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;
94 if (ra->cra_end != CL_PAGE_EOF)
95 ra->cra_contention = true;
103 * An implementation of cl_io_operations::cio_io_submit() method for osc
104 * layer. Iterates over pages in the in-queue, prepares each for io by calling
105 * cl_page_prep() and then either submits them through osc_io_submit_page()
106 * or, if page is already submitted, changes osc flags through
107 * osc_set_async_flags().
109 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
110 enum cl_req_type crt, struct cl_2queue *queue)
112 struct cl_page *page;
114 struct client_obd *cli = NULL;
115 struct osc_object *osc = NULL; /* to keep gcc happy */
116 struct osc_page *opg;
118 struct list_head list = LIST_HEAD_INIT(list);
120 struct cl_page_list *qin = &queue->c2_qin;
121 struct cl_page_list *qout = &queue->c2_qout;
122 unsigned int queued = 0;
125 unsigned int max_pages;
126 unsigned int ppc_bits; /* pages per chunk bits */
128 bool sync_queue = false;
130 LASSERT(qin->pl_nr > 0);
132 CDEBUG(D_CACHE|D_READA, "%d %d\n", qin->pl_nr, crt);
134 osc = cl2osc(ios->cis_obj);
136 max_pages = cli->cl_max_pages_per_rpc;
137 ppc_bits = cli->cl_chunkbits - PAGE_SHIFT;
140 brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
141 brw_flags |= crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
142 if (crt == CRT_READ && ios->cis_io->ci_ndelay)
143 brw_flags |= OBD_BRW_NDELAY;
146 * NOTE: here @page is a top-level page. This is done to avoid
147 * creation of sub-page-list.
149 cl_page_list_for_each_safe(page, tmp, qin) {
150 struct osc_async_page *oap;
156 opg = osc_cl_page_osc(page, osc);
158 LASSERT(osc == oap->oap_obj);
160 if (!list_empty(&oap->oap_pending_item) ||
161 !list_empty(&oap->oap_rpc_item)) {
162 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
168 result = cl_page_prep(env, io, page, crt);
171 if (result != -EALREADY)
174 * Handle -EALREADY error: for read case, the page is
175 * already in UPTODATE state; for write, the page
182 spin_lock(&oap->oap_lock);
183 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
184 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
185 spin_unlock(&oap->oap_lock);
187 osc_page_submit(env, opg, crt, brw_flags);
188 list_add_tail(&oap->oap_pending_item, &list);
190 if (page->cp_sync_io != NULL)
191 cl_page_list_move(qout, qin, page);
193 cl_page_list_del(env, qin, page);
196 if (queued == max_pages) {
198 } else if (crt == CRT_WRITE) {
200 unsigned int next_chunks;
202 chunks = (queued + ppc - 1) >> ppc_bits;
203 /* chunk number if add another page */
204 next_chunks = (queued + ppc) >> ppc_bits;
206 /* next page will excceed write chunk limit */
207 if (chunks == osc_max_write_chunks(cli) &&
208 next_chunks > chunks)
213 result = osc_queue_sync_pages(env, io, osc, &list,
223 result = osc_queue_sync_pages(env, io, osc, &list, brw_flags);
225 /* Update c/mtime for sync write. LU-7310 */
226 if (crt == CRT_WRITE && qout->pl_nr > 0 && result == 0) {
227 struct cl_object *obj = ios->cis_obj;
228 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
230 cl_object_attr_lock(obj);
231 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
232 cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
233 cl_object_attr_unlock(obj);
236 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
237 return qout->pl_nr > 0 ? 0 : result;
239 EXPORT_SYMBOL(osc_io_submit);
242 * This is called to update the attributes when modifying a specific page,
243 * both when making new pages and when doing updates to existing cached pages.
245 * Expand stripe KMS if necessary.
247 void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj,
248 pgoff_t idx, size_t to)
250 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
251 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
257 /* offset within stripe */
258 kms = cl_offset(obj, idx) + to;
260 cl_object_attr_lock(obj);
261 CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n",
262 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
263 loi->loi_lvb.lvb_size);
265 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
266 valid = CAT_MTIME | CAT_CTIME;
267 if (kms > loi->loi_kms) {
271 if (kms > loi->loi_lvb.lvb_size) {
272 attr->cat_size = kms;
275 cl_object_attr_update(env, obj, attr, valid);
276 cl_object_attr_unlock(obj);
281 int osc_io_commit_async(const struct lu_env *env,
282 const struct cl_io_slice *ios,
283 struct cl_page_list *qin, int from, int to,
286 struct cl_io *io = ios->cis_io;
287 struct osc_io *oio = cl2osc_io(env, ios);
288 struct osc_object *osc = cl2osc(ios->cis_obj);
289 struct cl_page *page;
290 struct cl_page *last_page;
291 struct osc_page *opg;
295 LASSERT(qin->pl_nr > 0);
297 /* Handle partial page cases */
298 last_page = cl_page_list_last(qin);
299 if (oio->oi_lockless) {
300 page = cl_page_list_first(qin);
301 if (page == last_page) {
302 cl_page_clip(env, page, from, to);
305 cl_page_clip(env, page, from, PAGE_SIZE);
307 cl_page_clip(env, last_page, 0, to);
311 while (qin->pl_nr > 0) {
312 struct osc_async_page *oap;
314 page = cl_page_list_first(qin);
315 opg = osc_cl_page_osc(page, osc);
318 LASSERTF(osc == oap->oap_obj,
319 "obj mismatch: %p / %p\n", osc, oap->oap_obj);
321 if (!list_empty(&oap->oap_rpc_item)) {
322 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
328 /* The page may be already in dirty cache. */
329 if (list_empty(&oap->oap_pending_item)) {
330 result = osc_page_cache_add(env, &opg->ops_cl, io);
335 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
336 page == last_page ? to : PAGE_SIZE);
338 cl_page_list_del(env, qin, page);
340 (*cb)(env, io, page);
341 /* Can't access page any more. Page can be in transfer and
342 * complete at any time. */
345 /* for sync write, kernel will wait for this page to be flushed before
346 * osc_io_end() is called, so release it earlier.
347 * for mkwrite(), it's known there is no further pages. */
348 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
349 osc_extent_release(env, oio->oi_active);
350 oio->oi_active = NULL;
353 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
356 EXPORT_SYMBOL(osc_io_commit_async);
358 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios)
360 struct osc_object *osc = cl2osc(ios->cis_obj);
361 struct obd_import *imp = osc_cli(osc)->cl_import;
362 struct osc_io *oio = osc_env_io(env);
365 spin_lock(&imp->imp_lock);
366 if (likely(!imp->imp_invalid)) {
367 atomic_inc(&osc->oo_nr_ios);
368 oio->oi_is_active = 1;
371 spin_unlock(&imp->imp_lock);
373 if (cfs_capable(CFS_CAP_SYS_RESOURCE))
374 oio->oi_cap_sys_resource = 1;
378 EXPORT_SYMBOL(osc_io_iter_init);
380 int osc_io_rw_iter_init(const struct lu_env *env,
381 const struct cl_io_slice *ios)
383 struct cl_io *io = ios->cis_io;
384 struct osc_io *oio = osc_env_io(env);
385 struct osc_object *osc = cl2osc(ios->cis_obj);
386 unsigned long npages;
389 if (cl_io_is_append(io))
390 RETURN(osc_io_iter_init(env, ios));
392 npages = io->u.ci_rw.crw_count >> PAGE_SHIFT;
393 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
396 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
398 RETURN(osc_io_iter_init(env, ios));
400 EXPORT_SYMBOL(osc_io_rw_iter_init);
402 void osc_io_iter_fini(const struct lu_env *env,
403 const struct cl_io_slice *ios)
405 struct osc_io *oio = osc_env_io(env);
407 if (oio->oi_is_active) {
408 struct osc_object *osc = cl2osc(ios->cis_obj);
410 oio->oi_is_active = 0;
411 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
412 if (atomic_dec_and_test(&osc->oo_nr_ios))
413 wake_up_all(&osc->oo_io_waitq);
416 EXPORT_SYMBOL(osc_io_iter_fini);
418 void osc_io_rw_iter_fini(const struct lu_env *env,
419 const struct cl_io_slice *ios)
421 struct osc_io *oio = osc_env_io(env);
422 struct osc_object *osc = cl2osc(ios->cis_obj);
424 if (oio->oi_lru_reserved > 0) {
425 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
426 oio->oi_lru_reserved = 0;
428 oio->oi_write_osclock = NULL;
430 osc_io_iter_fini(env, ios);
432 EXPORT_SYMBOL(osc_io_rw_iter_fini);
434 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios)
437 struct cl_fault_io *fio;
441 fio = &io->u.ci_fault;
442 CDEBUG(D_INFO, "%lu %d %zu\n",
443 fio->ft_index, fio->ft_writable, fio->ft_nob);
445 * If mapping is writeable, adjust kms to cover this page,
446 * but do not extend kms beyond actual file size.
449 if (fio->ft_writable)
450 osc_page_touch_at(env, ios->cis_obj,
451 fio->ft_index, fio->ft_nob);
454 EXPORT_SYMBOL(osc_io_fault_start);
457 static int osc_async_upcall(void *a, int rc)
459 struct osc_async_cbargs *args = a;
462 complete(&args->opc_sync);
467 * Checks that there are no pages being written in the extent being truncated.
469 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
470 struct osc_page *ops , void *cbdata)
472 struct cl_page *page = ops->ops_cl.cpl_page;
473 struct osc_async_page *oap;
474 __u64 start = *(__u64 *)cbdata;
477 if (oap->oap_cmd & OBD_BRW_WRITE &&
478 !list_empty(&oap->oap_pending_item))
479 CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n",
480 start, current->comm);
482 if (PageLocked(page->cp_vmpage))
483 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
484 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
486 return CLP_GANG_OKAY;
489 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
490 struct osc_io *oio, __u64 size)
492 struct cl_object *clob;
496 clob = oio->oi_cl.cis_obj;
497 start = cl_index(clob, size);
498 partial = cl_offset(clob, start) < size;
501 * Complain if there are pages in the truncated region.
503 osc_page_gang_lookup(env, io, cl2osc(clob),
504 start + partial, CL_PAGE_EOF,
505 trunc_check_cb, (void *)&size);
508 static int osc_io_setattr_start(const struct lu_env *env,
509 const struct cl_io_slice *slice)
511 struct cl_io *io = slice->cis_io;
512 struct osc_io *oio = cl2osc_io(env, slice);
513 struct cl_object *obj = slice->cis_obj;
514 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
515 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
516 struct obdo *oa = &oio->oi_oa;
517 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
518 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
519 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
520 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
524 /* truncate cache dirty pages first */
525 if (cl_io_is_trunc(io))
526 result = osc_cache_truncate_start(env, cl2osc(obj), size,
529 if (result == 0 && oio->oi_lockless == 0) {
530 cl_object_attr_lock(obj);
531 result = cl_object_attr_get(env, obj, attr);
533 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
534 unsigned int cl_valid = 0;
536 if (ia_avalid & ATTR_SIZE) {
537 attr->cat_size = size;
538 attr->cat_kms = size;
539 cl_valid = (CAT_SIZE | CAT_KMS);
541 if (ia_avalid & ATTR_MTIME_SET) {
542 attr->cat_mtime = lvb->lvb_mtime;
543 cl_valid |= CAT_MTIME;
545 if (ia_avalid & ATTR_ATIME_SET) {
546 attr->cat_atime = lvb->lvb_atime;
547 cl_valid |= CAT_ATIME;
549 if (ia_xvalid & OP_XVALID_CTIME_SET) {
550 attr->cat_ctime = lvb->lvb_ctime;
551 cl_valid |= CAT_CTIME;
553 result = cl_object_attr_update(env, obj, attr,
556 cl_object_attr_unlock(obj);
558 memset(oa, 0, sizeof(*oa));
560 oa->o_oi = loi->loi_oi;
561 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
562 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
563 oa->o_layout = io->u.ci_setattr.sa_layout;
564 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
566 if (ia_avalid & ATTR_CTIME) {
567 oa->o_valid |= OBD_MD_FLCTIME;
568 oa->o_ctime = attr->cat_ctime;
570 if (ia_avalid & ATTR_ATIME) {
571 oa->o_valid |= OBD_MD_FLATIME;
572 oa->o_atime = attr->cat_atime;
574 if (ia_avalid & ATTR_MTIME) {
575 oa->o_valid |= OBD_MD_FLMTIME;
576 oa->o_mtime = attr->cat_mtime;
578 if (ia_avalid & ATTR_SIZE) {
580 oa->o_blocks = OBD_OBJECT_EOF;
581 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
583 if (oio->oi_lockless) {
584 oa->o_flags = OBD_FL_SRVLOCK;
585 oa->o_valid |= OBD_MD_FLFLAGS;
588 if (io->ci_layout_version > 0) {
589 /* verify layout version */
590 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
591 oa->o_layout_version = io->ci_layout_version;
594 LASSERT(oio->oi_lockless == 0);
597 if (ia_xvalid & OP_XVALID_FLAGS) {
598 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
599 oa->o_valid |= OBD_MD_FLFLAGS;
602 init_completion(&cbargs->opc_sync);
604 if (ia_avalid & ATTR_SIZE)
605 result = osc_punch_send(osc_export(cl2osc(obj)),
606 oa, osc_async_upcall, cbargs);
608 result = osc_setattr_async(osc_export(cl2osc(obj)),
609 oa, osc_async_upcall,
610 cbargs, PTLRPCD_SET);
612 cbargs->opc_rpc_sent = result == 0;
618 void osc_io_setattr_end(const struct lu_env *env,
619 const struct cl_io_slice *slice)
621 struct cl_io *io = slice->cis_io;
622 struct osc_io *oio = cl2osc_io(env, slice);
623 struct cl_object *obj = slice->cis_obj;
624 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
625 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
626 struct obdo *oa = &oio->oi_oa;
627 unsigned int cl_valid = 0;
630 if (cbargs->opc_rpc_sent) {
631 wait_for_completion(&cbargs->opc_sync);
632 result = io->ci_result = cbargs->opc_rc;
636 if (oio->oi_lockless) {
637 /* lockless truncate */
638 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
640 LASSERT(cl_io_is_trunc(io));
641 /* XXX: Need a lock. */
642 osd->od_stats.os_lockless_truncates++;
646 if (cl_io_is_trunc(io)) {
647 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
648 cl_object_attr_lock(obj);
649 if (oa->o_valid & OBD_MD_FLBLOCKS) {
650 attr->cat_blocks = oa->o_blocks;
651 cl_valid |= CAT_BLOCKS;
654 cl_object_attr_update(env, obj, attr, cl_valid);
655 cl_object_attr_unlock(obj);
656 osc_trunc_check(env, io, oio, size);
657 osc_cache_truncate_end(env, oio->oi_trunc);
658 oio->oi_trunc = NULL;
661 EXPORT_SYMBOL(osc_io_setattr_end);
663 struct osc_data_version_args {
664 struct osc_io *dva_oio;
668 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
671 struct osc_data_version_args *dva = args;
672 struct osc_io *oio = dva->dva_oio;
673 const struct ost_body *body;
679 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
681 GOTO(out, rc = -EPROTO);
683 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
687 oio->oi_cbarg.opc_rc = rc;
688 complete(&oio->oi_cbarg.opc_sync);
693 static int osc_io_data_version_start(const struct lu_env *env,
694 const struct cl_io_slice *slice)
696 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
697 struct osc_io *oio = cl2osc_io(env, slice);
698 struct obdo *oa = &oio->oi_oa;
699 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
700 struct osc_object *obj = cl2osc(slice->cis_obj);
701 struct lov_oinfo *loi = obj->oo_oinfo;
702 struct obd_export *exp = osc_export(obj);
703 struct ptlrpc_request *req;
704 struct ost_body *body;
705 struct osc_data_version_args *dva;
709 memset(oa, 0, sizeof(*oa));
710 oa->o_oi = loi->loi_oi;
711 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
713 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
714 oa->o_valid |= OBD_MD_FLFLAGS;
715 oa->o_flags |= OBD_FL_SRVLOCK;
716 if (dv->dv_flags & LL_DV_WR_FLUSH)
717 oa->o_flags |= OBD_FL_FLUSH;
720 init_completion(&cbargs->opc_sync);
722 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
726 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
728 ptlrpc_request_free(req);
732 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
733 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
735 ptlrpc_request_set_replen(req);
736 req->rq_interpret_reply = osc_data_version_interpret;
737 dva = ptlrpc_req_async_args(dva, req);
740 ptlrpcd_add_req(req);
745 static void osc_io_data_version_end(const struct lu_env *env,
746 const struct cl_io_slice *slice)
748 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
749 struct osc_io *oio = cl2osc_io(env, slice);
750 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
753 wait_for_completion(&cbargs->opc_sync);
755 if (cbargs->opc_rc != 0) {
756 slice->cis_io->ci_result = cbargs->opc_rc;
758 slice->cis_io->ci_result = 0;
759 if (!(oio->oi_oa.o_valid &
760 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
761 slice->cis_io->ci_result = -ENOTSUPP;
763 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
764 dv->dv_layout_version = oio->oi_oa.o_layout_version;
765 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
766 dv->dv_data_version = oio->oi_oa.o_data_version;
772 int osc_io_read_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 if (!slice->cis_io->ci_noatime) {
781 cl_object_attr_lock(obj);
782 attr->cat_atime = ktime_get_real_seconds();
783 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
784 cl_object_attr_unlock(obj);
789 EXPORT_SYMBOL(osc_io_read_start);
791 int osc_io_write_start(const struct lu_env *env,
792 const struct cl_io_slice *slice)
794 struct cl_object *obj = slice->cis_obj;
795 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
799 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
800 cl_object_attr_lock(obj);
801 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
802 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
803 cl_object_attr_unlock(obj);
807 EXPORT_SYMBOL(osc_io_write_start);
809 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
810 struct cl_fsync_io *fio)
812 struct osc_io *oio = osc_env_io(env);
813 struct obdo *oa = &oio->oi_oa;
814 struct lov_oinfo *loi = obj->oo_oinfo;
815 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
819 memset(oa, 0, sizeof(*oa));
820 oa->o_oi = loi->loi_oi;
821 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
823 /* reload size abd blocks for start and end of sync range */
824 oa->o_size = fio->fi_start;
825 oa->o_blocks = fio->fi_end;
826 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
828 obdo_set_parent_fid(oa, fio->fi_fid);
830 init_completion(&cbargs->opc_sync);
832 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
835 EXPORT_SYMBOL(osc_fsync_ost);
837 int osc_io_fsync_start(const struct lu_env *env,
838 const struct cl_io_slice *slice)
840 struct cl_io *io = slice->cis_io;
841 struct cl_fsync_io *fio = &io->u.ci_fsync;
842 struct cl_object *obj = slice->cis_obj;
843 struct osc_object *osc = cl2osc(obj);
844 pgoff_t start = cl_index(obj, fio->fi_start);
845 pgoff_t end = cl_index(obj, fio->fi_end);
849 if (fio->fi_end == OBD_OBJECT_EOF)
852 result = osc_cache_writeback_range(env, osc, start, end, 0,
853 fio->fi_mode == CL_FSYNC_DISCARD);
855 fio->fi_nr_written += result;
858 if (fio->fi_mode == CL_FSYNC_ALL) {
861 /* we have to wait for writeback to finish before we can
862 * send OST_SYNC RPC. This is bad because it causes extents
863 * to be written osc by osc. However, we usually start
864 * writeback before CL_FSYNC_ALL so this won't have any real
866 rc = osc_cache_wait_range(env, osc, start, end);
869 rc = osc_fsync_ost(env, osc, fio);
877 void osc_io_fsync_end(const struct lu_env *env,
878 const struct cl_io_slice *slice)
880 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
881 struct cl_object *obj = slice->cis_obj;
882 pgoff_t start = cl_index(obj, fio->fi_start);
883 pgoff_t end = cl_index(obj, fio->fi_end);
886 if (fio->fi_mode == CL_FSYNC_LOCAL) {
887 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
888 } else if (fio->fi_mode == CL_FSYNC_ALL) {
889 struct osc_io *oio = cl2osc_io(env, slice);
890 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
892 wait_for_completion(&cbargs->opc_sync);
894 result = cbargs->opc_rc;
896 slice->cis_io->ci_result = result;
898 EXPORT_SYMBOL(osc_io_fsync_end);
900 static int osc_io_ladvise_start(const struct lu_env *env,
901 const struct cl_io_slice *slice)
904 struct cl_io *io = slice->cis_io;
905 struct osc_io *oio = cl2osc_io(env, slice);
906 struct cl_object *obj = slice->cis_obj;
907 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
908 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
909 struct obdo *oa = &oio->oi_oa;
910 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
911 struct lu_ladvise *ladvise;
912 struct ladvise_hdr *ladvise_hdr;
917 /* TODO: add multiple ladvise support in CLIO */
918 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
919 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
920 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
922 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
923 if (ladvise_hdr == NULL)
926 memset(ladvise_hdr, 0, buf_size);
927 ladvise_hdr->lah_magic = LADVISE_MAGIC;
928 ladvise_hdr->lah_count = num_advise;
929 ladvise_hdr->lah_flags = lio->li_flags;
931 memset(oa, 0, sizeof(*oa));
932 oa->o_oi = loi->loi_oi;
933 oa->o_valid = OBD_MD_FLID;
934 obdo_set_parent_fid(oa, lio->li_fid);
936 ladvise = ladvise_hdr->lah_advise;
937 ladvise->lla_start = lio->li_start;
938 ladvise->lla_end = lio->li_end;
939 ladvise->lla_advice = lio->li_advice;
941 if (lio->li_flags & LF_ASYNC) {
942 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
943 ladvise_hdr, NULL, NULL, NULL);
945 init_completion(&cbargs->opc_sync);
946 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
947 ladvise_hdr, osc_async_upcall,
948 cbargs, PTLRPCD_SET);
949 cbargs->opc_rpc_sent = result == 0;
954 static void osc_io_ladvise_end(const struct lu_env *env,
955 const struct cl_io_slice *slice)
957 struct cl_io *io = slice->cis_io;
958 struct osc_io *oio = cl2osc_io(env, slice);
959 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
961 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
963 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
964 wait_for_completion(&cbargs->opc_sync);
965 result = cbargs->opc_rc;
967 slice->cis_io->ci_result = result;
970 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice)
972 struct osc_io *oio = cl2osc_io(env, slice);
974 if (oio->oi_active) {
975 osc_extent_release(env, oio->oi_active);
976 oio->oi_active = NULL;
979 EXPORT_SYMBOL(osc_io_end);
981 static const struct cl_io_operations osc_io_ops = {
984 .cio_iter_init = osc_io_rw_iter_init,
985 .cio_iter_fini = osc_io_rw_iter_fini,
986 .cio_start = osc_io_read_start,
987 .cio_fini = osc_io_fini
990 .cio_iter_init = osc_io_rw_iter_init,
991 .cio_iter_fini = osc_io_rw_iter_fini,
992 .cio_start = osc_io_write_start,
993 .cio_end = osc_io_end,
994 .cio_fini = osc_io_fini
997 .cio_iter_init = osc_io_iter_init,
998 .cio_iter_fini = osc_io_iter_fini,
999 .cio_start = osc_io_setattr_start,
1000 .cio_end = osc_io_setattr_end
1002 [CIT_DATA_VERSION] = {
1003 .cio_start = osc_io_data_version_start,
1004 .cio_end = osc_io_data_version_end,
1007 .cio_iter_init = osc_io_iter_init,
1008 .cio_iter_fini = osc_io_iter_fini,
1009 .cio_start = osc_io_fault_start,
1010 .cio_end = osc_io_end,
1011 .cio_fini = osc_io_fini
1014 .cio_start = osc_io_fsync_start,
1015 .cio_end = osc_io_fsync_end,
1016 .cio_fini = osc_io_fini
1019 .cio_start = osc_io_ladvise_start,
1020 .cio_end = osc_io_ladvise_end,
1021 .cio_fini = osc_io_fini
1024 .cio_fini = osc_io_fini
1027 .cio_read_ahead = osc_io_read_ahead,
1028 .cio_submit = osc_io_submit,
1029 .cio_commit_async = osc_io_commit_async
1032 /*****************************************************************************
1034 * Transfer operations.
1038 int osc_io_init(const struct lu_env *env,
1039 struct cl_object *obj, struct cl_io *io)
1041 struct osc_io *oio = osc_env_io(env);
1043 CL_IO_SLICE_CLEAN(oio, oi_cl);
1044 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);