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>
42 #include <linux/pagevec.h>
44 #include "osc_internal.h"
50 /*****************************************************************************
56 static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
60 void osc_read_ahead_release(const struct lu_env *env, void *cbdata)
62 struct ldlm_lock *dlmlock = cbdata;
63 struct lustre_handle lockh;
65 ldlm_lock2handle(dlmlock, &lockh);
66 ldlm_lock_decref(&lockh, LCK_PR);
67 LDLM_LOCK_PUT(dlmlock);
69 EXPORT_SYMBOL(osc_read_ahead_release);
71 static int osc_io_read_ahead(const struct lu_env *env,
72 const struct cl_io_slice *ios,
73 pgoff_t start, struct cl_read_ahead *ra)
75 struct osc_object *osc = cl2osc(ios->cis_obj);
76 struct ldlm_lock *dlmlock;
77 int result = -ENODATA;
80 dlmlock = osc_dlmlock_at_pgoff(env, osc, start, 0);
81 if (dlmlock != NULL) {
82 LASSERT(dlmlock->l_ast_data == osc);
83 if (dlmlock->l_req_mode != LCK_PR) {
84 struct lustre_handle lockh;
85 ldlm_lock2handle(dlmlock, &lockh);
86 ldlm_lock_addref(&lockh, LCK_PR);
87 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
90 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
91 ra->cra_end_idx = cl_index(osc2cl(osc),
92 dlmlock->l_policy_data.l_extent.end);
93 ra->cra_release = osc_read_ahead_release;
94 ra->cra_cbdata = dlmlock;
95 if (ra->cra_end_idx != CL_PAGE_EOF)
96 ra->cra_contention = true;
104 * An implementation of cl_io_operations::cio_io_submit() method for osc
105 * layer. Iterates over pages in the in-queue, prepares each for io by calling
106 * cl_page_prep() and then either submits them through osc_io_submit_page()
107 * or, if page is already submitted, changes osc flags through
108 * osc_set_async_flags().
110 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
111 enum cl_req_type crt, struct cl_2queue *queue)
113 struct cl_page *page;
115 struct client_obd *cli = NULL;
116 struct osc_object *osc = NULL; /* to keep gcc happy */
117 struct osc_page *opg;
121 struct cl_page_list *qin = &queue->c2_qin;
122 struct cl_page_list *qout = &queue->c2_qout;
123 unsigned int queued = 0;
126 unsigned int max_pages;
127 unsigned int ppc_bits; /* pages per chunk bits */
129 bool sync_queue = false;
131 LASSERT(qin->pl_nr > 0);
133 CDEBUG(D_CACHE|D_READA, "%d %d\n", qin->pl_nr, crt);
135 osc = cl2osc(ios->cis_obj);
137 max_pages = cli->cl_max_pages_per_rpc;
138 ppc_bits = cli->cl_chunkbits - PAGE_SHIFT;
141 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
142 brw_flags |= crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
143 if (crt == CRT_READ && ios->cis_io->ci_ndelay)
144 brw_flags |= OBD_BRW_NDELAY;
146 page = cl_page_list_first(qin);
147 if (page->cp_type == CPT_TRANSIENT)
148 brw_flags |= OBD_BRW_NOCACHE;
151 * NOTE: here @page is a top-level page. This is done to avoid
152 * creation of sub-page-list.
154 cl_page_list_for_each_safe(page, tmp, qin) {
155 struct osc_async_page *oap;
161 opg = osc_cl_page_osc(page, osc);
163 LASSERT(osc == oap->oap_obj);
165 if (!list_empty(&oap->oap_pending_item) ||
166 !list_empty(&oap->oap_rpc_item)) {
167 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
173 result = cl_page_prep(env, io, page, crt);
176 if (result != -EALREADY)
179 * Handle -EALREADY error: for read case, the page is
180 * already in UPTODATE state; for write, the page
187 spin_lock(&oap->oap_lock);
188 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
189 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
190 spin_unlock(&oap->oap_lock);
192 osc_page_submit(env, opg, crt, brw_flags);
193 list_add_tail(&oap->oap_pending_item, &list);
195 if (page->cp_sync_io != NULL)
196 cl_page_list_move(qout, qin, page);
198 cl_page_list_del(env, qin, page);
201 if (queued == max_pages) {
203 } else if (crt == CRT_WRITE) {
205 unsigned int next_chunks;
207 chunks = (queued + ppc - 1) >> ppc_bits;
208 /* chunk number if add another page */
209 next_chunks = (queued + ppc) >> ppc_bits;
211 /* next page will excceed write chunk limit */
212 if (chunks == osc_max_write_chunks(cli) &&
213 next_chunks > chunks)
218 result = osc_queue_sync_pages(env, io, osc, &list,
228 result = osc_queue_sync_pages(env, io, osc, &list, brw_flags);
230 /* Update c/mtime for sync write. LU-7310 */
231 if (crt == CRT_WRITE && qout->pl_nr > 0 && result == 0) {
232 struct cl_object *obj = ios->cis_obj;
233 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
235 cl_object_attr_lock(obj);
236 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
237 cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
238 cl_object_attr_unlock(obj);
241 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
242 return qout->pl_nr > 0 ? 0 : result;
244 EXPORT_SYMBOL(osc_io_submit);
247 * This is called to update the attributes when modifying a specific page,
248 * both when making new pages and when doing updates to existing cached pages.
250 * Expand stripe KMS if necessary.
252 void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj,
253 pgoff_t idx, size_t to)
255 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
256 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
262 /* offset within stripe */
263 kms = cl_offset(obj, idx) + to;
265 cl_object_attr_lock(obj);
266 CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n",
267 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
268 loi->loi_lvb.lvb_size);
270 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
271 valid = CAT_MTIME | CAT_CTIME;
272 if (kms > loi->loi_kms) {
276 if (kms > loi->loi_lvb.lvb_size) {
277 attr->cat_size = kms;
280 cl_object_attr_update(env, obj, attr, valid);
281 cl_object_attr_unlock(obj);
286 int osc_io_commit_async(const struct lu_env *env,
287 const struct cl_io_slice *ios,
288 struct cl_page_list *qin, int from, int to,
291 struct cl_io *io = ios->cis_io;
292 struct osc_io *oio = cl2osc_io(env, ios);
293 struct osc_object *osc = cl2osc(ios->cis_obj);
294 struct cl_page *page;
295 struct cl_page *last_page;
296 struct osc_page *opg;
297 struct pagevec *pvec = &osc_env_info(env)->oti_pagevec;
301 LASSERT(qin->pl_nr > 0);
303 /* Handle partial page cases */
304 last_page = cl_page_list_last(qin);
305 if (oio->oi_lockless) {
306 page = cl_page_list_first(qin);
307 if (page == last_page) {
308 cl_page_clip(env, page, from, to);
311 cl_page_clip(env, page, from, PAGE_SIZE);
313 cl_page_clip(env, last_page, 0, to);
317 ll_pagevec_init(pvec, 0);
319 while (qin->pl_nr > 0) {
320 struct osc_async_page *oap;
322 page = cl_page_list_first(qin);
323 opg = osc_cl_page_osc(page, osc);
326 LASSERTF(osc == oap->oap_obj,
327 "obj mismatch: %p / %p\n", osc, oap->oap_obj);
329 if (!list_empty(&oap->oap_rpc_item)) {
330 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
336 /* The page may be already in dirty cache. */
337 if (list_empty(&oap->oap_pending_item)) {
338 result = osc_page_cache_add(env, opg, io, cb);
343 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
344 page == last_page ? to : PAGE_SIZE);
346 cl_page_list_del(env, qin, page);
348 /* if there are no more slots, do the callback & reinit */
349 if (pagevec_add(pvec, page->cp_vmpage) == 0) {
350 (*cb)(env, io, pvec);
351 pagevec_reinit(pvec);
355 /* Clean up any partially full pagevecs */
356 if (pagevec_count(pvec) != 0)
357 (*cb)(env, io, pvec);
359 /* Can't access these pages any more. Page can be in transfer and
360 * complete at any time. */
362 /* for sync write, kernel will wait for this page to be flushed before
363 * osc_io_end() is called, so release it earlier.
364 * for mkwrite(), it's known there is no further pages. */
365 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
366 osc_extent_release(env, oio->oi_active);
367 oio->oi_active = NULL;
370 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
373 EXPORT_SYMBOL(osc_io_commit_async);
375 static bool osc_import_not_healthy(struct obd_import *imp)
377 return imp->imp_invalid || imp->imp_deactive ||
378 !(imp->imp_state == LUSTRE_IMP_FULL ||
379 imp->imp_state == LUSTRE_IMP_IDLE);
382 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios)
384 struct osc_object *osc = cl2osc(ios->cis_obj);
385 struct obd_import *imp = osc_cli(osc)->cl_import;
386 struct osc_io *oio = osc_env_io(env);
390 spin_lock(&imp->imp_lock);
392 * check whether this OSC device is available for non-delay read,
393 * fast switching mirror if we haven't tried all mirrors.
395 if (ios->cis_io->ci_type == CIT_READ && ios->cis_io->ci_ndelay &&
396 !ios->cis_io->ci_tried_all_mirrors && osc_import_not_healthy(imp)) {
398 } else if (likely(!imp->imp_invalid)) {
399 atomic_inc(&osc->oo_nr_ios);
400 oio->oi_is_active = 1;
403 spin_unlock(&imp->imp_lock);
405 if (cfs_capable(CFS_CAP_SYS_RESOURCE))
406 oio->oi_cap_sys_resource = 1;
410 EXPORT_SYMBOL(osc_io_iter_init);
412 int osc_io_rw_iter_init(const struct lu_env *env,
413 const struct cl_io_slice *ios)
415 struct cl_io *io = ios->cis_io;
416 struct osc_io *oio = osc_env_io(env);
417 struct osc_object *osc = cl2osc(ios->cis_obj);
418 unsigned long npages;
421 if (cl_io_is_append(io))
422 RETURN(osc_io_iter_init(env, ios));
424 npages = io->u.ci_rw.crw_count >> PAGE_SHIFT;
425 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
428 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
430 RETURN(osc_io_iter_init(env, ios));
432 EXPORT_SYMBOL(osc_io_rw_iter_init);
434 void osc_io_iter_fini(const struct lu_env *env,
435 const struct cl_io_slice *ios)
437 struct osc_io *oio = osc_env_io(env);
439 if (oio->oi_is_active) {
440 struct osc_object *osc = cl2osc(ios->cis_obj);
442 oio->oi_is_active = 0;
443 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
444 if (atomic_dec_and_test(&osc->oo_nr_ios))
445 wake_up_all(&osc->oo_io_waitq);
448 EXPORT_SYMBOL(osc_io_iter_fini);
450 void osc_io_rw_iter_fini(const struct lu_env *env,
451 const struct cl_io_slice *ios)
453 struct osc_io *oio = osc_env_io(env);
454 struct osc_object *osc = cl2osc(ios->cis_obj);
456 if (oio->oi_lru_reserved > 0) {
457 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
458 oio->oi_lru_reserved = 0;
460 oio->oi_write_osclock = NULL;
462 osc_io_iter_fini(env, ios);
464 EXPORT_SYMBOL(osc_io_rw_iter_fini);
466 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios)
469 struct cl_fault_io *fio;
473 fio = &io->u.ci_fault;
474 CDEBUG(D_INFO, "%lu %d %zu\n",
475 fio->ft_index, fio->ft_writable, fio->ft_nob);
477 * If mapping is writeable, adjust kms to cover this page,
478 * but do not extend kms beyond actual file size.
481 if (fio->ft_writable)
482 osc_page_touch_at(env, ios->cis_obj,
483 fio->ft_index, fio->ft_nob);
486 EXPORT_SYMBOL(osc_io_fault_start);
489 static int osc_async_upcall(void *a, int rc)
491 struct osc_async_cbargs *args = a;
494 complete(&args->opc_sync);
499 * Checks that there are no pages being written in the extent being truncated.
501 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
502 struct osc_page *ops , void *cbdata)
504 struct cl_page *page = ops->ops_cl.cpl_page;
505 struct osc_async_page *oap;
506 __u64 start = *(__u64 *)cbdata;
509 if (oap->oap_cmd & OBD_BRW_WRITE &&
510 !list_empty(&oap->oap_pending_item))
511 CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n",
512 start, current->comm);
514 if (PageLocked(page->cp_vmpage))
515 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
516 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
518 return CLP_GANG_OKAY;
521 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
522 struct osc_io *oio, __u64 size)
524 struct cl_object *clob;
528 clob = oio->oi_cl.cis_obj;
529 start = cl_index(clob, size);
530 partial = cl_offset(clob, start) < size;
533 * Complain if there are pages in the truncated region.
535 osc_page_gang_lookup(env, io, cl2osc(clob),
536 start + partial, CL_PAGE_EOF,
537 trunc_check_cb, (void *)&size);
540 static int osc_io_setattr_start(const struct lu_env *env,
541 const struct cl_io_slice *slice)
543 struct cl_io *io = slice->cis_io;
544 struct osc_io *oio = cl2osc_io(env, slice);
545 struct cl_object *obj = slice->cis_obj;
546 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
547 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
548 struct obdo *oa = &oio->oi_oa;
549 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
550 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
551 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
552 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
556 /* truncate cache dirty pages first */
557 if (cl_io_is_trunc(io))
558 result = osc_cache_truncate_start(env, cl2osc(obj), size,
561 if (result == 0 && oio->oi_lockless == 0) {
562 cl_object_attr_lock(obj);
563 result = cl_object_attr_get(env, obj, attr);
565 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
566 unsigned int cl_valid = 0;
568 if (ia_avalid & ATTR_SIZE) {
569 attr->cat_size = size;
570 attr->cat_kms = size;
571 cl_valid = (CAT_SIZE | CAT_KMS);
573 if (ia_avalid & ATTR_MTIME_SET) {
574 attr->cat_mtime = lvb->lvb_mtime;
575 cl_valid |= CAT_MTIME;
577 if (ia_avalid & ATTR_ATIME_SET) {
578 attr->cat_atime = lvb->lvb_atime;
579 cl_valid |= CAT_ATIME;
581 if (ia_xvalid & OP_XVALID_CTIME_SET) {
582 attr->cat_ctime = lvb->lvb_ctime;
583 cl_valid |= CAT_CTIME;
585 result = cl_object_attr_update(env, obj, attr,
588 cl_object_attr_unlock(obj);
590 memset(oa, 0, sizeof(*oa));
592 oa->o_oi = loi->loi_oi;
593 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
594 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
595 oa->o_layout = io->u.ci_setattr.sa_layout;
596 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
598 if (ia_avalid & ATTR_CTIME) {
599 oa->o_valid |= OBD_MD_FLCTIME;
600 oa->o_ctime = attr->cat_ctime;
602 if (ia_avalid & ATTR_ATIME) {
603 oa->o_valid |= OBD_MD_FLATIME;
604 oa->o_atime = attr->cat_atime;
606 if (ia_avalid & ATTR_MTIME) {
607 oa->o_valid |= OBD_MD_FLMTIME;
608 oa->o_mtime = attr->cat_mtime;
610 if (ia_avalid & ATTR_SIZE) {
612 oa->o_blocks = OBD_OBJECT_EOF;
613 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
615 if (oio->oi_lockless) {
616 oa->o_flags = OBD_FL_SRVLOCK;
617 oa->o_valid |= OBD_MD_FLFLAGS;
620 if (io->ci_layout_version > 0) {
621 /* verify layout version */
622 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
623 oa->o_layout_version = io->ci_layout_version;
626 LASSERT(oio->oi_lockless == 0);
629 if (ia_xvalid & OP_XVALID_FLAGS) {
630 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
631 oa->o_valid |= OBD_MD_FLFLAGS;
634 init_completion(&cbargs->opc_sync);
636 if (ia_avalid & ATTR_SIZE)
637 result = osc_punch_send(osc_export(cl2osc(obj)),
638 oa, osc_async_upcall, cbargs);
640 result = osc_setattr_async(osc_export(cl2osc(obj)),
641 oa, osc_async_upcall,
642 cbargs, PTLRPCD_SET);
644 cbargs->opc_rpc_sent = result == 0;
650 void osc_io_setattr_end(const struct lu_env *env,
651 const struct cl_io_slice *slice)
653 struct cl_io *io = slice->cis_io;
654 struct osc_io *oio = cl2osc_io(env, slice);
655 struct cl_object *obj = slice->cis_obj;
656 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
657 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
658 struct obdo *oa = &oio->oi_oa;
659 unsigned int cl_valid = 0;
662 if (cbargs->opc_rpc_sent) {
663 wait_for_completion(&cbargs->opc_sync);
664 result = io->ci_result = cbargs->opc_rc;
668 if (oio->oi_lockless) {
669 /* lockless truncate */
670 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
672 LASSERT(cl_io_is_trunc(io));
673 /* XXX: Need a lock. */
674 osd->od_stats.os_lockless_truncates++;
678 if (cl_io_is_trunc(io)) {
679 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
680 cl_object_attr_lock(obj);
681 if (oa->o_valid & OBD_MD_FLBLOCKS) {
682 attr->cat_blocks = oa->o_blocks;
683 cl_valid |= CAT_BLOCKS;
686 cl_object_attr_update(env, obj, attr, cl_valid);
687 cl_object_attr_unlock(obj);
688 osc_trunc_check(env, io, oio, size);
689 osc_cache_truncate_end(env, oio->oi_trunc);
690 oio->oi_trunc = NULL;
693 EXPORT_SYMBOL(osc_io_setattr_end);
695 struct osc_data_version_args {
696 struct osc_io *dva_oio;
700 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
703 struct osc_data_version_args *dva = args;
704 struct osc_io *oio = dva->dva_oio;
705 const struct ost_body *body;
711 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
713 GOTO(out, rc = -EPROTO);
715 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
719 oio->oi_cbarg.opc_rc = rc;
720 complete(&oio->oi_cbarg.opc_sync);
725 static int osc_io_data_version_start(const struct lu_env *env,
726 const struct cl_io_slice *slice)
728 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
729 struct osc_io *oio = cl2osc_io(env, slice);
730 struct obdo *oa = &oio->oi_oa;
731 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
732 struct osc_object *obj = cl2osc(slice->cis_obj);
733 struct lov_oinfo *loi = obj->oo_oinfo;
734 struct obd_export *exp = osc_export(obj);
735 struct ptlrpc_request *req;
736 struct ost_body *body;
737 struct osc_data_version_args *dva;
741 memset(oa, 0, sizeof(*oa));
742 oa->o_oi = loi->loi_oi;
743 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
745 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
746 oa->o_valid |= OBD_MD_FLFLAGS;
747 oa->o_flags |= OBD_FL_SRVLOCK;
748 if (dv->dv_flags & LL_DV_WR_FLUSH)
749 oa->o_flags |= OBD_FL_FLUSH;
752 init_completion(&cbargs->opc_sync);
754 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
758 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
760 ptlrpc_request_free(req);
764 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
765 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
767 ptlrpc_request_set_replen(req);
768 req->rq_interpret_reply = osc_data_version_interpret;
769 dva = ptlrpc_req_async_args(dva, req);
772 ptlrpcd_add_req(req);
777 static void osc_io_data_version_end(const struct lu_env *env,
778 const struct cl_io_slice *slice)
780 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
781 struct osc_io *oio = cl2osc_io(env, slice);
782 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
785 wait_for_completion(&cbargs->opc_sync);
787 if (cbargs->opc_rc != 0) {
788 slice->cis_io->ci_result = cbargs->opc_rc;
790 slice->cis_io->ci_result = 0;
791 if (!(oio->oi_oa.o_valid &
792 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
793 slice->cis_io->ci_result = -ENOTSUPP;
795 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
796 dv->dv_layout_version = oio->oi_oa.o_layout_version;
797 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
798 dv->dv_data_version = oio->oi_oa.o_data_version;
804 int osc_io_read_start(const struct lu_env *env,
805 const struct cl_io_slice *slice)
807 struct cl_object *obj = slice->cis_obj;
808 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
812 if (!slice->cis_io->ci_noatime) {
813 cl_object_attr_lock(obj);
814 attr->cat_atime = ktime_get_real_seconds();
815 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
816 cl_object_attr_unlock(obj);
821 EXPORT_SYMBOL(osc_io_read_start);
823 int osc_io_write_start(const struct lu_env *env,
824 const struct cl_io_slice *slice)
826 struct cl_object *obj = slice->cis_obj;
827 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
831 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
832 cl_object_attr_lock(obj);
833 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
834 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
835 cl_object_attr_unlock(obj);
839 EXPORT_SYMBOL(osc_io_write_start);
841 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
842 struct cl_fsync_io *fio)
844 struct osc_io *oio = osc_env_io(env);
845 struct obdo *oa = &oio->oi_oa;
846 struct lov_oinfo *loi = obj->oo_oinfo;
847 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
851 memset(oa, 0, sizeof(*oa));
852 oa->o_oi = loi->loi_oi;
853 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
855 /* reload size abd blocks for start and end of sync range */
856 oa->o_size = fio->fi_start;
857 oa->o_blocks = fio->fi_end;
858 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
860 obdo_set_parent_fid(oa, fio->fi_fid);
862 init_completion(&cbargs->opc_sync);
864 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
867 EXPORT_SYMBOL(osc_fsync_ost);
869 int osc_io_fsync_start(const struct lu_env *env,
870 const struct cl_io_slice *slice)
872 struct cl_io *io = slice->cis_io;
873 struct cl_fsync_io *fio = &io->u.ci_fsync;
874 struct cl_object *obj = slice->cis_obj;
875 struct osc_object *osc = cl2osc(obj);
876 pgoff_t start = cl_index(obj, fio->fi_start);
877 pgoff_t end = cl_index(obj, fio->fi_end);
881 if (fio->fi_end == OBD_OBJECT_EOF)
884 result = osc_cache_writeback_range(env, osc, start, end, 0,
885 fio->fi_mode == CL_FSYNC_DISCARD);
887 fio->fi_nr_written += result;
890 if (fio->fi_mode == CL_FSYNC_ALL) {
893 /* we have to wait for writeback to finish before we can
894 * send OST_SYNC RPC. This is bad because it causes extents
895 * to be written osc by osc. However, we usually start
896 * writeback before CL_FSYNC_ALL so this won't have any real
898 rc = osc_cache_wait_range(env, osc, start, end);
901 rc = osc_fsync_ost(env, osc, fio);
909 void osc_io_fsync_end(const struct lu_env *env,
910 const struct cl_io_slice *slice)
912 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
913 struct cl_object *obj = slice->cis_obj;
914 pgoff_t start = cl_index(obj, fio->fi_start);
915 pgoff_t end = cl_index(obj, fio->fi_end);
918 if (fio->fi_mode == CL_FSYNC_LOCAL) {
919 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
920 } else if (fio->fi_mode == CL_FSYNC_ALL) {
921 struct osc_io *oio = cl2osc_io(env, slice);
922 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
924 wait_for_completion(&cbargs->opc_sync);
926 result = cbargs->opc_rc;
928 slice->cis_io->ci_result = result;
930 EXPORT_SYMBOL(osc_io_fsync_end);
932 static int osc_io_ladvise_start(const struct lu_env *env,
933 const struct cl_io_slice *slice)
936 struct cl_io *io = slice->cis_io;
937 struct osc_io *oio = cl2osc_io(env, slice);
938 struct cl_object *obj = slice->cis_obj;
939 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
940 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
941 struct obdo *oa = &oio->oi_oa;
942 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
943 struct lu_ladvise *ladvise;
944 struct ladvise_hdr *ladvise_hdr;
949 /* TODO: add multiple ladvise support in CLIO */
950 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
951 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
952 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
954 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
955 if (ladvise_hdr == NULL)
958 memset(ladvise_hdr, 0, buf_size);
959 ladvise_hdr->lah_magic = LADVISE_MAGIC;
960 ladvise_hdr->lah_count = num_advise;
961 ladvise_hdr->lah_flags = lio->li_flags;
963 memset(oa, 0, sizeof(*oa));
964 oa->o_oi = loi->loi_oi;
965 oa->o_valid = OBD_MD_FLID;
966 obdo_set_parent_fid(oa, lio->li_fid);
968 ladvise = ladvise_hdr->lah_advise;
969 ladvise->lla_start = lio->li_start;
970 ladvise->lla_end = lio->li_end;
971 ladvise->lla_advice = lio->li_advice;
973 if (lio->li_flags & LF_ASYNC) {
974 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
975 ladvise_hdr, NULL, NULL, NULL);
977 init_completion(&cbargs->opc_sync);
978 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
979 ladvise_hdr, osc_async_upcall,
980 cbargs, PTLRPCD_SET);
981 cbargs->opc_rpc_sent = result == 0;
986 static void osc_io_ladvise_end(const struct lu_env *env,
987 const struct cl_io_slice *slice)
989 struct cl_io *io = slice->cis_io;
990 struct osc_io *oio = cl2osc_io(env, slice);
991 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
993 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
995 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
996 wait_for_completion(&cbargs->opc_sync);
997 result = cbargs->opc_rc;
999 slice->cis_io->ci_result = result;
1002 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice)
1004 struct osc_io *oio = cl2osc_io(env, slice);
1006 if (oio->oi_active) {
1007 osc_extent_release(env, oio->oi_active);
1008 oio->oi_active = NULL;
1011 EXPORT_SYMBOL(osc_io_end);
1013 static const struct cl_io_operations osc_io_ops = {
1016 .cio_iter_init = osc_io_rw_iter_init,
1017 .cio_iter_fini = osc_io_rw_iter_fini,
1018 .cio_start = osc_io_read_start,
1019 .cio_fini = osc_io_fini
1022 .cio_iter_init = osc_io_rw_iter_init,
1023 .cio_iter_fini = osc_io_rw_iter_fini,
1024 .cio_start = osc_io_write_start,
1025 .cio_end = osc_io_end,
1026 .cio_fini = osc_io_fini
1029 .cio_iter_init = osc_io_iter_init,
1030 .cio_iter_fini = osc_io_iter_fini,
1031 .cio_start = osc_io_setattr_start,
1032 .cio_end = osc_io_setattr_end
1034 [CIT_DATA_VERSION] = {
1035 .cio_start = osc_io_data_version_start,
1036 .cio_end = osc_io_data_version_end,
1039 .cio_iter_init = osc_io_iter_init,
1040 .cio_iter_fini = osc_io_iter_fini,
1041 .cio_start = osc_io_fault_start,
1042 .cio_end = osc_io_end,
1043 .cio_fini = osc_io_fini
1046 .cio_start = osc_io_fsync_start,
1047 .cio_end = osc_io_fsync_end,
1048 .cio_fini = osc_io_fini
1051 .cio_start = osc_io_ladvise_start,
1052 .cio_end = osc_io_ladvise_end,
1053 .cio_fini = osc_io_fini
1056 .cio_fini = osc_io_fini
1059 .cio_read_ahead = osc_io_read_ahead,
1060 .cio_submit = osc_io_submit,
1061 .cio_commit_async = osc_io_commit_async
1064 /*****************************************************************************
1066 * Transfer operations.
1070 int osc_io_init(const struct lu_env *env,
1071 struct cl_object *obj, struct cl_io *io)
1073 struct osc_io *oio = osc_env_io(env);
1075 CL_IO_SLICE_CLEAN(oio, oi_cl);
1076 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);