X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosc%2Fosc_io.c;h=5778a4bca2c6758d6b00c493a02187836cd1b86c;hb=a77dd45edb772259813bcb44ba7bf7a2bf4a35ed;hp=bb5b6dd77a2dfeebf5250b2fb31c6d51e4a62484;hpb=1d8e1b31d547bb2fdaf2649df68256ede1f363af;p=fs%2Flustre-release.git diff --git a/lustre/osc/osc_io.c b/lustre/osc/osc_io.c index bb5b6dd..5778a4b 100644 --- a/lustre/osc/osc_io.c +++ b/lustre/osc/osc_io.c @@ -67,14 +67,18 @@ static struct osc_io *cl2osc_io(const struct lu_env *env, return oio; } -static struct osc_page *osc_cl_page_osc(struct cl_page *page) +static struct osc_page *osc_cl_page_osc(struct cl_page *page, + struct osc_object *osc) { - const struct cl_page_slice *slice; + const struct cl_page_slice *slice; - slice = cl_page_at(page, &osc_device_type); - LASSERT(slice != NULL); + if (osc != NULL) + slice = cl_object_page_slice(&osc->oo_cl, page); + else + slice = cl_page_at(page, &osc_device_type); + LASSERT(slice != NULL); - return cl2osc_page(slice); + return cl2osc_page(slice); } @@ -137,8 +141,8 @@ static int osc_io_submit(const struct lu_env *env, io = page->cp_owner; LASSERT(io != NULL); - opg = osc_cl_page_osc(page); - oap = &opg->ops_oap; + opg = osc_cl_page_osc(page, osc); + oap = &opg->ops_oap; LASSERT(osc == oap->oap_obj); if (!cfs_list_empty(&oap->oap_pending_item) || @@ -187,6 +191,13 @@ static int osc_io_submit(const struct lu_env *env, return qout->pl_nr > 0 ? 0 : result; } +/** + * 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). + * + * Expand stripe KMS if necessary. + */ static void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj, pgoff_t idx, unsigned to) { @@ -210,7 +221,8 @@ static void osc_page_touch_at(const struct lu_env *env, kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms, loi->loi_lvb.lvb_size); - valid = 0; + attr->cat_mtime = attr->cat_ctime = LTIME_S(CFS_CURRENT_TIME); + valid = CAT_MTIME | CAT_CTIME; if (kms > loi->loi_kms) { attr->cat_kms = kms; valid |= CAT_KMS; @@ -223,92 +235,77 @@ static void osc_page_touch_at(const struct lu_env *env, cl_object_attr_unlock(obj); } -/** - * 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. - * - * Expand stripe KMS if necessary. - */ -static void osc_page_touch(const struct lu_env *env, - struct osc_page *opage, unsigned to) +static 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_page *page = opage->ops_cl.cpl_page; - struct cl_object *obj = opage->ops_cl.cpl_obj; + 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; + int result = 0; + ENTRY; - osc_page_touch_at(env, obj, page->cp_index, to); -} + LASSERT(qin->pl_nr > 0); -/** - * 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) -{ - struct osc_device *dev = lu2osc_dev(slice->cpl_obj->co_lu.lo_dev); - struct obd_import *imp = class_exp2cliimp(dev->od_exp); - struct osc_io *oio = cl2osc_io(env, ios); - int result = 0; - ENTRY; + /* 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); + } + } - /* - * This implements OBD_BRW_CHECK logic from old client. - */ + while (qin->pl_nr > 0) { + struct osc_async_page *oap; - if (imp == NULL || imp->imp_invalid) - result = -EIO; - if (result == 0 && oio->oi_lockless) - /* this page contains `invalid' data, but who cares? - * nobody can access the invalid data. - * in osc_io_commit_write(), we're going to write exact - * [from, to) bytes of this page to OST. -jay */ - cl_page_export(env, slice->cpl_page, 1); + page = cl_page_list_first(qin); + opg = osc_cl_page_osc(page, osc); + oap = &opg->ops_oap; - RETURN(result); -} + if (!cfs_list_empty(&oap->oap_rpc_item)) { + CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n", + oap, opg); + result = -EBUSY; + break; + } -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) -{ - struct osc_io *oio = cl2osc_io(env, 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; + /* The page may be already in dirty cache. */ + if (cfs_list_empty(&oap->oap_pending_item)) { + result = osc_page_cache_add(env, &opg->ops_cl, io); + if (result != 0) + break; + } - 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; + osc_page_touch_at(env, osc2cl(osc), osc_index(opg), + page == last_page ? to : PAGE_SIZE); - if (oio->oi_lockless) - /* see osc_io_prepare_write() for lockless io handling. */ - cl_page_clip(env, slice->cpl_page, from, to); + cl_page_list_del(env, qin, page); - RETURN(0); + (*cb)(env, io, page); + /* Can't access page 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); } static int osc_io_rw_iter_init(const struct lu_env *env, @@ -334,15 +331,15 @@ static int osc_io_rw_iter_init(const struct lu_env *env, if (npages > max_pages) npages = max_pages; - c = cfs_atomic_read(cli->cl_lru_left); + c = atomic_read(cli->cl_lru_left); if (c < npages && osc_lru_reclaim(cli) > 0) - c = cfs_atomic_read(cli->cl_lru_left); + c = atomic_read(cli->cl_lru_left); while (c >= npages) { - if (c == cfs_atomic_cmpxchg(cli->cl_lru_left, c, c - npages)) { + if (c == atomic_cmpxchg(cli->cl_lru_left, c, c - npages)) { oio->oi_lru_reserved = npages; break; } - c = cfs_atomic_read(cli->cl_lru_left); + c = atomic_read(cli->cl_lru_left); } RETURN(0); @@ -356,7 +353,7 @@ static void osc_io_rw_iter_fini(const struct lu_env *env, struct client_obd *cli = osc_cli(osc); if (oio->oi_lru_reserved > 0) { - cfs_atomic_add(oio->oi_lru_reserved, cli->cl_lru_left); + atomic_add(oio->oi_lru_reserved, cli->cl_lru_left); oio->oi_lru_reserved = 0; } } @@ -398,31 +395,22 @@ static int osc_async_upcall(void *a, int rc) * Checks that there are no pages being written in the extent being truncated. */ static int trunc_check_cb(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, void *cbdata) + struct osc_page *ops , void *cbdata) { - const struct cl_page_slice *slice; - struct osc_page *ops; + struct cl_page *page = ops->ops_cl.cpl_page; struct osc_async_page *oap; __u64 start = *(__u64 *)cbdata; - slice = cl_page_at(page, &osc_device_type); - LASSERT(slice != NULL); - ops = cl2osc_page(slice); oap = &ops->ops_oap; - if (oap->oap_cmd & OBD_BRW_WRITE && !cfs_list_empty(&oap->oap_pending_item)) CL_PAGE_DEBUG(D_ERROR, env, page, "exists " LPU64 "/%s.\n", start, current->comm); #ifdef __linux__ - { - struct page *vmpage = cl_page_vmpage(env, page); - if (PageLocked(vmpage)) - CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n", - ops, page->cp_index, - (oap->oap_cmd & OBD_BRW_RWMASK)); - } + 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); #endif return CLP_GANG_OKAY; @@ -442,8 +430,9 @@ static void osc_trunc_check(const struct lu_env *env, struct cl_io *io, /* * Complain if there are pages in the truncated region. */ - cl_page_gang_lookup(env, clob, io, start + partial, CL_PAGE_EOF, - trunc_check_cb, (void *)&size); + osc_page_gang_lookup(env, io, cl2osc(clob), + start + partial, CL_PAGE_EOF, + trunc_check_cb, (void *)&size); } #else /* __KERNEL__ */ static void osc_trunc_check(const struct lu_env *env, struct cl_io *io, @@ -721,23 +710,23 @@ static void osc_io_end(const struct lu_env *env, } 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] = { + .op = { + [CIT_READ] = { + .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_start = osc_io_write_start, .cio_end = osc_io_end, - .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 + }, + [CIT_SETATTR] = { + .cio_start = osc_io_setattr_start, + .cio_end = osc_io_setattr_end + }, + [CIT_FAULT] = { .cio_start = osc_io_fault_start, .cio_end = osc_io_end, .cio_fini = osc_io_fini @@ -747,20 +736,12 @@ static const struct cl_io_operations osc_io_ops = { .cio_end = osc_io_fsync_end, .cio_fini = osc_io_fini }, - [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 + [CIT_MISC] = { + .cio_fini = osc_io_fini + } + }, + .cio_submit = osc_io_submit, + .cio_commit_async = osc_io_commit_async }; /***************************************************************************** @@ -828,18 +809,21 @@ static void osc_req_attr_set(const struct lu_env *env, oa->o_valid |= OBD_MD_FLID; } 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); + struct cl_object *subobj; + + 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, NULL); + subobj = opg->ops_cl.cpl_obj; + lock = cl_lock_at_pgoff(env, subobj, osc_index(opg), + NULL, 1, 1); + if (lock == NULL) { + struct cl_object_header *head; + struct cl_lock *scan; + + head = cl_object_header(subobj); cfs_list_for_each_entry(scan, &head->coh_locks, cll_linkage) CL_LOCK_DEBUG(D_ERROR, env, scan, @@ -880,18 +864,18 @@ int osc_io_init(const struct lu_env *env, } int osc_req_init(const struct lu_env *env, struct cl_device *dev, - struct cl_req *req) + struct cl_req *req) { - struct osc_req *or; - int result; - - OBD_SLAB_ALLOC_PTR_GFP(or, osc_req_kmem, __GFP_IO); - if (or != NULL) { - cl_req_slice_add(req, &or->or_cl, dev, &osc_req_ops); - result = 0; - } else - result = -ENOMEM; - return result; + struct osc_req *or; + int result; + + OBD_SLAB_ALLOC_PTR_GFP(or, osc_req_kmem, GFP_NOFS); + if (or != NULL) { + cl_req_slice_add(req, &or->or_cl, dev, &osc_req_ops); + result = 0; + } else + result = -ENOMEM; + return result; } /** @} osc */