X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosc%2Fosc_io.c;h=389363f9afdabd0f127c1b62ebf6b375c7c08cc8;hp=53366f8630eaf66bc22f8ca32289cdc188478196;hb=a56fefc535677b146b3c55752109d20b92c7be3e;hpb=f95393b0d0a59cf3dc2f29cffc35dcc4cc9d7728 diff --git a/lustre/osc/osc_io.c b/lustre/osc/osc_io.c index 53366f8..389363f 100644 --- a/lustre/osc/osc_io.c +++ b/lustre/osc/osc_io.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -17,17 +15,15 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ /* * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -36,47 +32,22 @@ * Implementation of cl_io for OSC layer. * * Author: Nikita Danilov + * Author: Jinshan Xiong */ #define DEBUG_SUBSYSTEM S_OSC -#include "osc_cl_internal.h" +#include +#include +#include +#include -/** \addtogroup osc - * @{ - */ +#include "osc_internal.h" -/***************************************************************************** - * - * Type conversions. - * +/** \addtogroup osc + * @{ */ -static struct osc_req *cl2osc_req(const struct cl_req_slice *slice) -{ - LINVRNT(slice->crs_dev->cd_lu_dev.ld_type == &osc_device_type); - return container_of0(slice, struct osc_req, or_cl); -} - -static struct osc_io *cl2osc_io(const struct lu_env *env, - const struct cl_io_slice *slice) -{ - struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl); - LINVRNT(oio == osc_env_io(env)); - return oio; -} - -static struct osc_page *osc_cl_page_osc(struct cl_page *page) -{ - const struct cl_page_slice *slice; - - slice = cl_page_at(page, &osc_device_type); - LASSERT(slice != NULL); - - return cl2osc_page(slice); -} - - /***************************************************************************** * * io operations. @@ -87,17 +58,47 @@ static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io) { } -struct cl_page *osc_oap2cl_page(struct osc_async_page *oap) +void osc_read_ahead_release(const struct lu_env *env, void *cbdata) { - return container_of(oap, struct osc_page, ops_oap)->ops_cl.cpl_page; + struct ldlm_lock *dlmlock = cbdata; + struct lustre_handle lockh; + + ldlm_lock2handle(dlmlock, &lockh); + ldlm_lock_decref(&lockh, LCK_PR); + LDLM_LOCK_PUT(dlmlock); } +EXPORT_SYMBOL(osc_read_ahead_release); -static void osc_io_unplug(const struct lu_env *env, struct osc_object *osc, - struct client_obd *cli) +static int osc_io_read_ahead(const struct lu_env *env, + const struct cl_io_slice *ios, + pgoff_t start, struct cl_read_ahead *ra) { - loi_list_maint(cli, osc->oo_oinfo); - osc_check_rpcs(env, cli); - client_obd_list_unlock(&cli->cl_loi_list_lock); + struct osc_object *osc = cl2osc(ios->cis_obj); + struct ldlm_lock *dlmlock; + int result = -ENODATA; + ENTRY; + + dlmlock = osc_dlmlock_at_pgoff(env, osc, start, 0); + if (dlmlock != NULL) { + LASSERT(dlmlock->l_ast_data == osc); + if (dlmlock->l_req_mode != LCK_PR) { + struct lustre_handle lockh; + ldlm_lock2handle(dlmlock, &lockh); + ldlm_lock_addref(&lockh, LCK_PR); + ldlm_lock_decref(&lockh, dlmlock->l_req_mode); + } + + ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc; + ra->cra_end_idx = cl_index(osc2cl(osc), + dlmlock->l_policy_data.l_extent.end); + ra->cra_release = osc_read_ahead_release; + ra->cra_cbdata = dlmlock; + if (ra->cra_end_idx != CL_PAGE_EOF) + ra->cra_contention = true; + result = 0; + } + + RETURN(result); } /** @@ -105,93 +106,73 @@ static void osc_io_unplug(const struct lu_env *env, struct osc_object *osc, * layer. Iterates over pages in the in-queue, prepares each for io by calling * cl_page_prep() and then either submits them through osc_io_submit_page() * or, if page is already submitted, changes osc flags through - * osc_set_async_flags_base(). + * osc_set_async_flags(). */ -static int osc_io_submit(const struct lu_env *env, - const struct cl_io_slice *ios, - enum cl_req_type crt, struct cl_2queue *queue, - enum cl_req_priority priority) +int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios, + enum cl_req_type crt, struct cl_2queue *queue) { - struct cl_page *page; - struct cl_page *tmp; - struct osc_object *osc0 = NULL; - struct client_obd *cli = NULL; - struct osc_object *osc = NULL; /* to keep gcc happy */ - struct osc_page *opg; - struct cl_io *io; - - struct cl_page_list *qin = &queue->c2_qin; - struct cl_page_list *qout = &queue->c2_qout; - int queued = 0; - int result = 0; - - LASSERT(qin->pl_nr > 0); - - CDEBUG(D_INFO, "%i %i\n", qin->pl_nr, crt); + struct cl_page *page; + struct cl_page *tmp; + struct client_obd *cli = NULL; + struct osc_object *osc = NULL; /* to keep gcc happy */ + struct osc_page *opg; + struct cl_io *io; + LIST_HEAD(list); + + struct cl_page_list *qin = &queue->c2_qin; + struct cl_page_list *qout = &queue->c2_qout; + unsigned int queued = 0; + int result = 0; + int brw_flags; + unsigned int max_pages; + unsigned int ppc_bits; /* pages per chunk bits */ + unsigned int ppc; + bool sync_queue = false; + + LASSERT(qin->pl_nr > 0); + + CDEBUG(D_CACHE|D_READA, "%d %d\n", qin->pl_nr, crt); + + osc = cl2osc(ios->cis_obj); + cli = osc_cli(osc); + max_pages = cli->cl_max_pages_per_rpc; + ppc_bits = cli->cl_chunkbits - PAGE_SHIFT; + ppc = 1 << ppc_bits; + + brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0; + brw_flags |= crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ; + if (crt == CRT_READ && ios->cis_io->ci_ndelay) + brw_flags |= OBD_BRW_NDELAY; + + page = cl_page_list_first(qin); + if (page->cp_type == CPT_TRANSIENT) + brw_flags |= OBD_BRW_NOCACHE; + /* * NOTE: here @page is a top-level page. This is done to avoid * creation of sub-page-list. */ cl_page_list_for_each_safe(page, tmp, qin) { struct osc_async_page *oap; - struct obd_export *exp; /* Top level IO. */ io = page->cp_owner; LASSERT(io != NULL); - opg = osc_cl_page_osc(page); - oap = &opg->ops_oap; - osc = cl2osc(opg->ops_cl.cpl_obj); - exp = osc_export(osc); + opg = osc_cl_page_osc(page, osc); + oap = &opg->ops_oap; + LASSERT(osc == oap->oap_obj); - if (priority > CRP_NORMAL) { - cfs_spin_lock(&oap->oap_lock); - oap->oap_async_flags |= ASYNC_HP; - cfs_spin_unlock(&oap->oap_lock); - } - /* - * This can be checked without cli->cl_loi_list_lock, because - * ->oap_*_item are always manipulated when the page is owned. - */ - if (!cfs_list_empty(&oap->oap_urgent_item) || - !cfs_list_empty(&oap->oap_rpc_item)) { + if (!list_empty(&oap->oap_pending_item) || + !list_empty(&oap->oap_rpc_item)) { + CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n", + oap, opg); result = -EBUSY; break; } - if (osc0 == NULL) { /* first iteration */ - cli = &exp->exp_obd->u.cli; - osc0 = osc; - } else /* check that all pages are against the same object - * (for now) */ - LASSERT(osc == osc0); - if (queued++ == 0) - client_obd_list_lock(&cli->cl_loi_list_lock); result = cl_page_prep(env, io, page, crt); - if (result == 0) { - cl_page_list_move(qout, qin, page); - if (cfs_list_empty(&oap->oap_pending_item)) { - osc_io_submit_page(env, cl2osc_io(env, ios), - opg, crt); - } else { - result = osc_set_async_flags_base(cli, - osc->oo_oinfo, - oap, - OSC_FLAGS); - /* - * bug 18881: we can't just break out here when - * error occurrs after cl_page_prep has been - * called against the page. The correct - * way is to call page's completion routine, - * as in osc_oap_interrupted. For simplicity, - * we just force osc_set_async_flags_base() to - * not return error. - */ - LASSERT(result == 0); - } - opg->ops_submit_time = cfs_time_current(); - } else { + if (result != 0) { LASSERT(result < 0); if (result != -EALREADY) break; @@ -201,507 +182,933 @@ static int osc_io_submit(const struct lu_env *env, * is not dirty. */ result = 0; + continue; } - /* - * We might hold client_obd_list_lock() for too long and cause - * soft-lockups (see bug 16651). But on the other hand, pages - * are queued here with ASYNC_URGENT flag, thus will be sent - * out immediately once osc_io_unplug() be called, possibly - * resulting sub-optimal RPCs. - * - * We think creating optimal-sized RPCs is more important than - * avoiding the transient soft-lockups, plus I believe the - * soft-locks only happen in full debug testing. - */ - } - - LASSERT(ergo(result == 0, cli != NULL)); - LASSERT(ergo(result == 0, osc == osc0)); - - if (queued > 0) - osc_io_unplug(env, osc, cli); - CDEBUG(D_INFO, "%i/%i %i\n", qin->pl_nr, qout->pl_nr, result); - return qout->pl_nr > 0 ? 0 : result; -} - -static void osc_page_touch_at(const struct lu_env *env, - struct cl_object *obj, pgoff_t idx, unsigned to) -{ - struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo; - struct cl_attr *attr = &osc_env_info(env)->oti_attr; - int valid; - __u64 kms; - - /* offset within stripe */ - kms = cl_offset(obj, idx) + to; - - cl_object_attr_lock(obj); - /* - * XXX old code used - * - * ll_inode_size_lock(inode, 0); lov_stripe_lock(lsm); - * - * here - */ - CDEBUG(D_INODE, "stripe KMS %sincreasing "LPU64"->"LPU64" "LPU64"\n", - kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms, - loi->loi_lvb.lvb_size); - - valid = 0; - if (kms > loi->loi_kms) { - attr->cat_kms = kms; - valid |= CAT_KMS; - } - if (kms > loi->loi_lvb.lvb_size) { - attr->cat_size = kms; - valid |= CAT_SIZE; - } - cl_object_attr_set(env, obj, attr, valid); - cl_object_attr_unlock(obj); + spin_lock(&oap->oap_lock); + oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY; + oap->oap_async_flags |= ASYNC_COUNT_STABLE; + spin_unlock(&oap->oap_lock); + + osc_page_submit(env, opg, crt, brw_flags); + list_add_tail(&oap->oap_pending_item, &list); + + if (page->cp_sync_io != NULL) + cl_page_list_move(qout, qin, page); + else /* async IO */ + cl_page_list_del(env, qin, page); + + queued++; + if (queued == max_pages) { + sync_queue = true; + } else if (crt == CRT_WRITE) { + unsigned int chunks; + unsigned int next_chunks; + + chunks = (queued + ppc - 1) >> ppc_bits; + /* chunk number if add another page */ + next_chunks = (queued + ppc) >> ppc_bits; + + /* next page will excceed write chunk limit */ + if (chunks == osc_max_write_chunks(cli) && + next_chunks > chunks) + sync_queue = true; + } + + if (sync_queue) { + result = osc_queue_sync_pages(env, io, osc, &list, + brw_flags); + if (result < 0) + break; + queued = 0; + sync_queue = false; + } + } + + if (queued > 0) + result = osc_queue_sync_pages(env, io, osc, &list, brw_flags); + + /* Update c/mtime for sync write. LU-7310 */ + if (crt == CRT_WRITE && qout->pl_nr > 0 && result == 0) { + struct cl_object *obj = ios->cis_obj; + struct cl_attr *attr = &osc_env_info(env)->oti_attr; + + cl_object_attr_lock(obj); + attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds(); + cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME); + cl_object_attr_unlock(obj); + } + + CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result); + return qout->pl_nr > 0 ? 0 : result; } +EXPORT_SYMBOL(osc_io_submit); /** - * This is called when a page is accessed within file in a way that creates - * new page, if one were missing (i.e., if there were a hole at that place in - * the file, or accessed page is beyond the current file size). Examples: - * ->commit_write() and ->nopage() methods. + * This is called to update the attributes when modifying a specific page, + * both when making new pages and when doing updates to existing cached pages. * * Expand stripe KMS if necessary. */ -static void osc_page_touch(const struct lu_env *env, - struct osc_page *opage, unsigned to) +void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj, + pgoff_t idx, size_t to) { - struct cl_page *page = opage->ops_cl.cpl_page; - struct cl_object *obj = opage->ops_cl.cpl_obj; + struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo; + struct cl_attr *attr = &osc_env_info(env)->oti_attr; + int valid; + __u64 kms; + + ENTRY; + + /* offset within stripe */ + kms = cl_offset(obj, idx) + to; + + cl_object_attr_lock(obj); + CDEBUG(D_INODE, "stripe KMS %sincreasing %llu->%llu %llu\n", + kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms, + loi->loi_lvb.lvb_size); + + attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds(); + valid = CAT_MTIME | CAT_CTIME; + if (kms > loi->loi_kms) { + attr->cat_kms = kms; + valid |= CAT_KMS; + } + if (kms > loi->loi_lvb.lvb_size) { + attr->cat_size = kms; + valid |= CAT_SIZE; + } + cl_object_attr_update(env, obj, attr, valid); + cl_object_attr_unlock(obj); + + EXIT; +} - osc_page_touch_at(env, obj, page->cp_index, to); +int osc_io_commit_async(const struct lu_env *env, + const struct cl_io_slice *ios, + struct cl_page_list *qin, int from, int to, + cl_commit_cbt cb) +{ + struct cl_io *io = ios->cis_io; + struct osc_io *oio = cl2osc_io(env, ios); + struct osc_object *osc = cl2osc(ios->cis_obj); + struct cl_page *page; + struct cl_page *last_page; + struct osc_page *opg; + struct pagevec *pvec = &osc_env_info(env)->oti_pagevec; + int result = 0; + ENTRY; + + LASSERT(qin->pl_nr > 0); + + /* Handle partial page cases */ + last_page = cl_page_list_last(qin); + if (oio->oi_lockless) { + page = cl_page_list_first(qin); + if (page == last_page) { + cl_page_clip(env, page, from, to); + } else { + if (from != 0) + cl_page_clip(env, page, from, PAGE_SIZE); + if (to != PAGE_SIZE) + cl_page_clip(env, last_page, 0, to); + } + } + + ll_pagevec_init(pvec, 0); + + while (qin->pl_nr > 0) { + struct osc_async_page *oap; + + page = cl_page_list_first(qin); + opg = osc_cl_page_osc(page, osc); + oap = &opg->ops_oap; + + LASSERTF(osc == oap->oap_obj, + "obj mismatch: %p / %p\n", osc, oap->oap_obj); + + if (!list_empty(&oap->oap_rpc_item)) { + CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n", + oap, opg); + result = -EBUSY; + break; + } + + /* The page may be already in dirty cache. */ + if (list_empty(&oap->oap_pending_item)) { + result = osc_page_cache_add(env, opg, io, cb); + if (result != 0) + break; + } + + osc_page_touch_at(env, osc2cl(osc), osc_index(opg), + page == last_page ? to : PAGE_SIZE); + + cl_page_list_del(env, qin, page); + + /* if there are no more slots, do the callback & reinit */ + if (pagevec_add(pvec, page->cp_vmpage) == 0) { + (*cb)(env, io, pvec); + pagevec_reinit(pvec); + } + } + + /* Clean up any partially full pagevecs */ + if (pagevec_count(pvec) != 0) + (*cb)(env, io, pvec); + + /* Can't access these pages any more. Page can be in transfer and + * complete at any time. */ + + /* for sync write, kernel will wait for this page to be flushed before + * osc_io_end() is called, so release it earlier. + * for mkwrite(), it's known there is no further pages. */ + if (cl_io_is_sync_write(io) && oio->oi_active != NULL) { + osc_extent_release(env, oio->oi_active); + oio->oi_active = NULL; + } + + CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result); + RETURN(result); } +EXPORT_SYMBOL(osc_io_commit_async); -/** - * Implements cl_io_operations::cio_prepare_write() method for osc layer. - * - * \retval -EIO transfer initiated against this osc will most likely fail - * \retval 0 transfer initiated against this osc will most likely succeed. - * - * The reason for this check is to immediately return an error to the caller - * in the case of a deactivated import. Note, that import can be deactivated - * later, while pages, dirtied by this IO, are still in the cache, but this is - * irrelevant, because that would still return an error to the application (if - * it does fsync), but many applications don't do fsync because of performance - * issues, and we wanted to return an -EIO at write time to notify the - * application. - */ -static int osc_io_prepare_write(const struct lu_env *env, - const struct cl_io_slice *ios, - const struct cl_page_slice *slice, - unsigned from, unsigned to) +static bool osc_import_not_healthy(struct obd_import *imp) { - struct osc_device *dev = lu2osc_dev(slice->cpl_obj->co_lu.lo_dev); - struct obd_import *imp = class_exp2cliimp(dev->od_exp); + return imp->imp_invalid || imp->imp_deactive || + !(imp->imp_state == LUSTRE_IMP_FULL || + imp->imp_state == LUSTRE_IMP_IDLE); +} - ENTRY; +int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios) +{ + struct osc_object *osc = cl2osc(ios->cis_obj); + struct obd_import *imp = osc_cli(osc)->cl_import; + struct osc_io *oio = osc_env_io(env); + int rc = -EIO; + ENTRY; + + spin_lock(&imp->imp_lock); + /** + * check whether this OSC device is available for non-delay read, + * fast switching mirror if we haven't tried all mirrors. + */ + if (ios->cis_io->ci_type == CIT_READ && ios->cis_io->ci_ndelay && + !ios->cis_io->ci_tried_all_mirrors && osc_import_not_healthy(imp)) { + rc = -EWOULDBLOCK; + } else if (likely(!imp->imp_invalid)) { + atomic_inc(&osc->oo_nr_ios); + oio->oi_is_active = 1; + rc = 0; + } + spin_unlock(&imp->imp_lock); + + if (cfs_capable(CFS_CAP_SYS_RESOURCE)) + oio->oi_cap_sys_resource = 1; + + RETURN(rc); +} +EXPORT_SYMBOL(osc_io_iter_init); - /* - * This implements OBD_BRW_CHECK logic from old client. - */ +int osc_io_rw_iter_init(const struct lu_env *env, + const struct cl_io_slice *ios) +{ + struct cl_io *io = ios->cis_io; + struct osc_io *oio = osc_env_io(env); + struct osc_object *osc = cl2osc(ios->cis_obj); + unsigned long npages; + ENTRY; - RETURN(imp == NULL || imp->imp_invalid ? -EIO : 0); + if (cl_io_is_append(io)) + RETURN(osc_io_iter_init(env, ios)); + + npages = io->u.ci_rw.crw_count >> PAGE_SHIFT; + if (io->u.ci_rw.crw_pos & ~PAGE_MASK) + ++npages; + + oio->oi_lru_reserved = osc_lru_reserve(osc_cli(osc), npages); + + RETURN(osc_io_iter_init(env, ios)); } +EXPORT_SYMBOL(osc_io_rw_iter_init); -static int osc_io_commit_write(const struct lu_env *env, - const struct cl_io_slice *ios, - const struct cl_page_slice *slice, - unsigned from, unsigned to) +void osc_io_iter_fini(const struct lu_env *env, + const struct cl_io_slice *ios) { - struct osc_page *opg = cl2osc_page(slice); - struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj); - struct osc_async_page *oap = &opg->ops_oap; - ENTRY; + struct osc_io *oio = osc_env_io(env); - LASSERT(to > 0); - /* - * XXX instead of calling osc_page_touch() here and in - * osc_io_fault_start() it might be more logical to introduce - * cl_page_touch() method, that generic cl_io_commit_write() and page - * fault code calls. - */ - osc_page_touch(env, cl2osc_page(slice), to); - if (!client_is_remote(osc_export(obj)) && - cfs_capable(CFS_CAP_SYS_RESOURCE)) - oap->oap_brw_flags |= OBD_BRW_NOQUOTA; + if (oio->oi_is_active) { + struct osc_object *osc = cl2osc(ios->cis_obj); - RETURN(0); + oio->oi_is_active = 0; + LASSERT(atomic_read(&osc->oo_nr_ios) > 0); + if (atomic_dec_and_test(&osc->oo_nr_ios)) + wake_up_all(&osc->oo_io_waitq); + } } +EXPORT_SYMBOL(osc_io_iter_fini); -static int osc_io_fault_start(const struct lu_env *env, - const struct cl_io_slice *ios) +void osc_io_rw_iter_fini(const struct lu_env *env, + const struct cl_io_slice *ios) { - struct cl_io *io; - struct cl_fault_io *fio; + struct osc_io *oio = osc_env_io(env); + struct osc_object *osc = cl2osc(ios->cis_obj); - ENTRY; + if (oio->oi_lru_reserved > 0) { + osc_lru_unreserve(osc_cli(osc), oio->oi_lru_reserved); + oio->oi_lru_reserved = 0; + } + oio->oi_write_osclock = NULL; - io = ios->cis_io; - fio = &io->u.ci_fault; - CDEBUG(D_INFO, "%lu %i %i\n", - fio->ft_index, fio->ft_writable, fio->ft_nob); - /* - * If mapping is writeable, adjust kms to cover this page, - * but do not extend kms beyond actual file size. - * See bug 10919. - */ - if (fio->ft_writable) - osc_page_touch_at(env, ios->cis_obj, - fio->ft_index, fio->ft_nob); - RETURN(0); + osc_io_iter_fini(env, ios); } +EXPORT_SYMBOL(osc_io_rw_iter_fini); -static int osc_setattr_upcall(void *a, int rc) +int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios) { - struct osc_setattr_cbargs *args = a; + struct cl_io *io; + struct cl_fault_io *fio; + ENTRY; + + io = ios->cis_io; + fio = &io->u.ci_fault; + CDEBUG(D_INFO, "%lu %d %zu\n", + fio->ft_index, fio->ft_writable, fio->ft_nob); + /* + * If mapping is writeable, adjust kms to cover this page, + * but do not extend kms beyond actual file size. + * See bug 10919. + */ + if (fio->ft_writable) + osc_page_touch_at(env, ios->cis_obj, + fio->ft_index, fio->ft_nob); + RETURN(0); +} +EXPORT_SYMBOL(osc_io_fault_start); + + +static int osc_async_upcall(void *a, int rc) +{ + struct osc_async_cbargs *args = a; args->opc_rc = rc; - cfs_complete(&args->opc_sync); + complete(&args->opc_sync); return 0; } -/* Disable osc_trunc_check() because it is naturally race between read and - * truncate. See bug 20645 for details. - */ -#if 0 && defined(__KERNEL__) /** * Checks that there are no pages being written in the extent being truncated. */ +static bool trunc_check_cb(const struct lu_env *env, struct cl_io *io, + struct osc_page *ops , void *cbdata) +{ + struct cl_page *page = ops->ops_cl.cpl_page; + struct osc_async_page *oap; + __u64 start = *(__u64 *)cbdata; + + oap = &ops->ops_oap; + if (oap->oap_cmd & OBD_BRW_WRITE && + !list_empty(&oap->oap_pending_item)) + CL_PAGE_DEBUG(D_ERROR, env, page, "exists %llu/%s.\n", + start, current->comm); + + if (PageLocked(page->cp_vmpage)) + CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n", + ops, osc_index(ops), oap->oap_cmd & OBD_BRW_RWMASK); + + return true; +} + static void osc_trunc_check(const struct lu_env *env, struct cl_io *io, - struct osc_io *oio, size_t size) + struct osc_io *oio, __u64 size) { - struct osc_page *cp; - struct osc_object *obj; - struct cl_object *clob; - struct cl_page *page; - struct cl_page_list *list; - int partial; - pgoff_t start; + struct cl_object *clob; + int partial; + pgoff_t start; clob = oio->oi_cl.cis_obj; - obj = cl2osc(clob); start = cl_index(clob, size); partial = cl_offset(clob, start) < size; - list = &osc_env_info(env)->oti_plist; /* * Complain if there are pages in the truncated region. - * - * XXX this is quite expensive check. */ - cl_page_list_init(list); - cl_page_gang_lookup(env, clob, io, start + partial, CL_PAGE_EOF, list, 0); + osc_page_gang_lookup(env, io, cl2osc(clob), + start + partial, CL_PAGE_EOF, + trunc_check_cb, (void *)&size); +} - cl_page_list_for_each(page, list) - CL_PAGE_DEBUG(D_ERROR, env, page, "exists %lu\n", start); +static int osc_io_setattr_start(const struct lu_env *env, + const struct cl_io_slice *slice) +{ + struct cl_io *io = slice->cis_io; + struct osc_io *oio = cl2osc_io(env, slice); + struct cl_object *obj = slice->cis_obj; + struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo; + struct cl_attr *attr = &osc_env_info(env)->oti_attr; + struct obdo *oa = &oio->oi_oa; + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + unsigned int ia_avalid = io->u.ci_setattr.sa_avalid; + enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid; + int result = 0; + __u64 size = io->u.ci_setattr.sa_attr.lvb_size; + __u64 end = OBD_OBJECT_EOF; + bool io_is_falloc = false; + + ENTRY; + /* truncate cache dirty pages first */ + if (cl_io_is_trunc(io)) { + result = osc_cache_truncate_start(env, cl2osc(obj), size, + &oio->oi_trunc); + } else if (cl_io_is_fallocate(io)) { + io_is_falloc = true; + size = io->u.ci_setattr.sa_falloc_offset; + end = io->u.ci_setattr.sa_falloc_end; + } + + if (result == 0 && oio->oi_lockless == 0) { + cl_object_attr_lock(obj); + result = cl_object_attr_get(env, obj, attr); + if (result == 0) { + struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr; + unsigned int cl_valid = 0; + + if (ia_avalid & ATTR_SIZE) { + attr->cat_size = size; + attr->cat_kms = size; + cl_valid = (CAT_SIZE | CAT_KMS); + } + if (ia_avalid & ATTR_MTIME_SET) { + attr->cat_mtime = lvb->lvb_mtime; + cl_valid |= CAT_MTIME; + } + if (ia_avalid & ATTR_ATIME_SET) { + attr->cat_atime = lvb->lvb_atime; + cl_valid |= CAT_ATIME; + } + if (ia_xvalid & OP_XVALID_CTIME_SET) { + attr->cat_ctime = lvb->lvb_ctime; + cl_valid |= CAT_CTIME; + } + result = cl_object_attr_update(env, obj, attr, + cl_valid); + } + cl_object_attr_unlock(obj); + } + memset(oa, 0, sizeof(*oa)); + if (result == 0) { + oa->o_oi = loi->loi_oi; + obdo_set_parent_fid(oa, io->u.ci_setattr.sa_parent_fid); + oa->o_stripe_idx = io->u.ci_setattr.sa_stripe_index; + oa->o_layout = io->u.ci_setattr.sa_layout; + oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP | + OBD_MD_FLOSTLAYOUT; + if (ia_avalid & ATTR_CTIME) { + oa->o_valid |= OBD_MD_FLCTIME; + oa->o_ctime = attr->cat_ctime; + } + if (ia_avalid & ATTR_ATIME) { + oa->o_valid |= OBD_MD_FLATIME; + oa->o_atime = attr->cat_atime; + } + if (ia_avalid & ATTR_MTIME) { + oa->o_valid |= OBD_MD_FLMTIME; + oa->o_mtime = attr->cat_mtime; + } + if (ia_avalid & ATTR_SIZE) { + if (io_is_falloc) { + oa->o_size = size; + oa->o_blocks = end; + oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + } else { + oa->o_size = size; + oa->o_blocks = OBD_OBJECT_EOF; + oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + } + + if (oio->oi_lockless) { + oa->o_flags = OBD_FL_SRVLOCK; + oa->o_valid |= OBD_MD_FLFLAGS; + } + + if (io->ci_layout_version > 0) { + /* verify layout version */ + oa->o_valid |= OBD_MD_LAYOUT_VERSION; + oa->o_layout_version = io->ci_layout_version; + } + } else { + LASSERT(oio->oi_lockless == 0); + } + + if (ia_xvalid & OP_XVALID_FLAGS) { + oa->o_flags = io->u.ci_setattr.sa_attr_flags; + oa->o_valid |= OBD_MD_FLFLAGS; + } + + init_completion(&cbargs->opc_sync); + + if (io_is_falloc) { + int falloc_mode = io->u.ci_setattr.sa_falloc_mode; + + result = osc_fallocate_base(osc_export(cl2osc(obj)), + oa, osc_async_upcall, + cbargs, falloc_mode); + } else if (ia_avalid & ATTR_SIZE) { + result = osc_punch_send(osc_export(cl2osc(obj)), + oa, osc_async_upcall, cbargs); + } else { + result = osc_setattr_async(osc_export(cl2osc(obj)), + oa, osc_async_upcall, + cbargs, PTLRPCD_SET); + } + cbargs->opc_rpc_sent = result == 0; + } + + RETURN(result); +} - cl_page_list_disown(env, io, list); - cl_page_list_fini(env, list); +void osc_io_setattr_end(const struct lu_env *env, + const struct cl_io_slice *slice) +{ + struct cl_io *io = slice->cis_io; + struct osc_io *oio = cl2osc_io(env, slice); + struct cl_object *obj = slice->cis_obj; + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + struct cl_attr *attr = &osc_env_info(env)->oti_attr; + struct obdo *oa = &oio->oi_oa; + unsigned int cl_valid = 0; + int result = 0; + + if (cbargs->opc_rpc_sent) { + wait_for_completion(&cbargs->opc_sync); + result = io->ci_result = cbargs->opc_rc; + } + + if (result == 0) { + if (oio->oi_lockless) { + /* lockless truncate */ + struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev); + + LASSERT(cl_io_is_trunc(io)); + LASSERT(cl_io_is_trunc(io) || cl_io_is_fallocate(io)); + /* XXX: Need a lock. */ + osd->od_stats.os_lockless_truncates++; + } + } + + if (cl_io_is_trunc(io)) { + __u64 size = io->u.ci_setattr.sa_attr.lvb_size; + + if (result == 0) { + cl_object_attr_lock(obj); + if (oa->o_valid & OBD_MD_FLBLOCKS) { + attr->cat_blocks = oa->o_blocks; + cl_valid |= CAT_BLOCKS; + } + + cl_object_attr_update(env, obj, attr, cl_valid); + cl_object_attr_unlock(obj); + } + osc_trunc_check(env, io, oio, size); + osc_cache_truncate_end(env, oio->oi_trunc); + oio->oi_trunc = NULL; + } + + if (cl_io_is_fallocate(io)) { + cl_object_attr_lock(obj); + + /* update blocks */ + if (oa->o_valid & OBD_MD_FLBLOCKS) { + attr->cat_blocks = oa->o_blocks; + cl_valid |= CAT_BLOCKS; + } + + /* update size */ + if (oa->o_valid & OBD_MD_FLSIZE) { + attr->cat_size = oa->o_size; + cl_valid |= CAT_SIZE; + } + + cl_object_attr_update(env, obj, attr, cl_valid); + cl_object_attr_unlock(obj); + } +} +EXPORT_SYMBOL(osc_io_setattr_end); - cfs_spin_lock(&obj->oo_seatbelt); - cfs_list_for_each_entry(cp, &obj->oo_inflight[CRT_WRITE], - ops_inflight) { - page = cp->ops_cl.cpl_page; - if (page->cp_index >= start + partial) { - cfs_task_t *submitter; +struct osc_data_version_args { + struct osc_io *dva_oio; +}; - submitter = cp->ops_submitter; - /* - * XXX Linux specific debugging stuff. - */ - CL_PAGE_DEBUG(D_ERROR, env, page, "%s/%i %lu\n", - submitter->comm, submitter->pid, start); - libcfs_debug_dumpstack(submitter); - } - } - cfs_spin_unlock(&obj->oo_seatbelt); +static int +osc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req, + void *args, int rc) +{ + struct osc_data_version_args *dva = args; + struct osc_io *oio = dva->dva_oio; + const struct ost_body *body; + + ENTRY; + if (rc < 0) + GOTO(out, rc); + + body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); + if (body == NULL) + GOTO(out, rc = -EPROTO); + + lustre_get_wire_obdo(&req->rq_import->imp_connect_data, &oio->oi_oa, + &body->oa); + EXIT; +out: + oio->oi_cbarg.opc_rc = rc; + complete(&oio->oi_cbarg.opc_sync); + + return 0; } -#else /* __KERNEL__ */ -# define osc_trunc_check(env, io, oio, size) do {;} while (0) -#endif -static int osc_io_setattr_start(const struct lu_env *env, - const struct cl_io_slice *slice) +static int osc_io_data_version_start(const struct lu_env *env, + const struct cl_io_slice *slice) { - struct cl_io *io = slice->cis_io; - struct osc_io *oio = cl2osc_io(env, slice); - struct cl_object *obj = slice->cis_obj; - struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo; - struct cl_attr *attr = &osc_env_info(env)->oti_attr; - struct obdo *oa = &oio->oi_oa; - struct osc_setattr_cbargs *cbargs = &oio->oi_setattr_cbarg; - loff_t size = io->u.ci_setattr.sa_attr.lvb_size; - unsigned int ia_valid = io->u.ci_setattr.sa_valid; - int result = 0; - struct obd_info oinfo = { { { 0 } } }; - - if (ia_valid & ATTR_SIZE) - osc_trunc_check(env, io, oio, size); - - if (oio->oi_lockless == 0) { - cl_object_attr_lock(obj); - result = cl_object_attr_get(env, obj, attr); - if (result == 0) { - unsigned int cl_valid = 0; - - if (ia_valid & ATTR_SIZE) { - attr->cat_size = attr->cat_kms = size; - cl_valid = (CAT_SIZE | CAT_KMS); - } - if (ia_valid & ATTR_MTIME_SET) { - attr->cat_mtime = io->u.ci_setattr.sa_attr.lvb_mtime; - cl_valid |= CAT_MTIME; - } - if (ia_valid & ATTR_ATIME_SET) { - attr->cat_atime = io->u.ci_setattr.sa_attr.lvb_atime; - cl_valid |= CAT_ATIME; - } - if (ia_valid & ATTR_CTIME_SET) { - attr->cat_ctime = io->u.ci_setattr.sa_attr.lvb_ctime; - cl_valid |= CAT_CTIME; - } - result = cl_object_attr_set(env, obj, attr, cl_valid); - } - cl_object_attr_unlock(obj); - } - memset(oa, 0, sizeof(*oa)); - if (result == 0) { - oa->o_id = loi->loi_id; - oa->o_seq = loi->loi_seq; - oa->o_mtime = attr->cat_mtime; - oa->o_atime = attr->cat_atime; - oa->o_ctime = attr->cat_ctime; - oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME | - OBD_MD_FLCTIME | OBD_MD_FLMTIME; - if (ia_valid & ATTR_SIZE) { - oa->o_size = size; - oa->o_blocks = OBD_OBJECT_EOF; - oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; - - if (oio->oi_lockless) { - oa->o_flags = OBD_FL_SRVLOCK; - oa->o_valid |= OBD_MD_FLFLAGS; - } - } else { - LASSERT(oio->oi_lockless == 0); - } - - oinfo.oi_oa = oa; - oinfo.oi_capa = io->u.ci_setattr.sa_capa; - cfs_init_completion(&cbargs->opc_sync); - - if (ia_valid & ATTR_SIZE) - result = osc_punch_base(osc_export(cl2osc(obj)), - &oinfo, osc_setattr_upcall, - cbargs, PTLRPCD_SET); - else - result = osc_setattr_async_base(osc_export(cl2osc(obj)), - &oinfo, NULL, - osc_setattr_upcall, - cbargs, PTLRPCD_SET); - } - return result; + struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version; + struct osc_io *oio = cl2osc_io(env, slice); + struct obdo *oa = &oio->oi_oa; + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + struct osc_object *obj = cl2osc(slice->cis_obj); + struct lov_oinfo *loi = obj->oo_oinfo; + struct obd_export *exp = osc_export(obj); + struct ptlrpc_request *req; + struct ost_body *body; + struct osc_data_version_args *dva; + int rc; + + ENTRY; + memset(oa, 0, sizeof(*oa)); + oa->o_oi = loi->loi_oi; + oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; + + if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) { + oa->o_valid |= OBD_MD_FLFLAGS; + oa->o_flags |= OBD_FL_SRVLOCK; + if (dv->dv_flags & LL_DV_WR_FLUSH) + oa->o_flags |= OBD_FL_FLUSH; + } + + init_completion(&cbargs->opc_sync); + + req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR); + if (req == NULL) + RETURN(-ENOMEM); + + rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR); + if (rc < 0) { + ptlrpc_request_free(req); + RETURN(rc); + } + + body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); + lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa); + + ptlrpc_request_set_replen(req); + req->rq_interpret_reply = osc_data_version_interpret; + dva = ptlrpc_req_async_args(dva, req); + dva->dva_oio = oio; + + ptlrpcd_add_req(req); + + RETURN(0); } -static void osc_io_setattr_end(const struct lu_env *env, - const struct cl_io_slice *slice) +static void osc_io_data_version_end(const struct lu_env *env, + const struct cl_io_slice *slice) { - struct cl_io *io = slice->cis_io; - struct osc_io *oio = cl2osc_io(env, slice); - struct osc_setattr_cbargs *cbargs = &oio->oi_setattr_cbarg; - int result; - - cfs_wait_for_completion(&cbargs->opc_sync); - - result = io->ci_result = cbargs->opc_rc; - if (result == 0) { - struct cl_object *obj = slice->cis_obj; - if (oio->oi_lockless) { - /* lockless truncate */ - struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev); - - LASSERT(cl_io_is_trunc(io)); - /* XXX: Need a lock. */ - osd->od_stats.os_lockless_truncates++; - } - } + struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version; + struct osc_io *oio = cl2osc_io(env, slice); + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + + ENTRY; + wait_for_completion(&cbargs->opc_sync); + + if (cbargs->opc_rc != 0) { + slice->cis_io->ci_result = cbargs->opc_rc; + } else { + slice->cis_io->ci_result = 0; + if (!(oio->oi_oa.o_valid & + (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION))) + slice->cis_io->ci_result = -ENOTSUPP; + + if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION) + dv->dv_layout_version = oio->oi_oa.o_layout_version; + if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION) + dv->dv_data_version = oio->oi_oa.o_data_version; + } + + EXIT; } -static int osc_io_read_start(const struct lu_env *env, - const struct cl_io_slice *slice) +int osc_io_read_start(const struct lu_env *env, + const struct cl_io_slice *slice) { - struct osc_io *oio = cl2osc_io(env, slice); - struct cl_object *obj = slice->cis_obj; - struct cl_attr *attr = &osc_env_info(env)->oti_attr; - int result = 0; - ENTRY; - - if (oio->oi_lockless == 0) { - cl_object_attr_lock(obj); - result = cl_object_attr_get(env, obj, attr); - if (result == 0) { - attr->cat_atime = LTIME_S(CFS_CURRENT_TIME); - result = cl_object_attr_set(env, obj, attr, - CAT_ATIME); - } - cl_object_attr_unlock(obj); - } - RETURN(result); + struct cl_object *obj = slice->cis_obj; + struct cl_attr *attr = &osc_env_info(env)->oti_attr; + int rc = 0; + ENTRY; + + if (!slice->cis_io->ci_noatime) { + cl_object_attr_lock(obj); + attr->cat_atime = ktime_get_real_seconds(); + rc = cl_object_attr_update(env, obj, attr, CAT_ATIME); + cl_object_attr_unlock(obj); + } + + RETURN(rc); } +EXPORT_SYMBOL(osc_io_read_start); -static int osc_io_write_start(const struct lu_env *env, - const struct cl_io_slice *slice) +int osc_io_write_start(const struct lu_env *env, + const struct cl_io_slice *slice) { - struct osc_io *oio = cl2osc_io(env, slice); - struct cl_object *obj = slice->cis_obj; - struct cl_attr *attr = &osc_env_info(env)->oti_attr; - int result = 0; - ENTRY; - - if (oio->oi_lockless == 0) { - cl_object_attr_lock(obj); - result = cl_object_attr_get(env, obj, attr); - if (result == 0) { - attr->cat_mtime = attr->cat_ctime = - LTIME_S(CFS_CURRENT_TIME); - result = cl_object_attr_set(env, obj, attr, - CAT_MTIME | CAT_CTIME); - } - cl_object_attr_unlock(obj); - } - RETURN(result); + struct cl_object *obj = slice->cis_obj; + struct cl_attr *attr = &osc_env_info(env)->oti_attr; + int rc = 0; + ENTRY; + + OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_SETTIME, 1); + cl_object_attr_lock(obj); + attr->cat_mtime = attr->cat_ctime = ktime_get_real_seconds(); + rc = cl_object_attr_update(env, obj, attr, CAT_MTIME | CAT_CTIME); + cl_object_attr_unlock(obj); + + RETURN(rc); } +EXPORT_SYMBOL(osc_io_write_start); -static const struct cl_io_operations osc_io_ops = { - .op = { - [CIT_READ] = { - .cio_start = osc_io_read_start, - .cio_fini = osc_io_fini - }, - [CIT_WRITE] = { - .cio_start = osc_io_write_start, - .cio_fini = osc_io_fini - }, - [CIT_SETATTR] = { - .cio_start = osc_io_setattr_start, - .cio_end = osc_io_setattr_end - }, - [CIT_FAULT] = { - .cio_fini = osc_io_fini, - .cio_start = osc_io_fault_start - }, - [CIT_MISC] = { - .cio_fini = osc_io_fini - } - }, - .req_op = { - [CRT_READ] = { - .cio_submit = osc_io_submit - }, - [CRT_WRITE] = { - .cio_submit = osc_io_submit - } - }, - .cio_prepare_write = osc_io_prepare_write, - .cio_commit_write = osc_io_commit_write -}; +int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj, + struct cl_fsync_io *fio) +{ + struct osc_io *oio = osc_env_io(env); + struct obdo *oa = &oio->oi_oa; + struct lov_oinfo *loi = obj->oo_oinfo; + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + int rc = 0; + ENTRY; -/***************************************************************************** - * - * Transfer operations. - * - */ + memset(oa, 0, sizeof(*oa)); + oa->o_oi = loi->loi_oi; + oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; + + /* reload size abd blocks for start and end of sync range */ + oa->o_size = fio->fi_start; + oa->o_blocks = fio->fi_end; + oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + + obdo_set_parent_fid(oa, fio->fi_fid); + + init_completion(&cbargs->opc_sync); -static int osc_req_prep(const struct lu_env *env, - const struct cl_req_slice *slice) + rc = osc_sync_base(obj, oa, osc_async_upcall, cbargs, PTLRPCD_SET); + RETURN(rc); +} +EXPORT_SYMBOL(osc_fsync_ost); + +int osc_io_fsync_start(const struct lu_env *env, + const struct cl_io_slice *slice) { - return 0; + struct cl_io *io = slice->cis_io; + struct cl_fsync_io *fio = &io->u.ci_fsync; + struct cl_object *obj = slice->cis_obj; + struct osc_object *osc = cl2osc(obj); + pgoff_t start = cl_index(obj, fio->fi_start); + pgoff_t end = cl_index(obj, fio->fi_end); + int result = 0; + ENTRY; + + if (fio->fi_end == OBD_OBJECT_EOF) + end = CL_PAGE_EOF; + + result = osc_cache_writeback_range(env, osc, start, end, 0, + fio->fi_mode == CL_FSYNC_DISCARD); + if (result > 0) { + fio->fi_nr_written += result; + result = 0; + } + if (fio->fi_mode == CL_FSYNC_ALL) { + int rc; + + /* we have to wait for writeback to finish before we can + * send OST_SYNC RPC. This is bad because it causes extents + * to be written osc by osc. However, we usually start + * writeback before CL_FSYNC_ALL so this won't have any real + * problem. */ + rc = osc_cache_wait_range(env, osc, start, end); + if (result == 0) + result = rc; + rc = osc_fsync_ost(env, osc, fio); + if (result == 0) + result = rc; + } + + RETURN(result); } -static void osc_req_completion(const struct lu_env *env, - const struct cl_req_slice *slice, int ioret) +void osc_io_fsync_end(const struct lu_env *env, + const struct cl_io_slice *slice) { - struct osc_req *or; + struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync; + struct cl_object *obj = slice->cis_obj; + pgoff_t start = cl_index(obj, fio->fi_start); + pgoff_t end = cl_index(obj, fio->fi_end); + int result = 0; + + if (fio->fi_mode == CL_FSYNC_LOCAL) { + result = osc_cache_wait_range(env, cl2osc(obj), start, end); + } else if (fio->fi_mode == CL_FSYNC_ALL) { + struct osc_io *oio = cl2osc_io(env, slice); + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + + wait_for_completion(&cbargs->opc_sync); + if (result == 0) + result = cbargs->opc_rc; + } + slice->cis_io->ci_result = result; +} +EXPORT_SYMBOL(osc_io_fsync_end); - or = cl2osc_req(slice); - OBD_SLAB_FREE_PTR(or, osc_req_kmem); +static int osc_io_ladvise_start(const struct lu_env *env, + const struct cl_io_slice *slice) +{ + int result = 0; + struct cl_io *io = slice->cis_io; + struct osc_io *oio = cl2osc_io(env, slice); + struct cl_object *obj = slice->cis_obj; + struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo; + struct cl_ladvise_io *lio = &io->u.ci_ladvise; + struct obdo *oa = &oio->oi_oa; + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + struct lu_ladvise *ladvise; + struct ladvise_hdr *ladvise_hdr; + int buf_size; + int num_advise = 1; + ENTRY; + + /* TODO: add multiple ladvise support in CLIO */ + buf_size = offsetof(typeof(*ladvise_hdr), lah_advise[num_advise]); + if (osc_env_info(env)->oti_ladvise_buf.lb_len < buf_size) + lu_buf_realloc(&osc_env_info(env)->oti_ladvise_buf, buf_size); + + ladvise_hdr = osc_env_info(env)->oti_ladvise_buf.lb_buf; + if (ladvise_hdr == NULL) + RETURN(-ENOMEM); + + memset(ladvise_hdr, 0, buf_size); + ladvise_hdr->lah_magic = LADVISE_MAGIC; + ladvise_hdr->lah_count = num_advise; + ladvise_hdr->lah_flags = lio->li_flags; + + memset(oa, 0, sizeof(*oa)); + oa->o_oi = loi->loi_oi; + oa->o_valid = OBD_MD_FLID; + obdo_set_parent_fid(oa, lio->li_fid); + + ladvise = ladvise_hdr->lah_advise; + ladvise->lla_start = lio->li_start; + ladvise->lla_end = lio->li_end; + ladvise->lla_advice = lio->li_advice; + + if (lio->li_flags & LF_ASYNC) { + result = osc_ladvise_base(osc_export(cl2osc(obj)), oa, + ladvise_hdr, NULL, NULL, NULL); + } else { + init_completion(&cbargs->opc_sync); + result = osc_ladvise_base(osc_export(cl2osc(obj)), oa, + ladvise_hdr, osc_async_upcall, + cbargs, PTLRPCD_SET); + cbargs->opc_rpc_sent = result == 0; + } + RETURN(result); } -/** - * Implementation of struct cl_req_operations::cro_attr_set() for osc - * layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq - * fields. - */ -static void osc_req_attr_set(const struct lu_env *env, - const struct cl_req_slice *slice, - const struct cl_object *obj, - struct cl_req_attr *attr, obd_valid flags) +static void osc_io_ladvise_end(const struct lu_env *env, + const struct cl_io_slice *slice) { - struct lov_oinfo *oinfo; - struct cl_req *clerq; - struct cl_page *apage; /* _some_ page in @clerq */ - struct cl_lock *lock; /* _some_ lock protecting @apage */ - struct osc_lock *olck; - struct osc_page *opg; - struct obdo *oa; - - oa = attr->cra_oa; - oinfo = cl2osc(obj)->oo_oinfo; - if (flags & OBD_MD_FLID) { - oa->o_id = oinfo->loi_id; - oa->o_valid |= OBD_MD_FLID; - } - if (flags & OBD_MD_FLGROUP) { - oa->o_seq = oinfo->loi_seq; - oa->o_valid |= OBD_MD_FLGROUP; - } - if (flags & OBD_MD_FLHANDLE) { - clerq = slice->crs_req; - LASSERT(!cfs_list_empty(&clerq->crq_pages)); - apage = container_of(clerq->crq_pages.next, - struct cl_page, cp_flight); - opg = osc_cl_page_osc(apage); - apage = opg->ops_cl.cpl_page; /* now apage is a sub-page */ - lock = cl_lock_at_page(env, apage->cp_obj, apage, NULL, 1, 1); - if (lock == NULL) { - struct cl_object_header *head; - struct cl_lock *scan; - - head = cl_object_header(apage->cp_obj); - cfs_list_for_each_entry(scan, &head->coh_locks, - cll_linkage) - CL_LOCK_DEBUG(D_ERROR, env, scan, - "no cover page!\n"); - CL_PAGE_DEBUG(D_ERROR, env, apage, - "dump uncover page!\n"); - libcfs_debug_dumpstack(NULL); - LBUG(); - } + struct cl_io *io = slice->cis_io; + struct osc_io *oio = cl2osc_io(env, slice); + struct osc_async_cbargs *cbargs = &oio->oi_cbarg; + int result = 0; + struct cl_ladvise_io *lio = &io->u.ci_ladvise; + + if ((!(lio->li_flags & LF_ASYNC)) && cbargs->opc_rpc_sent) { + wait_for_completion(&cbargs->opc_sync); + result = cbargs->opc_rc; + } + slice->cis_io->ci_result = result; +} - olck = osc_lock_at(lock); - LASSERT(olck != NULL); - LASSERT(ergo(opg->ops_srvlock, olck->ols_lock == NULL)); - /* check for lockless io. */ - if (olck->ols_lock != NULL) { - oa->o_handle = olck->ols_lock->l_remote_handle; - oa->o_valid |= OBD_MD_FLHANDLE; - } - cl_lock_put(env, lock); - } +void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice) +{ + struct osc_io *oio = cl2osc_io(env, slice); + + if (oio->oi_active) { + osc_extent_release(env, oio->oi_active); + oio->oi_active = NULL; + } } +EXPORT_SYMBOL(osc_io_end); -static const struct cl_req_operations osc_req_ops = { - .cro_prep = osc_req_prep, - .cro_attr_set = osc_req_attr_set, - .cro_completion = osc_req_completion +static const struct cl_io_operations osc_io_ops = { + .op = { + [CIT_READ] = { + .cio_iter_init = osc_io_rw_iter_init, + .cio_iter_fini = osc_io_rw_iter_fini, + .cio_start = osc_io_read_start, + .cio_fini = osc_io_fini + }, + [CIT_WRITE] = { + .cio_iter_init = osc_io_rw_iter_init, + .cio_iter_fini = osc_io_rw_iter_fini, + .cio_start = osc_io_write_start, + .cio_end = osc_io_end, + .cio_fini = osc_io_fini + }, + [CIT_SETATTR] = { + .cio_iter_init = osc_io_iter_init, + .cio_iter_fini = osc_io_iter_fini, + .cio_start = osc_io_setattr_start, + .cio_end = osc_io_setattr_end + }, + [CIT_DATA_VERSION] = { + .cio_start = osc_io_data_version_start, + .cio_end = osc_io_data_version_end, + }, + [CIT_FAULT] = { + .cio_iter_init = osc_io_iter_init, + .cio_iter_fini = osc_io_iter_fini, + .cio_start = osc_io_fault_start, + .cio_end = osc_io_end, + .cio_fini = osc_io_fini + }, + [CIT_FSYNC] = { + .cio_start = osc_io_fsync_start, + .cio_end = osc_io_fsync_end, + .cio_fini = osc_io_fini + }, + [CIT_LADVISE] = { + .cio_start = osc_io_ladvise_start, + .cio_end = osc_io_ladvise_end, + .cio_fini = osc_io_fini + }, + [CIT_MISC] = { + .cio_fini = osc_io_fini + } + }, + .cio_read_ahead = osc_io_read_ahead, + .cio_submit = osc_io_submit, + .cio_commit_async = osc_io_commit_async }; +/***************************************************************************** + * + * Transfer operations. + * + */ int osc_io_init(const struct lu_env *env, struct cl_object *obj, struct cl_io *io) @@ -713,19 +1120,4 @@ int osc_io_init(const struct lu_env *env, return 0; } -int osc_req_init(const struct lu_env *env, struct cl_device *dev, - struct cl_req *req) -{ - struct osc_req *or; - int result; - - OBD_SLAB_ALLOC_PTR_GFP(or, osc_req_kmem, CFS_ALLOC_IO); - if (or != NULL) { - cl_req_slice_add(req, &or->or_cl, dev, &osc_req_ops); - result = 0; - } else - result = -ENOMEM; - return result; -} - /** @} osc */