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;
124 unsigned int ppc_bits; /* pages per chunk bits */
126 bool sync_queue = false;
128 LASSERT(qin->pl_nr > 0);
130 CDEBUG(D_CACHE|D_READA, "%d %d\n", qin->pl_nr, crt);
132 osc = cl2osc(ios->cis_obj);
134 max_pages = cli->cl_max_pages_per_rpc;
135 ppc_bits = cli->cl_chunkbits - PAGE_SHIFT;
138 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
139 brw_flags |= crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
140 if (crt == CRT_READ && ios->cis_io->ci_ndelay)
141 brw_flags |= OBD_BRW_NDELAY;
144 * NOTE: here @page is a top-level page. This is done to avoid
145 * creation of sub-page-list.
147 cl_page_list_for_each_safe(page, tmp, qin) {
148 struct osc_async_page *oap;
154 opg = osc_cl_page_osc(page, osc);
156 LASSERT(osc == oap->oap_obj);
158 if (!list_empty(&oap->oap_pending_item) ||
159 !list_empty(&oap->oap_rpc_item)) {
160 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
166 result = cl_page_prep(env, io, page, crt);
169 if (result != -EALREADY)
172 * Handle -EALREADY error: for read case, the page is
173 * already in UPTODATE state; for write, the page
180 spin_lock(&oap->oap_lock);
181 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
182 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
183 spin_unlock(&oap->oap_lock);
185 osc_page_submit(env, opg, crt, brw_flags);
186 list_add_tail(&oap->oap_pending_item, &list);
188 if (page->cp_sync_io != NULL)
189 cl_page_list_move(qout, qin, page);
191 cl_page_list_del(env, qin, page);
194 if (queued == max_pages) {
196 } else if (crt == CRT_WRITE) {
198 unsigned int next_chunks;
200 chunks = (queued + ppc - 1) >> ppc_bits;
201 /* chunk number if add another page */
202 next_chunks = (queued + ppc) >> ppc_bits;
204 /* next page will excceed write chunk limit */
205 if (chunks == osc_max_write_chunks(cli) &&
206 next_chunks > chunks)
211 result = osc_queue_sync_pages(env, io, osc, &list,
221 result = osc_queue_sync_pages(env, io, osc, &list, brw_flags);
223 /* Update c/mtime for sync write. LU-7310 */
224 if (crt == CRT_WRITE && qout->pl_nr > 0 && result == 0) {
225 struct cl_object *obj = ios->cis_obj;
226 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
228 cl_object_attr_lock(obj);
229 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
230 cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
231 cl_object_attr_unlock(obj);
234 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
235 return qout->pl_nr > 0 ? 0 : result;
237 EXPORT_SYMBOL(osc_io_submit);
240 * This is called to update the attributes when modifying a specific page,
241 * both when making new pages and when doing updates to existing cached pages.
243 * Expand stripe KMS if necessary.
245 void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj,
246 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;
255 /* offset within stripe */
256 kms = cl_offset(obj, idx) + to;
258 cl_object_attr_lock(obj);
259 CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n",
260 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
261 loi->loi_lvb.lvb_size);
263 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
264 valid = CAT_MTIME | CAT_CTIME;
265 if (kms > loi->loi_kms) {
269 if (kms > loi->loi_lvb.lvb_size) {
270 attr->cat_size = kms;
273 cl_object_attr_update(env, obj, attr, valid);
274 cl_object_attr_unlock(obj);
279 int osc_io_commit_async(const struct lu_env *env,
280 const struct cl_io_slice *ios,
281 struct cl_page_list *qin, int from, int to,
284 struct cl_io *io = ios->cis_io;
285 struct osc_io *oio = cl2osc_io(env, ios);
286 struct osc_object *osc = cl2osc(ios->cis_obj);
287 struct cl_page *page;
288 struct cl_page *last_page;
289 struct osc_page *opg;
293 LASSERT(qin->pl_nr > 0);
295 /* Handle partial page cases */
296 last_page = cl_page_list_last(qin);
297 if (oio->oi_lockless) {
298 page = cl_page_list_first(qin);
299 if (page == last_page) {
300 cl_page_clip(env, page, from, to);
303 cl_page_clip(env, page, from, PAGE_SIZE);
305 cl_page_clip(env, last_page, 0, to);
309 while (qin->pl_nr > 0) {
310 struct osc_async_page *oap;
312 page = cl_page_list_first(qin);
313 opg = osc_cl_page_osc(page, osc);
316 LASSERTF(osc == oap->oap_obj,
317 "obj mismatch: %p / %p\n", osc, oap->oap_obj);
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);
354 EXPORT_SYMBOL(osc_io_commit_async);
356 int osc_io_iter_init(const struct lu_env *env, 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;
360 struct osc_io *oio = osc_env_io(env);
363 spin_lock(&imp->imp_lock);
364 if (likely(!imp->imp_invalid)) {
365 atomic_inc(&osc->oo_nr_ios);
366 oio->oi_is_active = 1;
369 spin_unlock(&imp->imp_lock);
371 if (cfs_capable(CFS_CAP_SYS_RESOURCE))
372 oio->oi_cap_sys_resource = 1;
376 EXPORT_SYMBOL(osc_io_iter_init);
378 int osc_io_write_iter_init(const struct lu_env *env,
379 const struct cl_io_slice *ios)
381 struct cl_io *io = ios->cis_io;
382 struct osc_io *oio = osc_env_io(env);
383 struct osc_object *osc = cl2osc(ios->cis_obj);
384 unsigned long npages;
387 if (cl_io_is_append(io))
388 RETURN(osc_io_iter_init(env, ios));
390 npages = io->u.ci_rw.crw_count >> PAGE_SHIFT;
391 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
394 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
396 RETURN(osc_io_iter_init(env, ios));
398 EXPORT_SYMBOL(osc_io_write_iter_init);
400 void osc_io_iter_fini(const struct lu_env *env,
401 const struct cl_io_slice *ios)
403 struct osc_io *oio = osc_env_io(env);
405 if (oio->oi_is_active) {
406 struct osc_object *osc = cl2osc(ios->cis_obj);
408 oio->oi_is_active = 0;
409 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
410 if (atomic_dec_and_test(&osc->oo_nr_ios))
411 wake_up_all(&osc->oo_io_waitq);
414 EXPORT_SYMBOL(osc_io_iter_fini);
416 void osc_io_write_iter_fini(const struct lu_env *env,
417 const struct cl_io_slice *ios)
419 struct osc_io *oio = osc_env_io(env);
420 struct osc_object *osc = cl2osc(ios->cis_obj);
422 if (oio->oi_lru_reserved > 0) {
423 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
424 oio->oi_lru_reserved = 0;
426 oio->oi_write_osclock = NULL;
428 osc_io_iter_fini(env, ios);
430 EXPORT_SYMBOL(osc_io_write_iter_fini);
432 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios)
435 struct cl_fault_io *fio;
439 fio = &io->u.ci_fault;
440 CDEBUG(D_INFO, "%lu %d %zu\n",
441 fio->ft_index, fio->ft_writable, fio->ft_nob);
443 * If mapping is writeable, adjust kms to cover this page,
444 * but do not extend kms beyond actual file size.
447 if (fio->ft_writable)
448 osc_page_touch_at(env, ios->cis_obj,
449 fio->ft_index, fio->ft_nob);
452 EXPORT_SYMBOL(osc_io_fault_start);
455 static int osc_async_upcall(void *a, int rc)
457 struct osc_async_cbargs *args = a;
460 complete(&args->opc_sync);
465 * Checks that there are no pages being written in the extent being truncated.
467 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
468 struct osc_page *ops , void *cbdata)
470 struct cl_page *page = ops->ops_cl.cpl_page;
471 struct osc_async_page *oap;
472 __u64 start = *(__u64 *)cbdata;
475 if (oap->oap_cmd & OBD_BRW_WRITE &&
476 !list_empty(&oap->oap_pending_item))
477 CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n",
478 start, current->comm);
480 if (PageLocked(page->cp_vmpage))
481 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
482 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
484 return CLP_GANG_OKAY;
487 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
488 struct osc_io *oio, __u64 size)
490 struct cl_object *clob;
494 clob = oio->oi_cl.cis_obj;
495 start = cl_index(clob, size);
496 partial = cl_offset(clob, start) < size;
499 * Complain if there are pages in the truncated region.
501 osc_page_gang_lookup(env, io, cl2osc(clob),
502 start + partial, CL_PAGE_EOF,
503 trunc_check_cb, (void *)&size);
506 static int osc_io_setattr_start(const struct lu_env *env,
507 const struct cl_io_slice *slice)
509 struct cl_io *io = slice->cis_io;
510 struct osc_io *oio = cl2osc_io(env, slice);
511 struct cl_object *obj = slice->cis_obj;
512 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
513 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
514 struct obdo *oa = &oio->oi_oa;
515 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
516 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
517 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
518 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
522 /* truncate cache dirty pages first */
523 if (cl_io_is_trunc(io))
524 result = osc_cache_truncate_start(env, cl2osc(obj), size,
527 if (result == 0 && oio->oi_lockless == 0) {
528 cl_object_attr_lock(obj);
529 result = cl_object_attr_get(env, obj, attr);
531 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
532 unsigned int cl_valid = 0;
534 if (ia_avalid & ATTR_SIZE) {
535 attr->cat_size = size;
536 attr->cat_kms = size;
537 cl_valid = (CAT_SIZE | CAT_KMS);
539 if (ia_avalid & ATTR_MTIME_SET) {
540 attr->cat_mtime = lvb->lvb_mtime;
541 cl_valid |= CAT_MTIME;
543 if (ia_avalid & ATTR_ATIME_SET) {
544 attr->cat_atime = lvb->lvb_atime;
545 cl_valid |= CAT_ATIME;
547 if (ia_xvalid & OP_XVALID_CTIME_SET) {
548 attr->cat_ctime = lvb->lvb_ctime;
549 cl_valid |= CAT_CTIME;
551 result = cl_object_attr_update(env, obj, attr,
554 cl_object_attr_unlock(obj);
556 memset(oa, 0, sizeof(*oa));
558 oa->o_oi = loi->loi_oi;
559 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
560 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
561 oa->o_layout = io->u.ci_setattr.sa_layout;
562 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
564 if (ia_avalid & ATTR_CTIME) {
565 oa->o_valid |= OBD_MD_FLCTIME;
566 oa->o_ctime = attr->cat_ctime;
568 if (ia_avalid & ATTR_ATIME) {
569 oa->o_valid |= OBD_MD_FLATIME;
570 oa->o_atime = attr->cat_atime;
572 if (ia_avalid & ATTR_MTIME) {
573 oa->o_valid |= OBD_MD_FLMTIME;
574 oa->o_mtime = attr->cat_mtime;
576 if (ia_avalid & ATTR_SIZE) {
578 oa->o_blocks = OBD_OBJECT_EOF;
579 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
581 if (oio->oi_lockless) {
582 oa->o_flags = OBD_FL_SRVLOCK;
583 oa->o_valid |= OBD_MD_FLFLAGS;
586 if (io->ci_layout_version > 0) {
587 /* verify layout version */
588 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
589 oa->o_layout_version = io->ci_layout_version;
592 LASSERT(oio->oi_lockless == 0);
595 if (ia_xvalid & OP_XVALID_FLAGS) {
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_avalid & ATTR_SIZE)
603 result = osc_punch_send(osc_export(cl2osc(obj)),
604 oa, osc_async_upcall, cbargs);
606 result = osc_setattr_async(osc_export(cl2osc(obj)),
607 oa, osc_async_upcall,
608 cbargs, PTLRPCD_SET);
610 cbargs->opc_rpc_sent = result == 0;
616 void osc_io_setattr_end(const struct lu_env *env,
617 const struct cl_io_slice *slice)
619 struct cl_io *io = slice->cis_io;
620 struct osc_io *oio = cl2osc_io(env, slice);
621 struct cl_object *obj = slice->cis_obj;
622 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
623 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
624 struct obdo *oa = &oio->oi_oa;
625 unsigned int cl_valid = 0;
628 if (cbargs->opc_rpc_sent) {
629 wait_for_completion(&cbargs->opc_sync);
630 result = io->ci_result = cbargs->opc_rc;
634 if (oio->oi_lockless) {
635 /* lockless truncate */
636 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
638 LASSERT(cl_io_is_trunc(io));
639 /* XXX: Need a lock. */
640 osd->od_stats.os_lockless_truncates++;
644 if (cl_io_is_trunc(io)) {
645 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
646 cl_object_attr_lock(obj);
647 if (oa->o_valid & OBD_MD_FLBLOCKS) {
648 attr->cat_blocks = oa->o_blocks;
649 cl_valid |= CAT_BLOCKS;
652 cl_object_attr_update(env, obj, attr, cl_valid);
653 cl_object_attr_unlock(obj);
654 osc_trunc_check(env, io, oio, size);
655 osc_cache_truncate_end(env, oio->oi_trunc);
656 oio->oi_trunc = NULL;
659 EXPORT_SYMBOL(osc_io_setattr_end);
661 struct osc_data_version_args {
662 struct osc_io *dva_oio;
666 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
669 struct osc_data_version_args *dva = args;
670 struct osc_io *oio = dva->dva_oio;
671 const struct ost_body *body;
677 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
679 GOTO(out, rc = -EPROTO);
681 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
685 oio->oi_cbarg.opc_rc = rc;
686 complete(&oio->oi_cbarg.opc_sync);
691 static int osc_io_data_version_start(const struct lu_env *env,
692 const struct cl_io_slice *slice)
694 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
695 struct osc_io *oio = cl2osc_io(env, slice);
696 struct obdo *oa = &oio->oi_oa;
697 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
698 struct osc_object *obj = cl2osc(slice->cis_obj);
699 struct lov_oinfo *loi = obj->oo_oinfo;
700 struct obd_export *exp = osc_export(obj);
701 struct ptlrpc_request *req;
702 struct ost_body *body;
703 struct osc_data_version_args *dva;
707 memset(oa, 0, sizeof(*oa));
708 oa->o_oi = loi->loi_oi;
709 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
711 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
712 oa->o_valid |= OBD_MD_FLFLAGS;
713 oa->o_flags |= OBD_FL_SRVLOCK;
714 if (dv->dv_flags & LL_DV_WR_FLUSH)
715 oa->o_flags |= OBD_FL_FLUSH;
718 init_completion(&cbargs->opc_sync);
720 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
724 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
726 ptlrpc_request_free(req);
730 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
731 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
733 ptlrpc_request_set_replen(req);
734 req->rq_interpret_reply = osc_data_version_interpret;
735 CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
736 dva = ptlrpc_req_async_args(req);
739 ptlrpcd_add_req(req);
744 static void osc_io_data_version_end(const struct lu_env *env,
745 const struct cl_io_slice *slice)
747 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
748 struct osc_io *oio = cl2osc_io(env, slice);
749 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
752 wait_for_completion(&cbargs->opc_sync);
754 if (cbargs->opc_rc != 0) {
755 slice->cis_io->ci_result = cbargs->opc_rc;
757 slice->cis_io->ci_result = 0;
758 if (!(oio->oi_oa.o_valid &
759 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
760 slice->cis_io->ci_result = -ENOTSUPP;
762 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
763 dv->dv_layout_version = oio->oi_oa.o_layout_version;
764 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
765 dv->dv_data_version = oio->oi_oa.o_data_version;
771 int osc_io_read_start(const struct lu_env *env,
772 const struct cl_io_slice *slice)
774 struct cl_object *obj = slice->cis_obj;
775 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
779 if (!slice->cis_io->ci_noatime) {
780 cl_object_attr_lock(obj);
781 attr->cat_atime = ktime_get_real_seconds();
782 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
783 cl_object_attr_unlock(obj);
788 EXPORT_SYMBOL(osc_io_read_start);
790 int osc_io_write_start(const struct lu_env *env,
791 const struct cl_io_slice *slice)
793 struct cl_object *obj = slice->cis_obj;
794 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
798 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
799 cl_object_attr_lock(obj);
800 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
801 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
802 cl_object_attr_unlock(obj);
806 EXPORT_SYMBOL(osc_io_write_start);
808 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
809 struct cl_fsync_io *fio)
811 struct osc_io *oio = osc_env_io(env);
812 struct obdo *oa = &oio->oi_oa;
813 struct lov_oinfo *loi = obj->oo_oinfo;
814 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
818 memset(oa, 0, sizeof(*oa));
819 oa->o_oi = loi->loi_oi;
820 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
822 /* reload size abd blocks for start and end of sync range */
823 oa->o_size = fio->fi_start;
824 oa->o_blocks = fio->fi_end;
825 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
827 obdo_set_parent_fid(oa, fio->fi_fid);
829 init_completion(&cbargs->opc_sync);
831 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
834 EXPORT_SYMBOL(osc_fsync_ost);
836 int osc_io_fsync_start(const struct lu_env *env,
837 const struct cl_io_slice *slice)
839 struct cl_io *io = slice->cis_io;
840 struct cl_fsync_io *fio = &io->u.ci_fsync;
841 struct cl_object *obj = slice->cis_obj;
842 struct osc_object *osc = cl2osc(obj);
843 pgoff_t start = cl_index(obj, fio->fi_start);
844 pgoff_t end = cl_index(obj, fio->fi_end);
848 if (fio->fi_end == OBD_OBJECT_EOF)
851 result = osc_cache_writeback_range(env, osc, start, end, 0,
852 fio->fi_mode == CL_FSYNC_DISCARD);
854 fio->fi_nr_written += result;
857 if (fio->fi_mode == CL_FSYNC_ALL) {
860 /* we have to wait for writeback to finish before we can
861 * send OST_SYNC RPC. This is bad because it causes extents
862 * to be written osc by osc. However, we usually start
863 * writeback before CL_FSYNC_ALL so this won't have any real
865 rc = osc_cache_wait_range(env, osc, start, end);
868 rc = osc_fsync_ost(env, osc, fio);
876 void osc_io_fsync_end(const struct lu_env *env,
877 const struct cl_io_slice *slice)
879 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
880 struct cl_object *obj = slice->cis_obj;
881 pgoff_t start = cl_index(obj, fio->fi_start);
882 pgoff_t end = cl_index(obj, fio->fi_end);
885 if (fio->fi_mode == CL_FSYNC_LOCAL) {
886 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
887 } else if (fio->fi_mode == CL_FSYNC_ALL) {
888 struct osc_io *oio = cl2osc_io(env, slice);
889 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
891 wait_for_completion(&cbargs->opc_sync);
893 result = cbargs->opc_rc;
895 slice->cis_io->ci_result = result;
897 EXPORT_SYMBOL(osc_io_fsync_end);
899 static int osc_io_ladvise_start(const struct lu_env *env,
900 const struct cl_io_slice *slice)
903 struct cl_io *io = slice->cis_io;
904 struct osc_io *oio = cl2osc_io(env, slice);
905 struct cl_object *obj = slice->cis_obj;
906 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
907 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
908 struct obdo *oa = &oio->oi_oa;
909 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
910 struct lu_ladvise *ladvise;
911 struct ladvise_hdr *ladvise_hdr;
916 /* TODO: add multiple ladvise support in CLIO */
917 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
918 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
919 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
921 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
922 if (ladvise_hdr == NULL)
925 memset(ladvise_hdr, 0, buf_size);
926 ladvise_hdr->lah_magic = LADVISE_MAGIC;
927 ladvise_hdr->lah_count = num_advise;
928 ladvise_hdr->lah_flags = lio->li_flags;
930 memset(oa, 0, sizeof(*oa));
931 oa->o_oi = loi->loi_oi;
932 oa->o_valid = OBD_MD_FLID;
933 obdo_set_parent_fid(oa, lio->li_fid);
935 ladvise = ladvise_hdr->lah_advise;
936 ladvise->lla_start = lio->li_start;
937 ladvise->lla_end = lio->li_end;
938 ladvise->lla_advice = lio->li_advice;
940 if (lio->li_flags & LF_ASYNC) {
941 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
942 ladvise_hdr, NULL, NULL, NULL);
944 init_completion(&cbargs->opc_sync);
945 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
946 ladvise_hdr, osc_async_upcall,
947 cbargs, PTLRPCD_SET);
948 cbargs->opc_rpc_sent = result == 0;
953 static void osc_io_ladvise_end(const struct lu_env *env,
954 const struct cl_io_slice *slice)
956 struct cl_io *io = slice->cis_io;
957 struct osc_io *oio = cl2osc_io(env, slice);
958 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
960 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
962 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
963 wait_for_completion(&cbargs->opc_sync);
964 result = cbargs->opc_rc;
966 slice->cis_io->ci_result = result;
969 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice)
971 struct osc_io *oio = cl2osc_io(env, slice);
973 if (oio->oi_active) {
974 osc_extent_release(env, oio->oi_active);
975 oio->oi_active = NULL;
978 EXPORT_SYMBOL(osc_io_end);
980 static const struct cl_io_operations osc_io_ops = {
983 .cio_iter_init = osc_io_iter_init,
984 .cio_iter_fini = osc_io_iter_fini,
985 .cio_start = osc_io_read_start,
986 .cio_fini = osc_io_fini
989 .cio_iter_init = osc_io_write_iter_init,
990 .cio_iter_fini = osc_io_write_iter_fini,
991 .cio_start = osc_io_write_start,
992 .cio_end = osc_io_end,
993 .cio_fini = osc_io_fini
996 .cio_iter_init = osc_io_iter_init,
997 .cio_iter_fini = osc_io_iter_fini,
998 .cio_start = osc_io_setattr_start,
999 .cio_end = osc_io_setattr_end
1001 [CIT_DATA_VERSION] = {
1002 .cio_start = osc_io_data_version_start,
1003 .cio_end = osc_io_data_version_end,
1006 .cio_iter_init = osc_io_iter_init,
1007 .cio_iter_fini = osc_io_iter_fini,
1008 .cio_start = osc_io_fault_start,
1009 .cio_end = osc_io_end,
1010 .cio_fini = osc_io_fini
1013 .cio_start = osc_io_fsync_start,
1014 .cio_end = osc_io_fsync_end,
1015 .cio_fini = osc_io_fini
1018 .cio_start = osc_io_ladvise_start,
1019 .cio_end = osc_io_ladvise_end,
1020 .cio_fini = osc_io_fini
1023 .cio_fini = osc_io_fini
1026 .cio_read_ahead = osc_io_read_ahead,
1027 .cio_submit = osc_io_submit,
1028 .cio_commit_async = osc_io_commit_async
1031 /*****************************************************************************
1033 * Transfer operations.
1037 int osc_io_init(const struct lu_env *env,
1038 struct cl_object *obj, struct cl_io *io)
1040 struct osc_io *oio = osc_env_io(env);
1042 CL_IO_SLICE_CLEAN(oio, oi_cl);
1043 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);