
575 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
576 index = lov_lsm_entry(lsm, lio->lis_pos - 1);
577 /* no entry found for such offset */
579 RETURN(io->ci_result = -ENODATA);
580 } else if (!lsm_entry_inited(lsm, index)) {
581 io->ci_need_write_intent = 1;
582 RETURN(io->ci_result = -ENODATA);
586 RETURN(lov_io_iter_init(env, ios));
589 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
590 int (*iofunc)(const struct lu_env *, struct cl_io *))
592 struct cl_io *parent = lio->lis_cl.cis_io;
593 struct lov_io_sub *sub;
597 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
598 rc = iofunc(sub->sub_env, &sub->sub_io);
602 if (parent->ci_result == 0)
603 parent->ci_result = sub->sub_io.ci_result;
608 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
611 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
614 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
617 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
620 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
624 * It's possible that lov_io_start() wasn't called against this
625 * sub-io, either because previous sub-io failed, or upper layer
628 if (io->ci_state == CIS_IO_GOING)
631 io->ci_state = CIS_IO_FINISHED;
635 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
637 cl_io_iter_fini(env, io);
641 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
643 cl_io_unlock(env, io);
647 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
651 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
656 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
658 struct lov_io *lio = cl2lov_io(env, ios);
659 struct cl_io *parent = lio->lis_cl.cis_io;
660 struct lov_io_sub *sub;
663 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
664 lov_io_end_wrapper(env, &sub->sub_io);
666 parent->u.ci_data_version.dv_data_version +=
667 sub->sub_io.u.ci_data_version.dv_data_version;
669 if (parent->ci_result == 0)
670 parent->ci_result = sub->sub_io.ci_result;
676 static void lov_io_iter_fini(const struct lu_env *env,
677 const struct cl_io_slice *ios)
679 struct lov_io *lio = cl2lov_io(env, ios);
683 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
685 while (!list_empty(&lio->lis_active))
686 list_del_init(lio->lis_active.next);
690 static void lov_io_unlock(const struct lu_env *env,
691 const struct cl_io_slice *ios)
696 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
701 static int lov_io_read_ahead(const struct lu_env *env,
702 const struct cl_io_slice *ios,
703 pgoff_t start, struct cl_read_ahead *ra)
705 struct lov_io *lio = cl2lov_io(env, ios);
706 struct lov_object *loo = lio->lis_object;
707 struct cl_object *obj = lov2cl(loo);
708 struct lov_layout_raid0 *r0;
709 struct lov_io_sub *sub;
713 unsigned int pps; /* pages per stripe */
719 offset = cl_offset(obj, start);
720 index = lov_lsm_entry(loo->lo_lsm, offset);
721 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index))
724 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
726 r0 = lov_r0(loo, index);
727 if (unlikely(r0->lo_sub[stripe] == NULL))
730 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
732 RETURN(PTR_ERR(sub));
734 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
735 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
736 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
739 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
740 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end, r0->lo_nr, rc);
745 * Adjust the stripe index by layout of comp. ra->cra_end is the
746 * maximum page index covered by an underlying DLM lock.
747 * This function converts cra_end from stripe level to file level, and
748 * make sure it's not beyond stripe and component boundary.
751 /* cra_end is stripe level, convert it into file level */
752 ra_end = ra->cra_end;
753 if (ra_end != CL_PAGE_EOF)
754 ra->cra_end = lov_stripe_pgoff(loo->lo_lsm, index,
757 /* boundary of current component */
758 ra_end = cl_index(obj, (loff_t)lov_lse(loo, index)->lsme_extent.e_end);
759 if (ra_end != CL_PAGE_EOF && ra->cra_end >= ra_end)
760 ra->cra_end = ra_end - 1;
762 if (r0->lo_nr == 1) /* single stripe file */
765 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
767 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %u, "
768 "stripe_size = %u, stripe no = %u, start index = %lu\n",
769 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end, pps, index,
770 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
772 /* never exceed the end of the stripe */
773 ra->cra_end = min_t(pgoff_t,
774 ra->cra_end, start + pps - start % pps - 1);
779 * lov implementation of cl_operations::cio_submit() method. It takes a list
780 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
781 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
784 * Major complication of this function is a need to handle memory cleansing:
785 * cl_io_submit() is called to write out pages as a part of VM memory
786 * reclamation, and hence it may not fail due to memory shortages (system
787 * dead-locks otherwise). To deal with this, some resources (sub-lists,
788 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
789 * not-memory cleansing context), and in case of memory shortage, these
790 * pre-allocated resources are used by lov_io_submit() under
791 * lov_device::ld_mutex mutex.
793 static int lov_io_submit(const struct lu_env *env,
794 const struct cl_io_slice *ios,
795 enum cl_req_type crt, struct cl_2queue *queue)
797 struct cl_page_list *qin = &queue->c2_qin;
798 struct lov_io *lio = cl2lov_io(env, ios);
799 struct lov_io_sub *sub;
800 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
801 struct cl_page *page;
806 if (lio->lis_nr_subios == 1) {
807 int idx = lio->lis_single_subio_index;
809 sub = lov_sub_get(env, lio, idx);
810 LASSERT(!IS_ERR(sub));
811 LASSERT(sub == &lio->lis_single_subio);
812 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
817 cl_page_list_init(plist);
818 while (qin->pl_nr > 0) {
819 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
821 cl_2queue_init(cl2q);
823 page = cl_page_list_first(qin);
824 cl_page_list_move(&cl2q->c2_qin, qin, page);
826 index = lov_page_index(page);
827 while (qin->pl_nr > 0) {
828 page = cl_page_list_first(qin);
829 if (index != lov_page_index(page))
832 cl_page_list_move(&cl2q->c2_qin, qin, page);
835 sub = lov_sub_get(env, lio, index);
837 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
843 cl_page_list_splice(&cl2q->c2_qin, plist);
844 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
845 cl_2queue_fini(env, cl2q);
851 cl_page_list_splice(plist, qin);
852 cl_page_list_fini(env, plist);
857 static int lov_io_commit_async(const struct lu_env *env,
858 const struct cl_io_slice *ios,
859 struct cl_page_list *queue, int from, int to,
862 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
863 struct lov_io *lio = cl2lov_io(env, ios);
864 struct lov_io_sub *sub;
865 struct cl_page *page;
869 if (lio->lis_nr_subios == 1) {
870 int idx = lio->lis_single_subio_index;
872 sub = lov_sub_get(env, lio, idx);
873 LASSERT(!IS_ERR(sub));
874 LASSERT(sub == &lio->lis_single_subio);
875 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
880 cl_page_list_init(plist);
881 while (queue->pl_nr > 0) {
885 LASSERT(plist->pl_nr == 0);
886 page = cl_page_list_first(queue);
887 cl_page_list_move(plist, queue, page);
889 index = lov_page_index(page);
890 while (queue->pl_nr > 0) {
891 page = cl_page_list_first(queue);
892 if (index != lov_page_index(page))
895 cl_page_list_move(plist, queue, page);
898 if (queue->pl_nr > 0) /* still has more pages */
899 stripe_to = PAGE_SIZE;
901 sub = lov_sub_get(env, lio, index);
903 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
904 plist, from, stripe_to, cb);
910 if (plist->pl_nr > 0) /* short write */
916 /* for error case, add the page back into the qin list */
917 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
918 while (plist->pl_nr > 0) {
919 /* error occurred, add the uncommitted pages back into queue */
920 page = cl_page_list_last(plist);
921 cl_page_list_move_head(queue, plist, page);
927 static int lov_io_fault_start(const struct lu_env *env,
928 const struct cl_io_slice *ios)
930 struct cl_fault_io *fio;
932 struct lov_io_sub *sub;
936 fio = &ios->cis_io->u.ci_fault;
937 lio = cl2lov_io(env, ios);
938 sub = lov_sub_get(env, lio, lov_page_index(fio->ft_page));
939 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
941 RETURN(lov_io_start(env, ios));
944 static void lov_io_fsync_end(const struct lu_env *env,
945 const struct cl_io_slice *ios)
947 struct lov_io *lio = cl2lov_io(env, ios);
948 struct lov_io_sub *sub;
949 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
953 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
954 struct cl_io *subio = &sub->sub_io;
956 lov_io_end_wrapper(sub->sub_env, subio);
958 if (subio->ci_result == 0)
959 *written += subio->u.ci_fsync.fi_nr_written;
964 static const struct cl_io_operations lov_io_ops = {
967 .cio_fini = lov_io_fini,
968 .cio_iter_init = lov_io_rw_iter_init,
969 .cio_iter_fini = lov_io_iter_fini,
970 .cio_lock = lov_io_lock,
971 .cio_unlock = lov_io_unlock,
972 .cio_start = lov_io_start,
973 .cio_end = lov_io_end
976 .cio_fini = lov_io_fini,
977 .cio_iter_init = lov_io_rw_iter_init,
978 .cio_iter_fini = lov_io_iter_fini,
979 .cio_lock = lov_io_lock,
980 .cio_unlock = lov_io_unlock,
981 .cio_start = lov_io_start,
982 .cio_end = lov_io_end
985 .cio_fini = lov_io_fini,
986 .cio_iter_init = lov_io_setattr_iter_init,
987 .cio_iter_fini = lov_io_iter_fini,
988 .cio_lock = lov_io_lock,
989 .cio_unlock = lov_io_unlock,
990 .cio_start = lov_io_start,
991 .cio_end = lov_io_end
993 [CIT_DATA_VERSION] = {
994 .cio_fini = lov_io_fini,
995 .cio_iter_init = lov_io_iter_init,
996 .cio_iter_fini = lov_io_iter_fini,
997 .cio_lock = lov_io_lock,
998 .cio_unlock = lov_io_unlock,
999 .cio_start = lov_io_start,
1000 .cio_end = lov_io_data_version_end,
1003 .cio_fini = lov_io_fini,
1004 .cio_iter_init = lov_io_iter_init,
1005 .cio_iter_fini = lov_io_iter_fini,
1006 .cio_lock = lov_io_lock,
1007 .cio_unlock = lov_io_unlock,
1008 .cio_start = lov_io_fault_start,
1009 .cio_end = lov_io_end
1012 .cio_fini = lov_io_fini,
1013 .cio_iter_init = lov_io_iter_init,
1014 .cio_iter_fini = lov_io_iter_fini,
1015 .cio_lock = lov_io_lock,
1016 .cio_unlock = lov_io_unlock,
1017 .cio_start = lov_io_start,
1018 .cio_end = lov_io_fsync_end
1021 .cio_fini = lov_io_fini,
1022 .cio_iter_init = lov_io_iter_init,
1023 .cio_iter_fini = lov_io_iter_fini,
1024 .cio_lock = lov_io_lock,
1025 .cio_unlock = lov_io_unlock,
1026 .cio_start = lov_io_start,
1027 .cio_end = lov_io_end
1030 .cio_fini = lov_io_fini,
1033 .cio_fini = lov_io_fini
1036 .cio_read_ahead = lov_io_read_ahead,
1037 .cio_submit = lov_io_submit,
1038 .cio_commit_async = lov_io_commit_async,
1041 /*****************************************************************************
1043 * Empty lov io operations.
1047 static void lov_empty_io_fini(const struct lu_env *env,
1048 const struct cl_io_slice *ios)
1050 struct lov_object *lov = cl2lov(ios->cis_obj);
1053 if (atomic_dec_and_test(&lov->lo_active_ios))
1054 wake_up_all(&lov->lo_waitq);
1058 static int lov_empty_io_submit(const struct lu_env *env,
1059 const struct cl_io_slice *ios,
1060 enum cl_req_type crt, struct cl_2queue *queue)
1065 static void lov_empty_impossible(const struct lu_env *env,
1066 struct cl_io_slice *ios)
1071 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1074 * An io operation vector for files without stripes.
1076 static const struct cl_io_operations lov_empty_io_ops = {
1079 .cio_fini = lov_empty_io_fini,
1081 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1082 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1083 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1084 .cio_end = LOV_EMPTY_IMPOSSIBLE
1088 .cio_fini = lov_empty_io_fini,
1089 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1090 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1091 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1092 .cio_end = LOV_EMPTY_IMPOSSIBLE
1095 .cio_fini = lov_empty_io_fini,
1096 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1097 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1098 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1099 .cio_end = LOV_EMPTY_IMPOSSIBLE
1102 .cio_fini = lov_empty_io_fini,
1103 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1104 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1105 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1106 .cio_end = LOV_EMPTY_IMPOSSIBLE
1109 .cio_fini = lov_empty_io_fini
1112 .cio_fini = lov_empty_io_fini
1115 .cio_fini = lov_empty_io_fini
1118 .cio_fini = lov_empty_io_fini
1121 .cio_submit = lov_empty_io_submit,
1122 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1125 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1128 struct lov_io *lio = lov_env_io(env);
1129 struct lov_object *lov = cl2lov(obj);
1132 INIT_LIST_HEAD(&lio->lis_active);
1133 io->ci_result = lov_io_slice_init(lio, lov, io);
1134 if (io->ci_result != 0)
1135 RETURN(io->ci_result);
1137 if (io->ci_result == 0) {
1138 io->ci_result = lov_io_subio_init(env, lio, io);
1139 if (io->ci_result == 0) {
1140 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1141 atomic_inc(&lov->lo_active_ios);
1144 RETURN(io->ci_result);
1147 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1150 struct lov_object *lov = cl2lov(obj);
1151 struct lov_io *lio = lov_env_io(env);
1155 lio->lis_object = lov;
1156 switch (io->ci_type) {
1167 case CIT_DATA_VERSION:
1175 CERROR("Page fault on a file without stripes: "DFID"\n",
1176 PFID(lu_object_fid(&obj->co_lu)));
1180 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1181 atomic_inc(&lov->lo_active_ios);
1184 io->ci_result = result < 0 ? result : 0;
1188 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1191 struct lov_object *lov = cl2lov(obj);
1192 struct lov_io *lio = lov_env_io(env);
1196 LASSERT(lov->lo_lsm != NULL);
1197 lio->lis_object = lov;
1199 switch (io->ci_type) {
1201 LASSERTF(0, "invalid type %d\n", io->ci_type);
1202 result = -EOPNOTSUPP;
1208 case CIT_DATA_VERSION:
1212 /* the truncate to 0 is managed by MDT:
1213 * - in open, for open O_TRUNC
1214 * - in setattr, for truncate
1216 /* the truncate is for size > 0 so triggers a restore */
1217 if (cl_io_is_trunc(io)) {
1218 io->ci_restore_needed = 1;
1226 io->ci_restore_needed = 1;
1232 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1233 atomic_inc(&lov->lo_active_ios);
1236 io->ci_result = result < 0 ? result : 0;