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, 2016, 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 when a page is accessed within file in a way that creates
218 * new page, if one were missing (i.e., if there were a hole at that place in
219 * the file, or accessed page is beyond the current file size).
221 * Expand stripe KMS if necessary.
223 static void osc_page_touch_at(const struct lu_env *env,
224 struct cl_object *obj, pgoff_t idx, size_t to)
226 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
227 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
231 /* offset within stripe */
232 kms = cl_offset(obj, idx) + to;
234 cl_object_attr_lock(obj);
238 * ll_inode_size_lock(inode, 0); lov_stripe_lock(lsm);
242 CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n",
243 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
244 loi->loi_lvb.lvb_size);
246 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
247 valid = CAT_MTIME | CAT_CTIME;
248 if (kms > loi->loi_kms) {
252 if (kms > loi->loi_lvb.lvb_size) {
253 attr->cat_size = kms;
256 cl_object_attr_update(env, obj, attr, valid);
257 cl_object_attr_unlock(obj);
260 int osc_io_commit_async(const struct lu_env *env,
261 const struct cl_io_slice *ios,
262 struct cl_page_list *qin, int from, int to,
265 struct cl_io *io = ios->cis_io;
266 struct osc_io *oio = cl2osc_io(env, ios);
267 struct osc_object *osc = cl2osc(ios->cis_obj);
268 struct cl_page *page;
269 struct cl_page *last_page;
270 struct osc_page *opg;
274 LASSERT(qin->pl_nr > 0);
276 /* Handle partial page cases */
277 last_page = cl_page_list_last(qin);
278 if (oio->oi_lockless) {
279 page = cl_page_list_first(qin);
280 if (page == last_page) {
281 cl_page_clip(env, page, from, to);
284 cl_page_clip(env, page, from, PAGE_SIZE);
286 cl_page_clip(env, last_page, 0, to);
290 while (qin->pl_nr > 0) {
291 struct osc_async_page *oap;
293 page = cl_page_list_first(qin);
294 opg = osc_cl_page_osc(page, osc);
297 LASSERTF(osc == oap->oap_obj,
298 "obj mismatch: %p / %p\n", osc, oap->oap_obj);
300 if (!list_empty(&oap->oap_rpc_item)) {
301 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
307 /* The page may be already in dirty cache. */
308 if (list_empty(&oap->oap_pending_item)) {
309 result = osc_page_cache_add(env, &opg->ops_cl, io);
314 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
315 page == last_page ? to : PAGE_SIZE);
317 cl_page_list_del(env, qin, page);
319 (*cb)(env, io, page);
320 /* Can't access page any more. Page can be in transfer and
321 * complete at any time. */
324 /* for sync write, kernel will wait for this page to be flushed before
325 * osc_io_end() is called, so release it earlier.
326 * for mkwrite(), it's known there is no further pages. */
327 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
328 osc_extent_release(env, oio->oi_active);
329 oio->oi_active = NULL;
332 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
335 EXPORT_SYMBOL(osc_io_commit_async);
337 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios)
339 struct osc_object *osc = cl2osc(ios->cis_obj);
340 struct obd_import *imp = osc_cli(osc)->cl_import;
343 spin_lock(&imp->imp_lock);
344 if (likely(!imp->imp_invalid)) {
345 struct osc_io *oio = osc_env_io(env);
347 atomic_inc(&osc->oo_nr_ios);
348 oio->oi_is_active = 1;
351 spin_unlock(&imp->imp_lock);
355 EXPORT_SYMBOL(osc_io_iter_init);
357 int osc_io_write_iter_init(const struct lu_env *env,
358 const struct cl_io_slice *ios)
360 struct cl_io *io = ios->cis_io;
361 struct osc_io *oio = osc_env_io(env);
362 struct osc_object *osc = cl2osc(ios->cis_obj);
363 unsigned long npages;
366 if (cl_io_is_append(io))
367 RETURN(osc_io_iter_init(env, ios));
369 npages = io->u.ci_rw.rw_range.cir_count >> PAGE_SHIFT;
370 if (io->u.ci_rw.rw_range.cir_pos & ~PAGE_MASK)
373 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
375 RETURN(osc_io_iter_init(env, ios));
377 EXPORT_SYMBOL(osc_io_write_iter_init);
379 void osc_io_iter_fini(const struct lu_env *env,
380 const struct cl_io_slice *ios)
382 struct osc_io *oio = osc_env_io(env);
384 if (oio->oi_is_active) {
385 struct osc_object *osc = cl2osc(ios->cis_obj);
387 oio->oi_is_active = 0;
388 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
389 if (atomic_dec_and_test(&osc->oo_nr_ios))
390 wake_up_all(&osc->oo_io_waitq);
393 EXPORT_SYMBOL(osc_io_iter_fini);
395 void osc_io_write_iter_fini(const struct lu_env *env,
396 const struct cl_io_slice *ios)
398 struct osc_io *oio = osc_env_io(env);
399 struct osc_object *osc = cl2osc(ios->cis_obj);
401 if (oio->oi_lru_reserved > 0) {
402 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
403 oio->oi_lru_reserved = 0;
405 oio->oi_write_osclock = NULL;
407 osc_io_iter_fini(env, ios);
409 EXPORT_SYMBOL(osc_io_write_iter_fini);
411 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios)
414 struct cl_fault_io *fio;
418 fio = &io->u.ci_fault;
419 CDEBUG(D_INFO, "%lu %d %zu\n",
420 fio->ft_index, fio->ft_writable, fio->ft_nob);
422 * If mapping is writeable, adjust kms to cover this page,
423 * but do not extend kms beyond actual file size.
426 if (fio->ft_writable)
427 osc_page_touch_at(env, ios->cis_obj,
428 fio->ft_index, fio->ft_nob);
431 EXPORT_SYMBOL(osc_io_fault_start);
434 static int osc_async_upcall(void *a, int rc)
436 struct osc_async_cbargs *args = a;
439 complete(&args->opc_sync);
444 * Checks that there are no pages being written in the extent being truncated.
446 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
447 struct osc_page *ops , void *cbdata)
449 struct cl_page *page = ops->ops_cl.cpl_page;
450 struct osc_async_page *oap;
451 __u64 start = *(__u64 *)cbdata;
454 if (oap->oap_cmd & OBD_BRW_WRITE &&
455 !list_empty(&oap->oap_pending_item))
456 CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n",
457 start, current->comm);
459 if (PageLocked(page->cp_vmpage))
460 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
461 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
463 return CLP_GANG_OKAY;
466 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
467 struct osc_io *oio, __u64 size)
469 struct cl_object *clob;
473 clob = oio->oi_cl.cis_obj;
474 start = cl_index(clob, size);
475 partial = cl_offset(clob, start) < size;
478 * Complain if there are pages in the truncated region.
480 osc_page_gang_lookup(env, io, cl2osc(clob),
481 start + partial, CL_PAGE_EOF,
482 trunc_check_cb, (void *)&size);
485 static int osc_io_setattr_start(const struct lu_env *env,
486 const struct cl_io_slice *slice)
488 struct cl_io *io = slice->cis_io;
489 struct osc_io *oio = cl2osc_io(env, slice);
490 struct cl_object *obj = slice->cis_obj;
491 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
492 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
493 struct obdo *oa = &oio->oi_oa;
494 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
495 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
496 unsigned int ia_valid = io->u.ci_setattr.sa_valid;
500 /* truncate cache dirty pages first */
501 if (cl_io_is_trunc(io))
502 result = osc_cache_truncate_start(env, cl2osc(obj), size,
505 if (result == 0 && oio->oi_lockless == 0) {
506 cl_object_attr_lock(obj);
507 result = cl_object_attr_get(env, obj, attr);
509 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
510 unsigned int cl_valid = 0;
512 if (ia_valid & ATTR_SIZE) {
513 attr->cat_size = attr->cat_kms = size;
514 cl_valid = (CAT_SIZE | CAT_KMS);
516 if (ia_valid & ATTR_MTIME_SET) {
517 attr->cat_mtime = lvb->lvb_mtime;
518 cl_valid |= CAT_MTIME;
520 if (ia_valid & ATTR_ATIME_SET) {
521 attr->cat_atime = lvb->lvb_atime;
522 cl_valid |= CAT_ATIME;
524 if (ia_valid & ATTR_CTIME_SET) {
525 attr->cat_ctime = lvb->lvb_ctime;
526 cl_valid |= CAT_CTIME;
528 result = cl_object_attr_update(env, obj, attr,
531 cl_object_attr_unlock(obj);
533 memset(oa, 0, sizeof(*oa));
535 oa->o_oi = loi->loi_oi;
536 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
537 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
538 oa->o_layout = io->u.ci_setattr.sa_layout;
539 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
541 if (ia_valid & ATTR_CTIME) {
542 oa->o_valid |= OBD_MD_FLCTIME;
543 oa->o_ctime = attr->cat_ctime;
545 if (ia_valid & ATTR_ATIME) {
546 oa->o_valid |= OBD_MD_FLATIME;
547 oa->o_atime = attr->cat_atime;
549 if (ia_valid & ATTR_MTIME) {
550 oa->o_valid |= OBD_MD_FLMTIME;
551 oa->o_mtime = attr->cat_mtime;
553 if (ia_valid & ATTR_SIZE) {
555 oa->o_blocks = OBD_OBJECT_EOF;
556 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
558 if (oio->oi_lockless) {
559 oa->o_flags = OBD_FL_SRVLOCK;
560 oa->o_valid |= OBD_MD_FLFLAGS;
563 if (io->ci_layout_version > 0) {
564 /* verify layout version */
565 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
566 oa->o_layout_version = io->ci_layout_version;
569 LASSERT(oio->oi_lockless == 0);
572 if (ia_valid & ATTR_ATTR_FLAG) {
573 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
574 oa->o_valid |= OBD_MD_FLFLAGS;
577 init_completion(&cbargs->opc_sync);
579 if (ia_valid & ATTR_SIZE)
580 result = osc_punch_send(osc_export(cl2osc(obj)),
581 oa, osc_async_upcall, cbargs);
583 result = osc_setattr_async(osc_export(cl2osc(obj)),
584 oa, osc_async_upcall,
585 cbargs, PTLRPCD_SET);
587 cbargs->opc_rpc_sent = result == 0;
593 void osc_io_setattr_end(const struct lu_env *env,
594 const struct cl_io_slice *slice)
596 struct cl_io *io = slice->cis_io;
597 struct osc_io *oio = cl2osc_io(env, slice);
598 struct cl_object *obj = slice->cis_obj;
599 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
602 if (cbargs->opc_rpc_sent) {
603 wait_for_completion(&cbargs->opc_sync);
604 result = io->ci_result = cbargs->opc_rc;
607 if (oio->oi_lockless) {
608 /* lockless truncate */
609 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
611 LASSERT(cl_io_is_trunc(io));
612 /* XXX: Need a lock. */
613 osd->od_stats.os_lockless_truncates++;
617 if (cl_io_is_trunc(io)) {
618 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
619 osc_trunc_check(env, io, oio, size);
620 osc_cache_truncate_end(env, oio->oi_trunc);
621 oio->oi_trunc = NULL;
624 EXPORT_SYMBOL(osc_io_setattr_end);
626 struct osc_data_version_args {
627 struct osc_io *dva_oio;
631 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
634 struct osc_data_version_args *dva = arg;
635 struct osc_io *oio = dva->dva_oio;
636 const struct ost_body *body;
642 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
644 GOTO(out, rc = -EPROTO);
646 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
650 oio->oi_cbarg.opc_rc = rc;
651 complete(&oio->oi_cbarg.opc_sync);
656 static int osc_io_data_version_start(const struct lu_env *env,
657 const struct cl_io_slice *slice)
659 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
660 struct osc_io *oio = cl2osc_io(env, slice);
661 struct obdo *oa = &oio->oi_oa;
662 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
663 struct osc_object *obj = cl2osc(slice->cis_obj);
664 struct lov_oinfo *loi = obj->oo_oinfo;
665 struct obd_export *exp = osc_export(obj);
666 struct ptlrpc_request *req;
667 struct ost_body *body;
668 struct osc_data_version_args *dva;
672 memset(oa, 0, sizeof(*oa));
673 oa->o_oi = loi->loi_oi;
674 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
676 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
677 oa->o_valid |= OBD_MD_FLFLAGS;
678 oa->o_flags |= OBD_FL_SRVLOCK;
679 if (dv->dv_flags & LL_DV_WR_FLUSH)
680 oa->o_flags |= OBD_FL_FLUSH;
683 init_completion(&cbargs->opc_sync);
685 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
689 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
691 ptlrpc_request_free(req);
695 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
696 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
698 ptlrpc_request_set_replen(req);
699 req->rq_interpret_reply = osc_data_version_interpret;
700 CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
701 dva = ptlrpc_req_async_args(req);
704 ptlrpcd_add_req(req);
709 static void osc_io_data_version_end(const struct lu_env *env,
710 const struct cl_io_slice *slice)
712 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
713 struct osc_io *oio = cl2osc_io(env, slice);
714 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
717 wait_for_completion(&cbargs->opc_sync);
719 if (cbargs->opc_rc != 0) {
720 slice->cis_io->ci_result = cbargs->opc_rc;
722 slice->cis_io->ci_result = 0;
723 if (!(oio->oi_oa.o_valid &
724 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
725 slice->cis_io->ci_result = -ENOTSUPP;
727 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
728 dv->dv_layout_version = oio->oi_oa.o_layout_version;
729 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
730 dv->dv_data_version = oio->oi_oa.o_data_version;
736 int osc_io_read_start(const struct lu_env *env,
737 const struct cl_io_slice *slice)
739 struct cl_object *obj = slice->cis_obj;
740 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
744 if (!slice->cis_io->ci_noatime) {
745 cl_object_attr_lock(obj);
746 attr->cat_atime = ktime_get_real_seconds();
747 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
748 cl_object_attr_unlock(obj);
753 EXPORT_SYMBOL(osc_io_read_start);
755 int osc_io_write_start(const struct lu_env *env,
756 const struct cl_io_slice *slice)
758 struct cl_object *obj = slice->cis_obj;
759 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
763 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
764 cl_object_attr_lock(obj);
765 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
766 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
767 cl_object_attr_unlock(obj);
771 EXPORT_SYMBOL(osc_io_write_start);
773 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
774 struct cl_fsync_io *fio)
776 struct osc_io *oio = osc_env_io(env);
777 struct obdo *oa = &oio->oi_oa;
778 struct lov_oinfo *loi = obj->oo_oinfo;
779 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
783 memset(oa, 0, sizeof(*oa));
784 oa->o_oi = loi->loi_oi;
785 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
787 /* reload size abd blocks for start and end of sync range */
788 oa->o_size = fio->fi_start;
789 oa->o_blocks = fio->fi_end;
790 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
792 obdo_set_parent_fid(oa, fio->fi_fid);
794 init_completion(&cbargs->opc_sync);
796 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
799 EXPORT_SYMBOL(osc_fsync_ost);
801 int osc_io_fsync_start(const struct lu_env *env,
802 const struct cl_io_slice *slice)
804 struct cl_io *io = slice->cis_io;
805 struct cl_fsync_io *fio = &io->u.ci_fsync;
806 struct cl_object *obj = slice->cis_obj;
807 struct osc_object *osc = cl2osc(obj);
808 pgoff_t start = cl_index(obj, fio->fi_start);
809 pgoff_t end = cl_index(obj, fio->fi_end);
813 if (fio->fi_end == OBD_OBJECT_EOF)
816 result = osc_cache_writeback_range(env, osc, start, end, 0,
817 fio->fi_mode == CL_FSYNC_DISCARD);
819 fio->fi_nr_written += result;
822 if (fio->fi_mode == CL_FSYNC_ALL) {
825 /* we have to wait for writeback to finish before we can
826 * send OST_SYNC RPC. This is bad because it causes extents
827 * to be written osc by osc. However, we usually start
828 * writeback before CL_FSYNC_ALL so this won't have any real
830 rc = osc_cache_wait_range(env, osc, start, end);
833 rc = osc_fsync_ost(env, osc, fio);
841 void osc_io_fsync_end(const struct lu_env *env,
842 const struct cl_io_slice *slice)
844 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
845 struct cl_object *obj = slice->cis_obj;
846 pgoff_t start = cl_index(obj, fio->fi_start);
847 pgoff_t end = cl_index(obj, fio->fi_end);
850 if (fio->fi_mode == CL_FSYNC_LOCAL) {
851 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
852 } else if (fio->fi_mode == CL_FSYNC_ALL) {
853 struct osc_io *oio = cl2osc_io(env, slice);
854 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
856 wait_for_completion(&cbargs->opc_sync);
858 result = cbargs->opc_rc;
860 slice->cis_io->ci_result = result;
862 EXPORT_SYMBOL(osc_io_fsync_end);
864 static int osc_io_ladvise_start(const struct lu_env *env,
865 const struct cl_io_slice *slice)
868 struct cl_io *io = slice->cis_io;
869 struct osc_io *oio = cl2osc_io(env, slice);
870 struct cl_object *obj = slice->cis_obj;
871 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
872 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
873 struct obdo *oa = &oio->oi_oa;
874 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
875 struct lu_ladvise *ladvise;
876 struct ladvise_hdr *ladvise_hdr;
881 /* TODO: add multiple ladvise support in CLIO */
882 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
883 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
884 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
886 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
887 if (ladvise_hdr == NULL)
890 memset(ladvise_hdr, 0, buf_size);
891 ladvise_hdr->lah_magic = LADVISE_MAGIC;
892 ladvise_hdr->lah_count = num_advise;
893 ladvise_hdr->lah_flags = lio->li_flags;
895 memset(oa, 0, sizeof(*oa));
896 oa->o_oi = loi->loi_oi;
897 oa->o_valid = OBD_MD_FLID;
898 obdo_set_parent_fid(oa, lio->li_fid);
900 ladvise = ladvise_hdr->lah_advise;
901 ladvise->lla_start = lio->li_start;
902 ladvise->lla_end = lio->li_end;
903 ladvise->lla_advice = lio->li_advice;
905 if (lio->li_flags & LF_ASYNC) {
906 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
907 ladvise_hdr, NULL, NULL, NULL);
909 init_completion(&cbargs->opc_sync);
910 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
911 ladvise_hdr, osc_async_upcall,
912 cbargs, PTLRPCD_SET);
913 cbargs->opc_rpc_sent = result == 0;
918 static void osc_io_ladvise_end(const struct lu_env *env,
919 const struct cl_io_slice *slice)
921 struct cl_io *io = slice->cis_io;
922 struct osc_io *oio = cl2osc_io(env, slice);
923 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
925 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
927 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
928 wait_for_completion(&cbargs->opc_sync);
929 result = cbargs->opc_rc;
931 slice->cis_io->ci_result = result;
934 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice)
936 struct osc_io *oio = cl2osc_io(env, slice);
938 if (oio->oi_active) {
939 osc_extent_release(env, oio->oi_active);
940 oio->oi_active = NULL;
943 EXPORT_SYMBOL(osc_io_end);
945 static const struct cl_io_operations osc_io_ops = {
948 .cio_iter_init = osc_io_iter_init,
949 .cio_iter_fini = osc_io_iter_fini,
950 .cio_start = osc_io_read_start,
951 .cio_fini = osc_io_fini
954 .cio_iter_init = osc_io_write_iter_init,
955 .cio_iter_fini = osc_io_write_iter_fini,
956 .cio_start = osc_io_write_start,
957 .cio_end = osc_io_end,
958 .cio_fini = osc_io_fini
961 .cio_iter_init = osc_io_iter_init,
962 .cio_iter_fini = osc_io_iter_fini,
963 .cio_start = osc_io_setattr_start,
964 .cio_end = osc_io_setattr_end
966 [CIT_DATA_VERSION] = {
967 .cio_start = osc_io_data_version_start,
968 .cio_end = osc_io_data_version_end,
971 .cio_iter_init = osc_io_iter_init,
972 .cio_iter_fini = osc_io_iter_fini,
973 .cio_start = osc_io_fault_start,
974 .cio_end = osc_io_end,
975 .cio_fini = osc_io_fini
978 .cio_start = osc_io_fsync_start,
979 .cio_end = osc_io_fsync_end,
980 .cio_fini = osc_io_fini
983 .cio_start = osc_io_ladvise_start,
984 .cio_end = osc_io_ladvise_end,
985 .cio_fini = osc_io_fini
988 .cio_fini = osc_io_fini
991 .cio_read_ahead = osc_io_read_ahead,
992 .cio_submit = osc_io_submit,
993 .cio_commit_async = osc_io_commit_async
996 /*****************************************************************************
998 * Transfer operations.
1002 int osc_io_init(const struct lu_env *env,
1003 struct cl_object *obj, struct cl_io *io)
1005 struct osc_io *oio = osc_env_io(env);
1007 CL_IO_SLICE_CLEAN(oio, oi_cl);
1008 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);