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_req *cl2osc_req(const struct cl_req_slice *slice)
58 LINVRNT(slice->crs_dev->cd_lu_dev.ld_type == &osc_device_type);
59 return container_of0(slice, struct osc_req, or_cl);
62 static struct osc_io *cl2osc_io(const struct lu_env *env,
63 const struct cl_io_slice *slice)
65 struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl);
66 LINVRNT(oio == osc_env_io(env));
70 static struct osc_page *osc_cl_page_osc(struct cl_page *page,
71 struct osc_object *osc)
73 const struct cl_page_slice *slice;
76 slice = cl_object_page_slice(&osc->oo_cl, page);
78 slice = cl_page_at(page, &osc_device_type);
79 LASSERT(slice != NULL);
81 return cl2osc_page(slice);
85 /*****************************************************************************
91 static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
95 static void osc_read_ahead_release(const struct lu_env *env,
98 struct ldlm_lock *dlmlock = cbdata;
99 struct lustre_handle lockh;
101 ldlm_lock2handle(dlmlock, &lockh);
102 ldlm_lock_decref(&lockh, LCK_PR);
103 LDLM_LOCK_PUT(dlmlock);
106 static int osc_io_read_ahead(const struct lu_env *env,
107 const struct cl_io_slice *ios,
108 pgoff_t start, struct cl_read_ahead *ra)
110 struct osc_object *osc = cl2osc(ios->cis_obj);
111 struct ldlm_lock *dlmlock;
112 int result = -ENODATA;
115 dlmlock = osc_dlmlock_at_pgoff(env, osc, start, 0);
116 if (dlmlock != NULL) {
117 if (dlmlock->l_req_mode != LCK_PR) {
118 struct lustre_handle lockh;
119 ldlm_lock2handle(dlmlock, &lockh);
120 ldlm_lock_addref(&lockh, LCK_PR);
121 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
124 ra->cra_end = cl_index(osc2cl(osc),
125 dlmlock->l_policy_data.l_extent.end);
126 ra->cra_release = osc_read_ahead_release;
127 ra->cra_cbdata = dlmlock;
135 * An implementation of cl_io_operations::cio_io_submit() method for osc
136 * layer. Iterates over pages in the in-queue, prepares each for io by calling
137 * cl_page_prep() and then either submits them through osc_io_submit_page()
138 * or, if page is already submitted, changes osc flags through
139 * osc_set_async_flags().
141 static int osc_io_submit(const struct lu_env *env,
142 const struct cl_io_slice *ios,
143 enum cl_req_type crt, struct cl_2queue *queue)
145 struct cl_page *page;
147 struct client_obd *cli = NULL;
148 struct osc_object *osc = NULL; /* to keep gcc happy */
149 struct osc_page *opg;
151 struct list_head list = LIST_HEAD_INIT(list);
153 struct cl_page_list *qin = &queue->c2_qin;
154 struct cl_page_list *qout = &queue->c2_qout;
155 unsigned int queued = 0;
159 unsigned int max_pages;
161 LASSERT(qin->pl_nr > 0);
163 CDEBUG(D_CACHE, "%d %d\n", qin->pl_nr, crt);
165 osc = cl2osc(ios->cis_obj);
167 max_pages = cli->cl_max_pages_per_rpc;
169 cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
170 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
173 * NOTE: here @page is a top-level page. This is done to avoid
174 * creation of sub-page-list.
176 cl_page_list_for_each_safe(page, tmp, qin) {
177 struct osc_async_page *oap;
183 opg = osc_cl_page_osc(page, osc);
185 LASSERT(osc == oap->oap_obj);
187 if (!list_empty(&oap->oap_pending_item) ||
188 !list_empty(&oap->oap_rpc_item)) {
189 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
195 result = cl_page_prep(env, io, page, crt);
198 if (result != -EALREADY)
201 * Handle -EALREADY error: for read case, the page is
202 * already in UPTODATE state; for write, the page
209 spin_lock(&oap->oap_lock);
210 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
211 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
212 spin_unlock(&oap->oap_lock);
214 osc_page_submit(env, opg, crt, brw_flags);
215 list_add_tail(&oap->oap_pending_item, &list);
217 if (page->cp_sync_io != NULL)
218 cl_page_list_move(qout, qin, page);
220 cl_page_list_del(env, qin, page);
222 if (++queued == max_pages) {
224 result = osc_queue_sync_pages(env, osc, &list, cmd,
232 result = osc_queue_sync_pages(env, osc, &list, cmd, brw_flags);
234 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
235 return qout->pl_nr > 0 ? 0 : result;
239 * This is called when a page is accessed within file in a way that creates
240 * new page, if one were missing (i.e., if there were a hole at that place in
241 * the file, or accessed page is beyond the current file size).
243 * Expand stripe KMS if necessary.
245 static void osc_page_touch_at(const struct lu_env *env,
246 struct cl_object *obj, pgoff_t idx, size_t to)
248 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
249 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
253 /* offset within stripe */
254 kms = cl_offset(obj, idx) + to;
256 cl_object_attr_lock(obj);
260 * ll_inode_size_lock(inode, 0); lov_stripe_lock(lsm);
264 CDEBUG(D_INODE, "stripe KMS %sincreasing "LPU64"->"LPU64" "LPU64"\n",
265 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
266 loi->loi_lvb.lvb_size);
268 attr->cat_mtime = attr->cat_ctime = LTIME_S(CFS_CURRENT_TIME);
269 valid = CAT_MTIME | CAT_CTIME;
270 if (kms > loi->loi_kms) {
274 if (kms > loi->loi_lvb.lvb_size) {
275 attr->cat_size = kms;
278 cl_object_attr_update(env, obj, attr, valid);
279 cl_object_attr_unlock(obj);
282 static int osc_io_commit_async(const struct lu_env *env,
283 const struct cl_io_slice *ios,
284 struct cl_page_list *qin, int from, int to,
287 struct cl_io *io = ios->cis_io;
288 struct osc_io *oio = cl2osc_io(env, ios);
289 struct osc_object *osc = cl2osc(ios->cis_obj);
290 struct cl_page *page;
291 struct cl_page *last_page;
292 struct osc_page *opg;
296 LASSERT(qin->pl_nr > 0);
298 /* Handle partial page cases */
299 last_page = cl_page_list_last(qin);
300 if (oio->oi_lockless) {
301 page = cl_page_list_first(qin);
302 if (page == last_page) {
303 cl_page_clip(env, page, from, to);
306 cl_page_clip(env, page, from, PAGE_SIZE);
308 cl_page_clip(env, last_page, 0, to);
312 while (qin->pl_nr > 0) {
313 struct osc_async_page *oap;
315 page = cl_page_list_first(qin);
316 opg = osc_cl_page_osc(page, osc);
319 if (!list_empty(&oap->oap_rpc_item)) {
320 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
326 /* The page may be already in dirty cache. */
327 if (list_empty(&oap->oap_pending_item)) {
328 result = osc_page_cache_add(env, &opg->ops_cl, io);
333 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
334 page == last_page ? to : PAGE_SIZE);
336 cl_page_list_del(env, qin, page);
338 (*cb)(env, io, page);
339 /* Can't access page any more. Page can be in transfer and
340 * complete at any time. */
343 /* for sync write, kernel will wait for this page to be flushed before
344 * osc_io_end() is called, so release it earlier.
345 * for mkwrite(), it's known there is no further pages. */
346 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
347 osc_extent_release(env, oio->oi_active);
348 oio->oi_active = NULL;
351 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
355 static int osc_io_iter_init(const struct lu_env *env,
356 const struct cl_io_slice *ios)
358 struct osc_object *osc = cl2osc(ios->cis_obj);
359 struct obd_import *imp = osc_cli(osc)->cl_import;
362 spin_lock(&imp->imp_lock);
363 if (likely(!imp->imp_invalid)) {
364 atomic_inc(&osc->oo_nr_ios);
367 spin_unlock(&imp->imp_lock);
372 static int osc_io_write_iter_init(const struct lu_env *env,
373 const struct cl_io_slice *ios)
375 struct cl_io *io = ios->cis_io;
376 struct osc_io *oio = osc_env_io(env);
377 struct osc_object *osc = cl2osc(ios->cis_obj);
378 struct client_obd *cli = osc_cli(osc);
380 unsigned long max_pages;
381 unsigned long npages;
384 if (cl_io_is_append(io))
385 RETURN(osc_io_iter_init(env, ios));
387 npages = io->u.ci_rw.crw_count >> PAGE_CACHE_SHIFT;
388 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
391 max_pages = cli->cl_max_pages_per_rpc * cli->cl_max_rpcs_in_flight;
392 if (npages > max_pages)
395 c = atomic_long_read(cli->cl_lru_left);
396 if (c < npages && osc_lru_reclaim(cli, npages) > 0)
397 c = atomic_long_read(cli->cl_lru_left);
398 while (c >= npages) {
399 if (c == atomic_long_cmpxchg(cli->cl_lru_left, c, c - npages)) {
400 oio->oi_lru_reserved = npages;
403 c = atomic_long_read(cli->cl_lru_left);
405 if (atomic_long_read(cli->cl_lru_left) < max_pages) {
406 /* If there aren't enough pages in the per-OSC LRU then
407 * wake up the LRU thread to try and clear out space, so
408 * we don't block if pages are being dirtied quickly. */
409 CDEBUG(D_CACHE, "%s: queue LRU, left: %lu/%ld.\n",
410 cli_name(cli), atomic_long_read(cli->cl_lru_left),
412 (void)ptlrpcd_queue_work(cli->cl_lru_work);
415 RETURN(osc_io_iter_init(env, ios));
418 static void osc_io_iter_fini(const struct lu_env *env,
419 const struct cl_io_slice *ios)
421 struct osc_object *osc = cl2osc(ios->cis_obj);
423 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
424 if (atomic_dec_and_test(&osc->oo_nr_ios))
425 wake_up_all(&osc->oo_io_waitq);
428 static void osc_io_write_iter_fini(const struct lu_env *env,
429 const struct cl_io_slice *ios)
431 struct osc_io *oio = osc_env_io(env);
432 struct osc_object *osc = cl2osc(ios->cis_obj);
433 struct client_obd *cli = osc_cli(osc);
435 if (oio->oi_lru_reserved > 0) {
436 atomic_long_add(oio->oi_lru_reserved, cli->cl_lru_left);
437 oio->oi_lru_reserved = 0;
439 oio->oi_write_osclock = NULL;
441 osc_io_iter_fini(env, ios);
444 static int osc_io_fault_start(const struct lu_env *env,
445 const struct cl_io_slice *ios)
448 struct cl_fault_io *fio;
452 fio = &io->u.ci_fault;
453 CDEBUG(D_INFO, "%lu %d %zu\n",
454 fio->ft_index, fio->ft_writable, fio->ft_nob);
456 * If mapping is writeable, adjust kms to cover this page,
457 * but do not extend kms beyond actual file size.
460 if (fio->ft_writable)
461 osc_page_touch_at(env, ios->cis_obj,
462 fio->ft_index, fio->ft_nob);
466 static int osc_async_upcall(void *a, int rc)
468 struct osc_async_cbargs *args = a;
471 complete(&args->opc_sync);
476 * Checks that there are no pages being written in the extent being truncated.
478 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
479 struct osc_page *ops , void *cbdata)
481 struct cl_page *page = ops->ops_cl.cpl_page;
482 struct osc_async_page *oap;
483 __u64 start = *(__u64 *)cbdata;
486 if (oap->oap_cmd & OBD_BRW_WRITE &&
487 !list_empty(&oap->oap_pending_item))
488 CL_PAGE_DEBUG(D_ERROR, env, page, "exists " LPU64 "/%s.\n",
489 start, current->comm);
491 if (PageLocked(page->cp_vmpage))
492 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
493 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
495 return CLP_GANG_OKAY;
498 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
499 struct osc_io *oio, __u64 size)
501 struct cl_object *clob;
505 clob = oio->oi_cl.cis_obj;
506 start = cl_index(clob, size);
507 partial = cl_offset(clob, start) < size;
510 * Complain if there are pages in the truncated region.
512 osc_page_gang_lookup(env, io, cl2osc(clob),
513 start + partial, CL_PAGE_EOF,
514 trunc_check_cb, (void *)&size);
517 static int osc_io_setattr_start(const struct lu_env *env,
518 const struct cl_io_slice *slice)
520 struct cl_io *io = slice->cis_io;
521 struct osc_io *oio = cl2osc_io(env, slice);
522 struct cl_object *obj = slice->cis_obj;
523 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
524 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
525 struct obdo *oa = &oio->oi_oa;
526 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
527 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
528 unsigned int ia_valid = io->u.ci_setattr.sa_valid;
531 /* truncate cache dirty pages first */
532 if (cl_io_is_trunc(io))
533 result = osc_cache_truncate_start(env, cl2osc(obj), size,
536 if (result == 0 && oio->oi_lockless == 0) {
537 cl_object_attr_lock(obj);
538 result = cl_object_attr_get(env, obj, attr);
540 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
541 unsigned int cl_valid = 0;
543 if (ia_valid & ATTR_SIZE) {
544 attr->cat_size = attr->cat_kms = size;
545 cl_valid = (CAT_SIZE | CAT_KMS);
547 if (ia_valid & ATTR_MTIME_SET) {
548 attr->cat_mtime = lvb->lvb_mtime;
549 cl_valid |= CAT_MTIME;
551 if (ia_valid & ATTR_ATIME_SET) {
552 attr->cat_atime = lvb->lvb_atime;
553 cl_valid |= CAT_ATIME;
555 if (ia_valid & ATTR_CTIME_SET) {
556 attr->cat_ctime = lvb->lvb_ctime;
557 cl_valid |= CAT_CTIME;
559 result = cl_object_attr_update(env, obj, attr,
562 cl_object_attr_unlock(obj);
564 memset(oa, 0, sizeof(*oa));
566 oa->o_oi = loi->loi_oi;
567 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
568 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
569 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
570 if (ia_valid & ATTR_CTIME) {
571 oa->o_valid |= OBD_MD_FLCTIME;
572 oa->o_ctime = attr->cat_ctime;
574 if (ia_valid & ATTR_ATIME) {
575 oa->o_valid |= OBD_MD_FLATIME;
576 oa->o_atime = attr->cat_atime;
578 if (ia_valid & ATTR_MTIME) {
579 oa->o_valid |= OBD_MD_FLMTIME;
580 oa->o_mtime = attr->cat_mtime;
582 if (ia_valid & ATTR_SIZE) {
584 oa->o_blocks = OBD_OBJECT_EOF;
585 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
587 if (oio->oi_lockless) {
588 oa->o_flags = OBD_FL_SRVLOCK;
589 oa->o_valid |= OBD_MD_FLFLAGS;
592 LASSERT(oio->oi_lockless == 0);
595 if (ia_valid & ATTR_ATTR_FLAG) {
596 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
597 oa->o_valid |= OBD_MD_FLFLAGS;
600 init_completion(&cbargs->opc_sync);
602 if (ia_valid & ATTR_SIZE)
603 result = osc_punch_base(osc_export(cl2osc(obj)),
604 oa, osc_async_upcall,
605 cbargs, PTLRPCD_SET);
607 result = osc_setattr_async(osc_export(cl2osc(obj)),
608 oa, osc_async_upcall,
609 cbargs, PTLRPCD_SET);
611 cbargs->opc_rpc_sent = result == 0;
617 static void osc_io_setattr_end(const struct lu_env *env,
618 const struct cl_io_slice *slice)
620 struct cl_io *io = slice->cis_io;
621 struct osc_io *oio = cl2osc_io(env, slice);
622 struct cl_object *obj = slice->cis_obj;
623 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
626 if (cbargs->opc_rpc_sent) {
627 wait_for_completion(&cbargs->opc_sync);
628 result = io->ci_result = cbargs->opc_rc;
631 if (oio->oi_lockless) {
632 /* lockless truncate */
633 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
635 LASSERT(cl_io_is_trunc(io));
636 /* XXX: Need a lock. */
637 osd->od_stats.os_lockless_truncates++;
641 if (cl_io_is_trunc(io)) {
642 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
643 osc_trunc_check(env, io, oio, size);
644 osc_cache_truncate_end(env, oio->oi_trunc);
645 oio->oi_trunc = NULL;
649 struct osc_data_version_args {
650 struct osc_io *dva_oio;
654 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
657 struct osc_data_version_args *dva = arg;
658 struct osc_io *oio = dva->dva_oio;
659 const struct ost_body *body;
665 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
667 GOTO(out, rc = -EPROTO);
669 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
673 oio->oi_cbarg.opc_rc = rc;
674 complete(&oio->oi_cbarg.opc_sync);
679 static int osc_io_data_version_start(const struct lu_env *env,
680 const struct cl_io_slice *slice)
682 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
683 struct osc_io *oio = cl2osc_io(env, slice);
684 struct obdo *oa = &oio->oi_oa;
685 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
686 struct osc_object *obj = cl2osc(slice->cis_obj);
687 struct lov_oinfo *loi = obj->oo_oinfo;
688 struct obd_export *exp = osc_export(obj);
689 struct ptlrpc_request *req;
690 struct ost_body *body;
691 struct osc_data_version_args *dva;
695 memset(oa, 0, sizeof(*oa));
696 oa->o_oi = loi->loi_oi;
697 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
699 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
700 oa->o_valid |= OBD_MD_FLFLAGS;
701 oa->o_flags |= OBD_FL_SRVLOCK;
702 if (dv->dv_flags & LL_DV_WR_FLUSH)
703 oa->o_flags |= OBD_FL_FLUSH;
706 init_completion(&cbargs->opc_sync);
708 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
712 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
714 ptlrpc_request_free(req);
718 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
719 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
721 ptlrpc_request_set_replen(req);
722 req->rq_interpret_reply = osc_data_version_interpret;
723 CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
724 dva = ptlrpc_req_async_args(req);
727 ptlrpcd_add_req(req);
732 static void osc_io_data_version_end(const struct lu_env *env,
733 const struct cl_io_slice *slice)
735 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
736 struct osc_io *oio = cl2osc_io(env, slice);
737 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
740 wait_for_completion(&cbargs->opc_sync);
742 if (cbargs->opc_rc != 0) {
743 slice->cis_io->ci_result = cbargs->opc_rc;
744 } else if (!(oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)) {
745 slice->cis_io->ci_result = -EOPNOTSUPP;
747 dv->dv_data_version = oio->oi_oa.o_data_version;
748 slice->cis_io->ci_result = 0;
754 static int osc_io_read_start(const struct lu_env *env,
755 const struct cl_io_slice *slice)
757 struct cl_object *obj = slice->cis_obj;
758 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
762 if (!slice->cis_io->ci_noatime) {
763 cl_object_attr_lock(obj);
764 attr->cat_atime = LTIME_S(CFS_CURRENT_TIME);
765 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
766 cl_object_attr_unlock(obj);
772 static int osc_io_write_start(const struct lu_env *env,
773 const struct cl_io_slice *slice)
775 struct cl_object *obj = slice->cis_obj;
776 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
780 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
781 cl_object_attr_lock(obj);
782 attr->cat_mtime = attr->cat_ctime = LTIME_S(CFS_CURRENT_TIME);
783 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
784 cl_object_attr_unlock(obj);
789 static int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
790 struct cl_fsync_io *fio)
792 struct osc_io *oio = osc_env_io(env);
793 struct obdo *oa = &oio->oi_oa;
794 struct lov_oinfo *loi = obj->oo_oinfo;
795 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
799 memset(oa, 0, sizeof(*oa));
800 oa->o_oi = loi->loi_oi;
801 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
803 /* reload size abd blocks for start and end of sync range */
804 oa->o_size = fio->fi_start;
805 oa->o_blocks = fio->fi_end;
806 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
808 obdo_set_parent_fid(oa, fio->fi_fid);
810 init_completion(&cbargs->opc_sync);
812 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
816 static int osc_io_fsync_start(const struct lu_env *env,
817 const struct cl_io_slice *slice)
819 struct cl_io *io = slice->cis_io;
820 struct cl_fsync_io *fio = &io->u.ci_fsync;
821 struct cl_object *obj = slice->cis_obj;
822 struct osc_object *osc = cl2osc(obj);
823 pgoff_t start = cl_index(obj, fio->fi_start);
824 pgoff_t end = cl_index(obj, fio->fi_end);
828 if (fio->fi_end == OBD_OBJECT_EOF)
831 result = osc_cache_writeback_range(env, osc, start, end, 0,
832 fio->fi_mode == CL_FSYNC_DISCARD);
834 fio->fi_nr_written += result;
837 if (fio->fi_mode == CL_FSYNC_ALL) {
840 /* we have to wait for writeback to finish before we can
841 * send OST_SYNC RPC. This is bad because it causes extents
842 * to be written osc by osc. However, we usually start
843 * writeback before CL_FSYNC_ALL so this won't have any real
845 rc = osc_cache_wait_range(env, osc, start, end);
848 rc = osc_fsync_ost(env, osc, fio);
856 static void osc_io_fsync_end(const struct lu_env *env,
857 const struct cl_io_slice *slice)
859 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
860 struct cl_object *obj = slice->cis_obj;
861 pgoff_t start = cl_index(obj, fio->fi_start);
862 pgoff_t end = cl_index(obj, fio->fi_end);
865 if (fio->fi_mode == CL_FSYNC_LOCAL) {
866 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
867 } else if (fio->fi_mode == CL_FSYNC_ALL) {
868 struct osc_io *oio = cl2osc_io(env, slice);
869 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
871 wait_for_completion(&cbargs->opc_sync);
873 result = cbargs->opc_rc;
875 slice->cis_io->ci_result = result;
878 static void osc_io_end(const struct lu_env *env,
879 const struct cl_io_slice *slice)
881 struct osc_io *oio = cl2osc_io(env, slice);
883 if (oio->oi_active) {
884 osc_extent_release(env, oio->oi_active);
885 oio->oi_active = NULL;
889 static const struct cl_io_operations osc_io_ops = {
892 .cio_iter_init = osc_io_iter_init,
893 .cio_iter_fini = osc_io_iter_fini,
894 .cio_start = osc_io_read_start,
895 .cio_fini = osc_io_fini
898 .cio_iter_init = osc_io_write_iter_init,
899 .cio_iter_fini = osc_io_write_iter_fini,
900 .cio_start = osc_io_write_start,
901 .cio_end = osc_io_end,
902 .cio_fini = osc_io_fini
905 .cio_iter_init = osc_io_iter_init,
906 .cio_iter_fini = osc_io_iter_fini,
907 .cio_start = osc_io_setattr_start,
908 .cio_end = osc_io_setattr_end
910 [CIT_DATA_VERSION] = {
911 .cio_start = osc_io_data_version_start,
912 .cio_end = osc_io_data_version_end,
915 .cio_iter_init = osc_io_iter_init,
916 .cio_iter_fini = osc_io_iter_fini,
917 .cio_start = osc_io_fault_start,
918 .cio_end = osc_io_end,
919 .cio_fini = osc_io_fini
922 .cio_start = osc_io_fsync_start,
923 .cio_end = osc_io_fsync_end,
924 .cio_fini = osc_io_fini
927 .cio_fini = osc_io_fini
930 .cio_read_ahead = osc_io_read_ahead,
931 .cio_submit = osc_io_submit,
932 .cio_commit_async = osc_io_commit_async
935 /*****************************************************************************
937 * Transfer operations.
941 static int osc_req_prep(const struct lu_env *env,
942 const struct cl_req_slice *slice)
947 static void osc_req_completion(const struct lu_env *env,
948 const struct cl_req_slice *slice, int ioret)
952 or = cl2osc_req(slice);
953 OBD_SLAB_FREE_PTR(or, osc_req_kmem);
957 * Implementation of struct cl_req_operations::cro_attr_set() for osc
958 * layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq
961 static void osc_req_attr_set(const struct lu_env *env,
962 const struct cl_req_slice *slice,
963 const struct cl_object *obj,
964 struct cl_req_attr *attr, u64 flags)
966 struct lov_oinfo *oinfo;
967 struct cl_req *clerq;
968 struct cl_page *apage; /* _some_ page in @clerq */
969 struct ldlm_lock *lock; /* _some_ lock protecting @apage */
970 struct osc_page *opg;
974 oinfo = cl2osc(obj)->oo_oinfo;
975 lvb = &oinfo->loi_lvb;
978 if ((flags & OBD_MD_FLMTIME) != 0) {
979 oa->o_mtime = lvb->lvb_mtime;
980 oa->o_valid |= OBD_MD_FLMTIME;
982 if ((flags & OBD_MD_FLATIME) != 0) {
983 oa->o_atime = lvb->lvb_atime;
984 oa->o_valid |= OBD_MD_FLATIME;
986 if ((flags & OBD_MD_FLCTIME) != 0) {
987 oa->o_ctime = lvb->lvb_ctime;
988 oa->o_valid |= OBD_MD_FLCTIME;
990 if (flags & OBD_MD_FLGROUP) {
991 ostid_set_seq(&oa->o_oi, ostid_seq(&oinfo->loi_oi));
992 oa->o_valid |= OBD_MD_FLGROUP;
994 if (flags & OBD_MD_FLID) {
995 ostid_set_id(&oa->o_oi, ostid_id(&oinfo->loi_oi));
996 oa->o_valid |= OBD_MD_FLID;
998 if (flags & OBD_MD_FLHANDLE) {
999 clerq = slice->crs_req;
1000 LASSERT(!list_empty(&clerq->crq_pages));
1001 apage = container_of(clerq->crq_pages.next,
1002 struct cl_page, cp_flight);
1003 opg = osc_cl_page_osc(apage, NULL);
1004 lock = osc_dlmlock_at_pgoff(env, cl2osc(obj), osc_index(opg),
1005 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_CANCELING);
1006 if (lock == NULL && !opg->ops_srvlock) {
1007 struct ldlm_resource *res;
1008 struct ldlm_res_id *resname;
1010 CL_PAGE_DEBUG(D_ERROR, env, apage, "uncovered page!\n");
1012 resname = &osc_env_info(env)->oti_resname;
1013 ostid_build_res_name(&oinfo->loi_oi, resname);
1014 res = ldlm_resource_get(
1015 osc_export(cl2osc(obj))->exp_obd->obd_namespace,
1016 NULL, resname, LDLM_EXTENT, 0);
1017 ldlm_resource_dump(D_ERROR, res);
1019 libcfs_debug_dumpstack(NULL);
1023 /* check for lockless io. */
1025 oa->o_handle = lock->l_remote_handle;
1026 oa->o_valid |= OBD_MD_FLHANDLE;
1027 LDLM_LOCK_PUT(lock);
1032 static const struct cl_req_operations osc_req_ops = {
1033 .cro_prep = osc_req_prep,
1034 .cro_attr_set = osc_req_attr_set,
1035 .cro_completion = osc_req_completion
1039 int osc_io_init(const struct lu_env *env,
1040 struct cl_object *obj, struct cl_io *io)
1042 struct osc_io *oio = osc_env_io(env);
1044 CL_IO_SLICE_CLEAN(oio, oi_cl);
1045 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);
1049 int osc_req_init(const struct lu_env *env, struct cl_device *dev,
1055 OBD_SLAB_ALLOC_PTR_GFP(or, osc_req_kmem, GFP_NOFS);
1057 cl_req_slice_add(req, &or->or_cl, dev, &osc_req_ops);