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.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
38 #define DEBUG_SUBSYSTEM S_CLASS
40 #include <linux/sched.h>
41 #include <linux/list.h>
42 #include <obd_class.h>
43 #include <obd_support.h>
44 #include <lustre_fid.h>
45 #include <cl_object.h>
46 #include "cl_internal.h"
48 /*****************************************************************************
54 #define cl_io_for_each(slice, io) \
55 list_for_each_entry((slice), &io->ci_layers, cis_linkage)
56 #define cl_io_for_each_reverse(slice, io) \
57 list_for_each_entry_reverse((slice), &io->ci_layers, cis_linkage)
59 static inline int cl_io_type_is_valid(enum cl_io_type type)
61 return CIT_READ <= type && type < CIT_OP_NR;
64 static inline int cl_io_is_loopable(const struct cl_io *io)
66 return cl_io_type_is_valid(io->ci_type) && io->ci_type != CIT_MISC;
70 * cl_io invariant that holds at all times when exported cl_io_*() functions
71 * are entered and left.
73 static int cl_io_invariant(const struct cl_io *io)
80 * io can own pages only when it is ongoing. Sub-io might
81 * still be in CIS_LOCKED state when top-io is in
84 ergo(io->ci_owned_nr > 0, io->ci_state == CIS_IO_GOING ||
85 (io->ci_state == CIS_LOCKED && up != NULL));
89 * Finalize \a io, by calling cl_io_operations::cio_fini() bottom-to-top.
91 void cl_io_fini(const struct lu_env *env, struct cl_io *io)
93 struct cl_io_slice *slice;
95 LINVRNT(cl_io_type_is_valid(io->ci_type));
96 LINVRNT(cl_io_invariant(io));
99 while (!list_empty(&io->ci_layers)) {
100 slice = container_of(io->ci_layers.prev, struct cl_io_slice,
102 list_del_init(&slice->cis_linkage);
103 if (slice->cis_iop->op[io->ci_type].cio_fini != NULL)
104 slice->cis_iop->op[io->ci_type].cio_fini(env, slice);
106 * Invalidate slice to catch use after free. This assumes that
107 * slices are allocated within session and can be touched
108 * after ->cio_fini() returns.
110 slice->cis_io = NULL;
112 io->ci_state = CIS_FINI;
114 /* sanity check for layout change */
115 switch(io->ci_type) {
118 case CIT_DATA_VERSION:
122 LASSERT(!io->ci_need_restart);
126 /* Check ignore layout change conf */
127 LASSERT(ergo(io->ci_ignore_layout || !io->ci_verify_layout,
128 !io->ci_need_restart));
137 EXPORT_SYMBOL(cl_io_fini);
139 static int cl_io_init0(const struct lu_env *env, struct cl_io *io,
140 enum cl_io_type iot, struct cl_object *obj)
142 struct cl_object *scan;
145 LINVRNT(io->ci_state == CIS_ZERO || io->ci_state == CIS_FINI);
146 LINVRNT(cl_io_type_is_valid(iot));
147 LINVRNT(cl_io_invariant(io));
151 INIT_LIST_HEAD(&io->ci_lockset.cls_todo);
152 INIT_LIST_HEAD(&io->ci_lockset.cls_done);
153 INIT_LIST_HEAD(&io->ci_layers);
156 cl_object_for_each(scan, obj) {
157 if (scan->co_ops->coo_io_init != NULL) {
158 result = scan->co_ops->coo_io_init(env, scan, io);
164 io->ci_state = CIS_INIT;
169 * Initialize sub-io, by calling cl_io_operations::cio_init() top-to-bottom.
171 * \pre obj != cl_object_top(obj)
173 int cl_io_sub_init(const struct lu_env *env, struct cl_io *io,
174 enum cl_io_type iot, struct cl_object *obj)
176 LASSERT(obj != cl_object_top(obj));
178 return cl_io_init0(env, io, iot, obj);
180 EXPORT_SYMBOL(cl_io_sub_init);
183 * Initialize \a io, by calling cl_io_operations::cio_init() top-to-bottom.
185 * Caller has to call cl_io_fini() after a call to cl_io_init(), no matter
186 * what the latter returned.
188 * \pre obj == cl_object_top(obj)
189 * \pre cl_io_type_is_valid(iot)
190 * \post cl_io_type_is_valid(io->ci_type) && io->ci_type == iot
192 int cl_io_init(const struct lu_env *env, struct cl_io *io,
193 enum cl_io_type iot, struct cl_object *obj)
195 LASSERT(obj == cl_object_top(obj));
197 return cl_io_init0(env, io, iot, obj);
199 EXPORT_SYMBOL(cl_io_init);
202 * Initialize read or write io.
204 * \pre iot == CIT_READ || iot == CIT_WRITE
206 int cl_io_rw_init(const struct lu_env *env, struct cl_io *io,
207 enum cl_io_type iot, loff_t pos, size_t count)
209 LINVRNT(iot == CIT_READ || iot == CIT_WRITE);
210 LINVRNT(io->ci_obj != NULL);
213 LU_OBJECT_HEADER(D_VFSTRACE, env, &io->ci_obj->co_lu,
214 "io range: %u [%llu, %llu) %u %u\n",
215 iot, (__u64)pos, (__u64)pos + count,
216 io->u.ci_rw.crw_nonblock, io->u.ci_wr.wr_append);
217 io->u.ci_rw.crw_pos = pos;
218 io->u.ci_rw.crw_count = count;
219 RETURN(cl_io_init(env, io, iot, io->ci_obj));
221 EXPORT_SYMBOL(cl_io_rw_init);
223 static int cl_lock_descr_sort(const struct cl_lock_descr *d0,
224 const struct cl_lock_descr *d1)
226 return lu_fid_cmp(lu_object_fid(&d0->cld_obj->co_lu),
227 lu_object_fid(&d1->cld_obj->co_lu));
231 * Sort locks in lexicographical order of their (fid, start-offset) pairs.
233 static void cl_io_locks_sort(struct cl_io *io)
238 /* hidden treasure: bubble sort for now. */
240 struct cl_io_lock_link *curr;
241 struct cl_io_lock_link *prev;
242 struct cl_io_lock_link *temp;
247 list_for_each_entry_safe(curr, temp, &io->ci_lockset.cls_todo,
250 switch (cl_lock_descr_sort(&prev->cill_descr,
251 &curr->cill_descr)) {
254 * IMPOSSIBLE: Identical locks are
261 list_move_tail(&curr->cill_linkage,
262 &prev->cill_linkage);
264 continue; /* don't change prev: it's
265 * still "previous" */
266 case -1: /* already in order */
276 static void cl_lock_descr_merge(struct cl_lock_descr *d0,
277 const struct cl_lock_descr *d1)
279 d0->cld_start = min(d0->cld_start, d1->cld_start);
280 d0->cld_end = max(d0->cld_end, d1->cld_end);
282 if (d1->cld_mode == CLM_WRITE && d0->cld_mode != CLM_WRITE)
283 d0->cld_mode = CLM_WRITE;
285 if (d1->cld_mode == CLM_GROUP && d0->cld_mode != CLM_GROUP)
286 d0->cld_mode = CLM_GROUP;
289 static int cl_lockset_merge(const struct cl_lockset *set,
290 const struct cl_lock_descr *need)
292 struct cl_io_lock_link *scan;
295 list_for_each_entry(scan, &set->cls_todo, cill_linkage) {
296 if (!cl_object_same(scan->cill_descr.cld_obj, need->cld_obj))
299 /* Merge locks for the same object because ldlm lock server
300 * may expand the lock extent, otherwise there is a deadlock
301 * case if two conflicted locks are queueud for the same object
302 * and lock server expands one lock to overlap the another.
303 * The side effect is that it can generate a multi-stripe lock
304 * that may cause casacading problem */
305 cl_lock_descr_merge(&scan->cill_descr, need);
306 CDEBUG(D_VFSTRACE, "lock: %d: [%lu, %lu]\n",
307 scan->cill_descr.cld_mode, scan->cill_descr.cld_start,
308 scan->cill_descr.cld_end);
314 static int cl_lockset_lock(const struct lu_env *env, struct cl_io *io,
315 struct cl_lockset *set)
317 struct cl_io_lock_link *link;
318 struct cl_io_lock_link *temp;
323 list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) {
324 result = cl_lock_request(env, io, &link->cill_lock);
328 list_move(&link->cill_linkage, &set->cls_done);
334 * Takes locks necessary for the current iteration of io.
336 * Calls cl_io_operations::cio_lock() top-to-bottom to collect locks required
337 * by layers for the current iteration. Then sort locks (to avoid dead-locks),
340 int cl_io_lock(const struct lu_env *env, struct cl_io *io)
342 const struct cl_io_slice *scan;
345 LINVRNT(cl_io_is_loopable(io));
346 LINVRNT(io->ci_state == CIS_IT_STARTED);
347 LINVRNT(cl_io_invariant(io));
350 cl_io_for_each(scan, io) {
351 if (scan->cis_iop->op[io->ci_type].cio_lock == NULL)
353 result = scan->cis_iop->op[io->ci_type].cio_lock(env, scan);
358 cl_io_locks_sort(io);
359 result = cl_lockset_lock(env, io, &io->ci_lockset);
362 cl_io_unlock(env, io);
364 io->ci_state = CIS_LOCKED;
367 EXPORT_SYMBOL(cl_io_lock);
370 * Release locks takes by io.
372 void cl_io_unlock(const struct lu_env *env, struct cl_io *io)
374 struct cl_lockset *set;
375 struct cl_io_lock_link *link;
376 struct cl_io_lock_link *temp;
377 const struct cl_io_slice *scan;
379 LASSERT(cl_io_is_loopable(io));
380 LASSERT(CIS_IT_STARTED <= io->ci_state && io->ci_state < CIS_UNLOCKED);
381 LINVRNT(cl_io_invariant(io));
384 set = &io->ci_lockset;
386 list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) {
387 list_del_init(&link->cill_linkage);
388 if (link->cill_fini != NULL)
389 link->cill_fini(env, link);
392 list_for_each_entry_safe(link, temp, &set->cls_done, cill_linkage) {
393 list_del_init(&link->cill_linkage);
394 cl_lock_release(env, &link->cill_lock);
395 if (link->cill_fini != NULL)
396 link->cill_fini(env, link);
399 cl_io_for_each_reverse(scan, io) {
400 if (scan->cis_iop->op[io->ci_type].cio_unlock != NULL)
401 scan->cis_iop->op[io->ci_type].cio_unlock(env, scan);
403 io->ci_state = CIS_UNLOCKED;
406 EXPORT_SYMBOL(cl_io_unlock);
409 * Prepares next iteration of io.
411 * Calls cl_io_operations::cio_iter_init() top-to-bottom. This exists to give
412 * layers a chance to modify io parameters, e.g., so that lov can restrict io
413 * to a single stripe.
415 int cl_io_iter_init(const struct lu_env *env, struct cl_io *io)
417 const struct cl_io_slice *scan;
420 LINVRNT(cl_io_is_loopable(io));
421 LINVRNT(io->ci_state == CIS_INIT || io->ci_state == CIS_IT_ENDED);
422 LINVRNT(cl_io_invariant(io));
426 cl_io_for_each(scan, io) {
427 if (scan->cis_iop->op[io->ci_type].cio_iter_init == NULL)
429 result = scan->cis_iop->op[io->ci_type].cio_iter_init(env,
435 io->ci_state = CIS_IT_STARTED;
438 EXPORT_SYMBOL(cl_io_iter_init);
441 * Finalizes io iteration.
443 * Calls cl_io_operations::cio_iter_fini() bottom-to-top.
445 void cl_io_iter_fini(const struct lu_env *env, struct cl_io *io)
447 const struct cl_io_slice *scan;
449 LINVRNT(cl_io_is_loopable(io));
450 LINVRNT(io->ci_state == CIS_UNLOCKED);
451 LINVRNT(cl_io_invariant(io));
454 cl_io_for_each_reverse(scan, io) {
455 if (scan->cis_iop->op[io->ci_type].cio_iter_fini != NULL)
456 scan->cis_iop->op[io->ci_type].cio_iter_fini(env, scan);
458 io->ci_state = CIS_IT_ENDED;
461 EXPORT_SYMBOL(cl_io_iter_fini);
464 * Records that read or write io progressed \a nob bytes forward.
466 void cl_io_rw_advance(const struct lu_env *env, struct cl_io *io, size_t nob)
468 const struct cl_io_slice *scan;
470 LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE ||
472 LINVRNT(cl_io_is_loopable(io));
473 LINVRNT(cl_io_invariant(io));
477 io->u.ci_rw.crw_pos += nob;
478 io->u.ci_rw.crw_count -= nob;
480 /* layers have to be notified. */
481 cl_io_for_each_reverse(scan, io) {
482 if (scan->cis_iop->op[io->ci_type].cio_advance != NULL)
483 scan->cis_iop->op[io->ci_type].cio_advance(env, scan,
490 * Adds a lock to a lockset.
492 int cl_io_lock_add(const struct lu_env *env, struct cl_io *io,
493 struct cl_io_lock_link *link)
498 if (cl_lockset_merge(&io->ci_lockset, &link->cill_descr))
501 list_add(&link->cill_linkage, &io->ci_lockset.cls_todo);
506 EXPORT_SYMBOL(cl_io_lock_add);
508 static void cl_free_io_lock_link(const struct lu_env *env,
509 struct cl_io_lock_link *link)
515 * Allocates new lock link, and uses it to add a lock to a lockset.
517 int cl_io_lock_alloc_add(const struct lu_env *env, struct cl_io *io,
518 struct cl_lock_descr *descr)
520 struct cl_io_lock_link *link;
526 link->cill_descr = *descr;
527 link->cill_fini = cl_free_io_lock_link;
528 result = cl_io_lock_add(env, io, link);
529 if (result) /* lock match */
530 link->cill_fini(env, link);
536 EXPORT_SYMBOL(cl_io_lock_alloc_add);
539 * Starts io by calling cl_io_operations::cio_start() top-to-bottom.
541 int cl_io_start(const struct lu_env *env, struct cl_io *io)
543 const struct cl_io_slice *scan;
546 LINVRNT(cl_io_is_loopable(io));
547 LINVRNT(io->ci_state == CIS_LOCKED);
548 LINVRNT(cl_io_invariant(io));
551 io->ci_state = CIS_IO_GOING;
552 cl_io_for_each(scan, io) {
553 if (scan->cis_iop->op[io->ci_type].cio_start == NULL)
555 result = scan->cis_iop->op[io->ci_type].cio_start(env, scan);
563 EXPORT_SYMBOL(cl_io_start);
566 * Wait until current io iteration is finished by calling
567 * cl_io_operations::cio_end() bottom-to-top.
569 void cl_io_end(const struct lu_env *env, struct cl_io *io)
571 const struct cl_io_slice *scan;
573 LINVRNT(cl_io_is_loopable(io));
574 LINVRNT(io->ci_state == CIS_IO_GOING);
575 LINVRNT(cl_io_invariant(io));
578 cl_io_for_each_reverse(scan, io) {
579 if (scan->cis_iop->op[io->ci_type].cio_end != NULL)
580 scan->cis_iop->op[io->ci_type].cio_end(env, scan);
581 /* TODO: error handling. */
583 io->ci_state = CIS_IO_FINISHED;
586 EXPORT_SYMBOL(cl_io_end);
589 * Called by read io, to decide the readahead extent
591 * \see cl_io_operations::cio_read_ahead()
593 int cl_io_read_ahead(const struct lu_env *env, struct cl_io *io,
594 pgoff_t start, struct cl_read_ahead *ra)
596 const struct cl_io_slice *scan;
599 LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_FAULT);
600 LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
601 LINVRNT(cl_io_invariant(io));
604 cl_io_for_each(scan, io) {
605 if (scan->cis_iop->cio_read_ahead == NULL)
608 result = scan->cis_iop->cio_read_ahead(env, scan, start, ra);
612 RETURN(result > 0 ? 0 : result);
614 EXPORT_SYMBOL(cl_io_read_ahead);
617 * Commit a list of contiguous pages into writeback cache.
619 * \returns 0 if all pages committed, or errcode if error occurred.
620 * \see cl_io_operations::cio_commit_async()
622 int cl_io_commit_async(const struct lu_env *env, struct cl_io *io,
623 struct cl_page_list *queue, int from, int to,
626 const struct cl_io_slice *scan;
630 cl_io_for_each(scan, io) {
631 if (scan->cis_iop->cio_commit_async == NULL)
633 result = scan->cis_iop->cio_commit_async(env, scan, queue,
640 EXPORT_SYMBOL(cl_io_commit_async);
643 * Submits a list of pages for immediate io.
645 * After the function gets returned, The submitted pages are moved to
646 * queue->c2_qout queue, and queue->c2_qin contain both the pages don't need
647 * to be submitted, and the pages are errant to submit.
649 * \returns 0 if at least one page was submitted, error code otherwise.
650 * \see cl_io_operations::cio_submit()
652 int cl_io_submit_rw(const struct lu_env *env, struct cl_io *io,
653 enum cl_req_type crt, struct cl_2queue *queue)
655 const struct cl_io_slice *scan;
659 cl_io_for_each(scan, io) {
660 if (scan->cis_iop->cio_submit == NULL)
662 result = scan->cis_iop->cio_submit(env, scan, crt, queue);
667 * If ->cio_submit() failed, no pages were sent.
669 LASSERT(ergo(result != 0, list_empty(&queue->c2_qout.pl_pages)));
672 EXPORT_SYMBOL(cl_io_submit_rw);
675 * Submit a sync_io and wait for the IO to be finished, or error happens.
676 * If \a timeout is zero, it means to wait for the IO unconditionally.
678 int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io,
679 enum cl_req_type iot, struct cl_2queue *queue,
682 struct cl_sync_io *anchor = &cl_env_info(env)->clt_anchor;
686 cl_page_list_for_each(pg, &queue->c2_qin) {
687 LASSERT(pg->cp_sync_io == NULL);
688 pg->cp_sync_io = anchor;
691 cl_sync_io_init(anchor, queue->c2_qin.pl_nr, &cl_sync_io_end);
692 rc = cl_io_submit_rw(env, io, iot, queue);
695 * If some pages weren't sent for any reason (e.g.,
696 * read found up-to-date pages in the cache, or write found
697 * clean pages), count them as completed to avoid infinite
700 cl_page_list_for_each(pg, &queue->c2_qin) {
701 pg->cp_sync_io = NULL;
702 cl_sync_io_note(env, anchor, 1);
705 /* wait for the IO to be finished. */
706 rc = cl_sync_io_wait(env, anchor, timeout);
707 cl_page_list_assume(env, io, &queue->c2_qout);
709 LASSERT(list_empty(&queue->c2_qout.pl_pages));
710 cl_page_list_for_each(pg, &queue->c2_qin)
711 pg->cp_sync_io = NULL;
715 EXPORT_SYMBOL(cl_io_submit_sync);
718 * Cancel an IO which has been submitted by cl_io_submit_rw.
720 int cl_io_cancel(const struct lu_env *env, struct cl_io *io,
721 struct cl_page_list *queue)
723 struct cl_page *page;
726 CERROR("Canceling ongoing page trasmission\n");
727 cl_page_list_for_each(page, queue) {
730 rc = cl_page_cancel(env, page);
731 result = result ?: rc;
739 * Pumps io through iterations calling
741 * - cl_io_iter_init()
751 * - cl_io_iter_fini()
753 * repeatedly until there is no more io to do.
755 int cl_io_loop(const struct lu_env *env, struct cl_io *io)
759 LINVRNT(cl_io_is_loopable(io));
766 result = cl_io_iter_init(env, io);
769 result = cl_io_lock(env, io);
772 * Notify layers that locks has been taken,
775 * - llite: kms, short read;
776 * - llite: generic_file_read();
778 result = cl_io_start(env, io);
780 * Send any remaining pending
783 * - llite: ll_rw_stats_tally.
786 cl_io_unlock(env, io);
787 cl_io_rw_advance(env, io, io->ci_nob - nob);
790 cl_io_iter_fini(env, io);
791 } while (result == 0 && io->ci_continue);
793 result = io->ci_result;
794 RETURN(result < 0 ? result : 0);
796 EXPORT_SYMBOL(cl_io_loop);
799 * Adds io slice to the cl_io.
801 * This is called by cl_object_operations::coo_io_init() methods to add a
802 * per-layer state to the io. New state is added at the end of
803 * cl_io::ci_layers list, that is, it is at the bottom of the stack.
805 * \see cl_lock_slice_add(), cl_req_slice_add(), cl_page_slice_add()
807 void cl_io_slice_add(struct cl_io *io, struct cl_io_slice *slice,
808 struct cl_object *obj,
809 const struct cl_io_operations *ops)
811 struct list_head *linkage = &slice->cis_linkage;
813 LASSERT((linkage->prev == NULL && linkage->next == NULL) ||
814 list_empty(linkage));
817 list_add_tail(linkage, &io->ci_layers);
819 slice->cis_obj = obj;
820 slice->cis_iop = ops;
823 EXPORT_SYMBOL(cl_io_slice_add);
827 * Initializes page list.
829 void cl_page_list_init(struct cl_page_list *plist)
833 INIT_LIST_HEAD(&plist->pl_pages);
834 plist->pl_owner = current;
837 EXPORT_SYMBOL(cl_page_list_init);
840 * Adds a page to a page list.
842 void cl_page_list_add(struct cl_page_list *plist, struct cl_page *page)
845 /* it would be better to check that page is owned by "current" io, but
846 * it is not passed here. */
847 LASSERT(page->cp_owner != NULL);
848 LINVRNT(plist->pl_owner == current);
850 LASSERT(list_empty(&page->cp_batch));
851 list_add_tail(&page->cp_batch, &plist->pl_pages);
853 lu_ref_add_at(&page->cp_reference, &page->cp_queue_ref, "queue", plist);
857 EXPORT_SYMBOL(cl_page_list_add);
860 * Removes a page from a page list.
862 void cl_page_list_del(const struct lu_env *env,
863 struct cl_page_list *plist, struct cl_page *page)
865 LASSERT(plist->pl_nr > 0);
866 LASSERT(cl_page_is_vmlocked(env, page));
867 LINVRNT(plist->pl_owner == current);
870 list_del_init(&page->cp_batch);
872 lu_ref_del_at(&page->cp_reference, &page->cp_queue_ref, "queue", plist);
873 cl_page_put(env, page);
876 EXPORT_SYMBOL(cl_page_list_del);
879 * Moves a page from one page list to another.
881 void cl_page_list_move(struct cl_page_list *dst, struct cl_page_list *src,
882 struct cl_page *page)
884 LASSERT(src->pl_nr > 0);
885 LINVRNT(dst->pl_owner == current);
886 LINVRNT(src->pl_owner == current);
889 list_move_tail(&page->cp_batch, &dst->pl_pages);
892 lu_ref_set_at(&page->cp_reference, &page->cp_queue_ref, "queue",
896 EXPORT_SYMBOL(cl_page_list_move);
899 * Moves a page from one page list to the head of another list.
901 void cl_page_list_move_head(struct cl_page_list *dst, struct cl_page_list *src,
902 struct cl_page *page)
904 LASSERT(src->pl_nr > 0);
905 LINVRNT(dst->pl_owner == current);
906 LINVRNT(src->pl_owner == current);
909 list_move(&page->cp_batch, &dst->pl_pages);
912 lu_ref_set_at(&page->cp_reference, &page->cp_queue_ref, "queue",
916 EXPORT_SYMBOL(cl_page_list_move_head);
919 * splice the cl_page_list, just as list head does
921 void cl_page_list_splice(struct cl_page_list *list, struct cl_page_list *head)
923 struct cl_page *page;
926 LINVRNT(list->pl_owner == current);
927 LINVRNT(head->pl_owner == current);
930 cl_page_list_for_each_safe(page, tmp, list)
931 cl_page_list_move(head, list, page);
934 EXPORT_SYMBOL(cl_page_list_splice);
937 * Disowns pages in a queue.
939 void cl_page_list_disown(const struct lu_env *env,
940 struct cl_io *io, struct cl_page_list *plist)
942 struct cl_page *page;
943 struct cl_page *temp;
945 LINVRNT(plist->pl_owner == current);
948 cl_page_list_for_each_safe(page, temp, plist) {
949 LASSERT(plist->pl_nr > 0);
951 list_del_init(&page->cp_batch);
954 * cl_page_disown0 rather than usual cl_page_disown() is used,
955 * because pages are possibly in CPS_FREEING state already due
956 * to the call to cl_page_list_discard().
959 * XXX cl_page_disown0() will fail if page is not locked.
961 cl_page_disown0(env, io, page);
962 lu_ref_del_at(&page->cp_reference, &page->cp_queue_ref, "queue",
964 cl_page_put(env, page);
968 EXPORT_SYMBOL(cl_page_list_disown);
971 * Releases pages from queue.
973 void cl_page_list_fini(const struct lu_env *env, struct cl_page_list *plist)
975 struct cl_page *page;
976 struct cl_page *temp;
978 LINVRNT(plist->pl_owner == current);
981 cl_page_list_for_each_safe(page, temp, plist)
982 cl_page_list_del(env, plist, page);
983 LASSERT(plist->pl_nr == 0);
986 EXPORT_SYMBOL(cl_page_list_fini);
989 * Assumes all pages in a queue.
991 void cl_page_list_assume(const struct lu_env *env,
992 struct cl_io *io, struct cl_page_list *plist)
994 struct cl_page *page;
996 LINVRNT(plist->pl_owner == current);
998 cl_page_list_for_each(page, plist)
999 cl_page_assume(env, io, page);
1003 * Discards all pages in a queue.
1005 void cl_page_list_discard(const struct lu_env *env, struct cl_io *io,
1006 struct cl_page_list *plist)
1008 struct cl_page *page;
1010 LINVRNT(plist->pl_owner == current);
1012 cl_page_list_for_each(page, plist)
1013 cl_page_discard(env, io, page);
1018 * Initialize dual page queue.
1020 void cl_2queue_init(struct cl_2queue *queue)
1023 cl_page_list_init(&queue->c2_qin);
1024 cl_page_list_init(&queue->c2_qout);
1027 EXPORT_SYMBOL(cl_2queue_init);
1030 * Add a page to the incoming page list of 2-queue.
1032 void cl_2queue_add(struct cl_2queue *queue, struct cl_page *page)
1035 cl_page_list_add(&queue->c2_qin, page);
1038 EXPORT_SYMBOL(cl_2queue_add);
1041 * Disown pages in both lists of a 2-queue.
1043 void cl_2queue_disown(const struct lu_env *env,
1044 struct cl_io *io, struct cl_2queue *queue)
1047 cl_page_list_disown(env, io, &queue->c2_qin);
1048 cl_page_list_disown(env, io, &queue->c2_qout);
1051 EXPORT_SYMBOL(cl_2queue_disown);
1054 * Discard (truncate) pages in both lists of a 2-queue.
1056 void cl_2queue_discard(const struct lu_env *env,
1057 struct cl_io *io, struct cl_2queue *queue)
1060 cl_page_list_discard(env, io, &queue->c2_qin);
1061 cl_page_list_discard(env, io, &queue->c2_qout);
1064 EXPORT_SYMBOL(cl_2queue_discard);
1067 * Assume to own the pages in cl_2queue
1069 void cl_2queue_assume(const struct lu_env *env,
1070 struct cl_io *io, struct cl_2queue *queue)
1072 cl_page_list_assume(env, io, &queue->c2_qin);
1073 cl_page_list_assume(env, io, &queue->c2_qout);
1077 * Finalize both page lists of a 2-queue.
1079 void cl_2queue_fini(const struct lu_env *env, struct cl_2queue *queue)
1082 cl_page_list_fini(env, &queue->c2_qout);
1083 cl_page_list_fini(env, &queue->c2_qin);
1086 EXPORT_SYMBOL(cl_2queue_fini);
1089 * Initialize a 2-queue to contain \a page in its incoming page list.
1091 void cl_2queue_init_page(struct cl_2queue *queue, struct cl_page *page)
1094 cl_2queue_init(queue);
1095 cl_2queue_add(queue, page);
1098 EXPORT_SYMBOL(cl_2queue_init_page);
1101 * Returns top-level io.
1103 * \see cl_object_top()
1105 struct cl_io *cl_io_top(struct cl_io *io)
1108 while (io->ci_parent != NULL)
1112 EXPORT_SYMBOL(cl_io_top);
1115 * Prints human readable representation of \a io to the \a f.
1117 void cl_io_print(const struct lu_env *env, void *cookie,
1118 lu_printer_t printer, const struct cl_io *io)
1123 * Fills in attributes that are passed to server together with transfer. Only
1124 * attributes from \a flags may be touched. This can be called multiple times
1125 * for the same request.
1127 void cl_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1128 struct cl_req_attr *attr)
1130 struct cl_object *scan;
1133 cl_object_for_each(scan, obj) {
1134 if (scan->co_ops->coo_req_attr_set != NULL)
1135 scan->co_ops->coo_req_attr_set(env, scan, attr);
1139 EXPORT_SYMBOL(cl_req_attr_set);
1141 /* cl_sync_io_callback assumes the caller must call cl_sync_io_wait() to
1142 * wait for the IO to finish. */
1143 void cl_sync_io_end(const struct lu_env *env, struct cl_sync_io *anchor)
1145 wake_up_all(&anchor->csi_waitq);
1147 /* it's safe to nuke or reuse anchor now */
1148 atomic_set(&anchor->csi_barrier, 0);
1150 EXPORT_SYMBOL(cl_sync_io_end);
1153 * Initialize synchronous io wait anchor
1155 void cl_sync_io_init(struct cl_sync_io *anchor, int nr,
1156 void (*end)(const struct lu_env *, struct cl_sync_io *))
1159 memset(anchor, 0, sizeof(*anchor));
1160 init_waitqueue_head(&anchor->csi_waitq);
1161 atomic_set(&anchor->csi_sync_nr, nr);
1162 atomic_set(&anchor->csi_barrier, nr > 0);
1163 anchor->csi_sync_rc = 0;
1164 anchor->csi_end_io = end;
1165 LASSERT(end != NULL);
1168 EXPORT_SYMBOL(cl_sync_io_init);
1171 * Wait until all IO completes. Transfer completion routine has to call
1172 * cl_sync_io_note() for every entity.
1174 int cl_sync_io_wait(const struct lu_env *env, struct cl_sync_io *anchor,
1177 struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout),
1182 LASSERT(timeout >= 0);
1184 rc = l_wait_event(anchor->csi_waitq,
1185 atomic_read(&anchor->csi_sync_nr) == 0,
1188 CERROR("IO failed: %d, still wait for %d remaining entries\n",
1189 rc, atomic_read(&anchor->csi_sync_nr));
1191 lwi = (struct l_wait_info) { 0 };
1192 (void)l_wait_event(anchor->csi_waitq,
1193 atomic_read(&anchor->csi_sync_nr) == 0,
1196 rc = anchor->csi_sync_rc;
1198 LASSERT(atomic_read(&anchor->csi_sync_nr) == 0);
1200 /* wait until cl_sync_io_note() has done wakeup */
1201 while (unlikely(atomic_read(&anchor->csi_barrier) != 0)) {
1206 EXPORT_SYMBOL(cl_sync_io_wait);
1209 * Indicate that transfer of a single page completed.
1211 void cl_sync_io_note(const struct lu_env *env, struct cl_sync_io *anchor,
1215 if (anchor->csi_sync_rc == 0 && ioret < 0)
1216 anchor->csi_sync_rc = ioret;
1218 * Synchronous IO done without releasing page lock (e.g., as a part of
1219 * ->{prepare,commit}_write(). Completion is used to signal the end of
1222 LASSERT(atomic_read(&anchor->csi_sync_nr) > 0);
1223 if (atomic_dec_and_test(&anchor->csi_sync_nr)) {
1224 LASSERT(anchor->csi_end_io != NULL);
1225 anchor->csi_end_io(env, anchor);
1226 /* Can't access anchor any more */
1230 EXPORT_SYMBOL(cl_sync_io_note);