1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
41 #define DEBUG_SUBSYSTEM S_CLASS
43 # define EXPORT_SYMTAB
46 #include <obd_class.h>
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <libcfs/list.h>
50 /* lu_time_global_{init,fini}() */
53 #include <cl_object.h>
54 #include "cl_internal.h"
56 /*****************************************************************************
62 #define cl_io_for_each(slice, io) \
63 list_for_each_entry((slice), &io->ci_layers, cis_linkage)
64 #define cl_io_for_each_reverse(slice, io) \
65 list_for_each_entry_reverse((slice), &io->ci_layers, cis_linkage)
67 static inline int cl_io_type_is_valid(enum cl_io_type type)
69 return CIT_READ <= type && type < CIT_OP_NR;
72 static inline int cl_io_is_loopable(const struct cl_io *io)
74 return cl_io_type_is_valid(io->ci_type) && io->ci_type != CIT_MISC;
78 * True, iff \a io is a sendfile().
80 int cl_io_is_sendfile(const struct cl_io *io)
82 return io->ci_type == CIT_READ && io->u.ci_rd.rd_is_sendfile;
84 EXPORT_SYMBOL(cl_io_is_sendfile);
87 * Returns true iff there is an IO ongoing in the given environment.
89 int cl_io_is_going(const struct lu_env *env)
91 return cl_env_info(env)->clt_current_io != NULL;
93 EXPORT_SYMBOL(cl_io_is_going);
96 * cl_io invariant that holds at all times when exported cl_io_*() functions
97 * are entered and left.
99 static int cl_io_invariant(const struct cl_io *io)
106 * io can own pages only when it is ongoing. Sub-io might
107 * still be in CIS_LOCKED state when top-io is in
110 ergo(io->ci_owned_nr > 0, io->ci_state == CIS_IO_GOING ||
111 (io->ci_state == CIS_LOCKED && up != NULL));
115 * Finalize \a io, by calling cl_io_operations::cio_fini() bottom-to-top.
117 void cl_io_fini(const struct lu_env *env, struct cl_io *io)
119 struct cl_io_slice *slice;
120 struct cl_thread_info *info;
122 LINVRNT(cl_io_type_is_valid(io->ci_type));
123 LINVRNT(cl_io_invariant(io));
126 while (!list_empty(&io->ci_layers)) {
127 slice = container_of(io->ci_layers.next, struct cl_io_slice,
129 list_del_init(&slice->cis_linkage);
130 if (slice->cis_iop->op[io->ci_type].cio_fini != NULL)
131 slice->cis_iop->op[io->ci_type].cio_fini(env, slice);
133 * Invalidate slice to catch use after free. This assumes that
134 * slices are allocated within session and can be touched
135 * after ->cio_fini() returns.
137 slice->cis_io = NULL;
139 io->ci_state = CIS_FINI;
140 info = cl_env_info(env);
141 if (info->clt_current_io == io)
142 info->clt_current_io = NULL;
145 EXPORT_SYMBOL(cl_io_fini);
147 static int cl_io_init0(const struct lu_env *env, struct cl_io *io,
148 enum cl_io_type iot, struct cl_object *obj)
150 struct cl_object *scan;
153 LINVRNT(io->ci_state == CIS_ZERO || io->ci_state == CIS_FINI);
154 LINVRNT(cl_io_type_is_valid(iot));
155 LINVRNT(cl_io_invariant(io));
159 CFS_INIT_LIST_HEAD(&io->ci_lockset.cls_todo);
160 CFS_INIT_LIST_HEAD(&io->ci_lockset.cls_curr);
161 CFS_INIT_LIST_HEAD(&io->ci_lockset.cls_done);
162 CFS_INIT_LIST_HEAD(&io->ci_layers);
165 cl_object_for_each(scan, obj) {
166 if (scan->co_ops->coo_io_init != NULL) {
167 result = scan->co_ops->coo_io_init(env, scan, io);
173 io->ci_state = CIS_INIT;
178 * Initialize sub-io, by calling cl_io_operations::cio_init() top-to-bottom.
180 * \pre obj != cl_object_top(obj)
182 int cl_io_sub_init(const struct lu_env *env, struct cl_io *io,
183 enum cl_io_type iot, struct cl_object *obj)
185 struct cl_thread_info *info = cl_env_info(env);
187 LASSERT(obj != cl_object_top(obj));
188 if (info->clt_current_io == NULL)
189 info->clt_current_io = io;
190 return cl_io_init0(env, io, iot, obj);
192 EXPORT_SYMBOL(cl_io_sub_init);
195 * Initialize \a io, by calling cl_io_operations::cio_init() top-to-bottom.
197 * Caller has to call cl_io_fini() after a call to cl_io_init(), no matter
198 * what the latter returned.
200 * \pre obj == cl_object_top(obj)
201 * \pre cl_io_type_is_valid(iot)
202 * \post cl_io_type_is_valid(io->ci_type) && io->ci_type == iot
204 int cl_io_init(const struct lu_env *env, struct cl_io *io,
205 enum cl_io_type iot, struct cl_object *obj)
207 struct cl_thread_info *info = cl_env_info(env);
209 LASSERT(obj == cl_object_top(obj));
210 LASSERT(info->clt_current_io == NULL);
212 info->clt_current_io = io;
213 return cl_io_init0(env, io, iot, obj);
215 EXPORT_SYMBOL(cl_io_init);
218 * Initialize read or write io.
220 * \pre iot == CIT_READ || iot == CIT_WRITE
222 int cl_io_rw_init(const struct lu_env *env, struct cl_io *io,
223 enum cl_io_type iot, loff_t pos, size_t count)
225 LINVRNT(iot == CIT_READ || iot == CIT_WRITE);
226 LINVRNT(io->ci_obj != NULL);
229 LU_OBJECT_HEADER(D_VFSTRACE, env, &io->ci_obj->co_lu,
230 "io range: %i [%llu, %llu) %i %i\n",
231 iot, (__u64)pos, (__u64)pos + count,
232 io->u.ci_rw.crw_nonblock, io->u.ci_wr.wr_append);
233 io->u.ci_rw.crw_pos = pos;
234 io->u.ci_rw.crw_count = count;
235 RETURN(cl_io_init(env, io, iot, io->ci_obj));
237 EXPORT_SYMBOL(cl_io_rw_init);
239 static inline const struct lu_fid *
240 cl_lock_descr_fid(const struct cl_lock_descr *descr)
242 return lu_object_fid(&descr->cld_obj->co_lu);
245 static int cl_lock_descr_cmp(const struct cl_lock_descr *d0,
246 const struct cl_lock_descr *d1)
248 return lu_fid_cmp(cl_lock_descr_fid(d0), cl_lock_descr_fid(d1)) ?:
249 __diff_normalize(d0->cld_start, d1->cld_start);
253 * Sort locks in lexicographical order of their (fid, start-offset) pairs.
255 static void cl_io_locks_sort(struct cl_io *io)
260 /* hidden treasure: bubble sort for now. */
262 struct cl_io_lock_link *curr;
263 struct cl_io_lock_link *prev;
264 struct cl_io_lock_link *temp;
269 list_for_each_entry_safe(curr, temp, &io->ci_lockset.cls_todo,
272 switch (cl_lock_descr_cmp(&prev->cill_descr,
273 &curr->cill_descr)) {
276 * IMPOSSIBLE: Identical locks are
283 list_move_tail(&curr->cill_linkage,
284 &prev->cill_linkage);
286 continue; /* don't change prev: it's
287 * still "previous" */
288 case -1: /* already in order */
299 * Check whether \a queue contains locks matching \a need.
301 * \retval +ve there is a matching lock in the \a queue
302 * \retval 0 there are no matching locks in the \a queue
304 int cl_queue_match(const struct list_head *queue,
305 const struct cl_lock_descr *need)
307 struct cl_io_lock_link *scan;
310 list_for_each_entry(scan, queue, cill_linkage) {
311 if (cl_lock_descr_match(&scan->cill_descr, need))
316 EXPORT_SYMBOL(cl_queue_match);
318 static int cl_lockset_match(const struct cl_lockset *set,
319 const struct cl_lock_descr *need, int all_queues)
321 return (all_queues ? cl_queue_match(&set->cls_todo, need) : 0) ||
322 cl_queue_match(&set->cls_curr, need) ||
323 cl_queue_match(&set->cls_done, need);
326 static int cl_lockset_lock_one(const struct lu_env *env,
327 struct cl_io *io, struct cl_lockset *set,
328 struct cl_io_lock_link *link)
330 struct cl_lock *lock;
335 lock = cl_lock_request(env, io, &link->cill_descr, link->cill_enq_flags,
338 link->cill_lock = lock;
339 list_move(&link->cill_linkage, &set->cls_curr);
340 if (!(link->cill_enq_flags & CEF_ASYNC)) {
341 result = cl_wait(env, lock);
343 list_move(&link->cill_linkage, &set->cls_done);
347 result = PTR_ERR(lock);
351 static void cl_lock_link_fini(const struct lu_env *env, struct cl_io *io,
352 struct cl_io_lock_link *link)
354 struct cl_lock *lock = link->cill_lock;
357 list_del_init(&link->cill_linkage);
359 cl_lock_release(env, lock, "io", io);
360 link->cill_lock = NULL;
362 if (link->cill_fini != NULL)
363 link->cill_fini(env, link);
367 static int cl_lockset_lock(const struct lu_env *env, struct cl_io *io,
368 struct cl_lockset *set)
370 struct cl_io_lock_link *link;
371 struct cl_io_lock_link *temp;
372 struct cl_lock *lock;
377 list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) {
378 if (!cl_lockset_match(set, &link->cill_descr, 0)) {
379 /* XXX some locking to guarantee that locks aren't
380 * expanded in between. */
381 result = cl_lockset_lock_one(env, io, set, link);
385 cl_lock_link_fini(env, io, link);
388 list_for_each_entry_safe(link, temp,
389 &set->cls_curr, cill_linkage) {
390 lock = link->cill_lock;
391 result = cl_wait(env, lock);
393 list_move(&link->cill_linkage, &set->cls_done);
402 * Takes locks necessary for the current iteration of io.
404 * Calls cl_io_operations::cio_lock() top-to-bottom to collect locks required
405 * by layers for the current iteration. Then sort locks (to avoid dead-locks),
408 int cl_io_lock(const struct lu_env *env, struct cl_io *io)
410 const struct cl_io_slice *scan;
413 LINVRNT(cl_io_is_loopable(io));
414 LINVRNT(io->ci_state == CIS_IT_STARTED);
415 LINVRNT(cl_io_invariant(io));
418 cl_io_for_each(scan, io) {
419 if (scan->cis_iop->op[io->ci_type].cio_lock == NULL)
421 result = scan->cis_iop->op[io->ci_type].cio_lock(env, scan);
426 cl_io_locks_sort(io);
427 result = cl_lockset_lock(env, io, &io->ci_lockset);
430 cl_io_unlock(env, io);
432 io->ci_state = CIS_LOCKED;
435 EXPORT_SYMBOL(cl_io_lock);
438 * Release locks takes by io.
440 void cl_io_unlock(const struct lu_env *env, struct cl_io *io)
442 struct cl_lockset *set;
443 struct cl_io_lock_link *link;
444 struct cl_io_lock_link *temp;
445 const struct cl_io_slice *scan;
447 LASSERT(cl_io_is_loopable(io));
448 LASSERT(CIS_IT_STARTED <= io->ci_state && io->ci_state < CIS_UNLOCKED);
449 LINVRNT(cl_io_invariant(io));
452 set = &io->ci_lockset;
454 list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage)
455 cl_lock_link_fini(env, io, link);
457 list_for_each_entry_safe(link, temp, &set->cls_curr, cill_linkage)
458 cl_lock_link_fini(env, io, link);
460 list_for_each_entry_safe(link, temp, &set->cls_done, cill_linkage) {
461 cl_unuse(env, link->cill_lock);
462 cl_lock_link_fini(env, io, link);
464 cl_io_for_each_reverse(scan, io) {
465 if (scan->cis_iop->op[io->ci_type].cio_unlock != NULL)
466 scan->cis_iop->op[io->ci_type].cio_unlock(env, scan);
468 io->ci_state = CIS_UNLOCKED;
469 LASSERT(cl_env_info(env)->clt_nr_locks_acquired == 0);
472 EXPORT_SYMBOL(cl_io_unlock);
475 * Prepares next iteration of io.
477 * Calls cl_io_operations::cio_iter_init() top-to-bottom. This exists to give
478 * layers a chance to modify io parameters, e.g., so that lov can restrict io
479 * to a single stripe.
481 int cl_io_iter_init(const struct lu_env *env, struct cl_io *io)
483 const struct cl_io_slice *scan;
486 LINVRNT(cl_io_is_loopable(io));
487 LINVRNT(io->ci_state == CIS_INIT || io->ci_state == CIS_IT_ENDED);
488 LINVRNT(cl_io_invariant(io));
492 cl_io_for_each(scan, io) {
493 if (scan->cis_iop->op[io->ci_type].cio_iter_init == NULL)
495 result = scan->cis_iop->op[io->ci_type].cio_iter_init(env,
501 io->ci_state = CIS_IT_STARTED;
504 EXPORT_SYMBOL(cl_io_iter_init);
507 * Finalizes io iteration.
509 * Calls cl_io_operations::cio_iter_fini() bottom-to-top.
511 void cl_io_iter_fini(const struct lu_env *env, struct cl_io *io)
513 const struct cl_io_slice *scan;
515 LINVRNT(cl_io_is_loopable(io));
516 LINVRNT(io->ci_state == CIS_UNLOCKED);
517 LINVRNT(cl_io_invariant(io));
520 cl_io_for_each_reverse(scan, io) {
521 if (scan->cis_iop->op[io->ci_type].cio_iter_fini != NULL)
522 scan->cis_iop->op[io->ci_type].cio_iter_fini(env, scan);
524 io->ci_state = CIS_IT_ENDED;
527 EXPORT_SYMBOL(cl_io_iter_fini);
530 * Records that read or write io progressed \a nob bytes forward.
532 void cl_io_rw_advance(const struct lu_env *env, struct cl_io *io, size_t nob)
534 const struct cl_io_slice *scan;
536 LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE ||
538 LINVRNT(cl_io_is_loopable(io));
539 LINVRNT(cl_io_invariant(io));
543 io->u.ci_rw.crw_pos += nob;
544 io->u.ci_rw.crw_count -= nob;
546 /* layers have to be notified. */
547 cl_io_for_each_reverse(scan, io) {
548 if (scan->cis_iop->op[io->ci_type].cio_advance != NULL)
549 scan->cis_iop->op[io->ci_type].cio_advance(env, scan,
554 EXPORT_SYMBOL(cl_io_rw_advance);
557 * Adds a lock to a lockset.
559 int cl_io_lock_add(const struct lu_env *env, struct cl_io *io,
560 struct cl_io_lock_link *link)
565 if (cl_lockset_match(&io->ci_lockset, &link->cill_descr, 1))
568 list_add(&link->cill_linkage, &io->ci_lockset.cls_todo);
573 EXPORT_SYMBOL(cl_io_lock_add);
575 static void cl_free_io_lock_link(const struct lu_env *env,
576 struct cl_io_lock_link *link)
582 * Allocates new lock link, and uses it to add a lock to a lockset.
584 int cl_io_lock_alloc_add(const struct lu_env *env, struct cl_io *io,
585 struct cl_lock_descr *descr, int enqflags)
587 struct cl_io_lock_link *link;
593 link->cill_descr = *descr;
594 link->cill_enq_flags = enqflags;
595 link->cill_fini = cl_free_io_lock_link;
596 result = cl_io_lock_add(env, io, link);
597 if (result) /* lock match */
598 link->cill_fini(env, link);
604 EXPORT_SYMBOL(cl_io_lock_alloc_add);
607 * Starts io by calling cl_io_operations::cio_start() top-to-bottom.
609 int cl_io_start(const struct lu_env *env, struct cl_io *io)
611 const struct cl_io_slice *scan;
614 LINVRNT(cl_io_is_loopable(io));
615 LINVRNT(io->ci_state == CIS_LOCKED);
616 LINVRNT(cl_io_invariant(io));
619 io->ci_state = CIS_IO_GOING;
620 cl_io_for_each(scan, io) {
621 if (scan->cis_iop->op[io->ci_type].cio_start == NULL)
623 result = scan->cis_iop->op[io->ci_type].cio_start(env, scan);
631 EXPORT_SYMBOL(cl_io_start);
634 * Wait until current io iteration is finished by calling
635 * cl_io_operations::cio_end() bottom-to-top.
637 void cl_io_end(const struct lu_env *env, struct cl_io *io)
639 const struct cl_io_slice *scan;
641 LINVRNT(cl_io_is_loopable(io));
642 LINVRNT(io->ci_state == CIS_IO_GOING);
643 LINVRNT(cl_io_invariant(io));
646 cl_io_for_each_reverse(scan, io) {
647 if (scan->cis_iop->op[io->ci_type].cio_end != NULL)
648 scan->cis_iop->op[io->ci_type].cio_end(env, scan);
649 /* TODO: error handling. */
651 io->ci_state = CIS_IO_FINISHED;
654 EXPORT_SYMBOL(cl_io_end);
656 static const struct cl_page_slice *
657 cl_io_slice_page(const struct cl_io_slice *ios, struct cl_page *page)
659 const struct cl_page_slice *slice;
661 slice = cl_page_at(page, ios->cis_obj->co_lu.lo_dev->ld_type);
662 LINVRNT(slice != NULL);
667 * True iff \a page is within \a io range.
669 static int cl_page_in_io(const struct cl_page *page, const struct cl_io *io)
676 idx = page->cp_index;
677 switch (io->ci_type) {
681 * check that [start, end) and [pos, pos + count) extents
684 start = cl_offset(page->cp_obj, idx);
685 end = cl_offset(page->cp_obj, idx + 1);
686 result = io->u.ci_rw.crw_pos < end &&
687 start < io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
690 result = io->u.ci_fault.ft_index == idx;
699 * Called by read io, when page has to be read from the server.
701 * \see cl_io_operations::cio_read_page()
703 int cl_io_read_page(const struct lu_env *env, struct cl_io *io,
704 struct cl_page *page)
706 const struct cl_io_slice *scan;
707 struct cl_2queue *queue;
710 LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_FAULT);
711 LINVRNT(cl_page_is_owned(page, io));
712 LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
713 LINVRNT(cl_page_in_io(page, io));
714 LINVRNT(cl_io_invariant(io));
717 queue = &io->ci_queue;
719 cl_2queue_init(queue);
721 * ->cio_read_page() methods called in the loop below are supposed to
722 * never block waiting for network (the only subtle point is the
723 * creation of new pages for read-ahead that might result in cache
724 * shrinking, but currently only clean pages are shrunk and this
725 * requires no network io).
727 * Should this ever starts blocking, retry loop would be needed for
728 * "parallel io" (see CLO_REPEAT loops in cl_lock.c).
730 cl_io_for_each(scan, io) {
731 if (scan->cis_iop->cio_read_page != NULL) {
732 const struct cl_page_slice *slice;
734 slice = cl_io_slice_page(scan, page);
735 LINVRNT(slice != NULL);
736 result = scan->cis_iop->cio_read_page(env, scan, slice);
742 result = cl_io_submit_rw(env, io, CRT_READ, queue);
744 * Unlock unsent pages in case of error.
746 cl_page_list_disown(env, io, &queue->c2_qin);
747 cl_2queue_fini(env, queue);
750 EXPORT_SYMBOL(cl_io_read_page);
753 * Called by write io to prepare page to receive data from user buffer.
755 * \see cl_io_operations::cio_prepare_write()
757 int cl_io_prepare_write(const struct lu_env *env, struct cl_io *io,
758 struct cl_page *page, unsigned from, unsigned to)
760 const struct cl_io_slice *scan;
763 LINVRNT(io->ci_type == CIT_WRITE);
764 LINVRNT(cl_page_is_owned(page, io));
765 LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
766 LINVRNT(cl_io_invariant(io));
767 LASSERT(cl_page_in_io(page, io));
770 cl_io_for_each_reverse(scan, io) {
771 if (scan->cis_iop->cio_prepare_write != NULL) {
772 const struct cl_page_slice *slice;
774 slice = cl_io_slice_page(scan, page);
775 result = scan->cis_iop->cio_prepare_write(env, scan,
784 EXPORT_SYMBOL(cl_io_prepare_write);
787 * Called by write io after user data were copied into a page.
789 * \see cl_io_operations::cio_commit_write()
791 int cl_io_commit_write(const struct lu_env *env, struct cl_io *io,
792 struct cl_page *page, unsigned from, unsigned to)
794 const struct cl_io_slice *scan;
797 LINVRNT(io->ci_type == CIT_WRITE);
798 LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
799 LINVRNT(cl_io_invariant(io));
801 * XXX Uh... not nice. Top level cl_io_commit_write() call (vvp->lov)
802 * already called cl_page_cache_add(), moving page into CPS_CACHED
803 * state. Better (and more general) way of dealing with such situation
806 LASSERT(cl_page_is_owned(page, io) || page->cp_parent != NULL);
807 LASSERT(cl_page_in_io(page, io));
810 cl_io_for_each(scan, io) {
811 if (scan->cis_iop->cio_commit_write != NULL) {
812 const struct cl_page_slice *slice;
814 slice = cl_io_slice_page(scan, page);
815 result = scan->cis_iop->cio_commit_write(env, scan,
822 LINVRNT(result <= 0);
825 EXPORT_SYMBOL(cl_io_commit_write);
828 * Submits a list of pages for immediate io.
830 * After the function gets returned, The submitted pages are moved to
831 * queue->c2_qout queue, and queue->c2_qin contain both the pages don't need
832 * to be submitted, and the pages are errant to submit.
834 * \returns 0 if at least one page was submitted, error code otherwise.
835 * \see cl_io_operations::cio_submit()
837 int cl_io_submit_rw(const struct lu_env *env, struct cl_io *io,
838 enum cl_req_type crt, struct cl_2queue *queue)
840 const struct cl_io_slice *scan;
843 LINVRNT(crt < ARRAY_SIZE(scan->cis_iop->req_op));
846 cl_io_for_each(scan, io) {
847 if (scan->cis_iop->req_op[crt].cio_submit == NULL)
849 result = scan->cis_iop->req_op[crt].cio_submit(env, scan, crt,
855 * If ->cio_submit() failed, no pages were sent.
857 LASSERT(ergo(result != 0, list_empty(&queue->c2_qout.pl_pages)));
860 EXPORT_SYMBOL(cl_io_submit_rw);
863 * Cancel an IO which has been submitted by cl_io_submit_rw.
865 int cl_io_cancel(const struct lu_env *env, struct cl_io *io,
866 struct cl_page_list *queue)
868 struct cl_page *page;
871 CERROR("Canceling ongoing page trasmission\n");
872 cl_page_list_for_each(page, queue) {
875 LINVRNT(cl_page_in_io(page, io));
876 rc = cl_page_cancel(env, page);
877 result = result ?: rc;
881 EXPORT_SYMBOL(cl_io_cancel);
886 * Pumps io through iterations calling
888 * - cl_io_iter_init()
898 * - cl_io_iter_fini()
900 * repeatedly until there is no more io to do.
902 int cl_io_loop(const struct lu_env *env, struct cl_io *io)
906 LINVRNT(cl_io_is_loopable(io));
913 result = cl_io_iter_init(env, io);
916 result = cl_io_lock(env, io);
919 * Notify layers that locks has been taken,
922 * - llite: kms, short read;
923 * - llite: generic_file_read();
925 result = cl_io_start(env, io);
927 * Send any remaining pending
930 * - llite: ll_rw_stats_tally.
933 cl_io_unlock(env, io);
934 cl_io_rw_advance(env, io, io->ci_nob - nob);
937 cl_io_iter_fini(env, io);
938 } while (result == 0 && io->ci_continue);
939 RETURN(result < 0 ? result : 0);
941 EXPORT_SYMBOL(cl_io_loop);
944 * Adds io slice to the cl_io.
946 * This is called by cl_object_operations::coo_io_init() methods to add a
947 * per-layer state to the io. New state is added at the end of
948 * cl_io::ci_layers list, that is, it is at the bottom of the stack.
950 * \see cl_lock_slice_add(), cl_req_slice_add(), cl_page_slice_add()
952 void cl_io_slice_add(struct cl_io *io, struct cl_io_slice *slice,
953 struct cl_object *obj,
954 const struct cl_io_operations *ops)
956 struct list_head *linkage = &slice->cis_linkage;
958 LASSERT((linkage->prev == NULL && linkage->next == NULL) ||
959 list_empty(linkage));
962 list_add_tail(linkage, &io->ci_layers);
964 slice->cis_obj = obj;
965 slice->cis_iop = ops;
968 EXPORT_SYMBOL(cl_io_slice_add);
972 * Initializes page list.
974 void cl_page_list_init(struct cl_page_list *plist)
978 CFS_INIT_LIST_HEAD(&plist->pl_pages);
979 plist->pl_owner = cfs_current();
982 EXPORT_SYMBOL(cl_page_list_init);
985 * Adds a page to a page list.
987 void cl_page_list_add(struct cl_page_list *plist, struct cl_page *page)
990 /* it would be better to check that page is owned by "current" io, but
991 * it is not passed here. */
992 LASSERT(page->cp_owner != NULL);
993 LINVRNT(plist->pl_owner == cfs_current());
996 mutex_lock(&page->cp_mutex);
998 LASSERT(list_empty(&page->cp_batch));
999 list_add_tail(&page->cp_batch, &plist->pl_pages);
1001 page->cp_queue_ref = lu_ref_add(&page->cp_reference, "queue", plist);
1005 EXPORT_SYMBOL(cl_page_list_add);
1008 * Removes a page from a page list.
1010 void cl_page_list_del(const struct lu_env *env,
1011 struct cl_page_list *plist, struct cl_page *page)
1013 LASSERT(plist->pl_nr > 0);
1014 LINVRNT(plist->pl_owner == cfs_current());
1017 list_del_init(&page->cp_batch);
1019 mutex_unlock(&page->cp_mutex);
1022 lu_ref_del_at(&page->cp_reference, page->cp_queue_ref, "queue", plist);
1023 cl_page_put(env, page);
1026 EXPORT_SYMBOL(cl_page_list_del);
1029 * Moves a page from one page list to another.
1031 void cl_page_list_move(struct cl_page_list *dst, struct cl_page_list *src,
1032 struct cl_page *page)
1034 LASSERT(src->pl_nr > 0);
1035 LINVRNT(dst->pl_owner == cfs_current());
1036 LINVRNT(src->pl_owner == cfs_current());
1039 list_move_tail(&page->cp_batch, &dst->pl_pages);
1042 lu_ref_set_at(&page->cp_reference,
1043 page->cp_queue_ref, "queue", src, dst);
1046 EXPORT_SYMBOL(cl_page_list_move);
1049 * splice the cl_page_list, just as list head does
1051 void cl_page_list_splice(struct cl_page_list *list, struct cl_page_list *head)
1053 struct cl_page *page;
1054 struct cl_page *tmp;
1056 LINVRNT(list->pl_owner == cfs_current());
1057 LINVRNT(head->pl_owner == cfs_current());
1060 cl_page_list_for_each_safe(page, tmp, list)
1061 cl_page_list_move(head, list, page);
1064 EXPORT_SYMBOL(cl_page_list_splice);
1066 void cl_page_disown0(const struct lu_env *env,
1067 struct cl_io *io, struct cl_page *pg);
1070 * Disowns pages in a queue.
1072 void cl_page_list_disown(const struct lu_env *env,
1073 struct cl_io *io, struct cl_page_list *plist)
1075 struct cl_page *page;
1076 struct cl_page *temp;
1078 LINVRNT(plist->pl_owner == cfs_current());
1081 cl_page_list_for_each_safe(page, temp, plist) {
1082 LASSERT(plist->pl_nr > 0);
1084 list_del_init(&page->cp_batch);
1086 mutex_unlock(&page->cp_mutex);
1090 * cl_page_disown0 rather than usual cl_page_disown() is used,
1091 * because pages are possibly in CPS_FREEING state already due
1092 * to the call to cl_page_list_discard().
1095 * XXX cl_page_disown0() will fail if page is not locked.
1097 cl_page_disown0(env, io, page);
1098 lu_ref_del(&page->cp_reference, "queue", plist);
1099 cl_page_put(env, page);
1103 EXPORT_SYMBOL(cl_page_list_disown);
1106 * Releases pages from queue.
1108 void cl_page_list_fini(const struct lu_env *env, struct cl_page_list *plist)
1110 struct cl_page *page;
1111 struct cl_page *temp;
1113 LINVRNT(plist->pl_owner == cfs_current());
1116 cl_page_list_for_each_safe(page, temp, plist)
1117 cl_page_list_del(env, plist, page);
1118 LASSERT(plist->pl_nr == 0);
1121 EXPORT_SYMBOL(cl_page_list_fini);
1124 * Owns all pages in a queue.
1126 int cl_page_list_own(const struct lu_env *env,
1127 struct cl_io *io, struct cl_page_list *plist)
1129 struct cl_page *page;
1130 struct cl_page *temp;
1133 LINVRNT(plist->pl_owner == cfs_current());
1137 cl_page_list_for_each_safe(page, temp, plist) {
1138 if (cl_page_own(env, io, page) == 0)
1139 result = result ?: page->cp_error;
1141 cl_page_list_del(env, plist, page);
1145 EXPORT_SYMBOL(cl_page_list_own);
1148 * Assumes all pages in a queue.
1150 void cl_page_list_assume(const struct lu_env *env,
1151 struct cl_io *io, struct cl_page_list *plist)
1153 struct cl_page *page;
1155 LINVRNT(plist->pl_owner == cfs_current());
1157 cl_page_list_for_each(page, plist)
1158 cl_page_assume(env, io, page);
1160 EXPORT_SYMBOL(cl_page_list_assume);
1163 * Discards all pages in a queue.
1165 void cl_page_list_discard(const struct lu_env *env, struct cl_io *io,
1166 struct cl_page_list *plist)
1168 struct cl_page *page;
1170 LINVRNT(plist->pl_owner == cfs_current());
1172 cl_page_list_for_each(page, plist)
1173 cl_page_discard(env, io, page);
1176 EXPORT_SYMBOL(cl_page_list_discard);
1179 * Unmaps all pages in a queue from user virtual memory.
1181 int cl_page_list_unmap(const struct lu_env *env, struct cl_io *io,
1182 struct cl_page_list *plist)
1184 struct cl_page *page;
1187 LINVRNT(plist->pl_owner == cfs_current());
1190 cl_page_list_for_each(page, plist) {
1191 result = cl_page_unmap(env, io, page);
1197 EXPORT_SYMBOL(cl_page_list_unmap);
1200 * Initialize dual page queue.
1202 void cl_2queue_init(struct cl_2queue *queue)
1205 cl_page_list_init(&queue->c2_qin);
1206 cl_page_list_init(&queue->c2_qout);
1209 EXPORT_SYMBOL(cl_2queue_init);
1212 * Add a page to the incoming page list of 2-queue.
1214 void cl_2queue_add(struct cl_2queue *queue, struct cl_page *page)
1217 cl_page_list_add(&queue->c2_qin, page);
1220 EXPORT_SYMBOL(cl_2queue_add);
1223 * Disown pages in both lists of a 2-queue.
1225 void cl_2queue_disown(const struct lu_env *env,
1226 struct cl_io *io, struct cl_2queue *queue)
1229 cl_page_list_disown(env, io, &queue->c2_qin);
1230 cl_page_list_disown(env, io, &queue->c2_qout);
1233 EXPORT_SYMBOL(cl_2queue_disown);
1236 * Discard (truncate) pages in both lists of a 2-queue.
1238 void cl_2queue_discard(const struct lu_env *env,
1239 struct cl_io *io, struct cl_2queue *queue)
1242 cl_page_list_discard(env, io, &queue->c2_qin);
1243 cl_page_list_discard(env, io, &queue->c2_qout);
1246 EXPORT_SYMBOL(cl_2queue_discard);
1249 * Assume to own the pages in cl_2queue
1251 void cl_2queue_assume(const struct lu_env *env,
1252 struct cl_io *io, struct cl_2queue *queue)
1254 cl_page_list_assume(env, io, &queue->c2_qin);
1255 cl_page_list_assume(env, io, &queue->c2_qout);
1257 EXPORT_SYMBOL(cl_2queue_assume);
1260 * Finalize both page lists of a 2-queue.
1262 void cl_2queue_fini(const struct lu_env *env, struct cl_2queue *queue)
1265 cl_page_list_fini(env, &queue->c2_qout);
1266 cl_page_list_fini(env, &queue->c2_qin);
1269 EXPORT_SYMBOL(cl_2queue_fini);
1272 * Initialize a 2-queue to contain \a page in its incoming page list.
1274 void cl_2queue_init_page(struct cl_2queue *queue, struct cl_page *page)
1277 cl_2queue_init(queue);
1278 cl_2queue_add(queue, page);
1281 EXPORT_SYMBOL(cl_2queue_init_page);
1284 * Returns top-level io.
1286 * \see cl_object_top(), cl_page_top().
1288 struct cl_io *cl_io_top(struct cl_io *io)
1291 while (io->ci_parent != NULL)
1295 EXPORT_SYMBOL(cl_io_top);
1298 * Prints human readable representation of \a io to the \a f.
1300 void cl_io_print(const struct lu_env *env, void *cookie,
1301 lu_printer_t printer, const struct cl_io *io)
1306 * Adds request slice to the compound request.
1308 * This is called by cl_device_operations::cdo_req_init() methods to add a
1309 * per-layer state to the request. New state is added at the end of
1310 * cl_req::crq_layers list, that is, it is at the bottom of the stack.
1312 * \see cl_lock_slice_add(), cl_page_slice_add(), cl_io_slice_add()
1314 void cl_req_slice_add(struct cl_req *req, struct cl_req_slice *slice,
1315 struct cl_device *dev,
1316 const struct cl_req_operations *ops)
1319 list_add_tail(&slice->crs_linkage, &req->crq_layers);
1320 slice->crs_dev = dev;
1321 slice->crs_ops = ops;
1322 slice->crs_req = req;
1325 EXPORT_SYMBOL(cl_req_slice_add);
1327 static void cl_req_free(const struct lu_env *env, struct cl_req *req)
1331 LASSERT(list_empty(&req->crq_pages));
1332 LASSERT(req->crq_nrpages == 0);
1333 LINVRNT(list_empty(&req->crq_layers));
1334 LINVRNT(equi(req->crq_nrobjs > 0, req->crq_o != NULL));
1337 if (req->crq_o != NULL) {
1338 for (i = 0; i < req->crq_nrobjs; ++i) {
1339 struct cl_object *obj = req->crq_o[i].ro_obj;
1341 lu_object_ref_del_at(&obj->co_lu,
1342 req->crq_o[i].ro_obj_ref,
1344 cl_object_put(env, obj);
1347 OBD_FREE(req->crq_o, req->crq_nrobjs * sizeof req->crq_o[0]);
1353 static int cl_req_init(const struct lu_env *env, struct cl_req *req,
1354 struct cl_page *page)
1356 struct cl_device *dev;
1357 struct cl_page_slice *slice;
1362 page = cl_page_top(page);
1364 list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
1365 dev = lu2cl_dev(slice->cpl_obj->co_lu.lo_dev);
1366 if (dev->cd_ops->cdo_req_init != NULL) {
1367 result = dev->cd_ops->cdo_req_init(env,
1373 page = page->cp_child;
1374 } while (page != NULL && result == 0);
1379 * Invokes per-request transfer completion call-backs
1380 * (cl_req_operations::cro_completion()) bottom-to-top.
1382 void cl_req_completion(const struct lu_env *env, struct cl_req *req, int rc)
1384 struct cl_req_slice *slice;
1388 * for the lack of list_for_each_entry_reverse_safe()...
1390 while (!list_empty(&req->crq_layers)) {
1391 slice = list_entry(req->crq_layers.prev,
1392 struct cl_req_slice, crs_linkage);
1393 list_del_init(&slice->crs_linkage);
1394 if (slice->crs_ops->cro_completion != NULL)
1395 slice->crs_ops->cro_completion(env, slice, rc);
1397 cl_req_free(env, req);
1400 EXPORT_SYMBOL(cl_req_completion);
1403 * Allocates new transfer request.
1405 struct cl_req *cl_req_alloc(const struct lu_env *env, struct cl_page *page,
1406 enum cl_req_type crt, int nr_objects)
1410 LINVRNT(nr_objects > 0);
1417 OBD_ALLOC(req->crq_o, nr_objects * sizeof req->crq_o[0]);
1418 if (req->crq_o != NULL) {
1419 req->crq_nrobjs = nr_objects;
1420 req->crq_type = crt;
1421 CFS_INIT_LIST_HEAD(&req->crq_pages);
1422 CFS_INIT_LIST_HEAD(&req->crq_layers);
1423 result = cl_req_init(env, req, page);
1427 cl_req_completion(env, req, result);
1428 req = ERR_PTR(result);
1431 req = ERR_PTR(-ENOMEM);
1434 EXPORT_SYMBOL(cl_req_alloc);
1437 * Adds a page to a request.
1439 void cl_req_page_add(const struct lu_env *env,
1440 struct cl_req *req, struct cl_page *page)
1442 struct cl_object *obj;
1443 struct cl_req_obj *rqo;
1447 page = cl_page_top(page);
1449 LINVRNT(cl_page_is_vmlocked(env, page));
1450 LASSERT(list_empty(&page->cp_flight));
1451 LASSERT(page->cp_req == NULL);
1453 list_add_tail(&page->cp_flight, &req->crq_pages);
1456 obj = cl_object_top(page->cp_obj);
1457 for (i = 0, rqo = req->crq_o; obj != rqo->ro_obj; ++i, ++rqo) {
1458 if (rqo->ro_obj == NULL) {
1461 rqo->ro_obj_ref = lu_object_ref_add(&obj->co_lu,
1466 LASSERT(i < req->crq_nrobjs);
1469 EXPORT_SYMBOL(cl_req_page_add);
1472 * Removes a page from a request.
1474 void cl_req_page_done(const struct lu_env *env, struct cl_page *page)
1476 struct cl_req *req = page->cp_req;
1479 page = cl_page_top(page);
1481 LINVRNT(cl_page_is_vmlocked(env, page));
1482 LASSERT(!list_empty(&page->cp_flight));
1483 LASSERT(req->crq_nrpages > 0);
1485 list_del_init(&page->cp_flight);
1487 page->cp_req = NULL;
1490 EXPORT_SYMBOL(cl_req_page_done);
1493 * Notifies layers that request is about to depart by calling
1494 * cl_req_operations::cro_prep() top-to-bottom.
1496 int cl_req_prep(const struct lu_env *env, struct cl_req *req)
1500 const struct cl_req_slice *slice;
1504 * Check that the caller of cl_req_alloc() didn't lie about the number
1507 for (i = 0; i < req->crq_nrobjs; ++i)
1508 LASSERT(req->crq_o[i].ro_obj != NULL);
1511 list_for_each_entry(slice, &req->crq_layers, crs_linkage) {
1512 if (slice->crs_ops->cro_prep != NULL) {
1513 result = slice->crs_ops->cro_prep(env, slice);
1520 EXPORT_SYMBOL(cl_req_prep);
1523 * Fills in attributes that are passed to server together with transfer. Only
1524 * attributes from \a flags may be touched. This can be called multiple times
1525 * for the same request.
1527 void cl_req_attr_set(const struct lu_env *env, struct cl_req *req,
1528 struct cl_req_attr *attr, obd_valid flags)
1530 const struct cl_req_slice *slice;
1531 struct cl_page *page;
1534 LASSERT(!list_empty(&req->crq_pages));
1537 /* Take any page to use as a model. */
1538 page = list_entry(req->crq_pages.next, struct cl_page, cp_flight);
1540 for (i = 0; i < req->crq_nrobjs; ++i) {
1541 list_for_each_entry(slice, &req->crq_layers, crs_linkage) {
1542 const struct cl_page_slice *scan;
1543 const struct cl_object *obj;
1545 scan = cl_page_at(page,
1546 slice->crs_dev->cd_lu_dev.ld_type);
1547 LASSERT(scan != NULL);
1548 obj = scan->cpl_obj;
1549 if (slice->crs_ops->cro_attr_set != NULL)
1550 slice->crs_ops->cro_attr_set(env, slice, obj,
1556 EXPORT_SYMBOL(cl_req_attr_set);
1558 /* XXX complete(), init_completion(), and wait_for_completion(), until they are
1559 * implemented in libcfs. */
1561 # include <linux/sched.h>
1562 #else /* __KERNEL__ */
1563 # include <liblustre.h>
1567 * Initialize synchronous io wait anchor, for transfer of \a nrpages pages.
1569 void cl_sync_io_init(struct cl_sync_io *anchor, int nrpages)
1572 init_completion(&anchor->csi_sync_completion);
1573 atomic_set(&anchor->csi_sync_nr, nrpages);
1574 anchor->csi_sync_rc = 0;
1577 EXPORT_SYMBOL(cl_sync_io_init);
1580 * Wait until all transfer completes. Transfer completion routine has to call
1581 * cl_sync_io_note() for every page.
1583 int cl_sync_io_wait(const struct lu_env *env, struct cl_io *io,
1584 struct cl_page_list *queue, struct cl_sync_io *anchor)
1589 rc = wait_for_completion_interruptible(&anchor->csi_sync_completion);
1592 rc2 = cl_io_cancel(env, io, queue);
1594 /* Too bad, some pages are still in IO. */
1595 CDEBUG(D_VFSTRACE, "Failed to cancel transfer (%i). "
1596 "Waiting for %i pages\n",
1597 rc2, atomic_read(&anchor->csi_sync_nr));
1598 wait_for_completion(&anchor->csi_sync_completion);
1601 rc = anchor->csi_sync_rc;
1602 LASSERT(atomic_read(&anchor->csi_sync_nr) == 0);
1603 cl_page_list_assume(env, io, queue);
1604 POISON(anchor, 0x5a, sizeof *anchor);
1607 EXPORT_SYMBOL(cl_sync_io_wait);
1610 * Indicate that transfer of a single page completed.
1612 void cl_sync_io_note(struct cl_sync_io *anchor, int ioret)
1615 if (anchor->csi_sync_rc == 0 && ioret < 0)
1616 anchor->csi_sync_rc = ioret;
1618 * Synchronous IO done without releasing page lock (e.g., as a part of
1619 * ->{prepare,commit}_write(). Completion is used to signal the end of
1622 if (atomic_dec_and_test(&anchor->csi_sync_nr))
1623 complete(&anchor->csi_sync_completion);
1626 EXPORT_SYMBOL(cl_sync_io_note);