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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Implementation of cl_io for OSC layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_OSC
44 #include <lustre_obdo.h>
46 #include "osc_cl_internal.h"
52 /*****************************************************************************
58 static struct osc_io *cl2osc_io(const struct lu_env *env,
59 const struct cl_io_slice *slice)
61 struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl);
62 LINVRNT(oio == osc_env_io(env));
66 /*****************************************************************************
72 static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
76 static void osc_read_ahead_release(const struct lu_env *env,
79 struct ldlm_lock *dlmlock = cbdata;
80 struct lustre_handle lockh;
82 ldlm_lock2handle(dlmlock, &lockh);
83 ldlm_lock_decref(&lockh, LCK_PR);
84 LDLM_LOCK_PUT(dlmlock);
87 static int osc_io_read_ahead(const struct lu_env *env,
88 const struct cl_io_slice *ios,
89 pgoff_t start, struct cl_read_ahead *ra)
91 struct osc_object *osc = cl2osc(ios->cis_obj);
92 struct ldlm_lock *dlmlock;
93 int result = -ENODATA;
96 dlmlock = osc_dlmlock_at_pgoff(env, osc, start, 0);
97 if (dlmlock != NULL) {
98 LASSERT(dlmlock->l_ast_data == osc);
99 if (dlmlock->l_req_mode != LCK_PR) {
100 struct lustre_handle lockh;
101 ldlm_lock2handle(dlmlock, &lockh);
102 ldlm_lock_addref(&lockh, LCK_PR);
103 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
106 ra->cra_rpc_size = osc_cli(osc)->cl_max_pages_per_rpc;
107 ra->cra_end = cl_index(osc2cl(osc),
108 dlmlock->l_policy_data.l_extent.end);
109 ra->cra_release = osc_read_ahead_release;
110 ra->cra_cbdata = dlmlock;
118 * An implementation of cl_io_operations::cio_io_submit() method for osc
119 * layer. Iterates over pages in the in-queue, prepares each for io by calling
120 * cl_page_prep() and then either submits them through osc_io_submit_page()
121 * or, if page is already submitted, changes osc flags through
122 * osc_set_async_flags().
124 static int osc_io_submit(const struct lu_env *env,
125 const struct cl_io_slice *ios,
126 enum cl_req_type crt, struct cl_2queue *queue)
128 struct cl_page *page;
130 struct client_obd *cli = NULL;
131 struct osc_object *osc = NULL; /* to keep gcc happy */
132 struct osc_page *opg;
134 struct list_head list = LIST_HEAD_INIT(list);
136 struct cl_page_list *qin = &queue->c2_qin;
137 struct cl_page_list *qout = &queue->c2_qout;
138 unsigned int queued = 0;
142 unsigned int max_pages;
144 LASSERT(qin->pl_nr > 0);
146 CDEBUG(D_CACHE|D_READA, "%d %d\n", qin->pl_nr, crt);
148 osc = cl2osc(ios->cis_obj);
150 max_pages = cli->cl_max_pages_per_rpc;
152 cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
153 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
156 * NOTE: here @page is a top-level page. This is done to avoid
157 * creation of sub-page-list.
159 cl_page_list_for_each_safe(page, tmp, qin) {
160 struct osc_async_page *oap;
166 opg = osc_cl_page_osc(page, osc);
168 LASSERT(osc == oap->oap_obj);
170 if (!list_empty(&oap->oap_pending_item) ||
171 !list_empty(&oap->oap_rpc_item)) {
172 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
178 result = cl_page_prep(env, io, page, crt);
181 if (result != -EALREADY)
184 * Handle -EALREADY error: for read case, the page is
185 * already in UPTODATE state; for write, the page
192 spin_lock(&oap->oap_lock);
193 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
194 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
195 spin_unlock(&oap->oap_lock);
197 osc_page_submit(env, opg, crt, brw_flags);
198 list_add_tail(&oap->oap_pending_item, &list);
200 if (page->cp_sync_io != NULL)
201 cl_page_list_move(qout, qin, page);
203 cl_page_list_del(env, qin, page);
205 if (++queued == max_pages) {
207 result = osc_queue_sync_pages(env, osc, &list, cmd,
215 result = osc_queue_sync_pages(env, osc, &list, cmd, brw_flags);
217 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
218 return qout->pl_nr > 0 ? 0 : result;
222 * This is called when a page is accessed within file in a way that creates
223 * new page, if one were missing (i.e., if there were a hole at that place in
224 * the file, or accessed page is beyond the current file size).
226 * Expand stripe KMS if necessary.
228 static void osc_page_touch_at(const struct lu_env *env,
229 struct cl_object *obj, pgoff_t idx, size_t to)
231 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
232 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
236 /* offset within stripe */
237 kms = cl_offset(obj, idx) + to;
239 cl_object_attr_lock(obj);
243 * ll_inode_size_lock(inode, 0); lov_stripe_lock(lsm);
247 CDEBUG(D_INODE, "stripe KMS %sincreasing "LPU64"->"LPU64" "LPU64"\n",
248 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
249 loi->loi_lvb.lvb_size);
251 attr->cat_mtime = attr->cat_ctime = LTIME_S(CURRENT_TIME);
252 valid = CAT_MTIME | CAT_CTIME;
253 if (kms > loi->loi_kms) {
257 if (kms > loi->loi_lvb.lvb_size) {
258 attr->cat_size = kms;
261 cl_object_attr_update(env, obj, attr, valid);
262 cl_object_attr_unlock(obj);
265 static int osc_io_commit_async(const struct lu_env *env,
266 const struct cl_io_slice *ios,
267 struct cl_page_list *qin, int from, int to,
270 struct cl_io *io = ios->cis_io;
271 struct osc_io *oio = cl2osc_io(env, ios);
272 struct osc_object *osc = cl2osc(ios->cis_obj);
273 struct cl_page *page;
274 struct cl_page *last_page;
275 struct osc_page *opg;
279 LASSERT(qin->pl_nr > 0);
281 /* Handle partial page cases */
282 last_page = cl_page_list_last(qin);
283 if (oio->oi_lockless) {
284 page = cl_page_list_first(qin);
285 if (page == last_page) {
286 cl_page_clip(env, page, from, to);
289 cl_page_clip(env, page, from, PAGE_SIZE);
291 cl_page_clip(env, last_page, 0, to);
295 while (qin->pl_nr > 0) {
296 struct osc_async_page *oap;
298 page = cl_page_list_first(qin);
299 opg = osc_cl_page_osc(page, osc);
302 if (!list_empty(&oap->oap_rpc_item)) {
303 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
309 /* The page may be already in dirty cache. */
310 if (list_empty(&oap->oap_pending_item)) {
311 result = osc_page_cache_add(env, &opg->ops_cl, io);
316 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
317 page == last_page ? to : PAGE_SIZE);
319 cl_page_list_del(env, qin, page);
321 (*cb)(env, io, page);
322 /* Can't access page any more. Page can be in transfer and
323 * complete at any time. */
326 /* for sync write, kernel will wait for this page to be flushed before
327 * osc_io_end() is called, so release it earlier.
328 * for mkwrite(), it's known there is no further pages. */
329 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
330 osc_extent_release(env, oio->oi_active);
331 oio->oi_active = NULL;
334 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
338 static int osc_io_iter_init(const struct lu_env *env,
339 const struct cl_io_slice *ios)
341 struct osc_object *osc = cl2osc(ios->cis_obj);
342 struct obd_import *imp = osc_cli(osc)->cl_import;
345 spin_lock(&imp->imp_lock);
346 if (likely(!imp->imp_invalid)) {
347 struct osc_io *oio = osc_env_io(env);
349 atomic_inc(&osc->oo_nr_ios);
350 oio->oi_is_active = 1;
353 spin_unlock(&imp->imp_lock);
358 static int osc_io_write_iter_init(const struct lu_env *env,
359 const struct cl_io_slice *ios)
361 struct cl_io *io = ios->cis_io;
362 struct osc_io *oio = osc_env_io(env);
363 struct osc_object *osc = cl2osc(ios->cis_obj);
364 unsigned long npages;
367 if (cl_io_is_append(io))
368 RETURN(osc_io_iter_init(env, ios));
370 npages = io->u.ci_rw.crw_count >> PAGE_CACHE_SHIFT;
371 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
374 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
376 RETURN(osc_io_iter_init(env, ios));
379 static void osc_io_iter_fini(const struct lu_env *env,
380 const struct cl_io_slice *ios)
382 struct osc_io *oio = osc_env_io(env);
384 if (oio->oi_is_active) {
385 struct osc_object *osc = cl2osc(ios->cis_obj);
387 oio->oi_is_active = 0;
388 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
389 if (atomic_dec_and_test(&osc->oo_nr_ios))
390 wake_up_all(&osc->oo_io_waitq);
394 static void osc_io_write_iter_fini(const struct lu_env *env,
395 const struct cl_io_slice *ios)
397 struct osc_io *oio = osc_env_io(env);
398 struct osc_object *osc = cl2osc(ios->cis_obj);
400 if (oio->oi_lru_reserved > 0) {
401 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
402 oio->oi_lru_reserved = 0;
404 oio->oi_write_osclock = NULL;
406 osc_io_iter_fini(env, ios);
409 static int osc_io_fault_start(const struct lu_env *env,
410 const struct cl_io_slice *ios)
413 struct cl_fault_io *fio;
417 fio = &io->u.ci_fault;
418 CDEBUG(D_INFO, "%lu %d %zu\n",
419 fio->ft_index, fio->ft_writable, fio->ft_nob);
421 * If mapping is writeable, adjust kms to cover this page,
422 * but do not extend kms beyond actual file size.
425 if (fio->ft_writable)
426 osc_page_touch_at(env, ios->cis_obj,
427 fio->ft_index, fio->ft_nob);
431 static int osc_async_upcall(void *a, int rc)
433 struct osc_async_cbargs *args = a;
436 complete(&args->opc_sync);
441 * Checks that there are no pages being written in the extent being truncated.
443 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
444 struct osc_page *ops , void *cbdata)
446 struct cl_page *page = ops->ops_cl.cpl_page;
447 struct osc_async_page *oap;
448 __u64 start = *(__u64 *)cbdata;
451 if (oap->oap_cmd & OBD_BRW_WRITE &&
452 !list_empty(&oap->oap_pending_item))
453 CL_PAGE_DEBUG(D_ERROR, env, page, "exists " LPU64 "/%s.\n",
454 start, current->comm);
456 if (PageLocked(page->cp_vmpage))
457 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
458 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
460 return CLP_GANG_OKAY;
463 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
464 struct osc_io *oio, __u64 size)
466 struct cl_object *clob;
470 clob = oio->oi_cl.cis_obj;
471 start = cl_index(clob, size);
472 partial = cl_offset(clob, start) < size;
475 * Complain if there are pages in the truncated region.
477 osc_page_gang_lookup(env, io, cl2osc(clob),
478 start + partial, CL_PAGE_EOF,
479 trunc_check_cb, (void *)&size);
482 static int osc_io_setattr_start(const struct lu_env *env,
483 const struct cl_io_slice *slice)
485 struct cl_io *io = slice->cis_io;
486 struct osc_io *oio = cl2osc_io(env, slice);
487 struct cl_object *obj = slice->cis_obj;
488 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
489 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
490 struct obdo *oa = &oio->oi_oa;
491 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
492 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
493 unsigned int ia_valid = io->u.ci_setattr.sa_valid;
496 /* truncate cache dirty pages first */
497 if (cl_io_is_trunc(io))
498 result = osc_cache_truncate_start(env, cl2osc(obj), size,
501 if (result == 0 && oio->oi_lockless == 0) {
502 cl_object_attr_lock(obj);
503 result = cl_object_attr_get(env, obj, attr);
505 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
506 unsigned int cl_valid = 0;
508 if (ia_valid & ATTR_SIZE) {
509 attr->cat_size = attr->cat_kms = size;
510 cl_valid = (CAT_SIZE | CAT_KMS);
512 if (ia_valid & ATTR_MTIME_SET) {
513 attr->cat_mtime = lvb->lvb_mtime;
514 cl_valid |= CAT_MTIME;
516 if (ia_valid & ATTR_ATIME_SET) {
517 attr->cat_atime = lvb->lvb_atime;
518 cl_valid |= CAT_ATIME;
520 if (ia_valid & ATTR_CTIME_SET) {
521 attr->cat_ctime = lvb->lvb_ctime;
522 cl_valid |= CAT_CTIME;
524 result = cl_object_attr_update(env, obj, attr,
527 cl_object_attr_unlock(obj);
529 memset(oa, 0, sizeof(*oa));
531 oa->o_oi = loi->loi_oi;
532 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
533 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
534 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
535 if (ia_valid & ATTR_CTIME) {
536 oa->o_valid |= OBD_MD_FLCTIME;
537 oa->o_ctime = attr->cat_ctime;
539 if (ia_valid & ATTR_ATIME) {
540 oa->o_valid |= OBD_MD_FLATIME;
541 oa->o_atime = attr->cat_atime;
543 if (ia_valid & ATTR_MTIME) {
544 oa->o_valid |= OBD_MD_FLMTIME;
545 oa->o_mtime = attr->cat_mtime;
547 if (ia_valid & ATTR_SIZE) {
549 oa->o_blocks = OBD_OBJECT_EOF;
550 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
552 if (oio->oi_lockless) {
553 oa->o_flags = OBD_FL_SRVLOCK;
554 oa->o_valid |= OBD_MD_FLFLAGS;
557 LASSERT(oio->oi_lockless == 0);
560 if (ia_valid & ATTR_ATTR_FLAG) {
561 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
562 oa->o_valid |= OBD_MD_FLFLAGS;
565 init_completion(&cbargs->opc_sync);
567 if (ia_valid & ATTR_SIZE)
568 result = osc_punch_base(osc_export(cl2osc(obj)),
569 oa, osc_async_upcall,
570 cbargs, PTLRPCD_SET);
572 result = osc_setattr_async(osc_export(cl2osc(obj)),
573 oa, osc_async_upcall,
574 cbargs, PTLRPCD_SET);
576 cbargs->opc_rpc_sent = result == 0;
582 static void osc_io_setattr_end(const struct lu_env *env,
583 const struct cl_io_slice *slice)
585 struct cl_io *io = slice->cis_io;
586 struct osc_io *oio = cl2osc_io(env, slice);
587 struct cl_object *obj = slice->cis_obj;
588 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
591 if (cbargs->opc_rpc_sent) {
592 wait_for_completion(&cbargs->opc_sync);
593 result = io->ci_result = cbargs->opc_rc;
596 if (oio->oi_lockless) {
597 /* lockless truncate */
598 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
600 LASSERT(cl_io_is_trunc(io));
601 /* XXX: Need a lock. */
602 osd->od_stats.os_lockless_truncates++;
606 if (cl_io_is_trunc(io)) {
607 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
608 osc_trunc_check(env, io, oio, size);
609 osc_cache_truncate_end(env, oio->oi_trunc);
610 oio->oi_trunc = NULL;
614 struct osc_data_version_args {
615 struct osc_io *dva_oio;
619 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
622 struct osc_data_version_args *dva = arg;
623 struct osc_io *oio = dva->dva_oio;
624 const struct ost_body *body;
630 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
632 GOTO(out, rc = -EPROTO);
634 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
638 oio->oi_cbarg.opc_rc = rc;
639 complete(&oio->oi_cbarg.opc_sync);
644 static int osc_io_data_version_start(const struct lu_env *env,
645 const struct cl_io_slice *slice)
647 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
648 struct osc_io *oio = cl2osc_io(env, slice);
649 struct obdo *oa = &oio->oi_oa;
650 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
651 struct osc_object *obj = cl2osc(slice->cis_obj);
652 struct lov_oinfo *loi = obj->oo_oinfo;
653 struct obd_export *exp = osc_export(obj);
654 struct ptlrpc_request *req;
655 struct ost_body *body;
656 struct osc_data_version_args *dva;
660 memset(oa, 0, sizeof(*oa));
661 oa->o_oi = loi->loi_oi;
662 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
664 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
665 oa->o_valid |= OBD_MD_FLFLAGS;
666 oa->o_flags |= OBD_FL_SRVLOCK;
667 if (dv->dv_flags & LL_DV_WR_FLUSH)
668 oa->o_flags |= OBD_FL_FLUSH;
671 init_completion(&cbargs->opc_sync);
673 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
677 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
679 ptlrpc_request_free(req);
683 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
684 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
686 ptlrpc_request_set_replen(req);
687 req->rq_interpret_reply = osc_data_version_interpret;
688 CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
689 dva = ptlrpc_req_async_args(req);
692 ptlrpcd_add_req(req);
697 static void osc_io_data_version_end(const struct lu_env *env,
698 const struct cl_io_slice *slice)
700 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
701 struct osc_io *oio = cl2osc_io(env, slice);
702 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
705 wait_for_completion(&cbargs->opc_sync);
707 if (cbargs->opc_rc != 0) {
708 slice->cis_io->ci_result = cbargs->opc_rc;
709 } else if (!(oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)) {
710 slice->cis_io->ci_result = -EOPNOTSUPP;
712 dv->dv_data_version = oio->oi_oa.o_data_version;
713 slice->cis_io->ci_result = 0;
719 static int osc_io_read_start(const struct lu_env *env,
720 const struct cl_io_slice *slice)
722 struct cl_object *obj = slice->cis_obj;
723 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
727 if (!slice->cis_io->ci_noatime) {
728 cl_object_attr_lock(obj);
729 attr->cat_atime = LTIME_S(CURRENT_TIME);
730 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
731 cl_object_attr_unlock(obj);
737 static int osc_io_write_start(const struct lu_env *env,
738 const struct cl_io_slice *slice)
740 struct cl_object *obj = slice->cis_obj;
741 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
745 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
746 cl_object_attr_lock(obj);
747 attr->cat_mtime = attr->cat_ctime = LTIME_S(CURRENT_TIME);
748 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
749 cl_object_attr_unlock(obj);
754 static int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
755 struct cl_fsync_io *fio)
757 struct osc_io *oio = osc_env_io(env);
758 struct obdo *oa = &oio->oi_oa;
759 struct lov_oinfo *loi = obj->oo_oinfo;
760 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
764 memset(oa, 0, sizeof(*oa));
765 oa->o_oi = loi->loi_oi;
766 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
768 /* reload size abd blocks for start and end of sync range */
769 oa->o_size = fio->fi_start;
770 oa->o_blocks = fio->fi_end;
771 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
773 obdo_set_parent_fid(oa, fio->fi_fid);
775 init_completion(&cbargs->opc_sync);
777 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
781 static int osc_io_fsync_start(const struct lu_env *env,
782 const struct cl_io_slice *slice)
784 struct cl_io *io = slice->cis_io;
785 struct cl_fsync_io *fio = &io->u.ci_fsync;
786 struct cl_object *obj = slice->cis_obj;
787 struct osc_object *osc = cl2osc(obj);
788 pgoff_t start = cl_index(obj, fio->fi_start);
789 pgoff_t end = cl_index(obj, fio->fi_end);
793 if (fio->fi_end == OBD_OBJECT_EOF)
796 result = osc_cache_writeback_range(env, osc, start, end, 0,
797 fio->fi_mode == CL_FSYNC_DISCARD);
799 fio->fi_nr_written += result;
802 if (fio->fi_mode == CL_FSYNC_ALL) {
805 /* we have to wait for writeback to finish before we can
806 * send OST_SYNC RPC. This is bad because it causes extents
807 * to be written osc by osc. However, we usually start
808 * writeback before CL_FSYNC_ALL so this won't have any real
810 rc = osc_cache_wait_range(env, osc, start, end);
813 rc = osc_fsync_ost(env, osc, fio);
821 static void osc_io_fsync_end(const struct lu_env *env,
822 const struct cl_io_slice *slice)
824 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
825 struct cl_object *obj = slice->cis_obj;
826 pgoff_t start = cl_index(obj, fio->fi_start);
827 pgoff_t end = cl_index(obj, fio->fi_end);
830 if (fio->fi_mode == CL_FSYNC_LOCAL) {
831 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
832 } else if (fio->fi_mode == CL_FSYNC_ALL) {
833 struct osc_io *oio = cl2osc_io(env, slice);
834 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
836 wait_for_completion(&cbargs->opc_sync);
838 result = cbargs->opc_rc;
840 slice->cis_io->ci_result = result;
843 static int osc_io_ladvise_start(const struct lu_env *env,
844 const struct cl_io_slice *slice)
847 struct cl_io *io = slice->cis_io;
848 struct osc_io *oio = cl2osc_io(env, slice);
849 struct cl_object *obj = slice->cis_obj;
850 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
851 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
852 struct obdo *oa = &oio->oi_oa;
853 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
854 struct lu_ladvise *ladvise;
855 struct ladvise_hdr *ladvise_hdr;
860 /* TODO: add multiple ladvise support in CLIO */
861 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
862 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
863 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
865 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
866 if (ladvise_hdr == NULL)
869 memset(ladvise_hdr, 0, buf_size);
870 ladvise_hdr->lah_magic = LADVISE_MAGIC;
871 ladvise_hdr->lah_count = num_advise;
872 ladvise_hdr->lah_flags = lio->li_flags;
874 memset(oa, 0, sizeof(*oa));
875 oa->o_oi = loi->loi_oi;
876 oa->o_valid = OBD_MD_FLID;
877 obdo_set_parent_fid(oa, lio->li_fid);
879 ladvise = ladvise_hdr->lah_advise;
880 ladvise->lla_start = lio->li_start;
881 ladvise->lla_end = lio->li_end;
882 ladvise->lla_advice = lio->li_advice;
884 if (lio->li_flags & LF_ASYNC) {
885 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
886 ladvise_hdr, NULL, NULL, NULL);
888 init_completion(&cbargs->opc_sync);
889 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
890 ladvise_hdr, osc_async_upcall,
891 cbargs, PTLRPCD_SET);
892 cbargs->opc_rpc_sent = result == 0;
897 static void osc_io_ladvise_end(const struct lu_env *env,
898 const struct cl_io_slice *slice)
900 struct cl_io *io = slice->cis_io;
901 struct osc_io *oio = cl2osc_io(env, slice);
902 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
904 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
906 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
907 wait_for_completion(&cbargs->opc_sync);
908 result = cbargs->opc_rc;
910 slice->cis_io->ci_result = result;
913 static void osc_io_end(const struct lu_env *env,
914 const struct cl_io_slice *slice)
916 struct osc_io *oio = cl2osc_io(env, slice);
918 if (oio->oi_active) {
919 osc_extent_release(env, oio->oi_active);
920 oio->oi_active = NULL;
924 static const struct cl_io_operations osc_io_ops = {
927 .cio_iter_init = osc_io_iter_init,
928 .cio_iter_fini = osc_io_iter_fini,
929 .cio_start = osc_io_read_start,
930 .cio_fini = osc_io_fini
933 .cio_iter_init = osc_io_write_iter_init,
934 .cio_iter_fini = osc_io_write_iter_fini,
935 .cio_start = osc_io_write_start,
936 .cio_end = osc_io_end,
937 .cio_fini = osc_io_fini
940 .cio_iter_init = osc_io_iter_init,
941 .cio_iter_fini = osc_io_iter_fini,
942 .cio_start = osc_io_setattr_start,
943 .cio_end = osc_io_setattr_end
945 [CIT_DATA_VERSION] = {
946 .cio_start = osc_io_data_version_start,
947 .cio_end = osc_io_data_version_end,
950 .cio_iter_init = osc_io_iter_init,
951 .cio_iter_fini = osc_io_iter_fini,
952 .cio_start = osc_io_fault_start,
953 .cio_end = osc_io_end,
954 .cio_fini = osc_io_fini
957 .cio_start = osc_io_fsync_start,
958 .cio_end = osc_io_fsync_end,
959 .cio_fini = osc_io_fini
962 .cio_start = osc_io_ladvise_start,
963 .cio_end = osc_io_ladvise_end,
964 .cio_fini = osc_io_fini
967 .cio_fini = osc_io_fini
970 .cio_read_ahead = osc_io_read_ahead,
971 .cio_submit = osc_io_submit,
972 .cio_commit_async = osc_io_commit_async
975 /*****************************************************************************
977 * Transfer operations.
981 int osc_io_init(const struct lu_env *env,
982 struct cl_object *obj, struct cl_io *io)
984 struct osc_io *oio = osc_env_io(env);
986 CL_IO_SLICE_CLEAN(oio, oi_cl);
987 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);