4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Implementation of cl_io for OSC layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_OSC
40 #include <lustre_obdo.h>
41 #include <lustre_osc.h>
43 #include "osc_internal.h"
49 /*****************************************************************************
55 static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
59 void osc_read_ahead_release(const struct lu_env *env, void *cbdata)
61 struct ldlm_lock *dlmlock = cbdata;
62 struct lustre_handle lockh;
64 ldlm_lock2handle(dlmlock, &lockh);
65 ldlm_lock_decref(&lockh, LCK_PR);
66 LDLM_LOCK_PUT(dlmlock);
68 EXPORT_SYMBOL(osc_read_ahead_release);
70 static int osc_io_read_ahead(const struct lu_env *env,
71 const struct cl_io_slice *ios,
72 pgoff_t start, struct cl_read_ahead *ra)
74 struct osc_object *osc = cl2osc(ios->cis_obj);
75 struct ldlm_lock *dlmlock;
76 int result = -ENODATA;
79 dlmlock = osc_dlmlock_at_pgoff(env, osc, start, 0);
80 if (dlmlock != NULL) {
81 LASSERT(dlmlock->l_ast_data == osc);
82 if (dlmlock->l_req_mode != LCK_PR) {
83 struct lustre_handle lockh;
84 ldlm_lock2handle(dlmlock, &lockh);
85 ldlm_lock_addref(&lockh, LCK_PR);
86 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
89 ra->cra_rpc_size = osc_cli(osc)->cl_max_pages_per_rpc;
90 ra->cra_end = cl_index(osc2cl(osc),
91 dlmlock->l_policy_data.l_extent.end);
92 ra->cra_release = osc_read_ahead_release;
93 ra->cra_cbdata = dlmlock;
101 * An implementation of cl_io_operations::cio_io_submit() method for osc
102 * layer. Iterates over pages in the in-queue, prepares each for io by calling
103 * cl_page_prep() and then either submits them through osc_io_submit_page()
104 * or, if page is already submitted, changes osc flags through
105 * osc_set_async_flags().
107 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
108 enum cl_req_type crt, struct cl_2queue *queue)
110 struct cl_page *page;
112 struct client_obd *cli = NULL;
113 struct osc_object *osc = NULL; /* to keep gcc happy */
114 struct osc_page *opg;
116 struct list_head list = LIST_HEAD_INIT(list);
118 struct cl_page_list *qin = &queue->c2_qin;
119 struct cl_page_list *qout = &queue->c2_qout;
120 unsigned int queued = 0;
123 unsigned int max_pages;
125 LASSERT(qin->pl_nr > 0);
127 CDEBUG(D_CACHE|D_READA, "%d %d\n", qin->pl_nr, crt);
129 osc = cl2osc(ios->cis_obj);
131 max_pages = cli->cl_max_pages_per_rpc;
133 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
134 brw_flags |= crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
135 if (crt == CRT_READ && ios->cis_io->ci_ndelay)
136 brw_flags |= OBD_BRW_NDELAY;
139 * NOTE: here @page is a top-level page. This is done to avoid
140 * creation of sub-page-list.
142 cl_page_list_for_each_safe(page, tmp, qin) {
143 struct osc_async_page *oap;
149 opg = osc_cl_page_osc(page, osc);
151 LASSERT(osc == oap->oap_obj);
153 if (!list_empty(&oap->oap_pending_item) ||
154 !list_empty(&oap->oap_rpc_item)) {
155 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
161 result = cl_page_prep(env, io, page, crt);
164 if (result != -EALREADY)
167 * Handle -EALREADY error: for read case, the page is
168 * already in UPTODATE state; for write, the page
175 spin_lock(&oap->oap_lock);
176 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
177 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
178 spin_unlock(&oap->oap_lock);
180 osc_page_submit(env, opg, crt, brw_flags);
181 list_add_tail(&oap->oap_pending_item, &list);
183 if (page->cp_sync_io != NULL)
184 cl_page_list_move(qout, qin, page);
186 cl_page_list_del(env, qin, page);
188 if (++queued == max_pages) {
190 result = osc_queue_sync_pages(env, io, osc, &list,
198 result = osc_queue_sync_pages(env, io, osc, &list, brw_flags);
200 /* Update c/mtime for sync write. LU-7310 */
201 if (crt == CRT_WRITE && qout->pl_nr > 0 && result == 0) {
202 struct cl_object *obj = ios->cis_obj;
203 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
205 cl_object_attr_lock(obj);
206 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
207 cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
208 cl_object_attr_unlock(obj);
211 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
212 return qout->pl_nr > 0 ? 0 : result;
214 EXPORT_SYMBOL(osc_io_submit);
217 * This is called to update the attributes when modifying a specific page,
218 * both when making new pages and when doing updates to existing cached pages.
220 * Expand stripe KMS if necessary.
222 void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj,
223 pgoff_t idx, size_t to)
225 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
226 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);
236 CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n",
237 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
238 loi->loi_lvb.lvb_size);
240 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
241 valid = CAT_MTIME | CAT_CTIME;
242 if (kms > loi->loi_kms) {
246 if (kms > loi->loi_lvb.lvb_size) {
247 attr->cat_size = kms;
250 cl_object_attr_update(env, obj, attr, valid);
251 cl_object_attr_unlock(obj);
256 int osc_io_commit_async(const struct lu_env *env,
257 const struct cl_io_slice *ios,
258 struct cl_page_list *qin, int from, int to,
261 struct cl_io *io = ios->cis_io;
262 struct osc_io *oio = cl2osc_io(env, ios);
263 struct osc_object *osc = cl2osc(ios->cis_obj);
264 struct cl_page *page;
265 struct cl_page *last_page;
266 struct osc_page *opg;
270 LASSERT(qin->pl_nr > 0);
272 /* Handle partial page cases */
273 last_page = cl_page_list_last(qin);
274 if (oio->oi_lockless) {
275 page = cl_page_list_first(qin);
276 if (page == last_page) {
277 cl_page_clip(env, page, from, to);
280 cl_page_clip(env, page, from, PAGE_SIZE);
282 cl_page_clip(env, last_page, 0, to);
286 while (qin->pl_nr > 0) {
287 struct osc_async_page *oap;
289 page = cl_page_list_first(qin);
290 opg = osc_cl_page_osc(page, osc);
293 LASSERTF(osc == oap->oap_obj,
294 "obj mismatch: %p / %p\n", osc, oap->oap_obj);
296 if (!list_empty(&oap->oap_rpc_item)) {
297 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
303 /* The page may be already in dirty cache. */
304 if (list_empty(&oap->oap_pending_item)) {
305 result = osc_page_cache_add(env, &opg->ops_cl, io);
310 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
311 page == last_page ? to : PAGE_SIZE);
313 cl_page_list_del(env, qin, page);
315 (*cb)(env, io, page);
316 /* Can't access page any more. Page can be in transfer and
317 * complete at any time. */
320 /* for sync write, kernel will wait for this page to be flushed before
321 * osc_io_end() is called, so release it earlier.
322 * for mkwrite(), it's known there is no further pages. */
323 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
324 osc_extent_release(env, oio->oi_active);
325 oio->oi_active = NULL;
328 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
331 EXPORT_SYMBOL(osc_io_commit_async);
333 static bool osc_import_not_healthy(struct obd_import *imp)
335 return imp->imp_invalid || imp->imp_deactive ||
336 !(imp->imp_state == LUSTRE_IMP_FULL ||
337 imp->imp_state == LUSTRE_IMP_IDLE);
340 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios)
342 struct osc_object *osc = cl2osc(ios->cis_obj);
343 struct obd_import *imp = osc_cli(osc)->cl_import;
344 struct osc_io *oio = osc_env_io(env);
348 spin_lock(&imp->imp_lock);
350 * check whether this OSC device is available for non-delay read,
351 * fast switching mirror if we haven't tried all mirrors.
353 if (ios->cis_io->ci_type == CIT_READ && ios->cis_io->ci_ndelay &&
354 !ios->cis_io->ci_tried_all_mirrors && osc_import_not_healthy(imp)) {
356 } else if (likely(!imp->imp_invalid)) {
357 atomic_inc(&osc->oo_nr_ios);
358 oio->oi_is_active = 1;
361 spin_unlock(&imp->imp_lock);
363 if (cfs_capable(CFS_CAP_SYS_RESOURCE))
364 oio->oi_cap_sys_resource = 1;
368 EXPORT_SYMBOL(osc_io_iter_init);
370 int osc_io_write_iter_init(const struct lu_env *env,
371 const struct cl_io_slice *ios)
373 struct cl_io *io = ios->cis_io;
374 struct osc_io *oio = osc_env_io(env);
375 struct osc_object *osc = cl2osc(ios->cis_obj);
376 unsigned long npages;
379 if (cl_io_is_append(io))
380 RETURN(osc_io_iter_init(env, ios));
382 npages = io->u.ci_rw.crw_count >> PAGE_SHIFT;
383 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
386 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
388 RETURN(osc_io_iter_init(env, ios));
390 EXPORT_SYMBOL(osc_io_write_iter_init);
392 void osc_io_iter_fini(const struct lu_env *env,
393 const struct cl_io_slice *ios)
395 struct osc_io *oio = osc_env_io(env);
397 if (oio->oi_is_active) {
398 struct osc_object *osc = cl2osc(ios->cis_obj);
400 oio->oi_is_active = 0;
401 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
402 if (atomic_dec_and_test(&osc->oo_nr_ios))
403 wake_up_all(&osc->oo_io_waitq);
406 EXPORT_SYMBOL(osc_io_iter_fini);
408 void osc_io_write_iter_fini(const struct lu_env *env,
409 const struct cl_io_slice *ios)
411 struct osc_io *oio = osc_env_io(env);
412 struct osc_object *osc = cl2osc(ios->cis_obj);
414 if (oio->oi_lru_reserved > 0) {
415 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
416 oio->oi_lru_reserved = 0;
418 oio->oi_write_osclock = NULL;
420 osc_io_iter_fini(env, ios);
422 EXPORT_SYMBOL(osc_io_write_iter_fini);
424 int osc_io_fault_start(const struct lu_env *env, 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);
444 EXPORT_SYMBOL(osc_io_fault_start);
447 static int osc_async_upcall(void *a, int rc)
449 struct osc_async_cbargs *args = a;
452 complete(&args->opc_sync);
457 * Checks that there are no pages being written in the extent being truncated.
459 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
460 struct osc_page *ops , void *cbdata)
462 struct cl_page *page = ops->ops_cl.cpl_page;
463 struct osc_async_page *oap;
464 __u64 start = *(__u64 *)cbdata;
467 if (oap->oap_cmd & OBD_BRW_WRITE &&
468 !list_empty(&oap->oap_pending_item))
469 CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n",
470 start, current->comm);
472 if (PageLocked(page->cp_vmpage))
473 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
474 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
476 return CLP_GANG_OKAY;
479 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
480 struct osc_io *oio, __u64 size)
482 struct cl_object *clob;
486 clob = oio->oi_cl.cis_obj;
487 start = cl_index(clob, size);
488 partial = cl_offset(clob, start) < size;
491 * Complain if there are pages in the truncated region.
493 osc_page_gang_lookup(env, io, cl2osc(clob),
494 start + partial, CL_PAGE_EOF,
495 trunc_check_cb, (void *)&size);
498 static int osc_io_setattr_start(const struct lu_env *env,
499 const struct cl_io_slice *slice)
501 struct cl_io *io = slice->cis_io;
502 struct osc_io *oio = cl2osc_io(env, slice);
503 struct cl_object *obj = slice->cis_obj;
504 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
505 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
506 struct obdo *oa = &oio->oi_oa;
507 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
508 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
509 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
510 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
514 /* truncate cache dirty pages first */
515 if (cl_io_is_trunc(io))
516 result = osc_cache_truncate_start(env, cl2osc(obj), size,
519 if (result == 0 && oio->oi_lockless == 0) {
520 cl_object_attr_lock(obj);
521 result = cl_object_attr_get(env, obj, attr);
523 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
524 unsigned int cl_valid = 0;
526 if (ia_avalid & ATTR_SIZE) {
527 attr->cat_size = size;
528 attr->cat_kms = size;
529 cl_valid = (CAT_SIZE | CAT_KMS);
531 if (ia_avalid & ATTR_MTIME_SET) {
532 attr->cat_mtime = lvb->lvb_mtime;
533 cl_valid |= CAT_MTIME;
535 if (ia_avalid & ATTR_ATIME_SET) {
536 attr->cat_atime = lvb->lvb_atime;
537 cl_valid |= CAT_ATIME;
539 if (ia_xvalid & OP_XVALID_CTIME_SET) {
540 attr->cat_ctime = lvb->lvb_ctime;
541 cl_valid |= CAT_CTIME;
543 result = cl_object_attr_update(env, obj, attr,
546 cl_object_attr_unlock(obj);
548 memset(oa, 0, sizeof(*oa));
550 oa->o_oi = loi->loi_oi;
551 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
552 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
553 oa->o_layout = io->u.ci_setattr.sa_layout;
554 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
556 if (ia_avalid & ATTR_CTIME) {
557 oa->o_valid |= OBD_MD_FLCTIME;
558 oa->o_ctime = attr->cat_ctime;
560 if (ia_avalid & ATTR_ATIME) {
561 oa->o_valid |= OBD_MD_FLATIME;
562 oa->o_atime = attr->cat_atime;
564 if (ia_avalid & ATTR_MTIME) {
565 oa->o_valid |= OBD_MD_FLMTIME;
566 oa->o_mtime = attr->cat_mtime;
568 if (ia_avalid & ATTR_SIZE) {
570 oa->o_blocks = OBD_OBJECT_EOF;
571 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
573 if (oio->oi_lockless) {
574 oa->o_flags = OBD_FL_SRVLOCK;
575 oa->o_valid |= OBD_MD_FLFLAGS;
578 if (io->ci_layout_version > 0) {
579 /* verify layout version */
580 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
581 oa->o_layout_version = io->ci_layout_version;
584 LASSERT(oio->oi_lockless == 0);
587 if (ia_xvalid & OP_XVALID_FLAGS) {
588 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
589 oa->o_valid |= OBD_MD_FLFLAGS;
592 init_completion(&cbargs->opc_sync);
594 if (ia_avalid & ATTR_SIZE)
595 result = osc_punch_send(osc_export(cl2osc(obj)),
596 oa, osc_async_upcall, cbargs);
598 result = osc_setattr_async(osc_export(cl2osc(obj)),
599 oa, osc_async_upcall,
600 cbargs, PTLRPCD_SET);
602 cbargs->opc_rpc_sent = result == 0;
608 void osc_io_setattr_end(const struct lu_env *env,
609 const struct cl_io_slice *slice)
611 struct cl_io *io = slice->cis_io;
612 struct osc_io *oio = cl2osc_io(env, slice);
613 struct cl_object *obj = slice->cis_obj;
614 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
615 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
616 struct obdo *oa = &oio->oi_oa;
617 unsigned int cl_valid = 0;
620 if (cbargs->opc_rpc_sent) {
621 wait_for_completion(&cbargs->opc_sync);
622 result = io->ci_result = cbargs->opc_rc;
626 if (oio->oi_lockless) {
627 /* lockless truncate */
628 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
630 LASSERT(cl_io_is_trunc(io));
631 /* XXX: Need a lock. */
632 osd->od_stats.os_lockless_truncates++;
636 if (cl_io_is_trunc(io)) {
637 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
638 cl_object_attr_lock(obj);
639 if (oa->o_valid & OBD_MD_FLBLOCKS) {
640 attr->cat_blocks = oa->o_blocks;
641 cl_valid |= CAT_BLOCKS;
644 cl_object_attr_update(env, obj, attr, cl_valid);
645 cl_object_attr_unlock(obj);
646 osc_trunc_check(env, io, oio, size);
647 osc_cache_truncate_end(env, oio->oi_trunc);
648 oio->oi_trunc = NULL;
651 EXPORT_SYMBOL(osc_io_setattr_end);
653 struct osc_data_version_args {
654 struct osc_io *dva_oio;
658 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
661 struct osc_data_version_args *dva = arg;
662 struct osc_io *oio = dva->dva_oio;
663 const struct ost_body *body;
669 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
671 GOTO(out, rc = -EPROTO);
673 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
677 oio->oi_cbarg.opc_rc = rc;
678 complete(&oio->oi_cbarg.opc_sync);
683 static int osc_io_data_version_start(const struct lu_env *env,
684 const struct cl_io_slice *slice)
686 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
687 struct osc_io *oio = cl2osc_io(env, slice);
688 struct obdo *oa = &oio->oi_oa;
689 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
690 struct osc_object *obj = cl2osc(slice->cis_obj);
691 struct lov_oinfo *loi = obj->oo_oinfo;
692 struct obd_export *exp = osc_export(obj);
693 struct ptlrpc_request *req;
694 struct ost_body *body;
695 struct osc_data_version_args *dva;
699 memset(oa, 0, sizeof(*oa));
700 oa->o_oi = loi->loi_oi;
701 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
703 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
704 oa->o_valid |= OBD_MD_FLFLAGS;
705 oa->o_flags |= OBD_FL_SRVLOCK;
706 if (dv->dv_flags & LL_DV_WR_FLUSH)
707 oa->o_flags |= OBD_FL_FLUSH;
710 init_completion(&cbargs->opc_sync);
712 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
716 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
718 ptlrpc_request_free(req);
722 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
723 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
725 ptlrpc_request_set_replen(req);
726 req->rq_interpret_reply = osc_data_version_interpret;
727 CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
728 dva = ptlrpc_req_async_args(req);
731 ptlrpcd_add_req(req);
736 static void osc_io_data_version_end(const struct lu_env *env,
737 const struct cl_io_slice *slice)
739 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
740 struct osc_io *oio = cl2osc_io(env, slice);
741 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
744 wait_for_completion(&cbargs->opc_sync);
746 if (cbargs->opc_rc != 0) {
747 slice->cis_io->ci_result = cbargs->opc_rc;
749 slice->cis_io->ci_result = 0;
750 if (!(oio->oi_oa.o_valid &
751 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
752 slice->cis_io->ci_result = -ENOTSUPP;
754 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
755 dv->dv_layout_version = oio->oi_oa.o_layout_version;
756 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
757 dv->dv_data_version = oio->oi_oa.o_data_version;
763 int osc_io_read_start(const struct lu_env *env,
764 const struct cl_io_slice *slice)
766 struct cl_object *obj = slice->cis_obj;
767 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
771 if (!slice->cis_io->ci_noatime) {
772 cl_object_attr_lock(obj);
773 attr->cat_atime = ktime_get_real_seconds();
774 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
775 cl_object_attr_unlock(obj);
780 EXPORT_SYMBOL(osc_io_read_start);
782 int osc_io_write_start(const struct lu_env *env,
783 const struct cl_io_slice *slice)
785 struct cl_object *obj = slice->cis_obj;
786 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
790 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
791 cl_object_attr_lock(obj);
792 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
793 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
794 cl_object_attr_unlock(obj);
798 EXPORT_SYMBOL(osc_io_write_start);
800 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
801 struct cl_fsync_io *fio)
803 struct osc_io *oio = osc_env_io(env);
804 struct obdo *oa = &oio->oi_oa;
805 struct lov_oinfo *loi = obj->oo_oinfo;
806 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
810 memset(oa, 0, sizeof(*oa));
811 oa->o_oi = loi->loi_oi;
812 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
814 /* reload size abd blocks for start and end of sync range */
815 oa->o_size = fio->fi_start;
816 oa->o_blocks = fio->fi_end;
817 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
819 obdo_set_parent_fid(oa, fio->fi_fid);
821 init_completion(&cbargs->opc_sync);
823 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
826 EXPORT_SYMBOL(osc_fsync_ost);
828 int osc_io_fsync_start(const struct lu_env *env,
829 const struct cl_io_slice *slice)
831 struct cl_io *io = slice->cis_io;
832 struct cl_fsync_io *fio = &io->u.ci_fsync;
833 struct cl_object *obj = slice->cis_obj;
834 struct osc_object *osc = cl2osc(obj);
835 pgoff_t start = cl_index(obj, fio->fi_start);
836 pgoff_t end = cl_index(obj, fio->fi_end);
840 if (fio->fi_end == OBD_OBJECT_EOF)
843 result = osc_cache_writeback_range(env, osc, start, end, 0,
844 fio->fi_mode == CL_FSYNC_DISCARD);
846 fio->fi_nr_written += result;
849 if (fio->fi_mode == CL_FSYNC_ALL) {
852 /* we have to wait for writeback to finish before we can
853 * send OST_SYNC RPC. This is bad because it causes extents
854 * to be written osc by osc. However, we usually start
855 * writeback before CL_FSYNC_ALL so this won't have any real
857 rc = osc_cache_wait_range(env, osc, start, end);
860 rc = osc_fsync_ost(env, osc, fio);
868 void osc_io_fsync_end(const struct lu_env *env,
869 const struct cl_io_slice *slice)
871 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
872 struct cl_object *obj = slice->cis_obj;
873 pgoff_t start = cl_index(obj, fio->fi_start);
874 pgoff_t end = cl_index(obj, fio->fi_end);
877 if (fio->fi_mode == CL_FSYNC_LOCAL) {
878 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
879 } else if (fio->fi_mode == CL_FSYNC_ALL) {
880 struct osc_io *oio = cl2osc_io(env, slice);
881 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
883 wait_for_completion(&cbargs->opc_sync);
885 result = cbargs->opc_rc;
887 slice->cis_io->ci_result = result;
889 EXPORT_SYMBOL(osc_io_fsync_end);
891 static int osc_io_ladvise_start(const struct lu_env *env,
892 const struct cl_io_slice *slice)
895 struct cl_io *io = slice->cis_io;
896 struct osc_io *oio = cl2osc_io(env, slice);
897 struct cl_object *obj = slice->cis_obj;
898 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
899 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
900 struct obdo *oa = &oio->oi_oa;
901 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
902 struct lu_ladvise *ladvise;
903 struct ladvise_hdr *ladvise_hdr;
908 /* TODO: add multiple ladvise support in CLIO */
909 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
910 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
911 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
913 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
914 if (ladvise_hdr == NULL)
917 memset(ladvise_hdr, 0, buf_size);
918 ladvise_hdr->lah_magic = LADVISE_MAGIC;
919 ladvise_hdr->lah_count = num_advise;
920 ladvise_hdr->lah_flags = lio->li_flags;
922 memset(oa, 0, sizeof(*oa));
923 oa->o_oi = loi->loi_oi;
924 oa->o_valid = OBD_MD_FLID;
925 obdo_set_parent_fid(oa, lio->li_fid);
927 ladvise = ladvise_hdr->lah_advise;
928 ladvise->lla_start = lio->li_start;
929 ladvise->lla_end = lio->li_end;
930 ladvise->lla_advice = lio->li_advice;
932 if (lio->li_flags & LF_ASYNC) {
933 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
934 ladvise_hdr, NULL, NULL, NULL);
936 init_completion(&cbargs->opc_sync);
937 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
938 ladvise_hdr, osc_async_upcall,
939 cbargs, PTLRPCD_SET);
940 cbargs->opc_rpc_sent = result == 0;
945 static void osc_io_ladvise_end(const struct lu_env *env,
946 const struct cl_io_slice *slice)
948 struct cl_io *io = slice->cis_io;
949 struct osc_io *oio = cl2osc_io(env, slice);
950 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
952 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
954 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
955 wait_for_completion(&cbargs->opc_sync);
956 result = cbargs->opc_rc;
958 slice->cis_io->ci_result = result;
961 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice)
963 struct osc_io *oio = cl2osc_io(env, slice);
965 if (oio->oi_active) {
966 osc_extent_release(env, oio->oi_active);
967 oio->oi_active = NULL;
970 EXPORT_SYMBOL(osc_io_end);
972 static const struct cl_io_operations osc_io_ops = {
975 .cio_iter_init = osc_io_iter_init,
976 .cio_iter_fini = osc_io_iter_fini,
977 .cio_start = osc_io_read_start,
978 .cio_fini = osc_io_fini
981 .cio_iter_init = osc_io_write_iter_init,
982 .cio_iter_fini = osc_io_write_iter_fini,
983 .cio_start = osc_io_write_start,
984 .cio_end = osc_io_end,
985 .cio_fini = osc_io_fini
988 .cio_iter_init = osc_io_iter_init,
989 .cio_iter_fini = osc_io_iter_fini,
990 .cio_start = osc_io_setattr_start,
991 .cio_end = osc_io_setattr_end
993 [CIT_DATA_VERSION] = {
994 .cio_start = osc_io_data_version_start,
995 .cio_end = osc_io_data_version_end,
998 .cio_iter_init = osc_io_iter_init,
999 .cio_iter_fini = osc_io_iter_fini,
1000 .cio_start = osc_io_fault_start,
1001 .cio_end = osc_io_end,
1002 .cio_fini = osc_io_fini
1005 .cio_start = osc_io_fsync_start,
1006 .cio_end = osc_io_fsync_end,
1007 .cio_fini = osc_io_fini
1010 .cio_start = osc_io_ladvise_start,
1011 .cio_end = osc_io_ladvise_end,
1012 .cio_fini = osc_io_fini
1015 .cio_fini = osc_io_fini
1018 .cio_read_ahead = osc_io_read_ahead,
1019 .cio_submit = osc_io_submit,
1020 .cio_commit_async = osc_io_commit_async
1023 /*****************************************************************************
1025 * Transfer operations.
1029 int osc_io_init(const struct lu_env *env,
1030 struct cl_object *obj, struct cl_io *io)
1032 struct osc_io *oio = osc_env_io(env);
1034 CL_IO_SLICE_CLEAN(oio, oi_cl);
1035 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);