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;
267 lio->lis_endpos = OBD_OBJECT_EOF;
269 if ((obj->lo_lsm->lsm_flags & LCM_FL_FLR_MASK) == LCM_FL_RDONLY)
270 RETURN(1); /* SoM is accurate, no need glimpse */
275 lio->lis_endpos = OBD_OBJECT_EOF;
285 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
287 struct lov_io *lio = cl2lov_io(env, ios);
288 struct lov_object *lov = cl2lov(ios->cis_obj);
292 LASSERT(list_empty(&lio->lis_active));
294 while (!list_empty(&lio->lis_subios)) {
295 struct lov_io_sub *sub = list_entry(lio->lis_subios.next,
299 list_del_init(&sub->sub_list);
300 lio->lis_nr_subios--;
302 lov_io_sub_fini(env, lio, sub);
303 lov_sub_free(lio, sub);
305 LASSERT(lio->lis_nr_subios == 0);
307 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
308 if (atomic_dec_and_test(&lov->lo_active_ios))
309 wake_up_all(&lov->lo_waitq);
313 static void lov_io_sub_inherit(struct lov_io_sub *sub, struct lov_io *lio,
314 loff_t start, loff_t end)
316 struct cl_io *io = &sub->sub_io;
317 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
318 struct cl_io *parent = lio->lis_cl.cis_io;
319 int index = lov_comp_entry(sub->sub_subio_index);
320 int stripe = lov_comp_stripe(sub->sub_subio_index);
322 io->ci_pio = parent->ci_pio;
323 switch (io->ci_type) {
325 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
326 io->u.ci_setattr.sa_attr_flags =
327 parent->u.ci_setattr.sa_attr_flags;
328 io->u.ci_setattr.sa_valid = parent->u.ci_setattr.sa_valid;
329 io->u.ci_setattr.sa_stripe_index = stripe;
330 io->u.ci_setattr.sa_parent_fid =
331 parent->u.ci_setattr.sa_parent_fid;
332 if (cl_io_is_trunc(io)) {
333 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
335 new_size = lov_size_to_stripe(lsm, index, new_size,
337 io->u.ci_setattr.sa_attr.lvb_size = new_size;
339 lov_lsm2layout(lsm, lsm->lsm_entries[index],
340 &io->u.ci_setattr.sa_layout);
343 case CIT_DATA_VERSION: {
344 io->u.ci_data_version.dv_data_version = 0;
345 io->u.ci_data_version.dv_flags =
346 parent->u.ci_data_version.dv_flags;
350 struct cl_object *obj = parent->ci_obj;
351 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
353 io->u.ci_fault = parent->u.ci_fault;
354 off = lov_size_to_stripe(lsm, index, off, stripe);
355 io->u.ci_fault.ft_index = cl_index(obj, off);
359 io->u.ci_fsync.fi_start = start;
360 io->u.ci_fsync.fi_end = end;
361 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
362 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
367 io->u.ci_rw.rw_ptask = parent->u.ci_rw.rw_ptask;
368 io->u.ci_rw.rw_iter = parent->u.ci_rw.rw_iter;
369 io->u.ci_rw.rw_iocb = parent->u.ci_rw.rw_iocb;
370 io->u.ci_rw.rw_file = parent->u.ci_rw.rw_file;
371 io->u.ci_rw.rw_sync = parent->u.ci_rw.rw_sync;
372 if (cl_io_is_append(parent)) {
373 io->u.ci_rw.rw_append = 1;
375 io->u.ci_rw.rw_range.cir_pos = start;
376 io->u.ci_rw.rw_range.cir_count = end - start;
381 io->u.ci_ladvise.li_start = start;
382 io->u.ci_ladvise.li_end = end;
383 io->u.ci_ladvise.li_fid = parent->u.ci_ladvise.li_fid;
384 io->u.ci_ladvise.li_advice = parent->u.ci_ladvise.li_advice;
385 io->u.ci_ladvise.li_flags = parent->u.ci_ladvise.li_flags;
395 static loff_t lov_offset_mod(loff_t val, int delta)
397 if (val != OBD_OBJECT_EOF)
402 static int lov_io_iter_init(const struct lu_env *env,
403 const struct cl_io_slice *ios)
405 struct cl_io *io = ios->cis_io;
406 struct lov_io *lio = cl2lov_io(env, ios);
407 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
408 struct lov_io_sub *sub;
409 struct lov_layout_entry *le;
410 struct lu_extent ext;
416 ext.e_start = lio->lis_pos;
417 ext.e_end = lio->lis_endpos;
420 lov_foreach_layout_entry(lio->lis_object, le) {
421 struct lov_layout_raid0 *r0 = &le->lle_raid0;
427 if (!lu_extent_is_overlapped(&ext, &le->lle_extent))
430 CDEBUG(D_VFSTRACE, "component[%d] flags %#x\n",
431 index - 1, lsm->lsm_entries[index - 1]->lsme_flags);
432 if (!lsm_entry_inited(lsm, index - 1)) {
433 /* truncate IO will trigger write intent as well, and
434 * it's handled in lov_io_setattr_iter_init() */
435 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io)) {
436 io->ci_need_write_intent = 1;
437 /* execute it in main thread */
443 /* Read from uninitialized components should return
444 * zero filled pages. */
448 for (stripe = 0; stripe < r0->lo_nr; stripe++) {
449 if (!lov_stripe_intersects(lsm, index - 1, stripe,
453 if (unlikely(r0->lo_sub[stripe] == NULL)) {
454 if (ios->cis_io->ci_type == CIT_READ ||
455 ios->cis_io->ci_type == CIT_WRITE ||
456 ios->cis_io->ci_type == CIT_FAULT)
462 end = lov_offset_mod(end, 1);
463 sub = lov_sub_get(env, lio,
464 lov_comp_index(index - 1, stripe));
470 lov_io_sub_inherit(sub, lio, start, end);
471 rc = cl_io_iter_init(sub->sub_env, &sub->sub_io);
473 cl_io_iter_fini(sub->sub_env, &sub->sub_io);
478 "shrink stripe: {%d, %d} range: [%llu, %llu)\n",
479 index, stripe, start, end);
481 list_add_tail(&sub->sub_linkage, &lio->lis_active);
489 static int lov_io_rw_iter_init(const struct lu_env *env,
490 const struct cl_io_slice *ios)
492 struct cl_io *io = ios->cis_io;
493 struct lov_io *lio = cl2lov_io(env, ios);
494 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
495 struct lov_stripe_md_entry *lse;
496 struct cl_io_range *range = &io->u.ci_rw.rw_range;
497 loff_t start = range->cir_pos;
501 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
504 if (cl_io_is_append(io))
505 RETURN(lov_io_iter_init(env, ios));
507 index = lov_lsm_entry(lsm, range->cir_pos);
508 if (index < 0) { /* non-existing layout component */
509 if (io->ci_type == CIT_READ) {
510 /* TODO: it needs to detect the next component and
511 * then set the next pos */
513 /* execute it in main thread */
516 RETURN(lov_io_iter_init(env, ios));
522 lse = lov_lse(lio->lis_object, index);
524 next = MAX_LFS_FILESIZE;
525 if (lse->lsme_stripe_count > 1) {
526 unsigned long ssize = lse->lsme_stripe_size;
528 lov_do_div64(start, ssize);
529 next = (start + 1) * ssize;
530 if (next <= start * ssize)
531 next = MAX_LFS_FILESIZE;
534 LASSERTF(range->cir_pos >= lse->lsme_extent.e_start,
535 "pos %lld, [%lld, %lld)\n", range->cir_pos,
536 lse->lsme_extent.e_start, lse->lsme_extent.e_end);
537 next = min_t(__u64, next, lse->lsme_extent.e_end);
538 next = min_t(loff_t, next, lio->lis_io_endpos);
540 io->ci_continue = next < lio->lis_io_endpos;
541 range->cir_count = next - range->cir_pos;
542 lio->lis_pos = range->cir_pos;
543 lio->lis_endpos = range->cir_pos + range->cir_count;
545 "stripe: {%d, %llu} range: [%llu, %llu) end: %llu, count: %zd\n",
546 index, start, lio->lis_pos, lio->lis_endpos,
547 lio->lis_io_endpos, range->cir_count);
549 if (!io->ci_continue) {
550 /* the last piece of IO, execute it in main thread */
555 /* it only splits IO here for parallel IO,
556 * there will be no actual IO going to occur,
557 * so it doesn't need to invoke lov_io_iter_init()
558 * to initialize sub IOs. */
559 if (!lsm_entry_inited(lsm, index)) {
560 io->ci_need_write_intent = 1;
567 * XXX The following call should be optimized: we know, that
568 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
570 RETURN(lov_io_iter_init(env, ios));
573 static int lov_io_setattr_iter_init(const struct lu_env *env,
574 const struct cl_io_slice *ios)
576 struct lov_io *lio = cl2lov_io(env, ios);
577 struct cl_io *io = ios->cis_io;
578 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
582 if (cl_io_is_trunc(io) && lio->lis_pos > 0) {
583 index = lov_lsm_entry(lsm, lio->lis_pos - 1);
584 /* no entry found for such offset */
586 RETURN(io->ci_result = -ENODATA);
587 } else if (!lsm_entry_inited(lsm, index)) {
588 io->ci_need_write_intent = 1;
589 RETURN(io->ci_result = -ENODATA);
593 RETURN(lov_io_iter_init(env, ios));
596 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
597 int (*iofunc)(const struct lu_env *, struct cl_io *))
599 struct cl_io *parent = lio->lis_cl.cis_io;
600 struct lov_io_sub *sub;
604 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
605 rc = iofunc(sub->sub_env, &sub->sub_io);
609 if (parent->ci_result == 0)
610 parent->ci_result = sub->sub_io.ci_result;
615 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
618 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_lock));
621 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
624 RETURN(lov_io_call(env, cl2lov_io(env, ios), cl_io_start));
627 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
631 * It's possible that lov_io_start() wasn't called against this
632 * sub-io, either because previous sub-io failed, or upper layer
635 if (io->ci_state == CIS_IO_GOING)
638 io->ci_state = CIS_IO_FINISHED;
642 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
644 cl_io_iter_fini(env, io);
648 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
650 cl_io_unlock(env, io);
654 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
658 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
663 lov_io_data_version_end(const struct lu_env *env, const struct cl_io_slice *ios)
665 struct lov_io *lio = cl2lov_io(env, ios);
666 struct cl_io *parent = lio->lis_cl.cis_io;
667 struct lov_io_sub *sub;
670 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
671 lov_io_end_wrapper(env, &sub->sub_io);
673 parent->u.ci_data_version.dv_data_version +=
674 sub->sub_io.u.ci_data_version.dv_data_version;
676 if (parent->ci_result == 0)
677 parent->ci_result = sub->sub_io.ci_result;
683 static void lov_io_iter_fini(const struct lu_env *env,
684 const struct cl_io_slice *ios)
686 struct lov_io *lio = cl2lov_io(env, ios);
690 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
692 while (!list_empty(&lio->lis_active))
693 list_del_init(lio->lis_active.next);
697 static void lov_io_unlock(const struct lu_env *env,
698 const struct cl_io_slice *ios)
703 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
708 static int lov_io_read_ahead(const struct lu_env *env,
709 const struct cl_io_slice *ios,
710 pgoff_t start, struct cl_read_ahead *ra)
712 struct lov_io *lio = cl2lov_io(env, ios);
713 struct lov_object *loo = lio->lis_object;
714 struct cl_object *obj = lov2cl(loo);
715 struct lov_layout_raid0 *r0;
716 struct lov_io_sub *sub;
720 unsigned int pps; /* pages per stripe */
726 offset = cl_offset(obj, start);
727 index = lov_lsm_entry(loo->lo_lsm, offset);
728 if (index < 0 || !lsm_entry_inited(loo->lo_lsm, index))
731 stripe = lov_stripe_number(loo->lo_lsm, index, offset);
733 r0 = lov_r0(loo, index);
734 if (unlikely(r0->lo_sub[stripe] == NULL))
737 sub = lov_sub_get(env, lio, lov_comp_index(index, stripe));
739 RETURN(PTR_ERR(sub));
741 lov_stripe_offset(loo->lo_lsm, index, offset, stripe, &suboff);
742 rc = cl_io_read_ahead(sub->sub_env, &sub->sub_io,
743 cl_index(lovsub2cl(r0->lo_sub[stripe]), suboff),
746 CDEBUG(D_READA, DFID " cra_end = %lu, stripes = %d, rc = %d\n",
747 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end, r0->lo_nr, rc);
752 * Adjust the stripe index by layout of comp. ra->cra_end is the
753 * maximum page index covered by an underlying DLM lock.
754 * This function converts cra_end from stripe level to file level, and
755 * make sure it's not beyond stripe and component boundary.
758 /* cra_end is stripe level, convert it into file level */
759 ra_end = ra->cra_end;
760 if (ra_end != CL_PAGE_EOF)
761 ra->cra_end = lov_stripe_pgoff(loo->lo_lsm, index,
764 /* boundary of current component */
765 ra_end = cl_index(obj, (loff_t)lov_lse(loo, index)->lsme_extent.e_end);
766 if (ra_end != CL_PAGE_EOF && ra->cra_end >= ra_end)
767 ra->cra_end = ra_end - 1;
769 if (r0->lo_nr == 1) /* single stripe file */
772 pps = lov_lse(loo, index)->lsme_stripe_size >> PAGE_SHIFT;
774 CDEBUG(D_READA, DFID " max_index = %lu, pps = %u, index = %u, "
775 "stripe_size = %u, stripe no = %u, start index = %lu\n",
776 PFID(lu_object_fid(lov2lu(loo))), ra->cra_end, pps, index,
777 lov_lse(loo, index)->lsme_stripe_size, stripe, start);
779 /* never exceed the end of the stripe */
780 ra->cra_end = min_t(pgoff_t,
781 ra->cra_end, start + pps - start % pps - 1);
786 * lov implementation of cl_operations::cio_submit() method. It takes a list
787 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
788 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
791 * Major complication of this function is a need to handle memory cleansing:
792 * cl_io_submit() is called to write out pages as a part of VM memory
793 * reclamation, and hence it may not fail due to memory shortages (system
794 * dead-locks otherwise). To deal with this, some resources (sub-lists,
795 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
796 * not-memory cleansing context), and in case of memory shortage, these
797 * pre-allocated resources are used by lov_io_submit() under
798 * lov_device::ld_mutex mutex.
800 static int lov_io_submit(const struct lu_env *env,
801 const struct cl_io_slice *ios,
802 enum cl_req_type crt, struct cl_2queue *queue)
804 struct cl_page_list *qin = &queue->c2_qin;
805 struct lov_io *lio = cl2lov_io(env, ios);
806 struct lov_io_sub *sub;
807 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
808 struct cl_page *page;
813 if (lio->lis_nr_subios == 1) {
814 int idx = lio->lis_single_subio_index;
816 sub = lov_sub_get(env, lio, idx);
817 LASSERT(!IS_ERR(sub));
818 LASSERT(sub == &lio->lis_single_subio);
819 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
824 cl_page_list_init(plist);
825 while (qin->pl_nr > 0) {
826 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
828 cl_2queue_init(cl2q);
830 page = cl_page_list_first(qin);
831 cl_page_list_move(&cl2q->c2_qin, qin, page);
833 index = lov_page_index(page);
834 while (qin->pl_nr > 0) {
835 page = cl_page_list_first(qin);
836 if (index != lov_page_index(page))
839 cl_page_list_move(&cl2q->c2_qin, qin, page);
842 sub = lov_sub_get(env, lio, index);
844 rc = cl_io_submit_rw(sub->sub_env, &sub->sub_io,
850 cl_page_list_splice(&cl2q->c2_qin, plist);
851 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
852 cl_2queue_fini(env, cl2q);
858 cl_page_list_splice(plist, qin);
859 cl_page_list_fini(env, plist);
864 static int lov_io_commit_async(const struct lu_env *env,
865 const struct cl_io_slice *ios,
866 struct cl_page_list *queue, int from, int to,
869 struct cl_page_list *plist = &lov_env_info(env)->lti_plist;
870 struct lov_io *lio = cl2lov_io(env, ios);
871 struct lov_io_sub *sub;
872 struct cl_page *page;
876 if (lio->lis_nr_subios == 1) {
877 int idx = lio->lis_single_subio_index;
879 sub = lov_sub_get(env, lio, idx);
880 LASSERT(!IS_ERR(sub));
881 LASSERT(sub == &lio->lis_single_subio);
882 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io, queue,
887 cl_page_list_init(plist);
888 while (queue->pl_nr > 0) {
892 LASSERT(plist->pl_nr == 0);
893 page = cl_page_list_first(queue);
894 cl_page_list_move(plist, queue, page);
896 index = lov_page_index(page);
897 while (queue->pl_nr > 0) {
898 page = cl_page_list_first(queue);
899 if (index != lov_page_index(page))
902 cl_page_list_move(plist, queue, page);
905 if (queue->pl_nr > 0) /* still has more pages */
906 stripe_to = PAGE_SIZE;
908 sub = lov_sub_get(env, lio, index);
910 rc = cl_io_commit_async(sub->sub_env, &sub->sub_io,
911 plist, from, stripe_to, cb);
917 if (plist->pl_nr > 0) /* short write */
923 /* for error case, add the page back into the qin list */
924 LASSERT(ergo(rc == 0, plist->pl_nr == 0));
925 while (plist->pl_nr > 0) {
926 /* error occurred, add the uncommitted pages back into queue */
927 page = cl_page_list_last(plist);
928 cl_page_list_move_head(queue, plist, page);
934 static int lov_io_fault_start(const struct lu_env *env,
935 const struct cl_io_slice *ios)
937 struct cl_fault_io *fio;
939 struct lov_io_sub *sub;
943 fio = &ios->cis_io->u.ci_fault;
944 lio = cl2lov_io(env, ios);
945 sub = lov_sub_get(env, lio, lov_page_index(fio->ft_page));
946 sub->sub_io.u.ci_fault.ft_nob = fio->ft_nob;
948 RETURN(lov_io_start(env, ios));
951 static void lov_io_fsync_end(const struct lu_env *env,
952 const struct cl_io_slice *ios)
954 struct lov_io *lio = cl2lov_io(env, ios);
955 struct lov_io_sub *sub;
956 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
960 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
961 struct cl_io *subio = &sub->sub_io;
963 lov_io_end_wrapper(sub->sub_env, subio);
965 if (subio->ci_result == 0)
966 *written += subio->u.ci_fsync.fi_nr_written;
971 static const struct cl_io_operations lov_io_ops = {
974 .cio_fini = lov_io_fini,
975 .cio_iter_init = lov_io_rw_iter_init,
976 .cio_iter_fini = lov_io_iter_fini,
977 .cio_lock = lov_io_lock,
978 .cio_unlock = lov_io_unlock,
979 .cio_start = lov_io_start,
980 .cio_end = lov_io_end
983 .cio_fini = lov_io_fini,
984 .cio_iter_init = lov_io_rw_iter_init,
985 .cio_iter_fini = lov_io_iter_fini,
986 .cio_lock = lov_io_lock,
987 .cio_unlock = lov_io_unlock,
988 .cio_start = lov_io_start,
989 .cio_end = lov_io_end
992 .cio_fini = lov_io_fini,
993 .cio_iter_init = lov_io_setattr_iter_init,
994 .cio_iter_fini = lov_io_iter_fini,
995 .cio_lock = lov_io_lock,
996 .cio_unlock = lov_io_unlock,
997 .cio_start = lov_io_start,
998 .cio_end = lov_io_end
1000 [CIT_DATA_VERSION] = {
1001 .cio_fini = lov_io_fini,
1002 .cio_iter_init = lov_io_iter_init,
1003 .cio_iter_fini = lov_io_iter_fini,
1004 .cio_lock = lov_io_lock,
1005 .cio_unlock = lov_io_unlock,
1006 .cio_start = lov_io_start,
1007 .cio_end = lov_io_data_version_end,
1010 .cio_fini = lov_io_fini,
1011 .cio_iter_init = lov_io_iter_init,
1012 .cio_iter_fini = lov_io_iter_fini,
1013 .cio_lock = lov_io_lock,
1014 .cio_unlock = lov_io_unlock,
1015 .cio_start = lov_io_fault_start,
1016 .cio_end = lov_io_end
1019 .cio_fini = lov_io_fini,
1020 .cio_iter_init = lov_io_iter_init,
1021 .cio_iter_fini = lov_io_iter_fini,
1022 .cio_lock = lov_io_lock,
1023 .cio_unlock = lov_io_unlock,
1024 .cio_start = lov_io_start,
1025 .cio_end = lov_io_fsync_end
1028 .cio_fini = lov_io_fini,
1029 .cio_iter_init = lov_io_iter_init,
1030 .cio_iter_fini = lov_io_iter_fini,
1031 .cio_lock = lov_io_lock,
1032 .cio_unlock = lov_io_unlock,
1033 .cio_start = lov_io_start,
1034 .cio_end = lov_io_end
1037 .cio_fini = lov_io_fini,
1040 .cio_fini = lov_io_fini
1043 .cio_read_ahead = lov_io_read_ahead,
1044 .cio_submit = lov_io_submit,
1045 .cio_commit_async = lov_io_commit_async,
1048 /*****************************************************************************
1050 * Empty lov io operations.
1054 static void lov_empty_io_fini(const struct lu_env *env,
1055 const struct cl_io_slice *ios)
1057 struct lov_object *lov = cl2lov(ios->cis_obj);
1060 if (atomic_dec_and_test(&lov->lo_active_ios))
1061 wake_up_all(&lov->lo_waitq);
1065 static int lov_empty_io_submit(const struct lu_env *env,
1066 const struct cl_io_slice *ios,
1067 enum cl_req_type crt, struct cl_2queue *queue)
1072 static void lov_empty_impossible(const struct lu_env *env,
1073 struct cl_io_slice *ios)
1078 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
1081 * An io operation vector for files without stripes.
1083 static const struct cl_io_operations lov_empty_io_ops = {
1086 .cio_fini = lov_empty_io_fini,
1088 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1089 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1090 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1091 .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,
1110 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
1111 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
1112 .cio_start = LOV_EMPTY_IMPOSSIBLE,
1113 .cio_end = LOV_EMPTY_IMPOSSIBLE
1116 .cio_fini = lov_empty_io_fini
1119 .cio_fini = lov_empty_io_fini
1122 .cio_fini = lov_empty_io_fini
1125 .cio_fini = lov_empty_io_fini
1128 .cio_submit = lov_empty_io_submit,
1129 .cio_commit_async = LOV_EMPTY_IMPOSSIBLE
1132 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
1135 struct lov_io *lio = lov_env_io(env);
1136 struct lov_object *lov = cl2lov(obj);
1140 INIT_LIST_HEAD(&lio->lis_active);
1141 result = lov_io_slice_init(lio, lov, io);
1145 result = lov_io_subio_init(env, lio, io);
1147 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
1148 atomic_inc(&lov->lo_active_ios);
1152 io->ci_result = result < 0 ? result : 0;
1156 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
1159 struct lov_object *lov = cl2lov(obj);
1160 struct lov_io *lio = lov_env_io(env);
1164 lio->lis_object = lov;
1165 switch (io->ci_type) {
1176 case CIT_DATA_VERSION:
1184 CERROR("Page fault on a file without stripes: "DFID"\n",
1185 PFID(lu_object_fid(&obj->co_lu)));
1189 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1190 atomic_inc(&lov->lo_active_ios);
1193 io->ci_result = result < 0 ? result : 0;
1197 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
1200 struct lov_object *lov = cl2lov(obj);
1201 struct lov_io *lio = lov_env_io(env);
1205 LASSERT(lov->lo_lsm != NULL);
1206 lio->lis_object = lov;
1208 switch (io->ci_type) {
1210 LASSERTF(0, "invalid type %d\n", io->ci_type);
1211 result = -EOPNOTSUPP;
1217 case CIT_DATA_VERSION:
1221 /* the truncate to 0 is managed by MDT:
1222 * - in open, for open O_TRUNC
1223 * - in setattr, for truncate
1225 /* the truncate is for size > 0 so triggers a restore */
1226 if (cl_io_is_trunc(io)) {
1227 io->ci_restore_needed = 1;
1235 io->ci_restore_needed = 1;
1241 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
1242 atomic_inc(&lov->lo_active_ios);
1245 io->ci_result = result < 0 ? result : 0;