projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/trivial
[linux-2.6-block.git] / drivers / staging / lustre / lustre / obdclass / cl_page.c
CommitLineData
d7e09d03
PT		 1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * Client Lustre Page.
37 *
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 */
40
41#define DEBUG_SUBSYSTEM S_CLASS
42
43#include <linux/libcfs/libcfs.h>
44#include <obd_class.h>
45#include <obd_support.h>
46#include <linux/list.h>
47
48#include <cl_object.h>
49#include "cl_internal.h"
50
51static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg,
52 int radix);
53
54# define PASSERT(env, page, expr) \
55 do { \
56 if (unlikely(!(expr))) { \
57 CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \
58 LASSERT(0); \
59 } \
60 } while (0)
61
62# define PINVRNT(env, page, exp) \
63 ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp))
64
65/* Disable page statistic by default due to huge performance penalty. */
66#define CS_PAGE_INC(o, item)
67#define CS_PAGE_DEC(o, item)
68#define CS_PAGESTATE_INC(o, state)
69#define CS_PAGESTATE_DEC(o, state)
70
71/**
72 * Internal version of cl_page_top, it should be called if the page is
73 * known to be not freed, says with page referenced, or radix tree lock held,
74 * or page owned.
75 */
76static struct cl_page *cl_page_top_trusted(struct cl_page *page)
77{
78 while (page->cp_parent != NULL)
79 page = page->cp_parent;
80 return page;
81}
82
83/**
84 * Internal version of cl_page_get().
85 *
86 * This function can be used to obtain initial reference to previously
87 * unreferenced cached object. It can be called only if concurrent page
88 * reclamation is somehow prevented, e.g., by locking page radix-tree
89 * (cl_object_header::hdr->coh_page_guard), or by keeping a lock on a VM page,
90 * associated with \a page.
91 *
92 * Use with care! Not exported.
93 */
94static void cl_page_get_trust(struct cl_page *page)
95{
96 LASSERT(atomic_read(&page->cp_ref) > 0);
97 atomic_inc(&page->cp_ref);
98}
99
100/**
101 * Returns a slice within a page, corresponding to the given layer in the
102 * device stack.
103 *
104 * \see cl_lock_at()
105 */
106static const struct cl_page_slice *
107cl_page_at_trusted(const struct cl_page *page,
108 const struct lu_device_type *dtype)
109{
110 const struct cl_page_slice *slice;
111 ENTRY;
112
113 page = cl_page_top_trusted((struct cl_page *)page);
114 do {
115 list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
116 if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype)
117 RETURN(slice);
118 }
119 page = page->cp_child;
120 } while (page != NULL);
121 RETURN(NULL);
122}
123
124/**
125 * Returns a page with given index in the given object, or NULL if no page is
126 * found. Acquires a reference on \a page.
127 *
128 * Locking: called under cl_object_header::coh_page_guard spin-lock.
129 */
130struct cl_page *cl_page_lookup(struct cl_object_header *hdr, pgoff_t index)
131{
132 struct cl_page *page;
133
134 LASSERT(spin_is_locked(&hdr->coh_page_guard));
135
136 page = radix_tree_lookup(&hdr->coh_tree, index);
137 if (page != NULL)
138 cl_page_get_trust(page);
139 return page;
140}
141EXPORT_SYMBOL(cl_page_lookup);
142
143/**
144 * Returns a list of pages by a given [start, end] of \a obj.
145 *
146 * \param resched If not NULL, then we give up before hogging CPU for too
147 * long and set *resched = 1, in that case caller should implement a retry
148 * logic.
149 *
150 * Gang tree lookup (radix_tree_gang_lookup()) optimization is absolutely
151 * crucial in the face of [offset, EOF] locks.
152 *
153 * Return at least one page in @queue unless there is no covered page.
154 */
155int cl_page_gang_lookup(const struct lu_env *env, struct cl_object *obj,
156 struct cl_io *io, pgoff_t start, pgoff_t end,
157 cl_page_gang_cb_t cb, void *cbdata)
158{
159 struct cl_object_header *hdr;
160 struct cl_page *page;
161 struct cl_page **pvec;
162 const struct cl_page_slice *slice;
163 const struct lu_device_type *dtype;
164 pgoff_t idx;
165 unsigned int nr;
166 unsigned int i;
167 unsigned int j;
168 int res = CLP_GANG_OKAY;
169 int tree_lock = 1;
170 ENTRY;
171
172 idx = start;
173 hdr = cl_object_header(obj);
174 pvec = cl_env_info(env)->clt_pvec;
175 dtype = cl_object_top(obj)->co_lu.lo_dev->ld_type;
176 spin_lock(&hdr->coh_page_guard);
177 while ((nr = radix_tree_gang_lookup(&hdr->coh_tree, (void **)pvec,
178 idx, CLT_PVEC_SIZE)) > 0) {
179 int end_of_region = 0;
180 idx = pvec[nr - 1]->cp_index + 1;
181 for (i = 0, j = 0; i < nr; ++i) {
182 page = pvec[i];
183 pvec[i] = NULL;
184
185 LASSERT(page->cp_type == CPT_CACHEABLE);
186 if (page->cp_index > end) {
187 end_of_region = 1;
188 break;
189 }
190 if (page->cp_state == CPS_FREEING)
191 continue;
192
193 slice = cl_page_at_trusted(page, dtype);
194 /*
195 * Pages for lsm-less file has no underneath sub-page
196 * for osc, in case of ...
197 */
198 PASSERT(env, page, slice != NULL);
199
200 page = slice->cpl_page;
201 /*
202 * Can safely call cl_page_get_trust() under
203 * radix-tree spin-lock.
204 *
205 * XXX not true, because @page is from object another
206 * than @hdr and protected by different tree lock.
207 */
208 cl_page_get_trust(page);
209 lu_ref_add_atomic(&page->cp_reference,
210 "gang_lookup", current);
211 pvec[j++] = page;
212 }
213
214 /*
215 * Here a delicate locking dance is performed. Current thread
216 * holds a reference to a page, but has to own it before it
217 * can be placed into queue. Owning implies waiting, so
218 * radix-tree lock is to be released. After a wait one has to
219 * check that pages weren't truncated (cl_page_own() returns
220 * error in the latter case).
221 */
222 spin_unlock(&hdr->coh_page_guard);
223 tree_lock = 0;
224
225 for (i = 0; i < j; ++i) {
226 page = pvec[i];
227 if (res == CLP_GANG_OKAY)
228 res = (*cb)(env, io, page, cbdata);
229 lu_ref_del(&page->cp_reference,
230 "gang_lookup", current);
231 cl_page_put(env, page);
232 }
233 if (nr < CLT_PVEC_SIZE || end_of_region)
234 break;
235
236 if (res == CLP_GANG_OKAY && need_resched())
237 res = CLP_GANG_RESCHED;
238 if (res != CLP_GANG_OKAY)
239 break;
240
241 spin_lock(&hdr->coh_page_guard);
242 tree_lock = 1;
243 }
244 if (tree_lock)
245 spin_unlock(&hdr->coh_page_guard);
246 RETURN(res);
247}
248EXPORT_SYMBOL(cl_page_gang_lookup);
249
250static void cl_page_free(const struct lu_env *env, struct cl_page *page)
251{
252 struct cl_object *obj = page->cp_obj;
253 int pagesize = cl_object_header(obj)->coh_page_bufsize;
254
255 PASSERT(env, page, list_empty(&page->cp_batch));
256 PASSERT(env, page, page->cp_owner == NULL);
257 PASSERT(env, page, page->cp_req == NULL);
258 PASSERT(env, page, page->cp_parent == NULL);
259 PASSERT(env, page, page->cp_state == CPS_FREEING);
260
261 ENTRY;
262 might_sleep();
263 while (!list_empty(&page->cp_layers)) {
264 struct cl_page_slice *slice;
265
266 slice = list_entry(page->cp_layers.next,
267 struct cl_page_slice, cpl_linkage);
268 list_del_init(page->cp_layers.next);
269 slice->cpl_ops->cpo_fini(env, slice);
270 }
271 CS_PAGE_DEC(obj, total);
272 CS_PAGESTATE_DEC(obj, page->cp_state);
273 lu_object_ref_del_at(&obj->co_lu, page->cp_obj_ref, "cl_page", page);
274 cl_object_put(env, obj);
275 lu_ref_fini(&page->cp_reference);
276 OBD_FREE(page, pagesize);
277 EXIT;
278}
279
280/**
281 * Helper function updating page state. This is the only place in the code
282 * where cl_page::cp_state field is mutated.
283 */
284static inline void cl_page_state_set_trust(struct cl_page *page,
285 enum cl_page_state state)
286{
287 /* bypass const. */
288 *(enum cl_page_state *)&page->cp_state = state;
289}
290
291static struct cl_page *cl_page_alloc(const struct lu_env *env,
292 struct cl_object *o, pgoff_t ind, struct page *vmpage,
293 enum cl_page_type type)
294{
295 struct cl_page *page;
296 struct lu_object_header *head;
297
298 ENTRY;
299 OBD_ALLOC_GFP(page, cl_object_header(o)->coh_page_bufsize,
300 __GFP_IO);
301 if (page != NULL) {
302 int result = 0;
303 atomic_set(&page->cp_ref, 1);
304 if (type == CPT_CACHEABLE) /* for radix tree */
305 atomic_inc(&page->cp_ref);
306 page->cp_obj = o;
307 cl_object_get(o);
308 page->cp_obj_ref = lu_object_ref_add(&o->co_lu, "cl_page",page);
309 page->cp_index = ind;
310 cl_page_state_set_trust(page, CPS_CACHED);
311 page->cp_type = type;
312 INIT_LIST_HEAD(&page->cp_layers);
313 INIT_LIST_HEAD(&page->cp_batch);
314 INIT_LIST_HEAD(&page->cp_flight);
315 mutex_init(&page->cp_mutex);
316 lu_ref_init(&page->cp_reference);
317 head = o->co_lu.lo_header;
318 list_for_each_entry(o, &head->loh_layers,
319 co_lu.lo_linkage) {
320 if (o->co_ops->coo_page_init != NULL) {
321 result = o->co_ops->coo_page_init(env, o,
322 page, vmpage);
323 if (result != 0) {
324 cl_page_delete0(env, page, 0);
325 cl_page_free(env, page);
326 page = ERR_PTR(result);
327 break;
328 }
329 }
330 }
331 if (result == 0) {
332 CS_PAGE_INC(o, total);
333 CS_PAGE_INC(o, create);
334 CS_PAGESTATE_DEC(o, CPS_CACHED);
335 }
336 } else {
337 page = ERR_PTR(-ENOMEM);
338 }
339 RETURN(page);
340}
341
342/**
343 * Returns a cl_page with index \a idx at the object \a o, and associated with
344 * the VM page \a vmpage.
345 *
346 * This is the main entry point into the cl_page caching interface. First, a
347 * cache (implemented as a per-object radix tree) is consulted. If page is
348 * found there, it is returned immediately. Otherwise new page is allocated
349 * and returned. In any case, additional reference to page is acquired.
350 *
351 * \see cl_object_find(), cl_lock_find()
352 */
353static struct cl_page *cl_page_find0(const struct lu_env *env,
354 struct cl_object *o,
355 pgoff_t idx, struct page *vmpage,
356 enum cl_page_type type,
357 struct cl_page *parent)
358{
359 struct cl_page *page = NULL;
360 struct cl_page *ghost = NULL;
361 struct cl_object_header *hdr;
362 int err;
363
364 LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT);
365 might_sleep();
366
367 ENTRY;
368
369 hdr = cl_object_header(o);
370 CS_PAGE_INC(o, lookup);
371
372 CDEBUG(D_PAGE, "%lu@"DFID" %p %lx %d\n",
373 idx, PFID(&hdr->coh_lu.loh_fid), vmpage, vmpage->private, type);
374 /* fast path. */
375 if (type == CPT_CACHEABLE) {
376 /* vmpage lock is used to protect the child/parent
377 * relationship */
378 KLASSERT(PageLocked(vmpage));
379 /*
380 * cl_vmpage_page() can be called here without any locks as
381 *
382 * - "vmpage" is locked (which prevents ->private from
383 * concurrent updates), and
384 *
385 * - "o" cannot be destroyed while current thread holds a
386 * reference on it.
387 */
388 page = cl_vmpage_page(vmpage, o);
389 PINVRNT(env, page,
390 ergo(page != NULL,
391 cl_page_vmpage(env, page) == vmpage &&
392 (void *)radix_tree_lookup(&hdr->coh_tree,
393 idx) == page));
394 }
395
396 if (page != NULL) {
397 CS_PAGE_INC(o, hit);
398 RETURN(page);
399 }
400
401 /* allocate and initialize cl_page */
402 page = cl_page_alloc(env, o, idx, vmpage, type);
403 if (IS_ERR(page))
404 RETURN(page);
405
406 if (type == CPT_TRANSIENT) {
407 if (parent) {
408 LASSERT(page->cp_parent == NULL);
409 page->cp_parent = parent;
410 parent->cp_child = page;
411 }
412 RETURN(page);
413 }
414
415 /*
416 * XXX optimization: use radix_tree_preload() here, and change tree
417 * gfp mask to GFP_KERNEL in cl_object_header_init().
418 */
419 spin_lock(&hdr->coh_page_guard);
420 err = radix_tree_insert(&hdr->coh_tree, idx, page);
421 if (err != 0) {
422 ghost = page;
423 /*
424 * Noted by Jay: a lock on \a vmpage protects cl_page_find()
425 * from this race, but
426 *
427 * 0. it's better to have cl_page interface "locally
428 * consistent" so that its correctness can be reasoned
429 * about without appealing to the (obscure world of) VM
430 * locking.
431 *
432 * 1. handling this race allows ->coh_tree to remain
433 * consistent even when VM locking is somehow busted,
434 * which is very useful during diagnosing and debugging.
435 */
436 page = ERR_PTR(err);
437 CL_PAGE_DEBUG(D_ERROR, env, ghost,
438 "fail to insert into radix tree: %d\n", err);
439 } else {
440 if (parent) {
441 LASSERT(page->cp_parent == NULL);
442 page->cp_parent = parent;
443 parent->cp_child = page;
444 }
445 hdr->coh_pages++;
446 }
447 spin_unlock(&hdr->coh_page_guard);
448
449 if (unlikely(ghost != NULL)) {
450 cl_page_delete0(env, ghost, 0);
451 cl_page_free(env, ghost);
452 }
453 RETURN(page);
454}
455
456struct cl_page *cl_page_find(const struct lu_env *env, struct cl_object *o,
457 pgoff_t idx, struct page *vmpage,
458 enum cl_page_type type)
459{
460 return cl_page_find0(env, o, idx, vmpage, type, NULL);
461}
462EXPORT_SYMBOL(cl_page_find);
463
464
465struct cl_page *cl_page_find_sub(const struct lu_env *env, struct cl_object *o,
466 pgoff_t idx, struct page *vmpage,
467 struct cl_page *parent)
468{
469 return cl_page_find0(env, o, idx, vmpage, parent->cp_type, parent);
470}
471EXPORT_SYMBOL(cl_page_find_sub);
472
473static inline int cl_page_invariant(const struct cl_page *pg)
474{
475 struct cl_object_header *header;
476 struct cl_page *parent;
477 struct cl_page *child;
478 struct cl_io *owner;
479
480 /*
481 * Page invariant is protected by a VM lock.
482 */
483 LINVRNT(cl_page_is_vmlocked(NULL, pg));
484
485 header = cl_object_header(pg->cp_obj);
486 parent = pg->cp_parent;
487 child = pg->cp_child;
488 owner = pg->cp_owner;
489
490 return cl_page_in_use(pg) &&
491 ergo(parent != NULL, parent->cp_child == pg) &&
492 ergo(child != NULL, child->cp_parent == pg) &&
493 ergo(child != NULL, pg->cp_obj != child->cp_obj) &&
494 ergo(parent != NULL, pg->cp_obj != parent->cp_obj) &&
495 ergo(owner != NULL && parent != NULL,
496 parent->cp_owner == pg->cp_owner->ci_parent) &&
497 ergo(owner != NULL && child != NULL,
498 child->cp_owner->ci_parent == owner) &&
499 /*
500 * Either page is early in initialization (has neither child
501 * nor parent yet), or it is in the object radix tree.
502 */
503 ergo(pg->cp_state < CPS_FREEING && pg->cp_type == CPT_CACHEABLE,
504 (void *)radix_tree_lookup(&header->coh_tree,
505 pg->cp_index) == pg ||
506 (child == NULL && parent == NULL));
507}
508
509static void cl_page_state_set0(const struct lu_env *env,
510 struct cl_page *page, enum cl_page_state state)
511{
512 enum cl_page_state old;
513
514 /*
515 * Matrix of allowed state transitions [old][new], for sanity
516 * checking.
517 */
518 static const int allowed_transitions[CPS_NR][CPS_NR] = {
519 [CPS_CACHED] = {
520 [CPS_CACHED] = 0,
521 [CPS_OWNED] = 1, /* io finds existing cached page */
522 [CPS_PAGEIN] = 0,
523 [CPS_PAGEOUT] = 1, /* write-out from the cache */
524 [CPS_FREEING] = 1, /* eviction on the memory pressure */
525 },
526 [CPS_OWNED] = {
527 [CPS_CACHED] = 1, /* release to the cache */
528 [CPS_OWNED] = 0,
529 [CPS_PAGEIN] = 1, /* start read immediately */
530 [CPS_PAGEOUT] = 1, /* start write immediately */
531 [CPS_FREEING] = 1, /* lock invalidation or truncate */
532 },
533 [CPS_PAGEIN] = {
534 [CPS_CACHED] = 1, /* io completion */
535 [CPS_OWNED] = 0,
536 [CPS_PAGEIN] = 0,
537 [CPS_PAGEOUT] = 0,
538 [CPS_FREEING] = 0,
539 },
540 [CPS_PAGEOUT] = {
541 [CPS_CACHED] = 1, /* io completion */
542 [CPS_OWNED] = 0,
543 [CPS_PAGEIN] = 0,
544 [CPS_PAGEOUT] = 0,
545 [CPS_FREEING] = 0,
546 },
547 [CPS_FREEING] = {
548 [CPS_CACHED] = 0,
549 [CPS_OWNED] = 0,
550 [CPS_PAGEIN] = 0,
551 [CPS_PAGEOUT] = 0,
552 [CPS_FREEING] = 0,
553 }
554 };
555
556 ENTRY;
557 old = page->cp_state;
558 PASSERT(env, page, allowed_transitions[old][state]);
559 CL_PAGE_HEADER(D_TRACE, env, page, "%d -> %d\n", old, state);
560 for (; page != NULL; page = page->cp_child) {
561 PASSERT(env, page, page->cp_state == old);
562 PASSERT(env, page,
563 equi(state == CPS_OWNED, page->cp_owner != NULL));
564
565 CS_PAGESTATE_DEC(page->cp_obj, page->cp_state);
566 CS_PAGESTATE_INC(page->cp_obj, state);
567 cl_page_state_set_trust(page, state);
568 }
569 EXIT;
570}
571
572static void cl_page_state_set(const struct lu_env *env,
573 struct cl_page *page, enum cl_page_state state)
574{
575 cl_page_state_set0(env, page, state);
576}
577
578/**
579 * Acquires an additional reference to a page.
580 *
581 * This can be called only by caller already possessing a reference to \a
582 * page.
583 *
584 * \see cl_object_get(), cl_lock_get().
585 */
586void cl_page_get(struct cl_page *page)
587{
588 ENTRY;
589 cl_page_get_trust(page);
590 EXIT;
591}
592EXPORT_SYMBOL(cl_page_get);
593
594/**
595 * Releases a reference to a page.
596 *
597 * When last reference is released, page is returned to the cache, unless it
598 * is in cl_page_state::CPS_FREEING state, in which case it is immediately
599 * destroyed.
600 *
601 * \see cl_object_put(), cl_lock_put().
602 */
603void cl_page_put(const struct lu_env *env, struct cl_page *page)
604{
605 PASSERT(env, page, atomic_read(&page->cp_ref) > !!page->cp_parent);
606
607 ENTRY;
608 CL_PAGE_HEADER(D_TRACE, env, page, "%d\n",
609 atomic_read(&page->cp_ref));
610
611 if (atomic_dec_and_test(&page->cp_ref)) {
612 LASSERT(page->cp_state == CPS_FREEING);
613
614 LASSERT(atomic_read(&page->cp_ref) == 0);
615 PASSERT(env, page, page->cp_owner == NULL);
616 PASSERT(env, page, list_empty(&page->cp_batch));
617 /*
618 * Page is no longer reachable by other threads. Tear
619 * it down.
620 */
621 cl_page_free(env, page);
622 }
623
624 EXIT;
625}
626EXPORT_SYMBOL(cl_page_put);
627
628/**
629 * Returns a VM page associated with a given cl_page.
630 */
631struct page *cl_page_vmpage(const struct lu_env *env, struct cl_page *page)
632{
633 const struct cl_page_slice *slice;
634
635 /*
636 * Find uppermost layer with ->cpo_vmpage() method, and return its
637 * result.
638 */
639 page = cl_page_top(page);
640 do {
641 list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
642 if (slice->cpl_ops->cpo_vmpage != NULL)
643 RETURN(slice->cpl_ops->cpo_vmpage(env, slice));
644 }
645 page = page->cp_child;
646 } while (page != NULL);
647 LBUG(); /* ->cpo_vmpage() has to be defined somewhere in the stack */
648}
649EXPORT_SYMBOL(cl_page_vmpage);
650
651/**
652 * Returns a cl_page associated with a VM page, and given cl_object.
653 */
654struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj)
655{
656 struct cl_page *top;
657 struct cl_page *page;
658
659 ENTRY;
660 KLASSERT(PageLocked(vmpage));
661
662 /*
663 * NOTE: absence of races and liveness of data are guaranteed by page
664 * lock on a "vmpage". That works because object destruction has
665 * bottom-to-top pass.
666 */
667
668 /*
669 * This loop assumes that ->private points to the top-most page. This
670 * can be rectified easily.
671 */
672 top = (struct cl_page *)vmpage->private;
673 if (top == NULL)
674 RETURN(NULL);
675
676 for (page = top; page != NULL; page = page->cp_child) {
677 if (cl_object_same(page->cp_obj, obj)) {
678 cl_page_get_trust(page);
679 break;
680 }
681 }
682 LASSERT(ergo(page, page->cp_type == CPT_CACHEABLE));
683 RETURN(page);
684}
685EXPORT_SYMBOL(cl_vmpage_page);
686
687/**
688 * Returns the top-page for a given page.
689 *
690 * \see cl_object_top(), cl_io_top()
691 */
692struct cl_page *cl_page_top(struct cl_page *page)
693{
694 return cl_page_top_trusted(page);
695}
696EXPORT_SYMBOL(cl_page_top);
697
698const struct cl_page_slice *cl_page_at(const struct cl_page *page,
699 const struct lu_device_type *dtype)
700{
701 return cl_page_at_trusted(page, dtype);
702}
703EXPORT_SYMBOL(cl_page_at);
704
705#define CL_PAGE_OP(opname) offsetof(struct cl_page_operations, opname)
706
707#define CL_PAGE_INVOKE(_env, _page, _op, _proto, ...) \
708({ \
709 const struct lu_env *__env = (_env); \
710 struct cl_page *__page = (_page); \
711 const struct cl_page_slice *__scan; \
712 int __result; \
713 ptrdiff_t __op = (_op); \
714 int (*__method)_proto; \
715 \
716 __result = 0; \
717 __page = cl_page_top(__page); \
718 do { \
719 list_for_each_entry(__scan, &__page->cp_layers, \
720 cpl_linkage) { \
721 __method = *(void **)((char *)__scan->cpl_ops + \
722 __op); \
723 if (__method != NULL) { \
724 __result = (*__method)(__env, __scan, \
725 ## __VA_ARGS__); \
726 if (__result != 0) \
727 break; \
728 } \
729 } \
730 __page = __page->cp_child; \
731 } while (__page != NULL && __result == 0); \
732 if (__result > 0) \
733 __result = 0; \
734 __result; \
735})
736
737#define CL_PAGE_INVOID(_env, _page, _op, _proto, ...) \
738do { \
739 const struct lu_env *__env = (_env); \
740 struct cl_page *__page = (_page); \
741 const struct cl_page_slice *__scan; \
742 ptrdiff_t __op = (_op); \
743 void (*__method)_proto; \
744 \
745 __page = cl_page_top(__page); \
746 do { \
747 list_for_each_entry(__scan, &__page->cp_layers, \
748 cpl_linkage) { \
749 __method = *(void **)((char *)__scan->cpl_ops + \
750 __op); \
751 if (__method != NULL) \
752 (*__method)(__env, __scan, \
753 ## __VA_ARGS__); \
754 } \
755 __page = __page->cp_child; \
756 } while (__page != NULL); \
757} while (0)
758
759#define CL_PAGE_INVOID_REVERSE(_env, _page, _op, _proto, ...) \
760do { \
761 const struct lu_env *__env = (_env); \
762 struct cl_page *__page = (_page); \
763 const struct cl_page_slice *__scan; \
764 ptrdiff_t __op = (_op); \
765 void (*__method)_proto; \
766 \
767 /* get to the bottom page. */ \
768 while (__page->cp_child != NULL) \
769 __page = __page->cp_child; \
770 do { \
771 list_for_each_entry_reverse(__scan, &__page->cp_layers, \
772 cpl_linkage) { \
773 __method = *(void **)((char *)__scan->cpl_ops + \
774 __op); \
775 if (__method != NULL) \
776 (*__method)(__env, __scan, \
777 ## __VA_ARGS__); \
778 } \
779 __page = __page->cp_parent; \
780 } while (__page != NULL); \
781} while (0)
782
783static int cl_page_invoke(const struct lu_env *env,
784 struct cl_io *io, struct cl_page *page, ptrdiff_t op)
785
786{
787 PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj));
788 ENTRY;
789 RETURN(CL_PAGE_INVOKE(env, page, op,
790 (const struct lu_env *,
791 const struct cl_page_slice *, struct cl_io *),
792 io));
793}
794
795static void cl_page_invoid(const struct lu_env *env,
796 struct cl_io *io, struct cl_page *page, ptrdiff_t op)
797
798{
799 PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj));
800 ENTRY;
801 CL_PAGE_INVOID(env, page, op,
802 (const struct lu_env *,
803 const struct cl_page_slice *, struct cl_io *), io);
804 EXIT;
805}
806
807static void cl_page_owner_clear(struct cl_page *page)
808{
809 ENTRY;
810 for (page = cl_page_top(page); page != NULL; page = page->cp_child) {
811 if (page->cp_owner != NULL) {
812 LASSERT(page->cp_owner->ci_owned_nr > 0);
813 page->cp_owner->ci_owned_nr--;
814 page->cp_owner = NULL;
815 page->cp_task = NULL;
816 }
817 }
818 EXIT;
819}
820
821static void cl_page_owner_set(struct cl_page *page)
822{
823 ENTRY;
824 for (page = cl_page_top(page); page != NULL; page = page->cp_child) {
825 LASSERT(page->cp_owner != NULL);
826 page->cp_owner->ci_owned_nr++;
827 }
828 EXIT;
829}
830
831void cl_page_disown0(const struct lu_env *env,
832 struct cl_io *io, struct cl_page *pg)
833{
834 enum cl_page_state state;
835
836 ENTRY;
837 state = pg->cp_state;
838 PINVRNT(env, pg, state == CPS_OWNED || state == CPS_FREEING);
839 PINVRNT(env, pg, cl_page_invariant(pg));
840 cl_page_owner_clear(pg);
841
842 if (state == CPS_OWNED)
843 cl_page_state_set(env, pg, CPS_CACHED);
844 /*
845 * Completion call-backs are executed in the bottom-up order, so that
846 * uppermost layer (llite), responsible for VFS/VM interaction runs
847 * last and can release locks safely.
848 */
849 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_disown),
850 (const struct lu_env *,
851 const struct cl_page_slice *, struct cl_io *),
852 io);
853 EXIT;
854}
855
856/**
857 * returns true, iff page is owned by the given io.
858 */
859int cl_page_is_owned(const struct cl_page *pg, const struct cl_io *io)
860{
861 LINVRNT(cl_object_same(pg->cp_obj, io->ci_obj));
862 ENTRY;
863 RETURN(pg->cp_state == CPS_OWNED && pg->cp_owner == io);
864}
865EXPORT_SYMBOL(cl_page_is_owned);
866
867/**
868 * Try to own a page by IO.
869 *
870 * Waits until page is in cl_page_state::CPS_CACHED state, and then switch it
871 * into cl_page_state::CPS_OWNED state.
872 *
873 * \pre !cl_page_is_owned(pg, io)
874 * \post result == 0 iff cl_page_is_owned(pg, io)
875 *
876 * \retval 0 success
877 *
878 * \retval -ve failure, e.g., page was destroyed (and landed in
879 * cl_page_state::CPS_FREEING instead of cl_page_state::CPS_CACHED).
880 * or, page was owned by another thread, or in IO.
881 *
882 * \see cl_page_disown()
883 * \see cl_page_operations::cpo_own()
884 * \see cl_page_own_try()
885 * \see cl_page_own
886 */
887static int cl_page_own0(const struct lu_env *env, struct cl_io *io,
888 struct cl_page *pg, int nonblock)
889{
890 int result;
891
892 PINVRNT(env, pg, !cl_page_is_owned(pg, io));
893
894 ENTRY;
895 pg = cl_page_top(pg);
896 io = cl_io_top(io);
897
898 if (pg->cp_state == CPS_FREEING) {
899 result = -ENOENT;
900 } else {
901 result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(cpo_own),
902 (const struct lu_env *,
903 const struct cl_page_slice *,
904 struct cl_io *, int),
905 io, nonblock);
906 if (result == 0) {
907 PASSERT(env, pg, pg->cp_owner == NULL);
908 PASSERT(env, pg, pg->cp_req == NULL);
909 pg->cp_owner = io;
910 pg->cp_task = current;
911 cl_page_owner_set(pg);
912 if (pg->cp_state != CPS_FREEING) {
913 cl_page_state_set(env, pg, CPS_OWNED);
914 } else {
915 cl_page_disown0(env, io, pg);
916 result = -ENOENT;
917 }
918 }
919 }
920 PINVRNT(env, pg, ergo(result == 0, cl_page_invariant(pg)));
921 RETURN(result);
922}
923
924/**
925 * Own a page, might be blocked.
926 *
927 * \see cl_page_own0()
928 */
929int cl_page_own(const struct lu_env *env, struct cl_io *io, struct cl_page *pg)
930{
931 return cl_page_own0(env, io, pg, 0);
932}
933EXPORT_SYMBOL(cl_page_own);
934
935/**
936 * Nonblock version of cl_page_own().
937 *
938 * \see cl_page_own0()
939 */
940int cl_page_own_try(const struct lu_env *env, struct cl_io *io,
941 struct cl_page *pg)
942{
943 return cl_page_own0(env, io, pg, 1);
944}
945EXPORT_SYMBOL(cl_page_own_try);
946
947
948/**
949 * Assume page ownership.
950 *
951 * Called when page is already locked by the hosting VM.
952 *
953 * \pre !cl_page_is_owned(pg, io)
954 * \post cl_page_is_owned(pg, io)
955 *
956 * \see cl_page_operations::cpo_assume()
957 */
958void cl_page_assume(const struct lu_env *env,
959 struct cl_io *io, struct cl_page *pg)
960{
961 PINVRNT(env, pg, cl_object_same(pg->cp_obj, io->ci_obj));
962
963 ENTRY;
964 pg = cl_page_top(pg);
965 io = cl_io_top(io);
966
967 cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_assume));
968 PASSERT(env, pg, pg->cp_owner == NULL);
969 pg->cp_owner = io;
970 pg->cp_task = current;
971 cl_page_owner_set(pg);
972 cl_page_state_set(env, pg, CPS_OWNED);
973 EXIT;
974}
975EXPORT_SYMBOL(cl_page_assume);
976
977/**
978 * Releases page ownership without unlocking the page.
979 *
980 * Moves page into cl_page_state::CPS_CACHED without releasing a lock on the
981 * underlying VM page (as VM is supposed to do this itself).
982 *
983 * \pre cl_page_is_owned(pg, io)
984 * \post !cl_page_is_owned(pg, io)
985 *
986 * \see cl_page_assume()
987 */
988void cl_page_unassume(const struct lu_env *env,
989 struct cl_io *io, struct cl_page *pg)
990{
991 PINVRNT(env, pg, cl_page_is_owned(pg, io));
992 PINVRNT(env, pg, cl_page_invariant(pg));
993
994 ENTRY;
995 pg = cl_page_top(pg);
996 io = cl_io_top(io);
997 cl_page_owner_clear(pg);
998 cl_page_state_set(env, pg, CPS_CACHED);
999 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_unassume),
1000 (const struct lu_env *,
1001 const struct cl_page_slice *, struct cl_io *),
1002 io);
1003 EXIT;
1004}
1005EXPORT_SYMBOL(cl_page_unassume);
1006
1007/**
1008 * Releases page ownership.
1009 *
1010 * Moves page into cl_page_state::CPS_CACHED.
1011 *
1012 * \pre cl_page_is_owned(pg, io)
1013 * \post !cl_page_is_owned(pg, io)
1014 *
1015 * \see cl_page_own()
1016 * \see cl_page_operations::cpo_disown()
1017 */
1018void cl_page_disown(const struct lu_env *env,
1019 struct cl_io *io, struct cl_page *pg)
1020{
1021 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1022
1023 ENTRY;
1024 pg = cl_page_top(pg);
1025 io = cl_io_top(io);
1026 cl_page_disown0(env, io, pg);
1027 EXIT;
1028}
1029EXPORT_SYMBOL(cl_page_disown);
1030
1031/**
1032 * Called when page is to be removed from the object, e.g., as a result of
1033 * truncate.
1034 *
1035 * Calls cl_page_operations::cpo_discard() top-to-bottom.
1036 *
1037 * \pre cl_page_is_owned(pg, io)
1038 *
1039 * \see cl_page_operations::cpo_discard()
1040 */
1041void cl_page_discard(const struct lu_env *env,
1042 struct cl_io *io, struct cl_page *pg)
1043{
1044 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1045 PINVRNT(env, pg, cl_page_invariant(pg));
1046
1047 cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_discard));
1048}
1049EXPORT_SYMBOL(cl_page_discard);
1050
1051/**
1052 * Version of cl_page_delete() that can be called for not fully constructed
1053 * pages, e.g,. in a error handling cl_page_find()->cl_page_delete0()
1054 * path. Doesn't check page invariant.
1055 */
1056static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg,
1057 int radix)
1058{
1059 struct cl_page *tmp = pg;
1060 ENTRY;
1061
1062 PASSERT(env, pg, pg == cl_page_top(pg));
1063 PASSERT(env, pg, pg->cp_state != CPS_FREEING);
1064
1065 /*
1066 * Severe all ways to obtain new pointers to @pg.
1067 */
1068 cl_page_owner_clear(pg);
1069
1070 /*
1071 * unexport the page firstly before freeing it so that
1072 * the page content is considered to be invalid.
1073 * We have to do this because a CPS_FREEING cl_page may
1074 * be NOT under the protection of a cl_lock.
1075 * Afterwards, if this page is found by other threads, then this
1076 * page will be forced to reread.
1077 */
1078 cl_page_export(env, pg, 0);
1079 cl_page_state_set0(env, pg, CPS_FREEING);
1080
1081 CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_delete),
1082 (const struct lu_env *, const struct cl_page_slice *));
1083
1084 if (tmp->cp_type == CPT_CACHEABLE) {
1085 if (!radix)
1086 /* !radix means that @pg is not yet in the radix tree,
1087 * skip removing it.
1088 */
1089 tmp = pg->cp_child;
1090 for (; tmp != NULL; tmp = tmp->cp_child) {
1091 void *value;
1092 struct cl_object_header *hdr;
1093
1094 hdr = cl_object_header(tmp->cp_obj);
1095 spin_lock(&hdr->coh_page_guard);
1096 value = radix_tree_delete(&hdr->coh_tree,
1097 tmp->cp_index);
1098 PASSERT(env, tmp, value == tmp);
1099 PASSERT(env, tmp, hdr->coh_pages > 0);
1100 hdr->coh_pages--;
1101 spin_unlock(&hdr->coh_page_guard);
1102 cl_page_put(env, tmp);
1103 }
1104 }
1105
1106 EXIT;
1107}
1108
1109/**
1110 * Called when a decision is made to throw page out of memory.
1111 *
1112 * Notifies all layers about page destruction by calling
1113 * cl_page_operations::cpo_delete() method top-to-bottom.
1114 *
1115 * Moves page into cl_page_state::CPS_FREEING state (this is the only place
1116 * where transition to this state happens).
1117 *
1118 * Eliminates all venues through which new references to the page can be
1119 * obtained:
1120 *
1121 * - removes page from the radix trees,
1122 *
1123 * - breaks linkage from VM page to cl_page.
1124 *
1125 * Once page reaches cl_page_state::CPS_FREEING, all remaining references will
1126 * drain after some time, at which point page will be recycled.
1127 *
1128 * \pre pg == cl_page_top(pg)
1129 * \pre VM page is locked
1130 * \post pg->cp_state == CPS_FREEING
1131 *
1132 * \see cl_page_operations::cpo_delete()
1133 */
1134void cl_page_delete(const struct lu_env *env, struct cl_page *pg)
1135{
1136 PINVRNT(env, pg, cl_page_invariant(pg));
1137 ENTRY;
1138 cl_page_delete0(env, pg, 1);
1139 EXIT;
1140}
1141EXPORT_SYMBOL(cl_page_delete);
1142
1143/**
1144 * Unmaps page from user virtual memory.
1145 *
1146 * Calls cl_page_operations::cpo_unmap() through all layers top-to-bottom. The
1147 * layer responsible for VM interaction has to unmap page from user space
1148 * virtual memory.
1149 *
1150 * \see cl_page_operations::cpo_unmap()
1151 */
1152int cl_page_unmap(const struct lu_env *env,
1153 struct cl_io *io, struct cl_page *pg)
1154{
1155 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1156 PINVRNT(env, pg, cl_page_invariant(pg));
1157
1158 return cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_unmap));
1159}
1160EXPORT_SYMBOL(cl_page_unmap);
1161
1162/**
1163 * Marks page up-to-date.
1164 *
1165 * Call cl_page_operations::cpo_export() through all layers top-to-bottom. The
1166 * layer responsible for VM interaction has to mark/clear page as up-to-date
1167 * by the \a uptodate argument.
1168 *
1169 * \see cl_page_operations::cpo_export()
1170 */
1171void cl_page_export(const struct lu_env *env, struct cl_page *pg, int uptodate)
1172{
1173 PINVRNT(env, pg, cl_page_invariant(pg));
1174 CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_export),
1175 (const struct lu_env *,
1176 const struct cl_page_slice *, int), uptodate);
1177}
1178EXPORT_SYMBOL(cl_page_export);
1179
1180/**
1181 * Returns true, iff \a pg is VM locked in a suitable sense by the calling
1182 * thread.
1183 */
1184int cl_page_is_vmlocked(const struct lu_env *env, const struct cl_page *pg)
1185{
1186 int result;
1187 const struct cl_page_slice *slice;
1188
1189 ENTRY;
1190 pg = cl_page_top_trusted((struct cl_page *)pg);
1191 slice = container_of(pg->cp_layers.next,
1192 const struct cl_page_slice, cpl_linkage);
1193 PASSERT(env, pg, slice->cpl_ops->cpo_is_vmlocked != NULL);
1194 /*
1195 * Call ->cpo_is_vmlocked() directly instead of going through
1196 * CL_PAGE_INVOKE(), because cl_page_is_vmlocked() is used by
1197 * cl_page_invariant().
1198 */
1199 result = slice->cpl_ops->cpo_is_vmlocked(env, slice);
1200 PASSERT(env, pg, result == -EBUSY || result == -ENODATA);
1201 RETURN(result == -EBUSY);
1202}
1203EXPORT_SYMBOL(cl_page_is_vmlocked);
1204
1205static enum cl_page_state cl_req_type_state(enum cl_req_type crt)
1206{
1207 ENTRY;
1208 RETURN(crt == CRT_WRITE ? CPS_PAGEOUT : CPS_PAGEIN);
1209}
1210
1211static void cl_page_io_start(const struct lu_env *env,
1212 struct cl_page *pg, enum cl_req_type crt)
1213{
1214 /*
1215 * Page is queued for IO, change its state.
1216 */
1217 ENTRY;
1218 cl_page_owner_clear(pg);
1219 cl_page_state_set(env, pg, cl_req_type_state(crt));
1220 EXIT;
1221}
1222
1223/**
1224 * Prepares page for immediate transfer. cl_page_operations::cpo_prep() is
1225 * called top-to-bottom. Every layer either agrees to submit this page (by
1226 * returning 0), or requests to omit this page (by returning -EALREADY). Layer
1227 * handling interactions with the VM also has to inform VM that page is under
1228 * transfer now.
1229 */
1230int cl_page_prep(const struct lu_env *env, struct cl_io *io,
1231 struct cl_page *pg, enum cl_req_type crt)
1232{
1233 int result;
1234
1235 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1236 PINVRNT(env, pg, cl_page_invariant(pg));
1237 PINVRNT(env, pg, crt < CRT_NR);
1238
1239 /*
1240 * XXX this has to be called bottom-to-top, so that llite can set up
1241 * PG_writeback without risking other layers deciding to skip this
1242 * page.
1243 */
1244 if (crt >= CRT_NR)
1245 return -EINVAL;
1246 result = cl_page_invoke(env, io, pg, CL_PAGE_OP(io[crt].cpo_prep));
1247 if (result == 0)
1248 cl_page_io_start(env, pg, crt);
1249
1250 KLASSERT(ergo(crt == CRT_WRITE && pg->cp_type == CPT_CACHEABLE,
1251 equi(result == 0,
1252 PageWriteback(cl_page_vmpage(env, pg)))));
1253 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1254 return result;
1255}
1256EXPORT_SYMBOL(cl_page_prep);
1257
1258/**
1259 * Notify layers about transfer completion.
1260 *
1261 * Invoked by transfer sub-system (which is a part of osc) to notify layers
1262 * that a transfer, of which this page is a part of has completed.
1263 *
1264 * Completion call-backs are executed in the bottom-up order, so that
1265 * uppermost layer (llite), responsible for the VFS/VM interaction runs last
1266 * and can release locks safely.
1267 *
1268 * \pre pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT
1269 * \post pg->cp_state == CPS_CACHED
1270 *
1271 * \see cl_page_operations::cpo_completion()
1272 */
1273void cl_page_completion(const struct lu_env *env,
1274 struct cl_page *pg, enum cl_req_type crt, int ioret)
1275{
1276 struct cl_sync_io *anchor = pg->cp_sync_io;
1277
1278 PASSERT(env, pg, crt < CRT_NR);
1279 /* cl_page::cp_req already cleared by the caller (osc_completion()) */
1280 PASSERT(env, pg, pg->cp_req == NULL);
1281 PASSERT(env, pg, pg->cp_state == cl_req_type_state(crt));
1282
1283 ENTRY;
1284 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, ioret);
1285 if (crt == CRT_READ && ioret == 0) {
1286 PASSERT(env, pg, !(pg->cp_flags & CPF_READ_COMPLETED));
1287 pg->cp_flags |= CPF_READ_COMPLETED;
1288 }
1289
1290 cl_page_state_set(env, pg, CPS_CACHED);
1291 if (crt >= CRT_NR)
1292 return;
1293 CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(io[crt].cpo_completion),
1294 (const struct lu_env *,
1295 const struct cl_page_slice *, int), ioret);
1296 if (anchor) {
1297 LASSERT(cl_page_is_vmlocked(env, pg));
1298 LASSERT(pg->cp_sync_io == anchor);
1299 pg->cp_sync_io = NULL;
1300 }
1301 /*
1302 * As page->cp_obj is pinned by a reference from page->cp_req, it is
1303 * safe to call cl_page_put() without risking object destruction in a
1304 * non-blocking context.
1305 */
1306 cl_page_put(env, pg);
1307
1308 if (anchor)
1309 cl_sync_io_note(anchor, ioret);
1310
1311 EXIT;
1312}
1313EXPORT_SYMBOL(cl_page_completion);
1314
1315/**
1316 * Notify layers that transfer formation engine decided to yank this page from
1317 * the cache and to make it a part of a transfer.
1318 *
1319 * \pre pg->cp_state == CPS_CACHED
1320 * \post pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT
1321 *
1322 * \see cl_page_operations::cpo_make_ready()
1323 */
1324int cl_page_make_ready(const struct lu_env *env, struct cl_page *pg,
1325 enum cl_req_type crt)
1326{
1327 int result;
1328
1329 PINVRNT(env, pg, crt < CRT_NR);
1330
1331 ENTRY;
1332 if (crt >= CRT_NR)
1333 RETURN(-EINVAL);
1334 result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(io[crt].cpo_make_ready),
1335 (const struct lu_env *,
1336 const struct cl_page_slice *));
1337 if (result == 0) {
1338 PASSERT(env, pg, pg->cp_state == CPS_CACHED);
1339 cl_page_io_start(env, pg, crt);
1340 }
1341 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1342 RETURN(result);
1343}
1344EXPORT_SYMBOL(cl_page_make_ready);
1345
1346/**
1347 * Notify layers that high level io decided to place this page into a cache
1348 * for future transfer.
1349 *
1350 * The layer implementing transfer engine (osc) has to register this page in
1351 * its queues.
1352 *
1353 * \pre cl_page_is_owned(pg, io)
1354 * \post cl_page_is_owned(pg, io)
1355 *
1356 * \see cl_page_operations::cpo_cache_add()
1357 */
1358int cl_page_cache_add(const struct lu_env *env, struct cl_io *io,
1359 struct cl_page *pg, enum cl_req_type crt)
1360{
1361 const struct cl_page_slice *scan;
1362 int result = 0;
1363
1364 PINVRNT(env, pg, crt < CRT_NR);
1365 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1366 PINVRNT(env, pg, cl_page_invariant(pg));
1367
1368 ENTRY;
1369
1370 if (crt >= CRT_NR)
1371 RETURN(-EINVAL);
1372
1373 list_for_each_entry(scan, &pg->cp_layers, cpl_linkage) {
1374 if (scan->cpl_ops->io[crt].cpo_cache_add == NULL)
1375 continue;
1376
1377 result = scan->cpl_ops->io[crt].cpo_cache_add(env, scan, io);
1378 if (result != 0)
1379 break;
1380 }
1381 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1382 RETURN(result);
1383}
1384EXPORT_SYMBOL(cl_page_cache_add);
1385
1386/**
1387 * Called if a pge is being written back by kernel's intention.
1388 *
1389 * \pre cl_page_is_owned(pg, io)
1390 * \post ergo(result == 0, pg->cp_state == CPS_PAGEOUT)
1391 *
1392 * \see cl_page_operations::cpo_flush()
1393 */
1394int cl_page_flush(const struct lu_env *env, struct cl_io *io,
1395 struct cl_page *pg)
1396{
1397 int result;
1398
1399 PINVRNT(env, pg, cl_page_is_owned(pg, io));
1400 PINVRNT(env, pg, cl_page_invariant(pg));
1401
1402 ENTRY;
1403
1404 result = cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_flush));
1405
1406 CL_PAGE_HEADER(D_TRACE, env, pg, "%d\n", result);
1407 RETURN(result);
1408}
1409EXPORT_SYMBOL(cl_page_flush);
1410
1411/**
1412 * Checks whether page is protected by any extent lock is at least required
1413 * mode.
1414 *
1415 * \return the same as in cl_page_operations::cpo_is_under_lock() method.
1416 * \see cl_page_operations::cpo_is_under_lock()
1417 */
1418int cl_page_is_under_lock(const struct lu_env *env, struct cl_io *io,
1419 struct cl_page *page)
1420{
1421 int rc;
1422
1423 PINVRNT(env, page, cl_page_invariant(page));
1424
1425 ENTRY;
1426 rc = CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_is_under_lock),
1427 (const struct lu_env *,
1428 const struct cl_page_slice *, struct cl_io *),
1429 io);
1430 PASSERT(env, page, rc != 0);
1431 RETURN(rc);
1432}
1433EXPORT_SYMBOL(cl_page_is_under_lock);
1434
1435static int page_prune_cb(const struct lu_env *env, struct cl_io *io,
1436 struct cl_page *page, void *cbdata)
1437{
1438 cl_page_own(env, io, page);
1439 cl_page_unmap(env, io, page);
1440 cl_page_discard(env, io, page);
1441 cl_page_disown(env, io, page);
1442 return CLP_GANG_OKAY;
1443}
1444
1445/**
1446 * Purges all cached pages belonging to the object \a obj.
1447 */
1448int cl_pages_prune(const struct lu_env *env, struct cl_object *clobj)
1449{
1450 struct cl_thread_info *info;
1451 struct cl_object *obj = cl_object_top(clobj);
1452 struct cl_io *io;
1453 int result;
1454
1455 ENTRY;
1456 info = cl_env_info(env);
1457 io = &info->clt_io;
1458
1459 /*
1460 * initialize the io. This is ugly since we never do IO in this
1461 * function, we just make cl_page_list functions happy. -jay
1462 */
1463 io->ci_obj = obj;
1464 io->ci_ignore_layout = 1;
1465 result = cl_io_init(env, io, CIT_MISC, obj);
1466 if (result != 0) {
1467 cl_io_fini(env, io);
1468 RETURN(io->ci_result);
1469 }
1470
1471 do {
1472 result = cl_page_gang_lookup(env, obj, io, 0, CL_PAGE_EOF,
1473 page_prune_cb, NULL);
1474 if (result == CLP_GANG_RESCHED)
1475 cond_resched();
1476 } while (result != CLP_GANG_OKAY);
1477
1478 cl_io_fini(env, io);
1479 RETURN(result);
1480}
1481EXPORT_SYMBOL(cl_pages_prune);
1482
1483/**
1484 * Tells transfer engine that only part of a page is to be transmitted.
1485 *
1486 * \see cl_page_operations::cpo_clip()
1487 */
1488void cl_page_clip(const struct lu_env *env, struct cl_page *pg,
1489 int from, int to)
1490{
1491 PINVRNT(env, pg, cl_page_invariant(pg));
1492
1493 CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", from, to);
1494 CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_clip),
1495 (const struct lu_env *,
1496 const struct cl_page_slice *,int, int),
1497 from, to);
1498}
1499EXPORT_SYMBOL(cl_page_clip);
1500
1501/**
1502 * Prints human readable representation of \a pg to the \a f.
1503 */
1504void cl_page_header_print(const struct lu_env *env, void *cookie,
1505 lu_printer_t printer, const struct cl_page *pg)
1506{
1507 (*printer)(env, cookie,
1508 "page@%p[%d %p:%lu ^%p_%p %d %d %d %p %p %#x]\n",
1509 pg, atomic_read(&pg->cp_ref), pg->cp_obj,
1510 pg->cp_index, pg->cp_parent, pg->cp_child,
1511 pg->cp_state, pg->cp_error, pg->cp_type,
1512 pg->cp_owner, pg->cp_req, pg->cp_flags);
1513}
1514EXPORT_SYMBOL(cl_page_header_print);
1515
1516/**
1517 * Prints human readable representation of \a pg to the \a f.
1518 */
1519void cl_page_print(const struct lu_env *env, void *cookie,
1520 lu_printer_t printer, const struct cl_page *pg)
1521{
1522 struct cl_page *scan;
1523
1524 for (scan = cl_page_top((struct cl_page *)pg);
1525 scan != NULL; scan = scan->cp_child)
1526 cl_page_header_print(env, cookie, printer, scan);
1527 CL_PAGE_INVOKE(env, (struct cl_page *)pg, CL_PAGE_OP(cpo_print),
1528 (const struct lu_env *env,
1529 const struct cl_page_slice *slice,
1530 void *cookie, lu_printer_t p), cookie, printer);
1531 (*printer)(env, cookie, "end page@%p\n", pg);
1532}
1533EXPORT_SYMBOL(cl_page_print);
1534
1535/**
1536 * Cancel a page which is still in a transfer.
1537 */
1538int cl_page_cancel(const struct lu_env *env, struct cl_page *page)
1539{
1540 return CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_cancel),
1541 (const struct lu_env *,
1542 const struct cl_page_slice *));
1543}
1544EXPORT_SYMBOL(cl_page_cancel);
1545
1546/**
1547 * Converts a byte offset within object \a obj into a page index.
1548 */
1549loff_t cl_offset(const struct cl_object *obj, pgoff_t idx)
1550{
1551 /*
1552 * XXX for now.
1553 */
1554 return (loff_t)idx << PAGE_CACHE_SHIFT;
1555}
1556EXPORT_SYMBOL(cl_offset);
1557
1558/**
1559 * Converts a page index into a byte offset within object \a obj.
1560 */
1561pgoff_t cl_index(const struct cl_object *obj, loff_t offset)
1562{
1563 /*
1564 * XXX for now.
1565 */
1566 return offset >> PAGE_CACHE_SHIFT;
1567}
1568EXPORT_SYMBOL(cl_index);
1569
1570int cl_page_size(const struct cl_object *obj)
1571{
1572 return 1 << PAGE_CACHE_SHIFT;
1573}
1574EXPORT_SYMBOL(cl_page_size);
1575
1576/**
1577 * Adds page slice to the compound page.
1578 *
1579 * This is called by cl_object_operations::coo_page_init() methods to add a
1580 * per-layer state to the page. New state is added at the end of
1581 * cl_page::cp_layers list, that is, it is at the bottom of the stack.
1582 *
1583 * \see cl_lock_slice_add(), cl_req_slice_add(), cl_io_slice_add()
1584 */
1585void cl_page_slice_add(struct cl_page *page, struct cl_page_slice *slice,
1586 struct cl_object *obj,
1587 const struct cl_page_operations *ops)
1588{
1589 ENTRY;
1590 list_add_tail(&slice->cpl_linkage, &page->cp_layers);
1591 slice->cpl_obj = obj;
1592 slice->cpl_ops = ops;
1593 slice->cpl_page = page;
1594 EXIT;
1595}
1596EXPORT_SYMBOL(cl_page_slice_add);
1597
1598int cl_page_init(void)
1599{
1600 return 0;
1601}
1602
1603void cl_page_fini(void)
1604{
1605}
Linux 6.x block layer and io_uring tree(s)