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;
147 * NOTE: here @page is a top-level page. This is done to avoid
148 * creation of sub-page-list.
150 cl_page_list_for_each_safe(page, tmp, qin) {
151 struct osc_async_page *oap;
157 opg = osc_cl_page_osc(page, osc);
159 LASSERT(osc == oap->oap_obj);
161 if (!list_empty(&oap->oap_pending_item) ||
162 !list_empty(&oap->oap_rpc_item)) {
163 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
169 result = cl_page_prep(env, io, page, crt);
172 if (result != -EALREADY)
175 * Handle -EALREADY error: for read case, the page is
176 * already in UPTODATE state; for write, the page
183 spin_lock(&oap->oap_lock);
184 oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
185 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
186 spin_unlock(&oap->oap_lock);
188 osc_page_submit(env, opg, crt, brw_flags);
189 list_add_tail(&oap->oap_pending_item, &list);
191 if (page->cp_sync_io != NULL)
192 cl_page_list_move(qout, qin, page);
194 cl_page_list_del(env, qin, page);
197 if (queued == max_pages) {
199 } else if (crt == CRT_WRITE) {
201 unsigned int next_chunks;
203 chunks = (queued + ppc - 1) >> ppc_bits;
204 /* chunk number if add another page */
205 next_chunks = (queued + ppc) >> ppc_bits;
207 /* next page will excceed write chunk limit */
208 if (chunks == osc_max_write_chunks(cli) &&
209 next_chunks > chunks)
214 result = osc_queue_sync_pages(env, io, osc, &list,
224 result = osc_queue_sync_pages(env, io, osc, &list, brw_flags);
226 /* Update c/mtime for sync write. LU-7310 */
227 if (crt == CRT_WRITE && qout->pl_nr > 0 && result == 0) {
228 struct cl_object *obj = ios->cis_obj;
229 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
231 cl_object_attr_lock(obj);
232 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
233 cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
234 cl_object_attr_unlock(obj);
237 CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
238 return qout->pl_nr > 0 ? 0 : result;
240 EXPORT_SYMBOL(osc_io_submit);
243 * This is called to update the attributes when modifying a specific page,
244 * both when making new pages and when doing updates to existing cached pages.
246 * Expand stripe KMS if necessary.
248 void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj,
249 pgoff_t idx, size_t to)
251 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
252 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
258 /* offset within stripe */
259 kms = cl_offset(obj, idx) + to;
261 cl_object_attr_lock(obj);
262 CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n",
263 kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms,
264 loi->loi_lvb.lvb_size);
266 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
267 valid = CAT_MTIME | CAT_CTIME;
268 if (kms > loi->loi_kms) {
272 if (kms > loi->loi_lvb.lvb_size) {
273 attr->cat_size = kms;
276 cl_object_attr_update(env, obj, attr, valid);
277 cl_object_attr_unlock(obj);
282 int osc_io_commit_async(const struct lu_env *env,
283 const struct cl_io_slice *ios,
284 struct cl_page_list *qin, int from, int to,
287 struct cl_io *io = ios->cis_io;
288 struct osc_io *oio = cl2osc_io(env, ios);
289 struct osc_object *osc = cl2osc(ios->cis_obj);
290 struct cl_page *page;
291 struct cl_page *last_page;
292 struct osc_page *opg;
293 struct pagevec *pvec = &osc_env_info(env)->oti_pagevec;
297 LASSERT(qin->pl_nr > 0);
299 /* Handle partial page cases */
300 last_page = cl_page_list_last(qin);
301 if (oio->oi_lockless) {
302 page = cl_page_list_first(qin);
303 if (page == last_page) {
304 cl_page_clip(env, page, from, to);
307 cl_page_clip(env, page, from, PAGE_SIZE);
309 cl_page_clip(env, last_page, 0, to);
313 ll_pagevec_init(pvec, 0);
315 while (qin->pl_nr > 0) {
316 struct osc_async_page *oap;
318 page = cl_page_list_first(qin);
319 opg = osc_cl_page_osc(page, osc);
322 LASSERTF(osc == oap->oap_obj,
323 "obj mismatch: %p / %p\n", osc, oap->oap_obj);
325 if (!list_empty(&oap->oap_rpc_item)) {
326 CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
332 /* The page may be already in dirty cache. */
333 if (list_empty(&oap->oap_pending_item)) {
334 result = osc_page_cache_add(env, opg, io, cb);
339 osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
340 page == last_page ? to : PAGE_SIZE);
342 cl_page_list_del(env, qin, page);
344 /* if there are no more slots, do the callback & reinit */
345 if (pagevec_add(pvec, page->cp_vmpage) == 0) {
346 (*cb)(env, io, pvec);
347 pagevec_reinit(pvec);
351 /* Clean up any partially full pagevecs */
352 if (pagevec_count(pvec) != 0)
353 (*cb)(env, io, pvec);
355 /* Can't access these pages any more. Page can be in transfer and
356 * complete at any time. */
358 /* for sync write, kernel will wait for this page to be flushed before
359 * osc_io_end() is called, so release it earlier.
360 * for mkwrite(), it's known there is no further pages. */
361 if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
362 osc_extent_release(env, oio->oi_active);
363 oio->oi_active = NULL;
366 CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
369 EXPORT_SYMBOL(osc_io_commit_async);
371 static bool osc_import_not_healthy(struct obd_import *imp)
373 return imp->imp_invalid || imp->imp_deactive ||
374 !(imp->imp_state == LUSTRE_IMP_FULL ||
375 imp->imp_state == LUSTRE_IMP_IDLE);
378 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios)
380 struct osc_object *osc = cl2osc(ios->cis_obj);
381 struct obd_import *imp = osc_cli(osc)->cl_import;
382 struct osc_io *oio = osc_env_io(env);
386 spin_lock(&imp->imp_lock);
388 * check whether this OSC device is available for non-delay read,
389 * fast switching mirror if we haven't tried all mirrors.
391 if (ios->cis_io->ci_type == CIT_READ && ios->cis_io->ci_ndelay &&
392 !ios->cis_io->ci_tried_all_mirrors && osc_import_not_healthy(imp)) {
394 } else if (likely(!imp->imp_invalid)) {
395 atomic_inc(&osc->oo_nr_ios);
396 oio->oi_is_active = 1;
399 spin_unlock(&imp->imp_lock);
401 if (cfs_capable(CFS_CAP_SYS_RESOURCE))
402 oio->oi_cap_sys_resource = 1;
406 EXPORT_SYMBOL(osc_io_iter_init);
408 int osc_io_rw_iter_init(const struct lu_env *env,
409 const struct cl_io_slice *ios)
411 struct cl_io *io = ios->cis_io;
412 struct osc_io *oio = osc_env_io(env);
413 struct osc_object *osc = cl2osc(ios->cis_obj);
414 unsigned long npages;
417 if (cl_io_is_append(io))
418 RETURN(osc_io_iter_init(env, ios));
420 npages = io->u.ci_rw.crw_count >> PAGE_SHIFT;
421 if (io->u.ci_rw.crw_pos & ~PAGE_MASK)
424 oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages);
426 RETURN(osc_io_iter_init(env, ios));
428 EXPORT_SYMBOL(osc_io_rw_iter_init);
430 void osc_io_iter_fini(const struct lu_env *env,
431 const struct cl_io_slice *ios)
433 struct osc_io *oio = osc_env_io(env);
435 if (oio->oi_is_active) {
436 struct osc_object *osc = cl2osc(ios->cis_obj);
438 oio->oi_is_active = 0;
439 LASSERT(atomic_read(&osc->oo_nr_ios) > 0);
440 if (atomic_dec_and_test(&osc->oo_nr_ios))
441 wake_up_all(&osc->oo_io_waitq);
444 EXPORT_SYMBOL(osc_io_iter_fini);
446 void osc_io_rw_iter_fini(const struct lu_env *env,
447 const struct cl_io_slice *ios)
449 struct osc_io *oio = osc_env_io(env);
450 struct osc_object *osc = cl2osc(ios->cis_obj);
452 if (oio->oi_lru_reserved > 0) {
453 osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved);
454 oio->oi_lru_reserved = 0;
456 oio->oi_write_osclock = NULL;
458 osc_io_iter_fini(env, ios);
460 EXPORT_SYMBOL(osc_io_rw_iter_fini);
462 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios)
465 struct cl_fault_io *fio;
469 fio = &io->u.ci_fault;
470 CDEBUG(D_INFO, "%lu %d %zu\n",
471 fio->ft_index, fio->ft_writable, fio->ft_nob);
473 * If mapping is writeable, adjust kms to cover this page,
474 * but do not extend kms beyond actual file size.
477 if (fio->ft_writable)
478 osc_page_touch_at(env, ios->cis_obj,
479 fio->ft_index, fio->ft_nob);
482 EXPORT_SYMBOL(osc_io_fault_start);
485 static int osc_async_upcall(void *a, int rc)
487 struct osc_async_cbargs *args = a;
490 complete(&args->opc_sync);
495 * Checks that there are no pages being written in the extent being truncated.
497 static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
498 struct osc_page *ops , void *cbdata)
500 struct cl_page *page = ops->ops_cl.cpl_page;
501 struct osc_async_page *oap;
502 __u64 start = *(__u64 *)cbdata;
505 if (oap->oap_cmd & OBD_BRW_WRITE &&
506 !list_empty(&oap->oap_pending_item))
507 CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n",
508 start, current->comm);
510 if (PageLocked(page->cp_vmpage))
511 CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
512 ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK);
514 return CLP_GANG_OKAY;
517 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
518 struct osc_io *oio, __u64 size)
520 struct cl_object *clob;
524 clob = oio->oi_cl.cis_obj;
525 start = cl_index(clob, size);
526 partial = cl_offset(clob, start) < size;
529 * Complain if there are pages in the truncated region.
531 osc_page_gang_lookup(env, io, cl2osc(clob),
532 start + partial, CL_PAGE_EOF,
533 trunc_check_cb, (void *)&size);
536 static int osc_io_setattr_start(const struct lu_env *env,
537 const struct cl_io_slice *slice)
539 struct cl_io *io = slice->cis_io;
540 struct osc_io *oio = cl2osc_io(env, slice);
541 struct cl_object *obj = slice->cis_obj;
542 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
543 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
544 struct obdo *oa = &oio->oi_oa;
545 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
546 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
547 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
548 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
552 /* truncate cache dirty pages first */
553 if (cl_io_is_trunc(io))
554 result = osc_cache_truncate_start(env, cl2osc(obj), size,
557 if (result == 0 && oio->oi_lockless == 0) {
558 cl_object_attr_lock(obj);
559 result = cl_object_attr_get(env, obj, attr);
561 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
562 unsigned int cl_valid = 0;
564 if (ia_avalid & ATTR_SIZE) {
565 attr->cat_size = size;
566 attr->cat_kms = size;
567 cl_valid = (CAT_SIZE | CAT_KMS);
569 if (ia_avalid & ATTR_MTIME_SET) {
570 attr->cat_mtime = lvb->lvb_mtime;
571 cl_valid |= CAT_MTIME;
573 if (ia_avalid & ATTR_ATIME_SET) {
574 attr->cat_atime = lvb->lvb_atime;
575 cl_valid |= CAT_ATIME;
577 if (ia_xvalid & OP_XVALID_CTIME_SET) {
578 attr->cat_ctime = lvb->lvb_ctime;
579 cl_valid |= CAT_CTIME;
581 result = cl_object_attr_update(env, obj, attr,
584 cl_object_attr_unlock(obj);
586 memset(oa, 0, sizeof(*oa));
588 oa->o_oi = loi->loi_oi;
589 obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid);
590 oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index;
591 oa->o_layout = io->u.ci_setattr.sa_layout;
592 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
594 if (ia_avalid & ATTR_CTIME) {
595 oa->o_valid |= OBD_MD_FLCTIME;
596 oa->o_ctime = attr->cat_ctime;
598 if (ia_avalid & ATTR_ATIME) {
599 oa->o_valid |= OBD_MD_FLATIME;
600 oa->o_atime = attr->cat_atime;
602 if (ia_avalid & ATTR_MTIME) {
603 oa->o_valid |= OBD_MD_FLMTIME;
604 oa->o_mtime = attr->cat_mtime;
606 if (ia_avalid & ATTR_SIZE) {
608 oa->o_blocks = OBD_OBJECT_EOF;
609 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
611 if (oio->oi_lockless) {
612 oa->o_flags = OBD_FL_SRVLOCK;
613 oa->o_valid |= OBD_MD_FLFLAGS;
616 if (io->ci_layout_version > 0) {
617 /* verify layout version */
618 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
619 oa->o_layout_version = io->ci_layout_version;
622 LASSERT(oio->oi_lockless == 0);
625 if (ia_xvalid & OP_XVALID_FLAGS) {
626 oa->o_flags = io->u.ci_setattr.sa_attr_flags;
627 oa->o_valid |= OBD_MD_FLFLAGS;
630 init_completion(&cbargs->opc_sync);
632 if (ia_avalid & ATTR_SIZE)
633 result = osc_punch_send(osc_export(cl2osc(obj)),
634 oa, osc_async_upcall, cbargs);
636 result = osc_setattr_async(osc_export(cl2osc(obj)),
637 oa, osc_async_upcall,
638 cbargs, PTLRPCD_SET);
640 cbargs->opc_rpc_sent = result == 0;
646 void osc_io_setattr_end(const struct lu_env *env,
647 const struct cl_io_slice *slice)
649 struct cl_io *io = slice->cis_io;
650 struct osc_io *oio = cl2osc_io(env, slice);
651 struct cl_object *obj = slice->cis_obj;
652 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
653 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
654 struct obdo *oa = &oio->oi_oa;
655 unsigned int cl_valid = 0;
658 if (cbargs->opc_rpc_sent) {
659 wait_for_completion(&cbargs->opc_sync);
660 result = io->ci_result = cbargs->opc_rc;
664 if (oio->oi_lockless) {
665 /* lockless truncate */
666 struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
668 LASSERT(cl_io_is_trunc(io));
669 /* XXX: Need a lock. */
670 osd->od_stats.os_lockless_truncates++;
674 if (cl_io_is_trunc(io)) {
675 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
676 cl_object_attr_lock(obj);
677 if (oa->o_valid & OBD_MD_FLBLOCKS) {
678 attr->cat_blocks = oa->o_blocks;
679 cl_valid |= CAT_BLOCKS;
682 cl_object_attr_update(env, obj, attr, cl_valid);
683 cl_object_attr_unlock(obj);
684 osc_trunc_check(env, io, oio, size);
685 osc_cache_truncate_end(env, oio->oi_trunc);
686 oio->oi_trunc = NULL;
689 EXPORT_SYMBOL(osc_io_setattr_end);
691 struct osc_data_version_args {
692 struct osc_io *dva_oio;
696 osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
699 struct osc_data_version_args *dva = args;
700 struct osc_io *oio = dva->dva_oio;
701 const struct ost_body *body;
707 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
709 GOTO(out, rc = -EPROTO);
711 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa,
715 oio->oi_cbarg.opc_rc = rc;
716 complete(&oio->oi_cbarg.opc_sync);
721 static int osc_io_data_version_start(const struct lu_env *env,
722 const struct cl_io_slice *slice)
724 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
725 struct osc_io *oio = cl2osc_io(env, slice);
726 struct obdo *oa = &oio->oi_oa;
727 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
728 struct osc_object *obj = cl2osc(slice->cis_obj);
729 struct lov_oinfo *loi = obj->oo_oinfo;
730 struct obd_export *exp = osc_export(obj);
731 struct ptlrpc_request *req;
732 struct ost_body *body;
733 struct osc_data_version_args *dva;
737 memset(oa, 0, sizeof(*oa));
738 oa->o_oi = loi->loi_oi;
739 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
741 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
742 oa->o_valid |= OBD_MD_FLFLAGS;
743 oa->o_flags |= OBD_FL_SRVLOCK;
744 if (dv->dv_flags & LL_DV_WR_FLUSH)
745 oa->o_flags |= OBD_FL_FLUSH;
748 init_completion(&cbargs->opc_sync);
750 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
754 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
756 ptlrpc_request_free(req);
760 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
761 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
763 ptlrpc_request_set_replen(req);
764 req->rq_interpret_reply = osc_data_version_interpret;
765 dva = ptlrpc_req_async_args(dva, req);
768 ptlrpcd_add_req(req);
773 static void osc_io_data_version_end(const struct lu_env *env,
774 const struct cl_io_slice *slice)
776 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
777 struct osc_io *oio = cl2osc_io(env, slice);
778 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
781 wait_for_completion(&cbargs->opc_sync);
783 if (cbargs->opc_rc != 0) {
784 slice->cis_io->ci_result = cbargs->opc_rc;
786 slice->cis_io->ci_result = 0;
787 if (!(oio->oi_oa.o_valid &
788 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
789 slice->cis_io->ci_result = -ENOTSUPP;
791 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
792 dv->dv_layout_version = oio->oi_oa.o_layout_version;
793 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
794 dv->dv_data_version = oio->oi_oa.o_data_version;
800 int osc_io_read_start(const struct lu_env *env,
801 const struct cl_io_slice *slice)
803 struct cl_object *obj = slice->cis_obj;
804 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
808 if (!slice->cis_io->ci_noatime) {
809 cl_object_attr_lock(obj);
810 attr->cat_atime = ktime_get_real_seconds();
811 rc = cl_object_attr_update(env, obj, attr, CAT_ATIME);
812 cl_object_attr_unlock(obj);
817 EXPORT_SYMBOL(osc_io_read_start);
819 int osc_io_write_start(const struct lu_env *env,
820 const struct cl_io_slice *slice)
822 struct cl_object *obj = slice->cis_obj;
823 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
827 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1);
828 cl_object_attr_lock(obj);
829 attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds();
830 rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME);
831 cl_object_attr_unlock(obj);
835 EXPORT_SYMBOL(osc_io_write_start);
837 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
838 struct cl_fsync_io *fio)
840 struct osc_io *oio = osc_env_io(env);
841 struct obdo *oa = &oio->oi_oa;
842 struct lov_oinfo *loi = obj->oo_oinfo;
843 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
847 memset(oa, 0, sizeof(*oa));
848 oa->o_oi = loi->loi_oi;
849 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
851 /* reload size abd blocks for start and end of sync range */
852 oa->o_size = fio->fi_start;
853 oa->o_blocks = fio->fi_end;
854 oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
856 obdo_set_parent_fid(oa, fio->fi_fid);
858 init_completion(&cbargs->opc_sync);
860 rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET);
863 EXPORT_SYMBOL(osc_fsync_ost);
865 int osc_io_fsync_start(const struct lu_env *env,
866 const struct cl_io_slice *slice)
868 struct cl_io *io = slice->cis_io;
869 struct cl_fsync_io *fio = &io->u.ci_fsync;
870 struct cl_object *obj = slice->cis_obj;
871 struct osc_object *osc = cl2osc(obj);
872 pgoff_t start = cl_index(obj, fio->fi_start);
873 pgoff_t end = cl_index(obj, fio->fi_end);
877 if (fio->fi_end == OBD_OBJECT_EOF)
880 result = osc_cache_writeback_range(env, osc, start, end, 0,
881 fio->fi_mode == CL_FSYNC_DISCARD);
883 fio->fi_nr_written += result;
886 if (fio->fi_mode == CL_FSYNC_ALL) {
889 /* we have to wait for writeback to finish before we can
890 * send OST_SYNC RPC. This is bad because it causes extents
891 * to be written osc by osc. However, we usually start
892 * writeback before CL_FSYNC_ALL so this won't have any real
894 rc = osc_cache_wait_range(env, osc, start, end);
897 rc = osc_fsync_ost(env, osc, fio);
905 void osc_io_fsync_end(const struct lu_env *env,
906 const struct cl_io_slice *slice)
908 struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
909 struct cl_object *obj = slice->cis_obj;
910 pgoff_t start = cl_index(obj, fio->fi_start);
911 pgoff_t end = cl_index(obj, fio->fi_end);
914 if (fio->fi_mode == CL_FSYNC_LOCAL) {
915 result = osc_cache_wait_range(env, cl2osc(obj), start, end);
916 } else if (fio->fi_mode == CL_FSYNC_ALL) {
917 struct osc_io *oio = cl2osc_io(env, slice);
918 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
920 wait_for_completion(&cbargs->opc_sync);
922 result = cbargs->opc_rc;
924 slice->cis_io->ci_result = result;
926 EXPORT_SYMBOL(osc_io_fsync_end);
928 static int osc_io_ladvise_start(const struct lu_env *env,
929 const struct cl_io_slice *slice)
932 struct cl_io *io = slice->cis_io;
933 struct osc_io *oio = cl2osc_io(env, slice);
934 struct cl_object *obj = slice->cis_obj;
935 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
936 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
937 struct obdo *oa = &oio->oi_oa;
938 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
939 struct lu_ladvise *ladvise;
940 struct ladvise_hdr *ladvise_hdr;
945 /* TODO: add multiple ladvise support in CLIO */
946 buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]);
947 if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size)
948 lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size);
950 ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf;
951 if (ladvise_hdr == NULL)
954 memset(ladvise_hdr, 0, buf_size);
955 ladvise_hdr->lah_magic = LADVISE_MAGIC;
956 ladvise_hdr->lah_count = num_advise;
957 ladvise_hdr->lah_flags = lio->li_flags;
959 memset(oa, 0, sizeof(*oa));
960 oa->o_oi = loi->loi_oi;
961 oa->o_valid = OBD_MD_FLID;
962 obdo_set_parent_fid(oa, lio->li_fid);
964 ladvise = ladvise_hdr->lah_advise;
965 ladvise->lla_start = lio->li_start;
966 ladvise->lla_end = lio->li_end;
967 ladvise->lla_advice = lio->li_advice;
969 if (lio->li_flags & LF_ASYNC) {
970 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
971 ladvise_hdr, NULL, NULL, NULL);
973 init_completion(&cbargs->opc_sync);
974 result = osc_ladvise_base(osc_export(cl2osc(obj)), oa,
975 ladvise_hdr, osc_async_upcall,
976 cbargs, PTLRPCD_SET);
977 cbargs->opc_rpc_sent = result == 0;
982 static void osc_io_ladvise_end(const struct lu_env *env,
983 const struct cl_io_slice *slice)
985 struct cl_io *io = slice->cis_io;
986 struct osc_io *oio = cl2osc_io(env, slice);
987 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
989 struct cl_ladvise_io *lio = &io->u.ci_ladvise;
991 if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) {
992 wait_for_completion(&cbargs->opc_sync);
993 result = cbargs->opc_rc;
995 slice->cis_io->ci_result = result;
998 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice)
1000 struct osc_io *oio = cl2osc_io(env, slice);
1002 if (oio->oi_active) {
1003 osc_extent_release(env, oio->oi_active);
1004 oio->oi_active = NULL;
1007 EXPORT_SYMBOL(osc_io_end);
1009 static const struct cl_io_operations osc_io_ops = {
1012 .cio_iter_init = osc_io_rw_iter_init,
1013 .cio_iter_fini = osc_io_rw_iter_fini,
1014 .cio_start = osc_io_read_start,
1015 .cio_fini = osc_io_fini
1018 .cio_iter_init = osc_io_rw_iter_init,
1019 .cio_iter_fini = osc_io_rw_iter_fini,
1020 .cio_start = osc_io_write_start,
1021 .cio_end = osc_io_end,
1022 .cio_fini = osc_io_fini
1025 .cio_iter_init = osc_io_iter_init,
1026 .cio_iter_fini = osc_io_iter_fini,
1027 .cio_start = osc_io_setattr_start,
1028 .cio_end = osc_io_setattr_end
1030 [CIT_DATA_VERSION] = {
1031 .cio_start = osc_io_data_version_start,
1032 .cio_end = osc_io_data_version_end,
1035 .cio_iter_init = osc_io_iter_init,
1036 .cio_iter_fini = osc_io_iter_fini,
1037 .cio_start = osc_io_fault_start,
1038 .cio_end = osc_io_end,
1039 .cio_fini = osc_io_fini
1042 .cio_start = osc_io_fsync_start,
1043 .cio_end = osc_io_fsync_end,
1044 .cio_fini = osc_io_fini
1047 .cio_start = osc_io_ladvise_start,
1048 .cio_end = osc_io_ladvise_end,
1049 .cio_fini = osc_io_fini
1052 .cio_fini = osc_io_fini
1055 .cio_read_ahead = osc_io_read_ahead,
1056 .cio_submit = osc_io_submit,
1057 .cio_commit_async = osc_io_commit_async
1060 /*****************************************************************************
1062 * Transfer operations.
1066 int osc_io_init(const struct lu_env *env,
1067 struct cl_object *obj, struct cl_io *io)
1069 struct osc_io *oio = osc_env_io(env);
1071 CL_IO_SLICE_CLEAN(oio, oi_cl);
1072 cl_io_slice_add(io, &oio->oi_cl, obj, &osc_io_ops);