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, 2014, 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 "osc_cl_internal.h"
50 /*****************************************************************************
56 static struct osc_io *cl2osc_io(const struct lu_env *env,
57 const struct cl_io_slice *slice)
59 struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl);
60 LINVRNT(oio == osc_env_io(env));
64 /*****************************************************************************
70 static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
74 static void osc_read_ahead_release(const struct lu_env *env,
77 struct ldlm_lock *dlmlock = cbdata;
78 struct lustre_handle lockh;
80 ldlm_lock2handle(dlmlock, &lockh);
81 ldlm_lock_decref(&lockh, LCK_PR);
82 LDLM_LOCK_PUT(dlmlock);
85 static int osc_io_read_ahead(const struct lu_env *env,
86 const struct cl_io_slice *ios,
87 pgoff_t start, struct cl_read_ahead *ra)
89 struct osc_object *osc = cl2osc(ios->cis_obj);
90 struct ldlm_lock *dlmlock;
91 int result = -ENODATA;
94 dlmlock = osc_dlmlock_at_pgoff(env, osc, start, 0);
95 if (dlmlock != NULL) {
96 if (dlmlock->l_req_mode != LCK_PR) {
97 struct lustre_handle lockh;
98 ldlm_lock2handle(dlmlock, &lockh);
99 ldlm_lock_addref(&lockh, LCK_PR);
100 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
103 ra->cra_end = cl_index(osc2cl(osc),
104 dlmlock->l_policy_data.l_extent.end);
105 ra->cra_release = osc_read_ahead_release;
106 ra->cra_cbdata = dlmlock;
114 * An implementation of cl_io_operations::cio_io_submit() method for osc
115 * layer. Iterates over pages in the in-queue, prepares each for io by calling
116 * cl_page_prep() and then either submits them through osc_io_submit_page()
117 * or, if page is already submitted, changes osc flags through
118 * osc_set_async_flags().
120 static int osc_io_submit(const struct lu_env *env,
121 const struct cl_io_slice *ios,
122 enum cl_req_type crt, struct cl_2queue *queue)
124 struct cl_page *page;
126 struct client_obd *cli = NULL;
127 struct osc_object *osc = NULL; /* to keep gcc happy */
128 struct osc_page *opg;
130 struct list_head list = LIST_HEAD_INIT(list);
132 struct cl_page_list *qin = &queue->c2_qin;
133 struct cl_page_list *qout = &queue->c2_qout;
134 unsigned int queued = 0;
138 unsigned int max_pages;
140 LASSERT(qin->pl_nr > 0);
142 CDEBUG(D_CACHE, "%d %d\n", qin->pl_nr, crt);
144 osc = cl2osc(ios->cis_obj);
146 max_pages = cli->cl_max_pages_per_rpc;
148 cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
149 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
152 * NOTE: here @page is a top-level page. This is done to avoid
153 * creation of sub-page-list.
155 cl_page_list_for_each_safe(page, tmp, qin) {
156 struct osc_async_page *oap;
162 opg = osc_cl_page_osc(page, osc);
164 LASSERT(osc == oap->oap_obj);
166 if (!list_empty(&oap->oap_pending_item) ||
167 !list_empty(&oap->oap_rpc_item)) {
168 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
174 result = cl_page_prep(env, io, page, crt);
177 if (result != -EALREADY)
180 * Handle -EALREADY error: for read case, the page is
181 * already in UPTODATE state; for write, the page
188 spin_lock(&oap->oap_lock);
189 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
190 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
191 spin_unlock(&oap->oap_lock);
193 osc_page_submit(env, opg, crt, brw_flags);
194 list_add_tail(&oap->oap_pending_item, &list);
196 if (page->cp_sync_io != NULL)
197 cl_page_list_move(qout, qin, page);
199 cl_page_list_del(env, qin, page);
201 if (++queued == max_pages) {
203 result = osc_queue_sync_pages(env, osc, &list, cmd,
211 result = osc_queue_sync_pages(env, osc, &list, cmd, brw_flags);
213 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
214 return qout->pl_nr > 0 ? 0 : result;
218 * This is called when a page is accessed within file in a way that creates
219 * new page, if one were missing (i.e., if there were a hole at that place in
220 * the file, or accessed page is beyond the current file size).
222 * Expand stripe KMS if necessary.
224 static void osc_page_touch_at(const struct lu_env *env,
225 struct cl_object *obj, pgoff_t idx, size_t to)
227 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
228 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
232 /* offset within stripe */
233 kms = cl_offset(obj, idx) + to;
235 cl_object_attr_lock(obj);
239 * ll_inode_size_lock(inode, 0); lov_stripe_lock(lsm);
243 CDEBUG(D_INODE, "stripe KMS %sincreasing "LPU64"->"LPU64" "LPU64"\n",
244 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
245 loi->loi_lvb.lvb_size);
247 attr->cat_mtime = attr->cat_ctime = LTIME_S(CFS_CURRENT_TIME);
248 valid = CAT_MTIME | CAT_CTIME;
249 if (kms > loi->loi_kms) {
253 if (kms > loi->loi_lvb.lvb_size) {
254 attr->cat_size = kms;
257 cl_object_attr_update(env, obj, attr, valid);
258 cl_object_attr_unlock(obj);
261 static int osc_io_commit_async(const struct lu_env *env,
262 const struct cl_io_slice *ios,
263 struct cl_page_list *qin, int from, int to,
266 struct cl_io *io = ios->cis_io;
267 struct osc_io *oio = cl2osc_io(env, ios);
268 struct osc_object *osc = cl2osc(ios->cis_obj);
269 struct cl_page *page;
270 struct cl_page *last_page;
271 struct osc_page *opg;
275 LASSERT(qin->pl_nr > 0);
277 /* Handle partial page cases */
278 last_page = cl_page_list_last(qin);
279 if (oio->oi_lockless) {
280 page = cl_page_list_first(qin);
281 if (page == last_page) {
282 cl_page_clip(env, page, from, to);
285 cl_page_clip(env, page, from, PAGE_SIZE);
287 cl_page_clip(env, last_page, 0, to);
291 while (qin->pl_nr > 0) {
292 struct osc_async_page *oap;
294 page = cl_page_list_first(qin);
295 opg = osc_cl_page_osc(page, osc);
298 if (!list_empty(&oap->oap_rpc_item)) {
299 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
305 /* The page may be already in dirty cache. */
306 if (list_empty(&oap->oap_pending_item)) {
307 result = osc_page_cache_add(env, &opg->ops_cl, io);
312 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
313 page == last_page ? to : PAGE_SIZE);
315 cl_page_list_del(env, qin, page);
317 (*cb)(env, io, page);
318 /* Can't access page any more. Page can be in transfer and
319 * complete at any time. */
322 /* for sync write, kernel will wait for this page to be flushed before
323 * osc_io_end() is called, so release it earlier.
324 * for mkwrite(), it's known there is no further pages. */
325 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
326 osc_extent_release(env, oio->oi_active);
327 oio->oi_active = NULL;
330 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
334 static int osc_io_iter_init(const struct lu_env *env,
335 const struct cl_io_slice *ios)
337 struct osc_object *osc = cl2osc(ios->cis_obj);
338 struct obd_import *imp = osc_cli(osc)->cl_import;
341 spin_lock(&imp->imp_lock);
342 if (likely(!imp->imp_invalid)) {
343 atomic_inc(&osc->oo_nr_ios);
346 spin_unlock(&imp->imp_lock);
351 static int osc_io_write_iter_init(const struct lu_env *env,
352 const struct cl_io_slice *ios)
354 struct cl_io *io = ios->cis_io;
355 struct osc_io *oio = osc_env_io(env);
356 struct osc_object *osc = cl2osc(ios->cis_obj);
357 struct client_obd *cli = osc_cli(osc);
359 unsigned long max_pages;
360 unsigned long npages;
363 if (cl_io_is_append(io))
364 RETURN(osc_io_iter_init(env, ios));
366 npages = io->u.ci_rw.crw_count >> PAGE_CACHE_SHIFT;
367 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
370 max_pages = cli->cl_max_pages_per_rpc * cli->cl_max_rpcs_in_flight;
371 if (npages > max_pages)
374 c = atomic_long_read(cli->cl_lru_left);
375 if (c < npages && osc_lru_reclaim(cli, npages) > 0)
376 c = atomic_long_read(cli->cl_lru_left);
377 while (c >= npages) {
378 if (c == atomic_long_cmpxchg(cli->cl_lru_left, c, c - npages)) {
379 oio->oi_lru_reserved = npages;
382 c = atomic_long_read(cli->cl_lru_left);
384 if (atomic_long_read(cli->cl_lru_left) < max_pages) {
385 /* If there aren't enough pages in the per-OSC LRU then
386 * wake up the LRU thread to try and clear out space, so
387 * we don't block if pages are being dirtied quickly. */
388 CDEBUG(D_CACHE, "%s: queue LRU, left: %lu/%ld.\n",
389 cli_name(cli), atomic_long_read(cli->cl_lru_left),
391 (void)ptlrpcd_queue_work(cli->cl_lru_work);
394 RETURN(osc_io_iter_init(env, ios));
397 static void osc_io_iter_fini(const struct lu_env *env,
398 const struct cl_io_slice *ios)
400 struct osc_object *osc = cl2osc(ios->cis_obj);
402 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
403 if (atomic_dec_and_test(&osc->oo_nr_ios))
404 wake_up_all(&osc->oo_io_waitq);
407 static void osc_io_write_iter_fini(const struct lu_env *env,
408 const struct cl_io_slice *ios)
410 struct osc_io *oio = osc_env_io(env);
411 struct osc_object *osc = cl2osc(ios->cis_obj);
412 struct client_obd *cli = osc_cli(osc);
414 if (oio->oi_lru_reserved > 0) {
415 atomic_long_add(oio->oi_lru_reserved, cli->cl_lru_left);
416 oio->oi_lru_reserved = 0;
418 oio->oi_write_osclock = NULL;
420 osc_io_iter_fini(env, ios);
423 static int osc_io_fault_start(const struct lu_env *env,
424 const struct cl_io_slice *ios)
427 struct cl_fault_io *fio;
431 fio = &io->u.ci_fault;
432 CDEBUG(D_INFO, "%lu %d %zu\n",
433 fio->ft_index, fio->ft_writable, fio->ft_nob);
435 * If mapping is writeable, adjust kms to cover this page,
436 * but do not extend kms beyond actual file size.
439 if (fio->ft_writable)
440 osc_page_touch_at(env, ios->cis_obj,
441 fio->ft_index, fio->ft_nob);
445 static int osc_async_upcall(void *a, int rc)
447 struct osc_async_cbargs *args = a;
450 complete(&args->opc_sync);
455 * Checks that there are no pages being written in the extent being truncated.
457 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
458 struct osc_page *ops , void *cbdata)
460 struct cl_page *page = ops->ops_cl.cpl_page;
461 struct osc_async_page *oap;
462 __u64 start = *(__u64 *)cbdata;
465 if (oap->oap_cmd & OBD_BRW_WRITE &&
466 !list_empty(&oap->oap_pending_item))
467 CL_PAGE_DEBUG(D_ERROR, env, page, "exists " LPU64 "/%s.\n",
468 start, current->comm);
470 if (PageLocked(page->cp_vmpage))
471 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
472 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
474 return CLP_GANG_OKAY;
477 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
478 struct osc_io *oio, __u64 size)
480 struct cl_object *clob;
484 clob = oio->oi_cl.cis_obj;
485 start = cl_index(clob, size);
486 partial = cl_offset(clob, start) < size;
489 * Complain if there are pages in the truncated region.
491 osc_page_gang_lookup(env, io, cl2osc(clob),
492 start + partial, CL_PAGE_EOF,
493 trunc_check_cb, (void *)&size);
496 static int osc_io_setattr_start(const struct lu_env *env,
497 const struct cl_io_slice *slice)
499 struct cl_io *io = slice->cis_io;
500 struct osc_io *oio = cl2osc_io(env, slice);
501 struct cl_object *obj = slice->cis_obj;
502 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
503 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
504 struct obdo *oa = &oio->oi_oa;
505 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
506 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
507 unsigned int ia_valid = io->u.ci_setattr.sa_valid;
510 /* truncate cache dirty pages first */
511 if (cl_io_is_trunc(io))
512 result = osc_cache_truncate_start(env, cl2osc(obj), size,
515 if (result == 0 && oio->oi_lockless == 0) {
516 cl_object_attr_lock(obj);
517 result = cl_object_attr_get(env, obj, attr);
519 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
520 unsigned int cl_valid = 0;
522 if (ia_valid & ATTR_SIZE) {
523 attr->cat_size = attr->cat_kms = size;
524 cl_valid = (CAT_SIZE | CAT_KMS);
526 if (ia_valid & ATTR_MTIME_SET) {
527 attr->cat_mtime = lvb->lvb_mtime;
528 cl_valid |= CAT_MTIME;
530 if (ia_valid & ATTR_ATIME_SET) {
531 attr->cat_atime = lvb->lvb_atime;
532 cl_valid |= CAT_ATIME;
534 if (ia_valid & ATTR_CTIME_SET) {
535 attr->cat_ctime = lvb->lvb_ctime;
536 cl_valid |= CAT_CTIME;
538 result = cl_object_attr_update(env, obj, attr,
541 cl_object_attr_unlock(obj);
543 memset(oa, 0, sizeof(*oa));
545 oa->o_oi = loi->loi_oi;
546 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
547 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
548 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
549 if (ia_valid & ATTR_CTIME) {
550 oa->o_valid |= OBD_MD_FLCTIME;
551 oa->o_ctime = attr->cat_ctime;
553 if (ia_valid & ATTR_ATIME) {
554 oa->o_valid |= OBD_MD_FLATIME;
555 oa->o_atime = attr->cat_atime;
557 if (ia_valid & ATTR_MTIME) {
558 oa->o_valid |= OBD_MD_FLMTIME;
559 oa->o_mtime = attr->cat_mtime;
561 if (ia_valid & ATTR_SIZE) {
563 oa->o_blocks = OBD_OBJECT_EOF;
564 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
566 if (oio->oi_lockless) {
567 oa->o_flags = OBD_FL_SRVLOCK;
568 oa->o_valid |= OBD_MD_FLFLAGS;
571 LASSERT(oio->oi_lockless == 0);
574 if (ia_valid & ATTR_ATTR_FLAG) {
575 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
576 oa->o_valid |= OBD_MD_FLFLAGS;
579 init_completion(&cbargs->opc_sync);
581 if (ia_valid & ATTR_SIZE)
582 result = osc_punch_base(osc_export(cl2osc(obj)),
583 oa, osc_async_upcall,
584 cbargs, PTLRPCD_SET);
586 result = osc_setattr_async(osc_export(cl2osc(obj)),
587 oa, osc_async_upcall,
588 cbargs, PTLRPCD_SET);
590 cbargs->opc_rpc_sent = result == 0;
596 static void osc_io_setattr_end(const struct lu_env *env,
597 const struct cl_io_slice *slice)
599 struct cl_io *io = slice->cis_io;
600 struct osc_io *oio = cl2osc_io(env, slice);
601 struct cl_object *obj = slice->cis_obj;
602 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
605 if (cbargs->opc_rpc_sent) {
606 wait_for_completion(&cbargs->opc_sync);
607 result = io->ci_result = cbargs->opc_rc;
610 if (oio->oi_lockless) {
611 /* lockless truncate */
612 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
614 LASSERT(cl_io_is_trunc(io));
615 /* XXX: Need a lock. */
616 osd->od_stats.os_lockless_truncates++;
620 if (cl_io_is_trunc(io)) {
621 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
622 osc_trunc_check(env, io, oio, size);
623 osc_cache_truncate_end(env, oio->oi_trunc);
624 oio->oi_trunc = NULL;
628 struct osc_data_version_args {
629 struct osc_io *dva_oio;
633 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
636 struct osc_data_version_args *dva = arg;
637 struct osc_io *oio = dva->dva_oio;
638 const struct ost_body *body;
644 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
646 GOTO(out, rc = -EPROTO);
648 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
652 oio->oi_cbarg.opc_rc = rc;
653 complete(&oio->oi_cbarg.opc_sync);
658 static int osc_io_data_version_start(const struct lu_env *env,
659 const struct cl_io_slice *slice)
661 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
662 struct osc_io *oio = cl2osc_io(env, slice);
663 struct obdo *oa = &oio->oi_oa;
664 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
665 struct osc_object *obj = cl2osc(slice->cis_obj);
666 struct lov_oinfo *loi = obj->oo_oinfo;
667 struct obd_export *exp = osc_export(obj);
668 struct ptlrpc_request *req;
669 struct ost_body *body;
670 struct osc_data_version_args *dva;
674 memset(oa, 0, sizeof(*oa));
675 oa->o_oi = loi->loi_oi;
676 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
678 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
679 oa->o_valid |= OBD_MD_FLFLAGS;
680 oa->o_flags |= OBD_FL_SRVLOCK;
681 if (dv->dv_flags & LL_DV_WR_FLUSH)
682 oa->o_flags |= OBD_FL_FLUSH;
685 init_completion(&cbargs->opc_sync);
687 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
691 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
693 ptlrpc_request_free(req);
697 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
698 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
700 ptlrpc_request_set_replen(req);
701 req->rq_interpret_reply = osc_data_version_interpret;
702 CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
703 dva = ptlrpc_req_async_args(req);
706 ptlrpcd_add_req(req);
711 static void osc_io_data_version_end(const struct lu_env *env,
712 const struct cl_io_slice *slice)
714 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
715 struct osc_io *oio = cl2osc_io(env, slice);
716 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
719 wait_for_completion(&cbargs->opc_sync);
721 if (cbargs->opc_rc != 0) {
722 slice->cis_io->ci_result = cbargs->opc_rc;
723 } else if (!(oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)) {
724 slice->cis_io->ci_result = -EOPNOTSUPP;
726 dv->dv_data_version = oio->oi_oa.o_data_version;
727 slice->cis_io->ci_result = 0;
733 static int osc_io_read_start(const struct lu_env *env,
734 const struct cl_io_slice *slice)
736 struct cl_object *obj = slice->cis_obj;
737 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
741 if (!slice->cis_io->ci_noatime) {
742 cl_object_attr_lock(obj);
743 attr->cat_atime = LTIME_S(CFS_CURRENT_TIME);
744 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
745 cl_object_attr_unlock(obj);
751 static int osc_io_write_start(const struct lu_env *env,
752 const struct cl_io_slice *slice)
754 struct cl_object *obj = slice->cis_obj;
755 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
759 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
760 cl_object_attr_lock(obj);
761 attr->cat_mtime = attr->cat_ctime = LTIME_S(CFS_CURRENT_TIME);
762 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
763 cl_object_attr_unlock(obj);
768 static int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
769 struct cl_fsync_io *fio)
771 struct osc_io *oio = osc_env_io(env);
772 struct obdo *oa = &oio->oi_oa;
773 struct lov_oinfo *loi = obj->oo_oinfo;
774 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
778 memset(oa, 0, sizeof(*oa));
779 oa->o_oi = loi->loi_oi;
780 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
782 /* reload size abd blocks for start and end of sync range */
783 oa->o_size = fio->fi_start;
784 oa->o_blocks = fio->fi_end;
785 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
787 obdo_set_parent_fid(oa, fio->fi_fid);
789 init_completion(&cbargs->opc_sync);
791 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
795 static int osc_io_fsync_start(const struct lu_env *env,
796 const struct cl_io_slice *slice)
798 struct cl_io *io = slice->cis_io;
799 struct cl_fsync_io *fio = &io->u.ci_fsync;
800 struct cl_object *obj = slice->cis_obj;
801 struct osc_object *osc = cl2osc(obj);
802 pgoff_t start = cl_index(obj, fio->fi_start);
803 pgoff_t end = cl_index(obj, fio->fi_end);
807 if (fio->fi_end == OBD_OBJECT_EOF)
810 result = osc_cache_writeback_range(env, osc, start, end, 0,
811 fio->fi_mode == CL_FSYNC_DISCARD);
813 fio->fi_nr_written += result;
816 if (fio->fi_mode == CL_FSYNC_ALL) {
819 /* we have to wait for writeback to finish before we can
820 * send OST_SYNC RPC. This is bad because it causes extents
821 * to be written osc by osc. However, we usually start
822 * writeback before CL_FSYNC_ALL so this won't have any real
824 rc = osc_cache_wait_range(env, osc, start, end);
827 rc = osc_fsync_ost(env, osc, fio);
835 static void osc_io_fsync_end(const struct lu_env *env,
836 const struct cl_io_slice *slice)
838 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
839 struct cl_object *obj = slice->cis_obj;
840 pgoff_t start = cl_index(obj, fio->fi_start);
841 pgoff_t end = cl_index(obj, fio->fi_end);
844 if (fio->fi_mode == CL_FSYNC_LOCAL) {
845 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
846 } else if (fio->fi_mode == CL_FSYNC_ALL) {
847 struct osc_io *oio = cl2osc_io(env, slice);
848 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
850 wait_for_completion(&cbargs->opc_sync);
852 result = cbargs->opc_rc;
854 slice->cis_io->ci_result = result;
857 static void osc_io_end(const struct lu_env *env,
858 const struct cl_io_slice *slice)
860 struct osc_io *oio = cl2osc_io(env, slice);
862 if (oio->oi_active) {
863 osc_extent_release(env, oio->oi_active);
864 oio->oi_active = NULL;
868 static const struct cl_io_operations osc_io_ops = {
871 .cio_iter_init = osc_io_iter_init,
872 .cio_iter_fini = osc_io_iter_fini,
873 .cio_start = osc_io_read_start,
874 .cio_fini = osc_io_fini
877 .cio_iter_init = osc_io_write_iter_init,
878 .cio_iter_fini = osc_io_write_iter_fini,
879 .cio_start = osc_io_write_start,
880 .cio_end = osc_io_end,
881 .cio_fini = osc_io_fini
884 .cio_iter_init = osc_io_iter_init,
885 .cio_iter_fini = osc_io_iter_fini,
886 .cio_start = osc_io_setattr_start,
887 .cio_end = osc_io_setattr_end
889 [CIT_DATA_VERSION] = {
890 .cio_start = osc_io_data_version_start,
891 .cio_end = osc_io_data_version_end,
894 .cio_iter_init = osc_io_iter_init,
895 .cio_iter_fini = osc_io_iter_fini,
896 .cio_start = osc_io_fault_start,
897 .cio_end = osc_io_end,
898 .cio_fini = osc_io_fini
901 .cio_start = osc_io_fsync_start,
902 .cio_end = osc_io_fsync_end,
903 .cio_fini = osc_io_fini
906 .cio_fini = osc_io_fini
909 .cio_read_ahead = osc_io_read_ahead,
910 .cio_submit = osc_io_submit,
911 .cio_commit_async = osc_io_commit_async
914 /*****************************************************************************
916 * Transfer operations.
920 int osc_io_init(const struct lu_env *env,
921 struct cl_object *obj, struct cl_io *io)
923 struct osc_io *oio = osc_env_io(env);
925 CL_IO_SLICE_CLEAN(oio, oi_cl);
926 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);