X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fcl_page.c;h=1d1c375fc83850f74348c762a4d7f8467f7fc554;hp=67ca346cb4b6c3cec62bf162aa415ba0d3b4b62d;hb=HEAD;hpb=5cae09a2409dcd396a1ee7be1eca7d3bbf77365e;ds=sidebyside diff --git a/lustre/obdclass/cl_page.c b/lustre/obdclass/cl_page.c index 67ca346..1efb083 100644 --- a/lustre/obdclass/cl_page.c +++ b/lustre/obdclass/cl_page.c @@ -15,11 +15,7 @@ * * 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 */ @@ -27,95 +23,87 @@ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * Client Lustre Page. * * Author: Nikita Danilov + * Author: Jinshan Xiong */ #define DEBUG_SUBSYSTEM S_CLASS +#include #include #include #include -#include #include #include "cl_internal.h" -static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, - int radix); - -static cfs_mem_cache_t *cl_page_kmem = NULL; - -static struct lu_kmem_descr cl_page_caches[] = { - { - .ckd_cache = &cl_page_kmem, - .ckd_name = "cl_page_kmem", - .ckd_size = sizeof (struct cl_page) - }, - { - .ckd_cache = NULL - } -}; +static void __cl_page_delete(const struct lu_env *env, struct cl_page *pg); +static DEFINE_MUTEX(cl_page_kmem_mutex); #ifdef LIBCFS_DEBUG # define PASSERT(env, page, expr) \ - do { \ - if (unlikely(!(expr))) { \ - CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \ - LASSERT(0); \ - } \ - } while (0) +do { \ + if (unlikely(!(expr))) { \ + CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \ + LASSERT(0); \ + } \ +} while (0) #else /* !LIBCFS_DEBUG */ -# define PASSERT(env, page, exp) \ - ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp)) +#define PASSERT(env, page, exp) \ + ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp)) #endif /* !LIBCFS_DEBUG */ -#ifdef INVARIANT_CHECK +#ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK # define PINVRNT(env, page, expr) \ - do { \ - if (unlikely(!(expr))) { \ - CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \ - LINVRNT(0); \ - } \ - } while (0) -#else /* !INVARIANT_CHECK */ +do { \ + if (unlikely(!(expr))) { \ + CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \ + LINVRNT(0); \ + } \ +} while (0) +#else /* !CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK */ # define PINVRNT(env, page, exp) \ - ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp)) -#endif /* !INVARIANT_CHECK */ + ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp)) +#endif /* !CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK */ /* Disable page statistic by default due to huge performance penalty. */ +static void cs_page_inc(const struct cl_object *obj, + enum cache_stats_item item) +{ #ifdef CONFIG_DEBUG_PAGESTATE_TRACKING -#define CS_PAGE_INC(o, item) \ - cfs_atomic_inc(&cl_object_site(o)->cs_pages.cs_stats[CS_##item]) -#define CS_PAGE_DEC(o, item) \ - cfs_atomic_dec(&cl_object_site(o)->cs_pages.cs_stats[CS_##item]) -#define CS_PAGESTATE_INC(o, state) \ - cfs_atomic_inc(&cl_object_site(o)->cs_pages_state[state]) -#define CS_PAGESTATE_DEC(o, state) \ - cfs_atomic_dec(&cl_object_site(o)->cs_pages_state[state]) -#else -#define CS_PAGE_INC(o, item) -#define CS_PAGE_DEC(o, item) -#define CS_PAGESTATE_INC(o, state) -#define CS_PAGESTATE_DEC(o, state) + atomic_inc(&cl_object_site(obj)->cs_pages.cs_stats[item]); #endif +} -/** - * Internal version of cl_page_top, it should be called with page referenced, - * or cp_lock held. - */ -static struct cl_page *cl_page_top_trusted(struct cl_page *page) +static void cs_page_dec(const struct cl_object *obj, + enum cache_stats_item item) +{ +#ifdef CONFIG_DEBUG_PAGESTATE_TRACKING + atomic_dec(&cl_object_site(obj)->cs_pages.cs_stats[item]); +#endif +} + +static void cs_pagestate_inc(const struct cl_object *obj, + enum cl_page_state state) { - while (page->cp_parent != NULL) - page = page->cp_parent; - return page; +#ifdef CONFIG_DEBUG_PAGESTATE_TRACKING + atomic_inc(&cl_object_site(obj)->cs_pages_state[state]); +#endif +} + +static void cs_pagestate_dec(const struct cl_object *obj, + enum cl_page_state state) +{ +#ifdef CONFIG_DEBUG_PAGESTATE_TRACKING + atomic_dec(&cl_object_site(obj)->cs_pages_state[state]); +#endif } /** @@ -123,269 +111,205 @@ static struct cl_page *cl_page_top_trusted(struct cl_page *page) * * This function can be used to obtain initial reference to previously * unreferenced cached object. It can be called only if concurrent page - * reclamation is somehow prevented, e.g., by locking page radix-tree - * (cl_object_header::hdr->coh_page_guard), or by keeping a lock on a VM page, + * reclamation is somehow prevented, e.g., by keeping a lock on a VM page, * associated with \a page. * * Use with care! Not exported. */ static void cl_page_get_trust(struct cl_page *page) { - /* - * Checkless version for trusted users. - */ - if (cfs_atomic_inc_return(&page->cp_ref) == 1) - CS_PAGE_INC(page->cp_obj, busy); + LASSERT(refcount_read(&page->cp_ref) > 0); + refcount_inc(&page->cp_ref); } -/** - * Returns a slice within a page, corresponding to the given layer in the - * device stack. - * - * \see cl_lock_at() - */ -static const struct cl_page_slice * -cl_page_at_trusted(const struct cl_page *page, - const struct lu_device_type *dtype) +static struct cl_page_slice * +cl_page_slice_get(const struct cl_page *cl_page, int index) { - const struct cl_page_slice *slice; - -#ifdef INVARIANT_CHECK - if (!cfs_atomic_read(&page->cp_ref)) - LASSERT_SPIN_LOCKED(&page->cp_lock); -#endif - ENTRY; - - page = cl_page_top_trusted((struct cl_page *)page); - do { - cfs_list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { - if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype) - RETURN(slice); - } - page = page->cp_child; - } while (page != NULL); - RETURN(NULL); + if (index < 0 || index >= cl_page->cp_layer_count) + return NULL; + + /* To get the cp_layer_offset values fit under 256 bytes, we + * use the offset beyond the end of struct cl_page. + */ + return (struct cl_page_slice *)((char *)cl_page + sizeof(*cl_page) + + cl_page->cp_layer_offset[index]); } -/** - * Returns a page with given index in the given object, or NULL if no page is - * found. Acquires a reference on \a page. - * - * Locking: called under cl_object_header::coh_page_guard spin-lock. - */ -struct cl_page *cl_page_lookup(struct cl_object_header *hdr, pgoff_t index) -{ - struct cl_page *page; +#define cl_page_slice_for_each(cl_page, slice, i) \ + for (i = 0, slice = cl_page_slice_get(cl_page, 0); \ + i < (cl_page)->cp_layer_count; \ + slice = cl_page_slice_get(cl_page, ++i)) - LASSERT_SPIN_LOCKED(&hdr->coh_page_guard); +#define cl_page_slice_for_each_reverse(cl_page, slice, i) \ + for (i = (cl_page)->cp_layer_count - 1, \ + slice = cl_page_slice_get(cl_page, i); i >= 0; \ + slice = cl_page_slice_get(cl_page, --i)) - page = radix_tree_lookup(&hdr->coh_tree, index); - if (page != NULL) { - cl_page_get_trust(page); - } - return page; -} -EXPORT_SYMBOL(cl_page_lookup); - -/** - * Returns a list of pages by a given [start, end] of \a obj. - * - * \param resched If not NULL, then we give up before hogging CPU for too - * long and set *resched = 1, in that case caller should implement a retry - * logic. - * - * Gang tree lookup (radix_tree_gang_lookup()) optimization is absolutely - * crucial in the face of [offset, EOF] locks. - * - * Return at least one page in @queue unless there is no covered page. - */ -int cl_page_gang_lookup(const struct lu_env *env, struct cl_object *obj, - struct cl_io *io, pgoff_t start, pgoff_t end, - cl_page_gang_cb_t cb, void *cbdata) +static void __cl_page_free(struct cl_page *cl_page, unsigned short bufsize) { - struct cl_object_header *hdr; - struct cl_page *page; - struct cl_page **pvec; - const struct cl_page_slice *slice; - const struct lu_device_type *dtype; - pgoff_t idx; - unsigned int nr; - unsigned int i; - unsigned int j; - int res = CLP_GANG_OKAY; - int tree_lock = 1; - ENTRY; - - idx = start; - hdr = cl_object_header(obj); - pvec = cl_env_info(env)->clt_pvec; - dtype = cl_object_top(obj)->co_lu.lo_dev->ld_type; - spin_lock(&hdr->coh_page_guard); - while ((nr = radix_tree_gang_lookup(&hdr->coh_tree, (void **)pvec, - idx, CLT_PVEC_SIZE)) > 0) { - int end_of_region = 0; - idx = pvec[nr - 1]->cp_index + 1; - for (i = 0, j = 0; i < nr; ++i) { - page = pvec[i]; - pvec[i] = NULL; - - LASSERT(page->cp_type == CPT_CACHEABLE); - if (page->cp_index > end) { - end_of_region = 1; - break; - } - if (page->cp_state == CPS_FREEING) - continue; - - slice = cl_page_at_trusted(page, dtype); - /* - * Pages for lsm-less file has no underneath sub-page - * for osc, in case of ... - */ - PASSERT(env, page, slice != NULL); - - page = slice->cpl_page; - /* - * Can safely call cl_page_get_trust() under - * radix-tree spin-lock. - * - * XXX not true, because @page is from object another - * than @hdr and protected by different tree lock. - */ - cl_page_get_trust(page); - lu_ref_add_atomic(&page->cp_reference, - "gang_lookup", cfs_current()); - pvec[j++] = page; - } - - /* - * Here a delicate locking dance is performed. Current thread - * holds a reference to a page, but has to own it before it - * can be placed into queue. Owning implies waiting, so - * radix-tree lock is to be released. After a wait one has to - * check that pages weren't truncated (cl_page_own() returns - * error in the latter case). - */ - spin_unlock(&hdr->coh_page_guard); - tree_lock = 0; - - for (i = 0; i < j; ++i) { - page = pvec[i]; - if (res == CLP_GANG_OKAY) - res = (*cb)(env, io, page, cbdata); - lu_ref_del(&page->cp_reference, - "gang_lookup", cfs_current()); - cl_page_put(env, page); - } - if (nr < CLT_PVEC_SIZE || end_of_region) - break; - - if (res == CLP_GANG_OKAY && cfs_need_resched()) - res = CLP_GANG_RESCHED; - if (res != CLP_GANG_OKAY) - break; - - spin_lock(&hdr->coh_page_guard); - tree_lock = 1; + int index = cl_page->cp_kmem_index; + + if (index >= 0) { + LASSERT(index < ARRAY_SIZE(cl_page_kmem_array)); + LASSERT(cl_page_kmem_size_array[index] == bufsize); + OBD_SLAB_FREE(cl_page, cl_page_kmem_array[index], bufsize); + } else { + OBD_FREE(cl_page, bufsize); } - if (tree_lock) - spin_unlock(&hdr->coh_page_guard); - RETURN(res); } -EXPORT_SYMBOL(cl_page_gang_lookup); -static void cl_page_free(const struct lu_env *env, struct cl_page *page) +static void cl_page_free(const struct lu_env *env, struct cl_page *cp, + struct folio_batch *fbatch) { - struct cl_object *obj = page->cp_obj; - - PASSERT(env, page, cfs_list_empty(&page->cp_batch)); - PASSERT(env, page, page->cp_owner == NULL); - PASSERT(env, page, page->cp_req == NULL); - PASSERT(env, page, page->cp_parent == NULL); - PASSERT(env, page, page->cp_state == CPS_FREEING); - - ENTRY; - cfs_might_sleep(); - while (!cfs_list_empty(&page->cp_layers)) { - struct cl_page_slice *slice; - - slice = cfs_list_entry(page->cp_layers.next, - struct cl_page_slice, cpl_linkage); - cfs_list_del_init(page->cp_layers.next); - slice->cpl_ops->cpo_fini(env, slice); - } - CS_PAGE_DEC(obj, total); - CS_PAGESTATE_DEC(obj, page->cp_state); - lu_object_ref_del_at(&obj->co_lu, page->cp_obj_ref, "cl_page", page); - cl_object_put(env, obj); - lu_ref_fini(&page->cp_reference); - OBD_SLAB_FREE_PTR(page, cl_page_kmem); - EXIT; + struct cl_object *obj = cp->cp_obj; + unsigned short bufsize = cl_object_header(obj)->coh_page_bufsize; + struct page *vmpage; + + ENTRY; + PASSERT(env, cp, list_empty(&cp->cp_batch)); + + if (cp->cp_type == CPT_CACHEABLE) { + PASSERT(env, cp, cp->cp_owner == NULL); + PASSERT(env, cp, cp->cp_state == CPS_FREEING); + /* vmpage->private was already cleared when page was + * moved into CPS_FREEING state. + */ + vmpage = cp->cp_vmpage; + LASSERT(vmpage != NULL); + LASSERT((struct cl_page *)vmpage->private != cp); + + if (fbatch != NULL) { + if (!folio_batch_add_page(fbatch, vmpage)) + folio_batch_release(fbatch); + } else { + put_page(vmpage); + } + } + + cp->cp_layer_count = 0; + cs_page_dec(obj, CS_total); + if (cp->cp_type != CPT_TRANSIENT) + cs_pagestate_dec(obj, cp->cp_state); + lu_object_ref_del_at(&obj->co_lu, &cp->cp_obj_ref, "cl_page", cp); + if (cp->cp_type != CPT_TRANSIENT) + cl_object_put(env, obj); + lu_ref_fini(&cp->cp_reference); + __cl_page_free(cp, bufsize); + EXIT; } -/** - * Helper function updating page state. This is the only place in the code - * where cl_page::cp_state field is mutated. - */ -static inline void cl_page_state_set_trust(struct cl_page *page, - enum cl_page_state state) +static struct cl_page *__cl_page_alloc(struct cl_object *o) { - /* bypass const. */ - *(enum cl_page_state *)&page->cp_state = state; + int i = 0; + struct cl_page *cl_page = NULL; + unsigned short bufsize = cl_object_header(o)->coh_page_bufsize; + + if (CFS_FAIL_CHECK(OBD_FAIL_LLITE_PAGE_ALLOC)) + return NULL; + +check: + /* the number of entries in cl_page_kmem_array is expected to + * only be 2-3 entries, so the lookup overhead should be low. + */ + for ( ; i < ARRAY_SIZE(cl_page_kmem_array); i++) { + if (smp_load_acquire(&cl_page_kmem_size_array[i]) == bufsize) { + OBD_SLAB_ALLOC_GFP(cl_page, cl_page_kmem_array[i], + bufsize, GFP_NOFS); + if (cl_page) + cl_page->cp_kmem_index = i; + return cl_page; + } + if (cl_page_kmem_size_array[i] == 0) + break; + } + + if (i < ARRAY_SIZE(cl_page_kmem_array)) { + char cache_name[32]; + + mutex_lock(&cl_page_kmem_mutex); + if (cl_page_kmem_size_array[i]) { + mutex_unlock(&cl_page_kmem_mutex); + goto check; + } + snprintf(cache_name, sizeof(cache_name), + "cl_page_kmem-%u", bufsize); + cl_page_kmem_array[i] = + kmem_cache_create(cache_name, bufsize, + 0, 0, NULL); + if (cl_page_kmem_array[i] == NULL) { + mutex_unlock(&cl_page_kmem_mutex); + return NULL; + } + smp_store_release(&cl_page_kmem_size_array[i], bufsize); + mutex_unlock(&cl_page_kmem_mutex); + goto check; + } else { + OBD_ALLOC_GFP(cl_page, bufsize, GFP_NOFS); + if (cl_page) + cl_page->cp_kmem_index = -1; + } + + return cl_page; } -static int cl_page_alloc(const struct lu_env *env, struct cl_object *o, - pgoff_t ind, struct page *vmpage, - enum cl_page_type type, struct cl_page **out) +struct cl_page *cl_page_alloc(const struct lu_env *env, struct cl_object *o, + pgoff_t ind, struct page *vmpage, + enum cl_page_type type) { - struct cl_page *page; - struct cl_page *err = NULL; - struct lu_object_header *head; - int result; - - ENTRY; - result = +1; - OBD_SLAB_ALLOC_PTR_GFP(page, cl_page_kmem, CFS_ALLOC_IO); - if (page != NULL) { - cfs_atomic_set(&page->cp_ref, 1); - page->cp_obj = o; - cl_object_get(o); - page->cp_obj_ref = lu_object_ref_add(&o->co_lu, - "cl_page", page); - page->cp_index = ind; - cl_page_state_set_trust(page, CPS_CACHED); - spin_lock_init(&page->cp_lock); - page->cp_type = type; - CFS_INIT_LIST_HEAD(&page->cp_layers); - CFS_INIT_LIST_HEAD(&page->cp_batch); - CFS_INIT_LIST_HEAD(&page->cp_flight); - mutex_init(&page->cp_mutex); - lu_ref_init(&page->cp_reference); - head = o->co_lu.lo_header; - cfs_list_for_each_entry(o, &head->loh_layers, - co_lu.lo_linkage) { - if (o->co_ops->coo_page_init != NULL) { - err = o->co_ops->coo_page_init(env, o, - page, vmpage); - if (err != NULL) { - cl_page_delete0(env, page, 0); - cl_page_free(env, page); - page = err; - break; - } - } - } - if (err == NULL) { - CS_PAGE_INC(o, busy); - CS_PAGE_INC(o, total); - CS_PAGE_INC(o, create); - CS_PAGESTATE_DEC(o, CPS_CACHED); - result = 0; - } - } else - page = ERR_PTR(-ENOMEM); - *out = page; - RETURN(result); + struct cl_page *cl_page; + struct cl_object *head; + + ENTRY; + + cl_page = __cl_page_alloc(o); + if (cl_page != NULL) { + int result = 0; + + /* Please fix cl_page:cp_state/type declaration if + * these assertions fail in the future. + */ + BUILD_BUG_ON((1 << CP_STATE_BITS) < CPS_NR); /* cp_state */ + BUILD_BUG_ON((1 << CP_TYPE_BITS) < CPT_NR); /* cp_type */ + refcount_set(&cl_page->cp_ref, 1); + cl_page->cp_obj = o; + if (type != CPT_TRANSIENT) + cl_object_get(o); + lu_object_ref_add_at(&o->co_lu, &cl_page->cp_obj_ref, + "cl_page", cl_page); + cl_page->cp_vmpage = vmpage; + if (cl_page->cp_type != CPT_TRANSIENT) + cl_page->cp_state = CPS_CACHED; + cl_page->cp_type = type; + if (type == CPT_TRANSIENT) + /* correct inode to be added in ll_direct_rw_pages */ + cl_page->cp_inode = NULL; + else + cl_page->cp_inode = page2inode(vmpage); + INIT_LIST_HEAD(&cl_page->cp_batch); + lu_ref_init(&cl_page->cp_reference); + head = o; + cl_page->cp_page_index = ind; + cl_object_for_each(o, head) { + if (o->co_ops->coo_page_init != NULL) { + result = o->co_ops->coo_page_init(env, o, + cl_page, ind); + if (result != 0) { + __cl_page_delete(env, cl_page); + cl_page_free(env, cl_page, NULL); + cl_page = ERR_PTR(result); + break; + } + } + } + if (result == 0) { + cs_page_inc(o, CS_total); + cs_page_inc(o, CS_create); + cs_pagestate_dec(o, CPS_CACHED); + } + } else { + cl_page = ERR_PTR(-ENOMEM); + } + RETURN(cl_page); } /** @@ -399,232 +323,119 @@ static int cl_page_alloc(const struct lu_env *env, struct cl_object *o, * * \see cl_object_find(), cl_lock_find() */ -static struct cl_page *cl_page_find0(const struct lu_env *env, - struct cl_object *o, - pgoff_t idx, struct page *vmpage, - enum cl_page_type type, - struct cl_page *parent) +struct cl_page *cl_page_find(const struct lu_env *env, + struct cl_object *o, + pgoff_t idx, struct page *vmpage, + enum cl_page_type type) { - struct cl_page *page = NULL; - struct cl_page *ghost = NULL; - struct cl_object_header *hdr; - int err; - - LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT); - cfs_might_sleep(); - - ENTRY; - - hdr = cl_object_header(o); - CS_PAGE_INC(o, lookup); - - CDEBUG(D_PAGE, "%lu@"DFID" %p %lx %d\n", - idx, PFID(&hdr->coh_lu.loh_fid), vmpage, vmpage->private, type); - /* fast path. */ - if (type == CPT_CACHEABLE) { - /* cl_page::cp_lock is used to protect the page state and - * refcount, but need an external lock to protect the - * child/parent relationship, so vmpage lock must be held for - * this purpose. */ - KLASSERT(PageLocked(vmpage)); - /* - * cl_vmpage_page() can be called here without any locks as - * - * - "vmpage" is locked (which prevents ->private from - * concurrent updates), and - * - * - "o" cannot be destroyed while current thread holds a - * reference on it. - */ - page = cl_vmpage_page(vmpage, o); - PINVRNT(env, page, - ergo(page != NULL, - cl_page_vmpage(env, page) == vmpage && - (void *)radix_tree_lookup(&hdr->coh_tree, - idx) == page)); - } - - if (page != NULL) { - CS_PAGE_INC(o, hit); - RETURN(page); - } - - /* allocate and initialize cl_page */ - err = cl_page_alloc(env, o, idx, vmpage, type, &page); - if (err != 0) - RETURN(page); - - if (type == CPT_TRANSIENT) { - if (parent) { - LASSERT(page->cp_parent == NULL); - page->cp_parent = parent; - parent->cp_child = page; - } - RETURN(page); - } - - /* - * XXX optimization: use radix_tree_preload() here, and change tree - * gfp mask to GFP_KERNEL in cl_object_header_init(). - */ - spin_lock(&hdr->coh_page_guard); - err = radix_tree_insert(&hdr->coh_tree, idx, page); - if (err != 0) { - ghost = page; - /* - * Noted by Jay: a lock on \a vmpage protects cl_page_find() - * from this race, but - * - * 0. it's better to have cl_page interface "locally - * consistent" so that its correctness can be reasoned - * about without appealing to the (obscure world of) VM - * locking. - * - * 1. handling this race allows ->coh_tree to remain - * consistent even when VM locking is somehow busted, - * which is very useful during diagnosing and debugging. - */ - page = ERR_PTR(err); - CL_PAGE_DEBUG(D_ERROR, env, ghost, - "fail to insert into radix tree: %d\n", err); - } else { - if (parent) { - LASSERT(page->cp_parent == NULL); - page->cp_parent = parent; - parent->cp_child = page; - } - hdr->coh_pages++; - } - spin_unlock(&hdr->coh_page_guard); - - if (unlikely(ghost != NULL)) { - CS_PAGE_DEC(o, busy); - cl_page_delete0(env, ghost, 0); - cl_page_free(env, ghost); - } - RETURN(page); -} + struct cl_page *page = NULL; + struct cl_object_header *hdr; -struct cl_page *cl_page_find(const struct lu_env *env, struct cl_object *o, - pgoff_t idx, struct page *vmpage, - enum cl_page_type type) -{ - return cl_page_find0(env, o, idx, vmpage, type, NULL); -} -EXPORT_SYMBOL(cl_page_find); + LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT); + might_sleep(); + ENTRY; -struct cl_page *cl_page_find_sub(const struct lu_env *env, struct cl_object *o, - pgoff_t idx, struct page *vmpage, - struct cl_page *parent) -{ - return cl_page_find0(env, o, idx, vmpage, parent->cp_type, parent); + hdr = cl_object_header(o); + cs_page_inc(o, CS_lookup); + + CDEBUG(D_PAGE, "%lu@"DFID" %p %lx %d\n", + idx, PFID(&hdr->coh_lu.loh_fid), vmpage, vmpage->private, type); + /* fast path. */ + if (type == CPT_CACHEABLE) { + /* vmpage lock used to protect the child/parent relationship */ + LASSERT(PageLocked(vmpage)); + /* + * cl_vmpage_page() can be called here without any locks as + * + * - "vmpage" is locked (which prevents ->private from + * concurrent updates), and + * + * - "o" cannot be destroyed while current thread holds a + * reference on it. + */ + page = cl_vmpage_page(vmpage, o); + if (page != NULL) { + cs_page_inc(o, CS_hit); + RETURN(page); + } + } + + /* allocate and initialize cl_page */ + page = cl_page_alloc(env, o, idx, vmpage, type); + RETURN(page); } -EXPORT_SYMBOL(cl_page_find_sub); +EXPORT_SYMBOL(cl_page_find); static inline int cl_page_invariant(const struct cl_page *pg) { - struct cl_object_header *header; - struct cl_page *parent; - struct cl_page *child; - struct cl_io *owner; - - /* - * Page invariant is protected by a VM lock. - */ - LINVRNT(cl_page_is_vmlocked(NULL, pg)); - - header = cl_object_header(pg->cp_obj); - parent = pg->cp_parent; - child = pg->cp_child; - owner = pg->cp_owner; - - return cfs_atomic_read(&pg->cp_ref) > 0 && - ergo(parent != NULL, parent->cp_child == pg) && - ergo(child != NULL, child->cp_parent == pg) && - ergo(child != NULL, pg->cp_obj != child->cp_obj) && - ergo(parent != NULL, pg->cp_obj != parent->cp_obj) && - ergo(owner != NULL && parent != NULL, - parent->cp_owner == pg->cp_owner->ci_parent) && - ergo(owner != NULL && child != NULL, - child->cp_owner->ci_parent == owner) && - /* - * Either page is early in initialization (has neither child - * nor parent yet), or it is in the object radix tree. - */ - ergo(pg->cp_state < CPS_FREEING && pg->cp_type == CPT_CACHEABLE, - (void *)radix_tree_lookup(&header->coh_tree, - pg->cp_index) == pg || - (child == NULL && parent == NULL)); + return cl_page_in_use_noref(pg); } -static void cl_page_state_set0(const struct lu_env *env, - struct cl_page *page, enum cl_page_state state) +static void __cl_page_state_set(const struct lu_env *env, + struct cl_page *cl_page, + enum cl_page_state state) { - enum cl_page_state old; - - /* - * Matrix of allowed state transitions [old][new], for sanity - * checking. - */ - static const int allowed_transitions[CPS_NR][CPS_NR] = { - [CPS_CACHED] = { - [CPS_CACHED] = 0, - [CPS_OWNED] = 1, /* io finds existing cached page */ - [CPS_PAGEIN] = 0, - [CPS_PAGEOUT] = 1, /* write-out from the cache */ - [CPS_FREEING] = 1, /* eviction on the memory pressure */ - }, - [CPS_OWNED] = { - [CPS_CACHED] = 1, /* release to the cache */ - [CPS_OWNED] = 0, - [CPS_PAGEIN] = 1, /* start read immediately */ - [CPS_PAGEOUT] = 1, /* start write immediately */ - [CPS_FREEING] = 1, /* lock invalidation or truncate */ - }, - [CPS_PAGEIN] = { - [CPS_CACHED] = 1, /* io completion */ - [CPS_OWNED] = 0, - [CPS_PAGEIN] = 0, - [CPS_PAGEOUT] = 0, - [CPS_FREEING] = 0, - }, - [CPS_PAGEOUT] = { - [CPS_CACHED] = 1, /* io completion */ - [CPS_OWNED] = 0, - [CPS_PAGEIN] = 0, - [CPS_PAGEOUT] = 0, - [CPS_FREEING] = 0, - }, - [CPS_FREEING] = { - [CPS_CACHED] = 0, - [CPS_OWNED] = 0, - [CPS_PAGEIN] = 0, - [CPS_PAGEOUT] = 0, - [CPS_FREEING] = 0, - } - }; - - ENTRY; - old = page->cp_state; - PASSERT(env, page, allowed_transitions[old][state]); - CL_PAGE_HEADER(D_TRACE, env, page, "%d -> %d\n", old, state); - for (; page != NULL; page = page->cp_child) { - PASSERT(env, page, page->cp_state == old); - PASSERT(env, page, - equi(state == CPS_OWNED, page->cp_owner != NULL)); - - CS_PAGESTATE_DEC(page->cp_obj, page->cp_state); - CS_PAGESTATE_INC(page->cp_obj, state); - cl_page_state_set_trust(page, state); - } - EXIT; + enum cl_page_state old; + + /* Matrix of allowed state transitions [old][new] for sanity checking */ + static const int allowed_transitions[CPS_NR][CPS_NR] = { + [CPS_CACHED] = { + [CPS_CACHED] = 0, + [CPS_OWNED] = 1, /* io finds existing cached page */ + [CPS_PAGEIN] = 0, + [CPS_PAGEOUT] = 1, /* write-out from the cache */ + [CPS_FREEING] = 1, /* eviction on the memory pressure */ + }, + [CPS_OWNED] = { + [CPS_CACHED] = 1, /* release to the cache */ + [CPS_OWNED] = 0, + [CPS_PAGEIN] = 1, /* start read immediately */ + [CPS_PAGEOUT] = 1, /* start write immediately */ + [CPS_FREEING] = 1, /* lock invalidation or truncate */ + }, + [CPS_PAGEIN] = { + [CPS_CACHED] = 1, /* io completion */ + [CPS_OWNED] = 0, + [CPS_PAGEIN] = 0, + [CPS_PAGEOUT] = 0, + [CPS_FREEING] = 0, + }, + [CPS_PAGEOUT] = { + [CPS_CACHED] = 1, /* io completion */ + [CPS_OWNED] = 0, + [CPS_PAGEIN] = 0, + [CPS_PAGEOUT] = 0, + [CPS_FREEING] = 0, + }, + [CPS_FREEING] = { + [CPS_CACHED] = 0, + [CPS_OWNED] = 0, + [CPS_PAGEIN] = 0, + [CPS_PAGEOUT] = 0, + [CPS_FREEING] = 0, + } + }; + + ENTRY; + old = cl_page->cp_state; + PASSERT(env, cl_page, allowed_transitions[old][state]); + CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d -> %d\n", old, state); + PASSERT(env, cl_page, cl_page->cp_state == old); + PASSERT(env, cl_page, equi(state == CPS_OWNED, + cl_page->cp_owner != NULL)); + + cs_pagestate_dec(cl_page->cp_obj, cl_page->cp_state); + cs_pagestate_inc(cl_page->cp_obj, state); + cl_page->cp_state = state; + EXIT; } static void cl_page_state_set(const struct lu_env *env, - struct cl_page *page, enum cl_page_state state) + struct cl_page *page, enum cl_page_state state) { - cl_page_state_set0(env, page, state); + LASSERT(page->cp_type != CPT_TRANSIENT); + __cl_page_state_set(env, page, state); } /** @@ -644,292 +455,124 @@ void cl_page_get(struct cl_page *page) EXPORT_SYMBOL(cl_page_get); /** - * Releases a reference to a page. + * Releases a reference to a page, use the folio_batch to release the pages + * in batch if provided. * - * When last reference is released, page is returned to the cache, unless it - * is in cl_page_state::CPS_FREEING state, in which case it is immediately - * destroyed. - * - * \see cl_object_put(), cl_lock_put(). + * Users need to do a final folio_batch_release() to release any trailing pages. */ -void cl_page_put(const struct lu_env *env, struct cl_page *page) +void cl_batch_put(const struct lu_env *env, struct cl_page *page, + struct folio_batch *fbatch) { - PASSERT(env, page, cfs_atomic_read(&page->cp_ref) > !!page->cp_parent); - - ENTRY; - CL_PAGE_HEADER(D_TRACE, env, page, "%d\n", - cfs_atomic_read(&page->cp_ref)); - - if (cfs_atomic_dec_and_lock(&page->cp_ref, &page->cp_lock)) { - CS_PAGE_DEC(page->cp_obj, busy); - /* We're going to access the page w/o a reference, but it's - * ok because we have grabbed the lock cp_lock, which - * means nobody is able to free this page behind us. - */ - if (page->cp_state == CPS_FREEING) { - /* We drop the page reference and check the page state - * inside the cp_lock. So that if it gets here, - * it is the REALLY last reference to this page. - */ - spin_unlock(&page->cp_lock); - - LASSERT(cfs_atomic_read(&page->cp_ref) == 0); - PASSERT(env, page, page->cp_owner == NULL); - PASSERT(env, page, cfs_list_empty(&page->cp_batch)); - /* - * Page is no longer reachable by other threads. Tear - * it down. - */ - cl_page_free(env, page); - - EXIT; - return; - } - spin_unlock(&page->cp_lock); + ENTRY; + CL_PAGE_HEADER(D_TRACE, env, page, "%d\n", + refcount_read(&page->cp_ref)); + + if (refcount_dec_and_test(&page->cp_ref)) { + if (page->cp_type != CPT_TRANSIENT) { + LASSERT(page->cp_state == CPS_FREEING); + PASSERT(env, page, page->cp_owner == NULL); + } + + LASSERT(refcount_read(&page->cp_ref) == 0); + PASSERT(env, page, list_empty(&page->cp_batch)); + /* Page is no longer reachable by other threads. Tear it down */ + cl_page_free(env, page, fbatch); } EXIT; } -EXPORT_SYMBOL(cl_page_put); +EXPORT_SYMBOL(cl_batch_put); /** - * Returns a VM page associated with a given cl_page. + * Releases a reference to a page, wrapper to cl_batch_put + * + * When last reference is released, page is returned to the cache, unless it + * is in cl_page_state::CPS_FREEING state, in which case it is immediately + * destroyed. + * + * \see cl_object_put(), cl_lock_put(). */ -cfs_page_t *cl_page_vmpage(const struct lu_env *env, struct cl_page *page) +void cl_page_put(const struct lu_env *env, struct cl_page *page) { - const struct cl_page_slice *slice; - - /* - * Find uppermost layer with ->cpo_vmpage() method, and return its - * result. - */ - page = cl_page_top(page); - do { - cfs_list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { - if (slice->cpl_ops->cpo_vmpage != NULL) - RETURN(slice->cpl_ops->cpo_vmpage(env, slice)); - } - page = page->cp_child; - } while (page != NULL); - LBUG(); /* ->cpo_vmpage() has to be defined somewhere in the stack */ + cl_batch_put(env, page, NULL); } -EXPORT_SYMBOL(cl_page_vmpage); +EXPORT_SYMBOL(cl_page_put); -/** - * Returns a cl_page associated with a VM page, and given cl_object. - */ -struct cl_page *cl_vmpage_page(cfs_page_t *vmpage, struct cl_object *obj) +/* Returns a cl_page associated with a VM page, and given cl_object. */ +struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj) { - struct cl_page *top; struct cl_page *page; - ENTRY; - KLASSERT(PageLocked(vmpage)); - - /* - * NOTE: absence of races and liveness of data are guaranteed by page - * lock on a "vmpage". That works because object destruction has - * bottom-to-top pass. - */ - - /* - * This loop assumes that ->private points to the top-most page. This - * can be rectified easily. - */ - top = (struct cl_page *)vmpage->private; - if (top == NULL) - RETURN(NULL); - - spin_lock(&top->cp_lock); - for (page = top; page != NULL; page = page->cp_child) { - if (cl_object_same(page->cp_obj, obj)) { - cl_page_get_trust(page); - break; - } + ENTRY; + LASSERT(PageLocked(vmpage)); + + /* + * NOTE: absence of races and liveness of data are guaranteed by page + * lock on a "vmpage". That works because object destruction has + * bottom-to-top pass. + */ + + page = (struct cl_page *)vmpage->private; + if (page != NULL) { + cl_page_get_trust(page); + LASSERT(page->cp_type == CPT_CACHEABLE); } - spin_unlock(&top->cp_lock); - LASSERT(ergo(page, page->cp_type == CPT_CACHEABLE)); RETURN(page); } EXPORT_SYMBOL(cl_vmpage_page); -/** - * Returns the top-page for a given page. - * - * \see cl_object_top(), cl_io_top() - */ -struct cl_page *cl_page_top(struct cl_page *page) -{ - return cl_page_top_trusted(page); -} -EXPORT_SYMBOL(cl_page_top); - -const struct cl_page_slice *cl_page_at(const struct cl_page *page, - const struct lu_device_type *dtype) -{ - return cl_page_at_trusted(page, dtype); -} -EXPORT_SYMBOL(cl_page_at); - -#define CL_PAGE_OP(opname) offsetof(struct cl_page_operations, opname) - -#define CL_PAGE_INVOKE(_env, _page, _op, _proto, ...) \ -({ \ - const struct lu_env *__env = (_env); \ - struct cl_page *__page = (_page); \ - const struct cl_page_slice *__scan; \ - int __result; \ - ptrdiff_t __op = (_op); \ - int (*__method)_proto; \ - \ - __result = 0; \ - __page = cl_page_top(__page); \ - do { \ - cfs_list_for_each_entry(__scan, &__page->cp_layers, \ - cpl_linkage) { \ - __method = *(void **)((char *)__scan->cpl_ops + \ - __op); \ - if (__method != NULL) { \ - __result = (*__method)(__env, __scan, \ - ## __VA_ARGS__); \ - if (__result != 0) \ - break; \ - } \ - } \ - __page = __page->cp_child; \ - } while (__page != NULL && __result == 0); \ - if (__result > 0) \ - __result = 0; \ - __result; \ -}) - -#define CL_PAGE_INVOID(_env, _page, _op, _proto, ...) \ -do { \ - const struct lu_env *__env = (_env); \ - struct cl_page *__page = (_page); \ - const struct cl_page_slice *__scan; \ - ptrdiff_t __op = (_op); \ - void (*__method)_proto; \ - \ - __page = cl_page_top(__page); \ - do { \ - cfs_list_for_each_entry(__scan, &__page->cp_layers, \ - cpl_linkage) { \ - __method = *(void **)((char *)__scan->cpl_ops + \ - __op); \ - if (__method != NULL) \ - (*__method)(__env, __scan, \ - ## __VA_ARGS__); \ - } \ - __page = __page->cp_child; \ - } while (__page != NULL); \ -} while (0) - -#define CL_PAGE_INVOID_REVERSE(_env, _page, _op, _proto, ...) \ -do { \ - const struct lu_env *__env = (_env); \ - struct cl_page *__page = (_page); \ - const struct cl_page_slice *__scan; \ - ptrdiff_t __op = (_op); \ - void (*__method)_proto; \ - \ - /* get to the bottom page. */ \ - while (__page->cp_child != NULL) \ - __page = __page->cp_child; \ - do { \ - cfs_list_for_each_entry_reverse(__scan, &__page->cp_layers, \ - cpl_linkage) { \ - __method = *(void **)((char *)__scan->cpl_ops + \ - __op); \ - if (__method != NULL) \ - (*__method)(__env, __scan, \ - ## __VA_ARGS__); \ - } \ - __page = __page->cp_parent; \ - } while (__page != NULL); \ -} while (0) - -static int cl_page_invoke(const struct lu_env *env, - struct cl_io *io, struct cl_page *page, ptrdiff_t op) - -{ - PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj)); - ENTRY; - RETURN(CL_PAGE_INVOKE(env, page, op, - (const struct lu_env *, - const struct cl_page_slice *, struct cl_io *), - io)); -} - -static void cl_page_invoid(const struct lu_env *env, - struct cl_io *io, struct cl_page *page, ptrdiff_t op) - -{ - PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj)); - ENTRY; - CL_PAGE_INVOID(env, page, op, - (const struct lu_env *, - const struct cl_page_slice *, struct cl_io *), io); - EXIT; -} - static void cl_page_owner_clear(struct cl_page *page) { - ENTRY; - for (page = cl_page_top(page); page != NULL; page = page->cp_child) { - if (page->cp_owner != NULL) { - LASSERT(page->cp_owner->ci_owned_nr > 0); - page->cp_owner->ci_owned_nr--; - page->cp_owner = NULL; - page->cp_task = NULL; - } - } - EXIT; + ENTRY; + if (page->cp_owner != NULL) { + LASSERT(page->cp_owner->ci_owned_nr > 0); + page->cp_owner->ci_owned_nr--; + page->cp_owner = NULL; + } + EXIT; } static void cl_page_owner_set(struct cl_page *page) { - ENTRY; - for (page = cl_page_top(page); page != NULL; page = page->cp_child) { - LASSERT(page->cp_owner != NULL); - page->cp_owner->ci_owned_nr++; - } - EXIT; + ENTRY; + LASSERT(page->cp_owner != NULL); + page->cp_owner->ci_owned_nr++; + EXIT; } -void cl_page_disown0(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) +void __cl_page_disown(const struct lu_env *env, struct cl_page *cp) { - enum cl_page_state state; - - ENTRY; - state = pg->cp_state; - PINVRNT(env, pg, state == CPS_OWNED || state == CPS_FREEING); - PINVRNT(env, pg, cl_page_invariant(pg)); - cl_page_owner_clear(pg); - - if (state == CPS_OWNED) - cl_page_state_set(env, pg, CPS_CACHED); - /* - * Completion call-backs are executed in the bottom-up order, so that - * uppermost layer (llite), responsible for VFS/VM interaction runs - * last and can release locks safely. - */ - CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_disown), - (const struct lu_env *, - const struct cl_page_slice *, struct cl_io *), - io); - EXIT; + struct page *vmpage; + enum cl_page_state state; + + ENTRY; + if (cp->cp_type == CPT_CACHEABLE) { + cl_page_owner_clear(cp); + state = cp->cp_state; + PINVRNT(env, cp, state == CPS_OWNED || state == CPS_FREEING); + PINVRNT(env, cp, cl_page_invariant(cp) || state == CPS_FREEING); + if (state == CPS_OWNED) + cl_page_state_set(env, cp, CPS_CACHED); + vmpage = cp->cp_vmpage; + LASSERT(vmpage != NULL); + LASSERT(PageLocked(vmpage)); + unlock_page(vmpage); + } + + EXIT; } -/** - * returns true, iff page is owned by the given io. - */ +/* returns true, iff page is owned by the given io. */ int cl_page_is_owned(const struct cl_page *pg, const struct cl_io *io) { - LINVRNT(cl_object_same(pg->cp_obj, io->ci_obj)); - ENTRY; - RETURN(pg->cp_state == CPS_OWNED && pg->cp_owner == io); + struct cl_io *top = cl_io_top((struct cl_io *)io); + + LINVRNT(cl_object_same(pg->cp_obj, top->ci_obj)); + ENTRY; + if (pg->cp_type != CPT_TRANSIENT) + RETURN(pg->cp_state == CPS_OWNED && pg->cp_owner == top); + else + RETURN(pg->cp_owner == top); } EXPORT_SYMBOL(cl_page_is_owned); @@ -939,77 +582,86 @@ EXPORT_SYMBOL(cl_page_is_owned); * Waits until page is in cl_page_state::CPS_CACHED state, and then switch it * into cl_page_state::CPS_OWNED state. * - * \pre !cl_page_is_owned(pg, io) - * \post result == 0 iff cl_page_is_owned(pg, io) + * \pre !cl_page_is_owned(cl_page, io) + * \post result == 0 iff cl_page_is_owned(cl_page, io) * * \retval 0 success * - * \retval -ve failure, e.g., page was destroyed (and landed in + * \retval -ve failure, e.g., cl_page was destroyed (and landed in * cl_page_state::CPS_FREEING instead of cl_page_state::CPS_CACHED). * or, page was owned by another thread, or in IO. * * \see cl_page_disown() - * \see cl_page_operations::cpo_own() * \see cl_page_own_try() * \see cl_page_own */ -static int cl_page_own0(const struct lu_env *env, struct cl_io *io, - struct cl_page *pg, int nonblock) +static int __cl_page_own(const struct lu_env *env, struct cl_io *io, + struct cl_page *cl_page, int nonblock) { - int result; - - PINVRNT(env, pg, !cl_page_is_owned(pg, io)); - - ENTRY; - pg = cl_page_top(pg); - io = cl_io_top(io); - - if (pg->cp_state == CPS_FREEING) { - result = -ENOENT; - } else { - result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(cpo_own), - (const struct lu_env *, - const struct cl_page_slice *, - struct cl_io *, int), - io, nonblock); - if (result == 0) { - PASSERT(env, pg, pg->cp_owner == NULL); - PASSERT(env, pg, pg->cp_req == NULL); - pg->cp_owner = io; - pg->cp_task = current; - cl_page_owner_set(pg); - if (pg->cp_state != CPS_FREEING) { - cl_page_state_set(env, pg, CPS_OWNED); - } else { - cl_page_disown0(env, io, pg); - result = -ENOENT; - } - } - } - PINVRNT(env, pg, ergo(result == 0, cl_page_invariant(pg))); - RETURN(result); + struct page *vmpage = cl_page->cp_vmpage; + int result; + + ENTRY; + PINVRNT(env, cl_page, !cl_page_is_owned(cl_page, io)); + + LASSERT(cl_page->cp_type != CPT_TRANSIENT); + + if (cl_page->cp_type != CPT_TRANSIENT && + cl_page->cp_state == CPS_FREEING) { + result = -ENOENT; + goto out; + } + + LASSERT(vmpage != NULL); + + if (nonblock) { + if (!trylock_page(vmpage)) { + result = -EAGAIN; + goto out; + } + + if (unlikely(PageWriteback(vmpage))) { + unlock_page(vmpage); + result = -EAGAIN; + goto out; + } + } else { + lock_page(vmpage); + wait_on_page_writeback(vmpage); + } + + PASSERT(env, cl_page, cl_page->cp_owner == NULL); + cl_page->cp_owner = cl_io_top(io); + cl_page_owner_set(cl_page); + + if (cl_page->cp_state == CPS_FREEING) { + __cl_page_disown(env, cl_page); + result = -ENOENT; + goto out; + } + + cl_page_state_set(env, cl_page, CPS_OWNED); + + result = 0; +out: + CDEBUG(D_INFO, "res %d\n", result); + PINVRNT(env, cl_page, ergo(result == 0, + cl_page_invariant(cl_page))); + RETURN(result); } -/** - * Own a page, might be blocked. - * - * \see cl_page_own0() - */ +/* Own a page, might be blocked. (see __cl_page_own()) */ int cl_page_own(const struct lu_env *env, struct cl_io *io, struct cl_page *pg) { - return cl_page_own0(env, io, pg, 0); + return __cl_page_own(env, io, pg, 0); } EXPORT_SYMBOL(cl_page_own); -/** - * Nonblock version of cl_page_own(). - * - * \see cl_page_own0() - */ +/* Nonblock version of cl_page_own(). (see __cl_page_own()) */ int cl_page_own_try(const struct lu_env *env, struct cl_io *io, - struct cl_page *pg) + struct cl_page *pg) { - return cl_page_own0(env, io, pg, 1); + return __cl_page_own(env, io, pg, 1); } EXPORT_SYMBOL(cl_page_own_try); @@ -1019,57 +671,59 @@ EXPORT_SYMBOL(cl_page_own_try); * * Called when page is already locked by the hosting VM. * - * \pre !cl_page_is_owned(pg, io) - * \post cl_page_is_owned(pg, io) - * - * \see cl_page_operations::cpo_assume() + * \pre !cl_page_is_owned(cp, io) + * \post cl_page_is_owned(cp, io) */ void cl_page_assume(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) + struct cl_io *io, struct cl_page *cp) { - PINVRNT(env, pg, cl_object_same(pg->cp_obj, io->ci_obj)); - - ENTRY; - pg = cl_page_top(pg); - io = cl_io_top(io); - - cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_assume)); - PASSERT(env, pg, pg->cp_owner == NULL); - pg->cp_owner = io; - pg->cp_task = current; - cl_page_owner_set(pg); - cl_page_state_set(env, pg, CPS_OWNED); - EXIT; + struct page *vmpage; + + ENTRY; + PINVRNT(env, cp, cl_object_same(cp->cp_obj, cl_io_top(io)->ci_obj)); + + LASSERT(cp->cp_type != CPT_TRANSIENT); + PASSERT(env, cp, cp->cp_owner == NULL); + + vmpage = cp->cp_vmpage; + LASSERT(vmpage != NULL); + LASSERT(PageLocked(vmpage)); + wait_on_page_writeback(vmpage); + cp->cp_owner = cl_io_top(io); + cl_page_owner_set(cp); + cl_page_state_set(env, cp, CPS_OWNED); + + EXIT; } EXPORT_SYMBOL(cl_page_assume); /** * Releases page ownership without unlocking the page. * - * Moves page into cl_page_state::CPS_CACHED without releasing a lock on the - * underlying VM page (as VM is supposed to do this itself). - * - * \pre cl_page_is_owned(pg, io) - * \post !cl_page_is_owned(pg, io) + * Moves cl_page into cl_page_state::CPS_CACHED without releasing a lock + * on the underlying VM page (as VM is supposed to do this itself). * - * \see cl_page_assume() + * \pre cl_page_is_owned(cp, io) + * \post !cl_page_is_owned(cp, io) */ void cl_page_unassume(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) + struct cl_io *io, struct cl_page *cp) { - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); - - ENTRY; - pg = cl_page_top(pg); - io = cl_io_top(io); - cl_page_owner_clear(pg); - cl_page_state_set(env, pg, CPS_CACHED); - CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_unassume), - (const struct lu_env *, - const struct cl_page_slice *, struct cl_io *), - io); - EXIT; + struct page *vmpage; + + ENTRY; + + LASSERT(cp->cp_type != CPT_TRANSIENT); + + PINVRNT(env, cp, cl_page_is_owned(cp, io)); + PINVRNT(env, cp, cl_page_invariant(cp)); + cl_page_owner_clear(cp); + cl_page_state_set(env, cp, CPS_CACHED); + vmpage = cp->cp_vmpage; + LASSERT(vmpage != NULL); + LASSERT(PageLocked(vmpage)); + + EXIT; } EXPORT_SYMBOL(cl_page_unassume); @@ -1082,95 +736,82 @@ EXPORT_SYMBOL(cl_page_unassume); * \post !cl_page_is_owned(pg, io) * * \see cl_page_own() - * \see cl_page_operations::cpo_disown() */ void cl_page_disown(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) + struct cl_io *io, struct cl_page *pg) { - PINVRNT(env, pg, cl_page_is_owned(pg, io)); + if (pg->cp_type != CPT_TRANSIENT) { + PINVRNT(env, pg, cl_page_is_owned(pg, cl_io_top(io)) || + pg->cp_state == CPS_FREEING); + } - ENTRY; - pg = cl_page_top(pg); - io = cl_io_top(io); - cl_page_disown0(env, io, pg); - EXIT; + __cl_page_disown(env, pg); } EXPORT_SYMBOL(cl_page_disown); /** - * Called when page is to be removed from the object, e.g., as a result of - * truncate. + * Called when cl_page is to be removed from the object, e.g., + * as a result of truncate. * * Calls cl_page_operations::cpo_discard() top-to-bottom. * - * \pre cl_page_is_owned(pg, io) + * \pre cl_page_is_owned(cl_page, io) * * \see cl_page_operations::cpo_discard() */ void cl_page_discard(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) + struct cl_io *io, struct cl_page *cp) { - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); + struct page *vmpage; + const struct cl_page_slice *slice; + int i; + + cl_page_slice_for_each(cp, slice, i) { + if (slice->cpl_ops->cpo_discard != NULL) + (*slice->cpl_ops->cpo_discard)(env, slice, io); + } - cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_discard)); + if (cp->cp_type == CPT_CACHEABLE) { + PINVRNT(env, cp, cl_page_is_owned(cp, io)); + PINVRNT(env, cp, cl_page_invariant(cp)); + vmpage = cp->cp_vmpage; + LASSERT(vmpage != NULL); + LASSERT(PageLocked(vmpage)); + generic_error_remove_folio(vmpage->mapping, page_folio(vmpage)); + } else { + cl_page_delete(env, cp); + } } EXPORT_SYMBOL(cl_page_discard); /** * Version of cl_page_delete() that can be called for not fully constructed - * pages, e.g,. in a error handling cl_page_find()->cl_page_delete0() - * path. Doesn't check page invariant. + * cl_pages, e.g. in an error handling cl_page_find()->__cl_page_delete() + * path. Doesn't check cl_page invariant. */ -static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, - int radix) +static void __cl_page_delete(const struct lu_env *env, struct cl_page *cp) { - struct cl_page *tmp = pg; - ENTRY; - - PASSERT(env, pg, pg == cl_page_top(pg)); - PASSERT(env, pg, pg->cp_state != CPS_FREEING); - - /* - * Severe all ways to obtain new pointers to @pg. - */ - cl_page_owner_clear(pg); - - /* - * unexport the page firstly before freeing it so that - * the page content is considered to be invalid. - * We have to do this because a CPS_FREEING cl_page may - * be NOT under the protection of a cl_lock. - * Afterwards, if this page is found by other threads, then this - * page will be forced to reread. - */ - cl_page_export(env, pg, 0); - cl_page_state_set0(env, pg, CPS_FREEING); - - if (tmp->cp_type == CPT_CACHEABLE) { - if (!radix) - /* !radix means that @pg is not yet in the radix tree, - * skip removing it. - */ - tmp = pg->cp_child; - for (; tmp != NULL; tmp = tmp->cp_child) { - void *value; - struct cl_object_header *hdr; - - hdr = cl_object_header(tmp->cp_obj); - spin_lock(&hdr->coh_page_guard); - value = radix_tree_delete(&hdr->coh_tree, - tmp->cp_index); - PASSERT(env, tmp, value == tmp); - PASSERT(env, tmp, hdr->coh_pages > 0); - hdr->coh_pages--; - spin_unlock(&hdr->coh_page_guard); - } - } - - CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_delete), - (const struct lu_env *, const struct cl_page_slice *)); - EXIT; + const struct cl_page_slice *slice; + int i; + + ENTRY; + + /* + * Severe all ways to obtain new pointers to @pg. + * Transient pages already can't be found because they're not in cache. + */ + if (cp->cp_type != CPT_TRANSIENT) { + PASSERT(env, cp, cp->cp_state != CPS_FREEING); + cl_page_owner_clear(cp); + __cl_page_state_set(env, cp, CPS_FREEING); + } + + cl_page_slice_for_each_reverse(cp, slice, i) { + if (slice->cpl_ops->cpo_delete != NULL) + (*slice->cpl_ops->cpo_delete)(env, slice); + } + + EXIT; } /** @@ -1192,7 +833,6 @@ static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, * Once page reaches cl_page_state::CPS_FREEING, all remaining references will * drain after some time, at which point page will be recycled. * - * \pre pg == cl_page_top(pg) * \pre VM page is locked * \post pg->cp_state == CPS_FREEING * @@ -1200,123 +840,85 @@ static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, */ void cl_page_delete(const struct lu_env *env, struct cl_page *pg) { - PINVRNT(env, pg, cl_page_invariant(pg)); - ENTRY; - cl_page_delete0(env, pg, 1); - EXIT; + PINVRNT(env, pg, cl_page_invariant(pg)); + ENTRY; + __cl_page_delete(env, pg); + EXIT; } EXPORT_SYMBOL(cl_page_delete); -/** - * Unmaps page from user virtual memory. - * - * Calls cl_page_operations::cpo_unmap() through all layers top-to-bottom. The - * layer responsible for VM interaction has to unmap page from user space - * virtual memory. - * - * \see cl_page_operations::cpo_unmap() - */ -int cl_page_unmap(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) +void cl_page_touch(const struct lu_env *env, + const struct cl_page *cl_page, size_t to) { - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); + const struct cl_page_slice *slice; + int i; - return cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_unmap)); -} -EXPORT_SYMBOL(cl_page_unmap); + ENTRY; -/** - * Marks page up-to-date. - * - * Call cl_page_operations::cpo_export() through all layers top-to-bottom. The - * layer responsible for VM interaction has to mark/clear page as up-to-date - * by the \a uptodate argument. - * - * \see cl_page_operations::cpo_export() - */ -void cl_page_export(const struct lu_env *env, struct cl_page *pg, int uptodate) -{ - PINVRNT(env, pg, cl_page_invariant(pg)); - CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_export), - (const struct lu_env *, - const struct cl_page_slice *, int), uptodate); -} -EXPORT_SYMBOL(cl_page_export); + cl_page_slice_for_each(cl_page, slice, i) { + if (slice->cpl_ops->cpo_page_touch != NULL) + (*slice->cpl_ops->cpo_page_touch)(env, slice, to); + } -/** - * Returns true, iff \a pg is VM locked in a suitable sense by the calling - * thread. - */ -int cl_page_is_vmlocked(const struct lu_env *env, const struct cl_page *pg) -{ - int result; - const struct cl_page_slice *slice; - - ENTRY; - pg = cl_page_top_trusted((struct cl_page *)pg); - slice = container_of(pg->cp_layers.next, - const struct cl_page_slice, cpl_linkage); - PASSERT(env, pg, slice->cpl_ops->cpo_is_vmlocked != NULL); - /* - * Call ->cpo_is_vmlocked() directly instead of going through - * CL_PAGE_INVOKE(), because cl_page_is_vmlocked() is used by - * cl_page_invariant(). - */ - result = slice->cpl_ops->cpo_is_vmlocked(env, slice); - PASSERT(env, pg, result == -EBUSY || result == -ENODATA); - RETURN(result == -EBUSY); + EXIT; } -EXPORT_SYMBOL(cl_page_is_vmlocked); +EXPORT_SYMBOL(cl_page_touch); static enum cl_page_state cl_req_type_state(enum cl_req_type crt) { - ENTRY; - RETURN(crt == CRT_WRITE ? CPS_PAGEOUT : CPS_PAGEIN); + ENTRY; + RETURN(crt == CRT_WRITE ? CPS_PAGEOUT : CPS_PAGEIN); } static void cl_page_io_start(const struct lu_env *env, - struct cl_page *pg, enum cl_req_type crt) + struct cl_page *pg, enum cl_req_type crt) { - /* - * Page is queued for IO, change its state. - */ - ENTRY; - cl_page_owner_clear(pg); - cl_page_state_set(env, pg, cl_req_type_state(crt)); - EXIT; + /* Page is queued for IO, change its state. */ + ENTRY; + if (pg->cp_type != CPT_TRANSIENT) { + cl_page_owner_clear(pg); + cl_page_state_set(env, pg, cl_req_type_state(crt)); + } + EXIT; } /** - * Prepares page for immediate transfer. cl_page_operations::cpo_prep() is - * called top-to-bottom. Every layer either agrees to submit this page (by - * returning 0), or requests to omit this page (by returning -EALREADY). Layer - * handling interactions with the VM also has to inform VM that page is under - * transfer now. + * Prepares page for immediate transfer. Return -EALREADY if this page + * should be omitted from transfer. */ int cl_page_prep(const struct lu_env *env, struct cl_io *io, - struct cl_page *pg, enum cl_req_type crt) + struct cl_page *cp, enum cl_req_type crt) { - int result; - - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); - PINVRNT(env, pg, crt < CRT_NR); - - /* - * XXX this has to be called bottom-to-top, so that llite can set up - * PG_writeback without risking other layers deciding to skip this - * page. - */ - result = cl_page_invoke(env, io, pg, CL_PAGE_OP(io[crt].cpo_prep)); - if (result == 0) - cl_page_io_start(env, pg, crt); - - KLASSERT(ergo(crt == CRT_WRITE && pg->cp_type == CPT_CACHEABLE, - equi(result == 0, - PageWriteback(cl_page_vmpage(env, pg))))); - CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result); - return result; + struct page *vmpage = cp->cp_vmpage; + int rc; + + if (cp->cp_type == CPT_TRANSIENT) + GOTO(start, rc = 0); + PASSERT(env, cp, crt < CRT_NR); + PINVRNT(env, cp, cl_page_is_owned(cp, io)); + PINVRNT(env, cp, cl_page_invariant(cp)); + + if (crt == CRT_READ) { + if (PageUptodate(vmpage)) + GOTO(out, rc = -EALREADY); + } else { + LASSERT(PageLocked(vmpage)); + LASSERT(!PageDirty(vmpage)); + + /* ll_writepage path is not a sync write, so need to + * set page writeback flag + */ + if (cp->cp_sync_io == NULL) + set_page_writeback(vmpage); + } +start: + + cl_page_io_start(env, cp, crt); + rc = 0; +out: + CL_PAGE_HEADER(D_TRACE, env, cp, "%d %d\n", crt, rc); + + return rc; } EXPORT_SYMBOL(cl_page_prep); @@ -1330,39 +932,43 @@ EXPORT_SYMBOL(cl_page_prep); * uppermost layer (llite), responsible for the VFS/VM interaction runs last * and can release locks safely. * - * \pre pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT - * \post pg->cp_state == CPS_CACHED + * \pre cl_page->cp_state == CPS_PAGEIN || cl_page->cp_state == CPS_PAGEOUT + * \post cl_page->cl_page_state == CPS_CACHED * * \see cl_page_operations::cpo_completion() */ void cl_page_completion(const struct lu_env *env, - struct cl_page *pg, enum cl_req_type crt, int ioret) + struct cl_page *cl_page, enum cl_req_type crt, + int ioret) { - struct cl_sync_io *anchor = pg->cp_sync_io; - - PASSERT(env, pg, crt < CRT_NR); - /* cl_page::cp_req already cleared by the caller (osc_completion()) */ - PASSERT(env, pg, pg->cp_req == NULL); - PASSERT(env, pg, pg->cp_state == cl_req_type_state(crt)); - - ENTRY; - CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, ioret); - if (crt == CRT_READ && ioret == 0) { - PASSERT(env, pg, !(pg->cp_flags & CPF_READ_COMPLETED)); - pg->cp_flags |= CPF_READ_COMPLETED; - } - - cl_page_state_set(env, pg, CPS_CACHED); - CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(io[crt].cpo_completion), - (const struct lu_env *, - const struct cl_page_slice *, int), ioret); - if (anchor) { - LASSERT(cl_page_is_vmlocked(env, pg)); - LASSERT(pg->cp_sync_io == anchor); - pg->cp_sync_io = NULL; - cl_sync_io_note(anchor, ioret); - } - EXIT; + const struct cl_page_slice *slice; + struct cl_sync_io *anchor = cl_page->cp_sync_io; + int i; + + ENTRY; + PASSERT(env, cl_page, crt < CRT_NR); + if (cl_page->cp_type != CPT_TRANSIENT) + PASSERT(env, cl_page, + cl_page->cp_state == cl_req_type_state(crt)); + + CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d %d\n", crt, ioret); + if (cl_page->cp_type != CPT_TRANSIENT) + cl_page_state_set(env, cl_page, CPS_CACHED); + if (crt >= CRT_NR) + return; + + cl_page_slice_for_each_reverse(cl_page, slice, i) { + if (slice->cpl_ops->io[crt].cpo_completion != NULL) + (*slice->cpl_ops->io[crt].cpo_completion)(env, slice, + ioret); + } + + if (anchor != NULL) { + LASSERT(cl_page->cp_sync_io == anchor); + cl_page->cp_sync_io = NULL; + cl_sync_io_note(env, anchor, ioret); + } + EXIT; } EXPORT_SYMBOL(cl_page_completion); @@ -1370,257 +976,160 @@ EXPORT_SYMBOL(cl_page_completion); * Notify layers that transfer formation engine decided to yank this page from * the cache and to make it a part of a transfer. * - * \pre pg->cp_state == CPS_CACHED - * \post pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT - * - * \see cl_page_operations::cpo_make_ready() + * \pre cl_page->cp_state == CPS_CACHED + * \post cl_page->cp_state == CPS_PAGEIN || cl_page->cp_state == CPS_PAGEOUT */ -int cl_page_make_ready(const struct lu_env *env, struct cl_page *pg, - enum cl_req_type crt) +int cl_page_make_ready(const struct lu_env *env, struct cl_page *cp, + enum cl_req_type crt) { - int result; - - PINVRNT(env, pg, crt < CRT_NR); - - ENTRY; - result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(io[crt].cpo_make_ready), - (const struct lu_env *, - const struct cl_page_slice *)); - if (result == 0) { - PASSERT(env, pg, pg->cp_state == CPS_CACHED); - cl_page_io_start(env, pg, crt); - } - CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result); - RETURN(result); -} -EXPORT_SYMBOL(cl_page_make_ready); + struct page *vmpage = cp->cp_vmpage; + bool unlock = false; + int rc; -/** - * Notify layers that high level io decided to place this page into a cache - * for future transfer. - * - * The layer implementing transfer engine (osc) has to register this page in - * its queues. - * - * \pre cl_page_is_owned(pg, io) - * \post cl_page_is_owned(pg, io) - * - * \see cl_page_operations::cpo_cache_add() - */ -int cl_page_cache_add(const struct lu_env *env, struct cl_io *io, - struct cl_page *pg, enum cl_req_type crt) -{ - const struct cl_page_slice *scan; - int result = 0; + ENTRY; + PASSERT(env, cp, crt == CRT_WRITE); + + if (cp->cp_type == CPT_TRANSIENT) + GOTO(out, rc = 0); + + lock_page(vmpage); + PASSERT(env, cp, PageUptodate(vmpage)); + unlock = true; + + if (clear_page_dirty_for_io(vmpage)) { + LASSERT(cp->cp_state == CPS_CACHED); + /* This actually clears the dirty bit in the radix tree */ + set_page_writeback(vmpage); + CL_PAGE_HEADER(D_PAGE, env, cp, "readied\n"); + rc = 0; + } else if (cp->cp_state == CPS_PAGEOUT) { + /* is it possible for osc_flush_async_page() + * to already make it ready? + */ + rc = -EALREADY; + } else { + CL_PAGE_DEBUG(D_ERROR, env, cp, + "unexpecting page state %d\n", + cp->cp_state); + LBUG(); + } - PINVRNT(env, pg, crt < CRT_NR); - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); +out: + if (rc == 0) { + PASSERT(env, cp, cp->cp_state == CPS_CACHED); + cl_page_io_start(env, cp, crt); + } - ENTRY; + if (unlock) + unlock_page(vmpage); - cfs_list_for_each_entry(scan, &pg->cp_layers, cpl_linkage) { - if (scan->cpl_ops->io[crt].cpo_cache_add == NULL) - continue; + CL_PAGE_HEADER(D_TRACE, env, cp, "%d %d\n", crt, rc); - result = scan->cpl_ops->io[crt].cpo_cache_add(env, scan, io); - if (result != 0) - break; - } - CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result); - RETURN(result); + return rc; } -EXPORT_SYMBOL(cl_page_cache_add); +EXPORT_SYMBOL(cl_page_make_ready); /** - * Called if a pge is being written back by kernel's intention. + * Called if a page is being written back by kernel's intention. * - * \pre cl_page_is_owned(pg, io) - * \post ergo(result == 0, pg->cp_state == CPS_PAGEOUT) + * \pre cl_page_is_owned(cl_page, io) + * \post ergo(result == 0, cl_page->cp_state == CPS_PAGEOUT) * * \see cl_page_operations::cpo_flush() */ int cl_page_flush(const struct lu_env *env, struct cl_io *io, - struct cl_page *pg) + struct cl_page *cl_page) { - int result; - - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); + const struct cl_page_slice *slice; + int result = 0; + int i; ENTRY; + LASSERT(cl_page->cp_type != CPT_TRANSIENT); + PINVRNT(env, cl_page, cl_page_is_owned(cl_page, io)); + PINVRNT(env, cl_page, cl_page_invariant(cl_page)); - result = cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_flush)); + cl_page_slice_for_each(cl_page, slice, i) { + if (slice->cpl_ops->cpo_flush != NULL) + result = (*slice->cpl_ops->cpo_flush)(env, slice, io); + if (result != 0) + break; + } + if (result > 0) + result = 0; - CL_PAGE_HEADER(D_TRACE, env, pg, "%d\n", result); + CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d\n", result); RETURN(result); } EXPORT_SYMBOL(cl_page_flush); /** - * Checks whether page is protected by any extent lock is at least required - * mode. - * - * \return the same as in cl_page_operations::cpo_is_under_lock() method. - * \see cl_page_operations::cpo_is_under_lock() - */ -int cl_page_is_under_lock(const struct lu_env *env, struct cl_io *io, - struct cl_page *page) -{ - int rc; - - PINVRNT(env, page, cl_page_invariant(page)); - - ENTRY; - rc = CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_is_under_lock), - (const struct lu_env *, - const struct cl_page_slice *, struct cl_io *), - io); - PASSERT(env, page, rc != 0); - RETURN(rc); -} -EXPORT_SYMBOL(cl_page_is_under_lock); - -static int page_prune_cb(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, void *cbdata) -{ - cl_page_own(env, io, page); - cl_page_unmap(env, io, page); - cl_page_discard(env, io, page); - cl_page_disown(env, io, page); - return CLP_GANG_OKAY; -} - -/** - * Purges all cached pages belonging to the object \a obj. - */ -int cl_pages_prune(const struct lu_env *env, struct cl_object *clobj) -{ - struct cl_thread_info *info; - struct cl_object *obj = cl_object_top(clobj); - struct cl_io *io; - int result; - - ENTRY; - info = cl_env_info(env); - io = &info->clt_io; - - /* - * initialize the io. This is ugly since we never do IO in this - * function, we just make cl_page_list functions happy. -jay - */ - io->ci_obj = obj; - io->ci_ignore_layout = 1; - result = cl_io_init(env, io, CIT_MISC, obj); - if (result != 0) { - cl_io_fini(env, io); - RETURN(io->ci_result); - } - - do { - result = cl_page_gang_lookup(env, obj, io, 0, CL_PAGE_EOF, - page_prune_cb, NULL); - if (result == CLP_GANG_RESCHED) - cfs_cond_resched(); - } while (result != CLP_GANG_OKAY); - - cl_io_fini(env, io); - RETURN(result); -} -EXPORT_SYMBOL(cl_pages_prune); - -/** * Tells transfer engine that only part of a page is to be transmitted. * * \see cl_page_operations::cpo_clip() */ -void cl_page_clip(const struct lu_env *env, struct cl_page *pg, - int from, int to) +void cl_page_clip(const struct lu_env *env, struct cl_page *cl_page, + int from, int to) { - PINVRNT(env, pg, cl_page_invariant(pg)); + const struct cl_page_slice *slice; + int i; - CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", from, to); - CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_clip), - (const struct lu_env *, - const struct cl_page_slice *,int, int), - from, to); + PINVRNT(env, cl_page, cl_page_invariant(cl_page)); + + CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d %d\n", from, to); + cl_page_slice_for_each(cl_page, slice, i) { + if (slice->cpl_ops->cpo_clip != NULL) + (*slice->cpl_ops->cpo_clip)(env, slice, from, to); + } } EXPORT_SYMBOL(cl_page_clip); -/** - * Prints human readable representation of \a pg to the \a f. - */ +/* Prints human readable representation of \a pg to the \a f. */ void cl_page_header_print(const struct lu_env *env, void *cookie, - lu_printer_t printer, const struct cl_page *pg) + lu_printer_t printer, const struct cl_page *pg) { - (*printer)(env, cookie, - "page@%p[%d %p:%lu ^%p_%p %d %d %d %p %p %#x]\n", - pg, cfs_atomic_read(&pg->cp_ref), pg->cp_obj, - pg->cp_index, pg->cp_parent, pg->cp_child, - pg->cp_state, pg->cp_error, pg->cp_type, - pg->cp_owner, pg->cp_req, pg->cp_flags); + (*printer)(env, cookie, + "page@%p[%d %p %d %d %p]\n", + pg, refcount_read(&pg->cp_ref), pg->cp_obj, + pg->cp_state, pg->cp_type, + pg->cp_owner); } EXPORT_SYMBOL(cl_page_header_print); -/** - * Prints human readable representation of \a pg to the \a f. - */ +/* Prints human readable representation of \a cl_page to the \a f. */ void cl_page_print(const struct lu_env *env, void *cookie, - lu_printer_t printer, const struct cl_page *pg) + lu_printer_t printer, const struct cl_page *cp) { - struct cl_page *scan; - - for (scan = cl_page_top((struct cl_page *)pg); - scan != NULL; scan = scan->cp_child) - cl_page_header_print(env, cookie, printer, scan); - CL_PAGE_INVOKE(env, (struct cl_page *)pg, CL_PAGE_OP(cpo_print), - (const struct lu_env *env, - const struct cl_page_slice *slice, - void *cookie, lu_printer_t p), cookie, printer); - (*printer)(env, cookie, "end page@%p\n", pg); -} -EXPORT_SYMBOL(cl_page_print); + struct page *vmpage = cp->cp_vmpage; + const struct cl_page_slice *slice; + int result = 0; + int i; -/** - * Cancel a page which is still in a transfer. - */ -int cl_page_cancel(const struct lu_env *env, struct cl_page *page) -{ - return CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_cancel), - (const struct lu_env *, - const struct cl_page_slice *)); -} -EXPORT_SYMBOL(cl_page_cancel); + cl_page_header_print(env, cookie, printer, cp); -/** - * Converts a byte offset within object \a obj into a page index. - */ -loff_t cl_offset(const struct cl_object *obj, pgoff_t idx) -{ - /* - * XXX for now. - */ - return (loff_t)idx << CFS_PAGE_SHIFT; -} -EXPORT_SYMBOL(cl_offset); + (*printer)(env, cookie, "vmpage @%p", vmpage); -/** - * Converts a page index into a byte offset within object \a obj. - */ -pgoff_t cl_index(const struct cl_object *obj, loff_t offset) -{ - /* - * XXX for now. - */ - return offset >> CFS_PAGE_SHIFT; -} -EXPORT_SYMBOL(cl_index); + if (vmpage != NULL) { + (*printer)(env, cookie, " %lx %d:%d %lx %lu %slru", + (long)vmpage->flags, page_count(vmpage), + page_mapcount(vmpage), vmpage->private, + page_index(vmpage), + list_empty(&vmpage->lru) ? "not-" : ""); + } -int cl_page_size(const struct cl_object *obj) -{ - return 1 << CFS_PAGE_SHIFT; + (*printer)(env, cookie, "\n"); + + cl_page_slice_for_each(cp, slice, i) { + if (slice->cpl_ops->cpo_print != NULL) + result = (*slice->cpl_ops->cpo_print)(env, slice, + cookie, printer); + if (result != 0) + break; + } + + (*printer)(env, cookie, "end page@%p\n", cp); } -EXPORT_SYMBOL(cl_page_size); +EXPORT_SYMBOL(cl_page_print); /** * Adds page slice to the compound page. @@ -1631,25 +1140,64 @@ EXPORT_SYMBOL(cl_page_size); * * \see cl_lock_slice_add(), cl_req_slice_add(), cl_io_slice_add() */ -void cl_page_slice_add(struct cl_page *page, struct cl_page_slice *slice, - struct cl_object *obj, - const struct cl_page_operations *ops) +void cl_page_slice_add(struct cl_page *cl_page, struct cl_page_slice *slice, + struct cl_object *obj, + const struct cl_page_operations *ops) { - ENTRY; - cfs_list_add_tail(&slice->cpl_linkage, &page->cp_layers); - slice->cpl_obj = obj; - slice->cpl_ops = ops; - slice->cpl_page = page; - EXIT; + unsigned int offset = (char *)slice - + ((char *)cl_page + sizeof(*cl_page)); + + ENTRY; + LASSERT(cl_page->cp_layer_count < CP_MAX_LAYER); + LASSERT(offset < (1 << sizeof(cl_page->cp_layer_offset[0]) * 8)); + cl_page->cp_layer_offset[cl_page->cp_layer_count++] = offset; + slice->cpl_ops = ops; + slice->cpl_page = cl_page; + + EXIT; } EXPORT_SYMBOL(cl_page_slice_add); -int cl_page_init(void) +/* Allocate and initialize cl_cache, called by ll_init_sbi(). */ +struct cl_client_cache *cl_cache_init(unsigned long lru_page_max) { - return lu_kmem_init(cl_page_caches); + struct cl_client_cache *cache = NULL; + + ENTRY; + OBD_ALLOC(cache, sizeof(*cache)); + if (cache == NULL) + RETURN(NULL); + + /* Initialize cache data */ + refcount_set(&cache->ccc_users, 1); + cache->ccc_lru_max = lru_page_max; + atomic_long_set(&cache->ccc_lru_left, lru_page_max); + spin_lock_init(&cache->ccc_lru_lock); + INIT_LIST_HEAD(&cache->ccc_lru); + + cache->ccc_unstable_check = 1; + atomic_long_set(&cache->ccc_unstable_nr, 0); + mutex_init(&cache->ccc_max_cache_mb_lock); + + RETURN(cache); } +EXPORT_SYMBOL(cl_cache_init); -void cl_page_fini(void) +/* Increase cl_cache refcount */ +void cl_cache_incref(struct cl_client_cache *cache) +{ + refcount_inc(&cache->ccc_users); +} +EXPORT_SYMBOL(cl_cache_incref); + +/** + * Decrease cl_cache refcount and free the cache if refcount=0. + * Since llite, lov and osc all hold cl_cache refcount, + * the free will not cause race. (LU-6173) + */ +void cl_cache_decref(struct cl_client_cache *cache) { - lu_kmem_fini(cl_page_caches); + if (refcount_dec_and_test(&cache->ccc_users)) + OBD_FREE(cache, sizeof(*cache)); } +EXPORT_SYMBOL(cl_cache_decref);