4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
42 #define DEBUG_SUBSYSTEM S_CLASS
44 #include <libcfs/libcfs.h>
45 #include <obd_class.h>
46 #include <obd_support.h>
47 #include <libcfs/list.h>
49 #include <cl_object.h>
50 #include "cl_internal.h"
52 static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg);
55 # define PASSERT(env, page, expr) \
57 if (unlikely(!(expr))) { \
58 CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \
62 #else /* !LIBCFS_DEBUG */
63 # define PASSERT(env, page, exp) \
64 ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp))
65 #endif /* !LIBCFS_DEBUG */
67 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
68 # define PINVRNT(env, page, expr) \
70 if (unlikely(!(expr))) { \
71 CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \
75 #else /* !CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK */
76 # define PINVRNT(env, page, exp) \
77 ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp))
78 #endif /* !CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK */
80 /* Disable page statistic by default due to huge performance penalty. */
81 #ifdef CONFIG_DEBUG_PAGESTATE_TRACKING
82 #define CS_PAGE_INC(o, item) \
83 atomic_inc(&cl_object_site(o)->cs_pages.cs_stats[CS_##item])
84 #define CS_PAGE_DEC(o, item) \
85 atomic_dec(&cl_object_site(o)->cs_pages.cs_stats[CS_##item])
86 #define CS_PAGESTATE_INC(o, state) \
87 atomic_inc(&cl_object_site(o)->cs_pages_state[state])
88 #define CS_PAGESTATE_DEC(o, state) \
89 atomic_dec(&cl_object_site(o)->cs_pages_state[state])
91 #define CS_PAGE_INC(o, item)
92 #define CS_PAGE_DEC(o, item)
93 #define CS_PAGESTATE_INC(o, state)
94 #define CS_PAGESTATE_DEC(o, state)
98 * Internal version of cl_page_get().
100 * This function can be used to obtain initial reference to previously
101 * unreferenced cached object. It can be called only if concurrent page
102 * reclamation is somehow prevented, e.g., by keeping a lock on a VM page,
103 * associated with \a page.
105 * Use with care! Not exported.
107 static void cl_page_get_trust(struct cl_page *page)
109 LASSERT(atomic_read(&page->cp_ref) > 0);
110 atomic_inc(&page->cp_ref);
114 * Returns a slice within a page, corresponding to the given layer in the
119 static const struct cl_page_slice *
120 cl_page_at_trusted(const struct cl_page *page,
121 const struct lu_device_type *dtype)
123 const struct cl_page_slice *slice;
126 list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
127 if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype)
133 static void cl_page_free(const struct lu_env *env, struct cl_page *page)
135 struct cl_object *obj = page->cp_obj;
136 int pagesize = cl_object_header(obj)->coh_page_bufsize;
138 PASSERT(env, page, list_empty(&page->cp_batch));
139 PASSERT(env, page, page->cp_owner == NULL);
140 PASSERT(env, page, page->cp_req == NULL);
141 PASSERT(env, page, page->cp_state == CPS_FREEING);
144 while (!list_empty(&page->cp_layers)) {
145 struct cl_page_slice *slice;
147 slice = list_entry(page->cp_layers.next,
148 struct cl_page_slice, cpl_linkage);
149 list_del_init(page->cp_layers.next);
150 if (unlikely(slice->cpl_ops->cpo_fini != NULL))
151 slice->cpl_ops->cpo_fini(env, slice);
153 CS_PAGE_DEC(obj, total);
154 CS_PAGESTATE_DEC(obj, page->cp_state);
155 lu_object_ref_del_at(&obj->co_lu, &page->cp_obj_ref, "cl_page", page);
156 cl_object_put(env, obj);
157 lu_ref_fini(&page->cp_reference);
158 OBD_FREE(page, pagesize);
163 * Helper function updating page state. This is the only place in the code
164 * where cl_page::cp_state field is mutated.
166 static inline void cl_page_state_set_trust(struct cl_page *page,
167 enum cl_page_state state)
170 *(enum cl_page_state *)&page->cp_state = state;
173 struct cl_page *cl_page_alloc(const struct lu_env *env,
174 struct cl_object *o, pgoff_t ind, struct page *vmpage,
175 enum cl_page_type type)
177 struct cl_page *page;
178 struct lu_object_header *head;
181 OBD_ALLOC_GFP(page, cl_object_header(o)->coh_page_bufsize,
185 atomic_set(&page->cp_ref, 1);
188 lu_object_ref_add_at(&o->co_lu, &page->cp_obj_ref, "cl_page",
190 page->cp_vmpage = vmpage;
191 cl_page_state_set_trust(page, CPS_CACHED);
192 page->cp_type = type;
193 INIT_LIST_HEAD(&page->cp_layers);
194 INIT_LIST_HEAD(&page->cp_batch);
195 INIT_LIST_HEAD(&page->cp_flight);
196 lu_ref_init(&page->cp_reference);
197 head = o->co_lu.lo_header;
198 list_for_each_entry(o, &head->loh_layers,
200 if (o->co_ops->coo_page_init != NULL) {
201 result = o->co_ops->coo_page_init(env, o, page,
204 cl_page_delete0(env, page);
205 cl_page_free(env, page);
206 page = ERR_PTR(result);
212 CS_PAGE_INC(o, total);
213 CS_PAGE_INC(o, create);
214 CS_PAGESTATE_DEC(o, CPS_CACHED);
217 page = ERR_PTR(-ENOMEM);
221 EXPORT_SYMBOL(cl_page_alloc);
224 * Returns a cl_page with index \a idx at the object \a o, and associated with
225 * the VM page \a vmpage.
227 * This is the main entry point into the cl_page caching interface. First, a
228 * cache (implemented as a per-object radix tree) is consulted. If page is
229 * found there, it is returned immediately. Otherwise new page is allocated
230 * and returned. In any case, additional reference to page is acquired.
232 * \see cl_object_find(), cl_lock_find()
234 struct cl_page *cl_page_find(const struct lu_env *env,
236 pgoff_t idx, struct page *vmpage,
237 enum cl_page_type type)
239 struct cl_page *page = NULL;
240 struct cl_object_header *hdr;
242 LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT);
247 hdr = cl_object_header(o);
248 CS_PAGE_INC(o, lookup);
250 CDEBUG(D_PAGE, "%lu@"DFID" %p %lx %d\n",
251 idx, PFID(&hdr->coh_lu.loh_fid), vmpage, vmpage->private, type);
253 if (type == CPT_CACHEABLE) {
254 /* vmpage lock is used to protect the child/parent
256 KLASSERT(PageLocked(vmpage));
258 * cl_vmpage_page() can be called here without any locks as
260 * - "vmpage" is locked (which prevents ->private from
261 * concurrent updates), and
263 * - "o" cannot be destroyed while current thread holds a
266 page = cl_vmpage_page(vmpage, o);
273 /* allocate and initialize cl_page */
274 page = cl_page_alloc(env, o, idx, vmpage, type);
277 EXPORT_SYMBOL(cl_page_find);
279 static inline int cl_page_invariant(const struct cl_page *pg)
281 return cl_page_in_use_noref(pg);
284 static void cl_page_state_set0(const struct lu_env *env,
285 struct cl_page *page, enum cl_page_state state)
287 enum cl_page_state old;
290 * Matrix of allowed state transitions [old][new], for sanity
293 static const int allowed_transitions[CPS_NR][CPS_NR] = {
296 [CPS_OWNED] = 1, /* io finds existing cached page */
298 [CPS_PAGEOUT] = 1, /* write-out from the cache */
299 [CPS_FREEING] = 1, /* eviction on the memory pressure */
302 [CPS_CACHED] = 1, /* release to the cache */
304 [CPS_PAGEIN] = 1, /* start read immediately */
305 [CPS_PAGEOUT] = 1, /* start write immediately */
306 [CPS_FREEING] = 1, /* lock invalidation or truncate */
309 [CPS_CACHED] = 1, /* io completion */
316 [CPS_CACHED] = 1, /* io completion */
332 old = page->cp_state;
333 PASSERT(env, page, allowed_transitions[old][state]);
334 CL_PAGE_HEADER(D_TRACE, env, page, "%d -> %d\n", old, state);
335 PASSERT(env, page, page->cp_state == old);
336 PASSERT(env, page, equi(state == CPS_OWNED, page->cp_owner != NULL));
338 CS_PAGESTATE_DEC(page->cp_obj, page->cp_state);
339 CS_PAGESTATE_INC(page->cp_obj, state);
340 cl_page_state_set_trust(page, state);
344 static void cl_page_state_set(const struct lu_env *env,
345 struct cl_page *page, enum cl_page_state state)
347 cl_page_state_set0(env, page, state);
351 * Acquires an additional reference to a page.
353 * This can be called only by caller already possessing a reference to \a
356 * \see cl_object_get(), cl_lock_get().
358 void cl_page_get(struct cl_page *page)
361 cl_page_get_trust(page);
364 EXPORT_SYMBOL(cl_page_get);
367 * Releases a reference to a page.
369 * When last reference is released, page is returned to the cache, unless it
370 * is in cl_page_state::CPS_FREEING state, in which case it is immediately
373 * \see cl_object_put(), cl_lock_put().
375 void cl_page_put(const struct lu_env *env, struct cl_page *page)
378 CL_PAGE_HEADER(D_TRACE, env, page, "%d\n",
379 atomic_read(&page->cp_ref));
381 if (atomic_dec_and_test(&page->cp_ref)) {
382 LASSERT(page->cp_state == CPS_FREEING);
384 LASSERT(atomic_read(&page->cp_ref) == 0);
385 PASSERT(env, page, page->cp_owner == NULL);
386 PASSERT(env, page, list_empty(&page->cp_batch));
388 * Page is no longer reachable by other threads. Tear
391 cl_page_free(env, page);
396 EXPORT_SYMBOL(cl_page_put);
399 * Returns a cl_page associated with a VM page, and given cl_object.
401 struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj)
403 struct cl_page *page;
406 KLASSERT(PageLocked(vmpage));
409 * NOTE: absence of races and liveness of data are guaranteed by page
410 * lock on a "vmpage". That works because object destruction has
411 * bottom-to-top pass.
414 page = (struct cl_page *)vmpage->private;
416 cl_page_get_trust(page);
417 LASSERT(page->cp_type == CPT_CACHEABLE);
421 EXPORT_SYMBOL(cl_vmpage_page);
423 const struct cl_page_slice *cl_page_at(const struct cl_page *page,
424 const struct lu_device_type *dtype)
426 return cl_page_at_trusted(page, dtype);
428 EXPORT_SYMBOL(cl_page_at);
430 #define CL_PAGE_OP(opname) offsetof(struct cl_page_operations, opname)
432 #define CL_PAGE_INVOKE(_env, _page, _op, _proto, ...) \
434 const struct lu_env *__env = (_env); \
435 struct cl_page *__page = (_page); \
436 const struct cl_page_slice *__scan; \
438 ptrdiff_t __op = (_op); \
439 int (*__method)_proto; \
442 list_for_each_entry(__scan, &__page->cp_layers, cpl_linkage) { \
443 __method = *(void **)((char *)__scan->cpl_ops + __op); \
444 if (__method != NULL) { \
445 __result = (*__method)(__env, __scan, ## __VA_ARGS__); \
455 #define CL_PAGE_INVOKE_REVERSE(_env, _page, _op, _proto, ...) \
457 const struct lu_env *__env = (_env); \
458 struct cl_page *__page = (_page); \
459 const struct cl_page_slice *__scan; \
461 ptrdiff_t __op = (_op); \
462 int (*__method)_proto; \
465 list_for_each_entry_reverse(__scan, &__page->cp_layers, \
467 __method = *(void **)((char *)__scan->cpl_ops + __op); \
468 if (__method != NULL) { \
469 __result = (*__method)(__env, __scan, ## __VA_ARGS__); \
479 #define CL_PAGE_INVOID(_env, _page, _op, _proto, ...) \
481 const struct lu_env *__env = (_env); \
482 struct cl_page *__page = (_page); \
483 const struct cl_page_slice *__scan; \
484 ptrdiff_t __op = (_op); \
485 void (*__method)_proto; \
487 list_for_each_entry(__scan, &__page->cp_layers, cpl_linkage) { \
488 __method = *(void **)((char *)__scan->cpl_ops + __op); \
489 if (__method != NULL) \
490 (*__method)(__env, __scan, ## __VA_ARGS__); \
494 #define CL_PAGE_INVOID_REVERSE(_env, _page, _op, _proto, ...) \
496 const struct lu_env *__env = (_env); \
497 struct cl_page *__page = (_page); \
498 const struct cl_page_slice *__scan; \
499 ptrdiff_t __op = (_op); \
500 void (*__method)_proto; \
502 /* get to the bottom page. */ \
503 list_for_each_entry_reverse(__scan, &__page->cp_layers, \
505 __method = *(void **)((char *)__scan->cpl_ops + __op); \
506 if (__method != NULL) \
507 (*__method)(__env, __scan, ## __VA_ARGS__); \
511 static int cl_page_invoke(const struct lu_env *env,
512 struct cl_io *io, struct cl_page *page, ptrdiff_t op)
515 PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj));
517 RETURN(CL_PAGE_INVOKE(env, page, op,
518 (const struct lu_env *,
519 const struct cl_page_slice *, struct cl_io *),
523 static void cl_page_invoid(const struct lu_env *env,
524 struct cl_io *io, struct cl_page *page, ptrdiff_t op)
527 PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj));
529 CL_PAGE_INVOID(env, page, op,
530 (const struct lu_env *,
531 const struct cl_page_slice *, struct cl_io *), io);
535 static void cl_page_owner_clear(struct cl_page *page)
538 if (page->cp_owner != NULL) {
539 LASSERT(page->cp_owner->ci_owned_nr > 0);
540 page->cp_owner->ci_owned_nr--;
541 page->cp_owner = NULL;
546 static void cl_page_owner_set(struct cl_page *page)
549 LASSERT(page->cp_owner != NULL);
550 page->cp_owner->ci_owned_nr++;
554 void cl_page_disown0(const struct lu_env *env,
555 struct cl_io *io, struct cl_page *pg)
557 enum cl_page_state state;
560 state = pg->cp_state;
561 PINVRNT(env, pg, state == CPS_OWNED || state == CPS_FREEING);
562 PINVRNT(env, pg, cl_page_invariant(pg) || state == CPS_FREEING);
563 cl_page_owner_clear(pg);
565 if (state == CPS_OWNED)
566 cl_page_state_set(env, pg, CPS_CACHED);
568 * Completion call-backs are executed in the bottom-up order, so that
569 * uppermost layer (llite), responsible for VFS/VM interaction runs
570 * last and can release locks safely.
572 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_disown),
573 (const struct lu_env *,
574 const struct cl_page_slice *, struct cl_io *),
580 * returns true, iff page is owned by the given io.
582 int cl_page_is_owned(const struct cl_page *pg, const struct cl_io *io)
584 struct cl_io *top = cl_io_top((struct cl_io *)io);
585 LINVRNT(cl_object_same(pg->cp_obj, io->ci_obj));
587 RETURN(pg->cp_state == CPS_OWNED && pg->cp_owner == top);
589 EXPORT_SYMBOL(cl_page_is_owned);
592 * Try to own a page by IO.
594 * Waits until page is in cl_page_state::CPS_CACHED state, and then switch it
595 * into cl_page_state::CPS_OWNED state.
597 * \pre !cl_page_is_owned(pg, io)
598 * \post result == 0 iff cl_page_is_owned(pg, io)
602 * \retval -ve failure, e.g., page was destroyed (and landed in
603 * cl_page_state::CPS_FREEING instead of cl_page_state::CPS_CACHED).
604 * or, page was owned by another thread, or in IO.
606 * \see cl_page_disown()
607 * \see cl_page_operations::cpo_own()
608 * \see cl_page_own_try()
611 static int cl_page_own0(const struct lu_env *env, struct cl_io *io,
612 struct cl_page *pg, int nonblock)
616 PINVRNT(env, pg, !cl_page_is_owned(pg, io));
621 if (pg->cp_state == CPS_FREEING) {
624 result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(cpo_own),
625 (const struct lu_env *,
626 const struct cl_page_slice *,
627 struct cl_io *, int),
630 PASSERT(env, pg, pg->cp_owner == NULL);
631 PASSERT(env, pg, pg->cp_req == NULL);
632 pg->cp_owner = cl_io_top(io);;
633 cl_page_owner_set(pg);
634 if (pg->cp_state != CPS_FREEING) {
635 cl_page_state_set(env, pg, CPS_OWNED);
637 cl_page_disown0(env, io, pg);
642 PINVRNT(env, pg, ergo(result == 0, cl_page_invariant(pg)));
647 * Own a page, might be blocked.
649 * \see cl_page_own0()
651 int cl_page_own(const struct lu_env *env, struct cl_io *io, struct cl_page *pg)
653 return cl_page_own0(env, io, pg, 0);
655 EXPORT_SYMBOL(cl_page_own);
658 * Nonblock version of cl_page_own().
660 * \see cl_page_own0()
662 int cl_page_own_try(const struct lu_env *env, struct cl_io *io,
665 return cl_page_own0(env, io, pg, 1);
667 EXPORT_SYMBOL(cl_page_own_try);
671 * Assume page ownership.
673 * Called when page is already locked by the hosting VM.
675 * \pre !cl_page_is_owned(pg, io)
676 * \post cl_page_is_owned(pg, io)
678 * \see cl_page_operations::cpo_assume()
680 void cl_page_assume(const struct lu_env *env,
681 struct cl_io *io, struct cl_page *pg)
683 PINVRNT(env, pg, cl_object_same(pg->cp_obj, io->ci_obj));
688 cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_assume));
689 PASSERT(env, pg, pg->cp_owner == NULL);
690 pg->cp_owner = cl_io_top(io);
691 cl_page_owner_set(pg);
692 cl_page_state_set(env, pg, CPS_OWNED);
695 EXPORT_SYMBOL(cl_page_assume);
698 * Releases page ownership without unlocking the page.
700 * Moves page into cl_page_state::CPS_CACHED without releasing a lock on the
701 * underlying VM page (as VM is supposed to do this itself).
703 * \pre cl_page_is_owned(pg, io)
704 * \post !cl_page_is_owned(pg, io)
706 * \see cl_page_assume()
708 void cl_page_unassume(const struct lu_env *env,
709 struct cl_io *io, struct cl_page *pg)
711 PINVRNT(env, pg, cl_page_is_owned(pg, io));
712 PINVRNT(env, pg, cl_page_invariant(pg));
716 cl_page_owner_clear(pg);
717 cl_page_state_set(env, pg, CPS_CACHED);
718 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_unassume),
719 (const struct lu_env *,
720 const struct cl_page_slice *, struct cl_io *),
724 EXPORT_SYMBOL(cl_page_unassume);
727 * Releases page ownership.
729 * Moves page into cl_page_state::CPS_CACHED.
731 * \pre cl_page_is_owned(pg, io)
732 * \post !cl_page_is_owned(pg, io)
735 * \see cl_page_operations::cpo_disown()
737 void cl_page_disown(const struct lu_env *env,
738 struct cl_io *io, struct cl_page *pg)
740 PINVRNT(env, pg, cl_page_is_owned(pg, io) ||
741 pg->cp_state == CPS_FREEING);
745 cl_page_disown0(env, io, pg);
748 EXPORT_SYMBOL(cl_page_disown);
751 * Called when page is to be removed from the object, e.g., as a result of
754 * Calls cl_page_operations::cpo_discard() top-to-bottom.
756 * \pre cl_page_is_owned(pg, io)
758 * \see cl_page_operations::cpo_discard()
760 void cl_page_discard(const struct lu_env *env,
761 struct cl_io *io, struct cl_page *pg)
763 PINVRNT(env, pg, cl_page_is_owned(pg, io));
764 PINVRNT(env, pg, cl_page_invariant(pg));
766 cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_discard));
768 EXPORT_SYMBOL(cl_page_discard);
771 * Version of cl_page_delete() that can be called for not fully constructed
772 * pages, e.g,. in a error handling cl_page_find()->cl_page_delete0()
773 * path. Doesn't check page invariant.
775 static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg)
779 PASSERT(env, pg, pg->cp_state != CPS_FREEING);
782 * Severe all ways to obtain new pointers to @pg.
784 cl_page_owner_clear(pg);
786 cl_page_state_set0(env, pg, CPS_FREEING);
788 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_delete),
789 (const struct lu_env *, const struct cl_page_slice *));
795 * Called when a decision is made to throw page out of memory.
797 * Notifies all layers about page destruction by calling
798 * cl_page_operations::cpo_delete() method top-to-bottom.
800 * Moves page into cl_page_state::CPS_FREEING state (this is the only place
801 * where transition to this state happens).
803 * Eliminates all venues through which new references to the page can be
806 * - removes page from the radix trees,
808 * - breaks linkage from VM page to cl_page.
810 * Once page reaches cl_page_state::CPS_FREEING, all remaining references will
811 * drain after some time, at which point page will be recycled.
813 * \pre VM page is locked
814 * \post pg->cp_state == CPS_FREEING
816 * \see cl_page_operations::cpo_delete()
818 void cl_page_delete(const struct lu_env *env, struct cl_page *pg)
820 PINVRNT(env, pg, cl_page_invariant(pg));
822 cl_page_delete0(env, pg);
825 EXPORT_SYMBOL(cl_page_delete);
828 * Marks page up-to-date.
830 * Call cl_page_operations::cpo_export() through all layers top-to-bottom. The
831 * layer responsible for VM interaction has to mark/clear page as up-to-date
832 * by the \a uptodate argument.
834 * \see cl_page_operations::cpo_export()
836 void cl_page_export(const struct lu_env *env, struct cl_page *pg, int uptodate)
838 PINVRNT(env, pg, cl_page_invariant(pg));
839 CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_export),
840 (const struct lu_env *,
841 const struct cl_page_slice *, int), uptodate);
843 EXPORT_SYMBOL(cl_page_export);
846 * Returns true, iff \a pg is VM locked in a suitable sense by the calling
849 int cl_page_is_vmlocked(const struct lu_env *env, const struct cl_page *pg)
852 const struct cl_page_slice *slice;
855 slice = container_of(pg->cp_layers.next,
856 const struct cl_page_slice, cpl_linkage);
857 PASSERT(env, pg, slice->cpl_ops->cpo_is_vmlocked != NULL);
859 * Call ->cpo_is_vmlocked() directly instead of going through
860 * CL_PAGE_INVOKE(), because cl_page_is_vmlocked() is used by
861 * cl_page_invariant().
863 result = slice->cpl_ops->cpo_is_vmlocked(env, slice);
864 PASSERT(env, pg, result == -EBUSY || result == -ENODATA);
865 RETURN(result == -EBUSY);
867 EXPORT_SYMBOL(cl_page_is_vmlocked);
869 static enum cl_page_state cl_req_type_state(enum cl_req_type crt)
872 RETURN(crt == CRT_WRITE ? CPS_PAGEOUT : CPS_PAGEIN);
875 static void cl_page_io_start(const struct lu_env *env,
876 struct cl_page *pg, enum cl_req_type crt)
879 * Page is queued for IO, change its state.
882 cl_page_owner_clear(pg);
883 cl_page_state_set(env, pg, cl_req_type_state(crt));
888 * Prepares page for immediate transfer. cl_page_operations::cpo_prep() is
889 * called top-to-bottom. Every layer either agrees to submit this page (by
890 * returning 0), or requests to omit this page (by returning -EALREADY). Layer
891 * handling interactions with the VM also has to inform VM that page is under
894 int cl_page_prep(const struct lu_env *env, struct cl_io *io,
895 struct cl_page *pg, enum cl_req_type crt)
899 PINVRNT(env, pg, cl_page_is_owned(pg, io));
900 PINVRNT(env, pg, cl_page_invariant(pg));
901 PINVRNT(env, pg, crt < CRT_NR);
904 * XXX this has to be called bottom-to-top, so that llite can set up
905 * PG_writeback without risking other layers deciding to skip this
910 result = cl_page_invoke(env, io, pg, CL_PAGE_OP(io[crt].cpo_prep));
912 cl_page_io_start(env, pg, crt);
914 KLASSERT(ergo(crt == CRT_WRITE && pg->cp_type == CPT_CACHEABLE,
916 PageWriteback(cl_page_vmpage(pg)))));
917 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
920 EXPORT_SYMBOL(cl_page_prep);
923 * Notify layers about transfer completion.
925 * Invoked by transfer sub-system (which is a part of osc) to notify layers
926 * that a transfer, of which this page is a part of has completed.
928 * Completion call-backs are executed in the bottom-up order, so that
929 * uppermost layer (llite), responsible for the VFS/VM interaction runs last
930 * and can release locks safely.
932 * \pre pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT
933 * \post pg->cp_state == CPS_CACHED
935 * \see cl_page_operations::cpo_completion()
937 void cl_page_completion(const struct lu_env *env,
938 struct cl_page *pg, enum cl_req_type crt, int ioret)
940 struct cl_sync_io *anchor = pg->cp_sync_io;
942 PASSERT(env, pg, crt < CRT_NR);
943 /* cl_page::cp_req already cleared by the caller (osc_completion()) */
944 PASSERT(env, pg, pg->cp_req == NULL);
945 PASSERT(env, pg, pg->cp_state == cl_req_type_state(crt));
948 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, ioret);
949 cl_page_state_set(env, pg, CPS_CACHED);
952 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(io[crt].cpo_completion),
953 (const struct lu_env *,
954 const struct cl_page_slice *, int), ioret);
956 LASSERT(pg->cp_sync_io == anchor);
957 pg->cp_sync_io = NULL;
960 * As page->cp_obj is pinned by a reference from page->cp_req, it is
961 * safe to call cl_page_put() without risking object destruction in a
962 * non-blocking context.
964 cl_page_put(env, pg);
967 cl_sync_io_note(anchor, ioret);
971 EXPORT_SYMBOL(cl_page_completion);
974 * Notify layers that transfer formation engine decided to yank this page from
975 * the cache and to make it a part of a transfer.
977 * \pre pg->cp_state == CPS_CACHED
978 * \post pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT
980 * \see cl_page_operations::cpo_make_ready()
982 int cl_page_make_ready(const struct lu_env *env, struct cl_page *pg,
983 enum cl_req_type crt)
987 PINVRNT(env, pg, crt < CRT_NR);
992 result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(io[crt].cpo_make_ready),
993 (const struct lu_env *,
994 const struct cl_page_slice *));
996 PASSERT(env, pg, pg->cp_state == CPS_CACHED);
997 cl_page_io_start(env, pg, crt);
999 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1002 EXPORT_SYMBOL(cl_page_make_ready);
1005 * Called if a pge is being written back by kernel's intention.
1007 * \pre cl_page_is_owned(pg, io)
1008 * \post ergo(result == 0, pg->cp_state == CPS_PAGEOUT)
1010 * \see cl_page_operations::cpo_flush()
1012 int cl_page_flush(const struct lu_env *env, struct cl_io *io,
1017 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1018 PINVRNT(env, pg, cl_page_invariant(pg));
1022 result = cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_flush));
1024 CL_PAGE_HEADER(D_TRACE, env, pg, "%d\n", result);
1027 EXPORT_SYMBOL(cl_page_flush);
1030 * Checks whether page is protected by any extent lock is at least required
1033 * \return the same as in cl_page_operations::cpo_is_under_lock() method.
1034 * \see cl_page_operations::cpo_is_under_lock()
1036 int cl_page_is_under_lock(const struct lu_env *env, struct cl_io *io,
1037 struct cl_page *page, pgoff_t *max_index)
1041 PINVRNT(env, page, cl_page_invariant(page));
1044 rc = CL_PAGE_INVOKE_REVERSE(env, page, CL_PAGE_OP(cpo_is_under_lock),
1045 (const struct lu_env *,
1046 const struct cl_page_slice *,
1047 struct cl_io *, pgoff_t *),
1051 EXPORT_SYMBOL(cl_page_is_under_lock);
1054 * Tells transfer engine that only part of a page is to be transmitted.
1056 * \see cl_page_operations::cpo_clip()
1058 void cl_page_clip(const struct lu_env *env, struct cl_page *pg,
1061 PINVRNT(env, pg, cl_page_invariant(pg));
1063 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", from, to);
1064 CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_clip),
1065 (const struct lu_env *,
1066 const struct cl_page_slice *,int, int),
1069 EXPORT_SYMBOL(cl_page_clip);
1072 * Prints human readable representation of \a pg to the \a f.
1074 void cl_page_header_print(const struct lu_env *env, void *cookie,
1075 lu_printer_t printer, const struct cl_page *pg)
1077 (*printer)(env, cookie,
1078 "page@%p[%d %p %d %d %d %p %p]\n",
1079 pg, atomic_read(&pg->cp_ref), pg->cp_obj,
1080 pg->cp_state, pg->cp_error, pg->cp_type,
1081 pg->cp_owner, pg->cp_req);
1083 EXPORT_SYMBOL(cl_page_header_print);
1086 * Prints human readable representation of \a pg to the \a f.
1088 void cl_page_print(const struct lu_env *env, void *cookie,
1089 lu_printer_t printer, const struct cl_page *pg)
1091 cl_page_header_print(env, cookie, printer, pg);
1092 CL_PAGE_INVOKE(env, (struct cl_page *)pg, CL_PAGE_OP(cpo_print),
1093 (const struct lu_env *env,
1094 const struct cl_page_slice *slice,
1095 void *cookie, lu_printer_t p), cookie, printer);
1096 (*printer)(env, cookie, "end page@%p\n", pg);
1098 EXPORT_SYMBOL(cl_page_print);
1101 * Cancel a page which is still in a transfer.
1103 int cl_page_cancel(const struct lu_env *env, struct cl_page *page)
1105 return CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_cancel),
1106 (const struct lu_env *,
1107 const struct cl_page_slice *));
1109 EXPORT_SYMBOL(cl_page_cancel);
1112 * Converts a byte offset within object \a obj into a page index.
1114 loff_t cl_offset(const struct cl_object *obj, pgoff_t idx)
1116 return (loff_t)idx << PAGE_CACHE_SHIFT;
1118 EXPORT_SYMBOL(cl_offset);
1121 * Converts a page index into a byte offset within object \a obj.
1123 pgoff_t cl_index(const struct cl_object *obj, loff_t offset)
1125 return offset >> PAGE_CACHE_SHIFT;
1127 EXPORT_SYMBOL(cl_index);
1129 int cl_page_size(const struct cl_object *obj)
1131 return 1 << PAGE_CACHE_SHIFT;
1133 EXPORT_SYMBOL(cl_page_size);
1136 * Adds page slice to the compound page.
1138 * This is called by cl_object_operations::coo_page_init() methods to add a
1139 * per-layer state to the page. New state is added at the end of
1140 * cl_page::cp_layers list, that is, it is at the bottom of the stack.
1142 * \see cl_lock_slice_add(), cl_req_slice_add(), cl_io_slice_add()
1144 void cl_page_slice_add(struct cl_page *page, struct cl_page_slice *slice,
1145 struct cl_object *obj, pgoff_t index,
1146 const struct cl_page_operations *ops)
1149 list_add_tail(&slice->cpl_linkage, &page->cp_layers);
1150 slice->cpl_obj = obj;
1151 slice->cpl_index = index;
1152 slice->cpl_ops = ops;
1153 slice->cpl_page = page;
1156 EXPORT_SYMBOL(cl_page_slice_add);
1158 int cl_page_init(void)
1163 void cl_page_fini(void)