projects / deliverable / linux.git / blob
? search:


1a6df43dfd4fefc306eda85d944bb3a5593fecc3
[deliverable/linux.git] / drivers / staging / lustre / lustre / osc / osc_cache.c
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.gnu.org/licenses/gpl-2.0.html
19 *
20 * GPL HEADER END
21 */
22 /*
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
25 *
26 * Copyright (c) 2012, 2015, Intel Corporation.
27 *
28 */
29 /*
30 * This file is part of Lustre, http://www.lustre.org/
31 * Lustre is a trademark of Sun Microsystems, Inc.
32 *
33 * osc cache management.
34 *
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
36 */
37
38 #define DEBUG_SUBSYSTEM S_OSC
39
40 #include "osc_cl_internal.h"
41 #include "osc_internal.h"
42
43 static int extent_debug; /* set it to be true for more debug */
44
45 static void osc_update_pending(struct osc_object *obj, int cmd, int delta);
46 static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
47 int state);
48 static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
49 struct osc_async_page *oap, int sent, int rc);
50 static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap,
51 int cmd);
52 static int osc_refresh_count(const struct lu_env *env,
53 struct osc_async_page *oap, int cmd);
54 static int osc_io_unplug_async(const struct lu_env *env,
55 struct client_obd *cli, struct osc_object *osc);
56 static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages,
57 unsigned int lost_grant);
58
59 static void osc_extent_tree_dump0(int level, struct osc_object *obj,
60 const char *func, int line);
61 #define osc_extent_tree_dump(lvl, obj) \
62 osc_extent_tree_dump0(lvl, obj, __func__, __LINE__)
63
64 /** \addtogroup osc
65 * @{
66 */
67
68 /* ------------------ osc extent ------------------ */
69 static inline char *ext_flags(struct osc_extent *ext, char *flags)
70 {
71 char *buf = flags;
72 *buf++ = ext->oe_rw ? 'r' : 'w';
73 if (ext->oe_intree)
74 *buf++ = 'i';
75 if (ext->oe_sync)
76 *buf++ = 'S';
77 if (ext->oe_srvlock)
78 *buf++ = 's';
79 if (ext->oe_hp)
80 *buf++ = 'h';
81 if (ext->oe_urgent)
82 *buf++ = 'u';
83 if (ext->oe_memalloc)
84 *buf++ = 'm';
85 if (ext->oe_trunc_pending)
86 *buf++ = 't';
87 if (ext->oe_fsync_wait)
88 *buf++ = 'Y';
89 *buf = 0;
90 return flags;
91 }
92
93 static inline char list_empty_marker(struct list_head *list)
94 {
95 return list_empty(list) ? '-' : '+';
96 }
97
98 #define EXTSTR "[%lu -> %lu/%lu]"
99 #define EXTPARA(ext) (ext)->oe_start, (ext)->oe_end, (ext)->oe_max_end
100 static const char *oes_strings[] = {
101 "inv", "active", "cache", "locking", "lockdone", "rpc", "trunc", NULL };
102
103 #define OSC_EXTENT_DUMP(lvl, extent, fmt, ...) do { \
104 struct osc_extent *__ext = (extent); \
105 char __buf[16]; \
106 \
107 CDEBUG(lvl, \
108 "extent %p@{" EXTSTR ", " \
109 "[%d|%d|%c|%s|%s|%p], [%d|%d|%c|%c|%p|%u|%p]} " fmt, \
110 /* ----- extent part 0 ----- */ \
111 __ext, EXTPARA(__ext), \
112 /* ----- part 1 ----- */ \
113 atomic_read(&__ext->oe_refc), \
114 atomic_read(&__ext->oe_users), \
115 list_empty_marker(&__ext->oe_link), \
116 oes_strings[__ext->oe_state], ext_flags(__ext, __buf), \
117 __ext->oe_obj, \
118 /* ----- part 2 ----- */ \
119 __ext->oe_grants, __ext->oe_nr_pages, \
120 list_empty_marker(&__ext->oe_pages), \
121 waitqueue_active(&__ext->oe_waitq) ? '+' : '-', \
122 __ext->oe_dlmlock, __ext->oe_mppr, __ext->oe_owner, \
123 /* ----- part 4 ----- */ \
124 ## __VA_ARGS__); \
125 if (lvl == D_ERROR && __ext->oe_dlmlock) \
126 LDLM_ERROR(__ext->oe_dlmlock, "extent: %p\n", __ext); \
127 else \
128 LDLM_DEBUG(__ext->oe_dlmlock, "extent: %p\n", __ext); \
129 } while (0)
130
131 #undef EASSERTF
132 #define EASSERTF(expr, ext, fmt, args...) do { \
133 if (!(expr)) { \
134 OSC_EXTENT_DUMP(D_ERROR, (ext), fmt, ##args); \
135 osc_extent_tree_dump(D_ERROR, (ext)->oe_obj); \
136 LASSERT(expr); \
137 } \
138 } while (0)
139
140 #undef EASSERT
141 #define EASSERT(expr, ext) EASSERTF(expr, ext, "\n")
142
143 static inline struct osc_extent *rb_extent(struct rb_node *n)
144 {
145 if (!n)
146 return NULL;
147
148 return container_of(n, struct osc_extent, oe_node);
149 }
150
151 static inline struct osc_extent *next_extent(struct osc_extent *ext)
152 {
153 if (!ext)
154 return NULL;
155
156 LASSERT(ext->oe_intree);
157 return rb_extent(rb_next(&ext->oe_node));
158 }
159
160 static inline struct osc_extent *prev_extent(struct osc_extent *ext)
161 {
162 if (!ext)
163 return NULL;
164
165 LASSERT(ext->oe_intree);
166 return rb_extent(rb_prev(&ext->oe_node));
167 }
168
169 static inline struct osc_extent *first_extent(struct osc_object *obj)
170 {
171 return rb_extent(rb_first(&obj->oo_root));
172 }
173
174 /* object must be locked by caller. */
175 static int osc_extent_sanity_check0(struct osc_extent *ext,
176 const char *func, const int line)
177 {
178 struct osc_object *obj = ext->oe_obj;
179 struct osc_async_page *oap;
180 int page_count;
181 int rc = 0;
182
183 if (!osc_object_is_locked(obj)) {
184 rc = 9;
185 goto out;
186 }
187
188 if (ext->oe_state >= OES_STATE_MAX) {
189 rc = 10;
190 goto out;
191 }
192
193 if (atomic_read(&ext->oe_refc) <= 0) {
194 rc = 20;
195 goto out;
196 }
197
198 if (atomic_read(&ext->oe_refc) < atomic_read(&ext->oe_users)) {
199 rc = 30;
200 goto out;
201 }
202
203 switch (ext->oe_state) {
204 case OES_INV:
205 if (ext->oe_nr_pages > 0 || !list_empty(&ext->oe_pages))
206 rc = 35;
207 else
208 rc = 0;
209 goto out;
210 case OES_ACTIVE:
211 if (atomic_read(&ext->oe_users) == 0) {
212 rc = 40;
213 goto out;
214 }
215 if (ext->oe_hp) {
216 rc = 50;
217 goto out;
218 }
219 if (ext->oe_fsync_wait && !ext->oe_urgent) {
220 rc = 55;
221 goto out;
222 }
223 break;
224 case OES_CACHE:
225 if (ext->oe_grants == 0) {
226 rc = 60;
227 goto out;
228 }
229 if (ext->oe_fsync_wait && !ext->oe_urgent && !ext->oe_hp) {
230 rc = 65;
231 goto out;
232 }
233 default:
234 if (atomic_read(&ext->oe_users) > 0) {
235 rc = 70;
236 goto out;
237 }
238 }
239
240 if (ext->oe_max_end < ext->oe_end || ext->oe_end < ext->oe_start) {
241 rc = 80;
242 goto out;
243 }
244
245 if (ext->oe_sync && ext->oe_grants > 0) {
246 rc = 90;
247 goto out;
248 }
249
250 if (ext->oe_dlmlock) {
251 struct ldlm_extent *extent;
252
253 extent = &ext->oe_dlmlock->l_policy_data.l_extent;
254 if (!(extent->start <= cl_offset(osc2cl(obj), ext->oe_start) &&
255 extent->end >= cl_offset(osc2cl(obj), ext->oe_max_end))) {
256 rc = 100;
257 goto out;
258 }
259
260 if (!(ext->oe_dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))) {
261 rc = 102;
262 goto out;
263 }
264 }
265
266 if (ext->oe_nr_pages > ext->oe_mppr) {
267 rc = 105;
268 goto out;
269 }
270
271 /* Do not verify page list if extent is in RPC. This is because an
272 * in-RPC extent is supposed to be exclusively accessible w/o lock.
273 */
274 if (ext->oe_state > OES_CACHE) {
275 rc = 0;
276 goto out;
277 }
278
279 if (!extent_debug) {
280 rc = 0;
281 goto out;
282 }
283
284 page_count = 0;
285 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
286 pgoff_t index = osc_index(oap2osc(oap));
287 ++page_count;
288 if (index > ext->oe_end || index < ext->oe_start) {
289 rc = 110;
290 goto out;
291 }
292 }
293 if (page_count != ext->oe_nr_pages) {
294 rc = 120;
295 goto out;
296 }
297
298 out:
299 if (rc != 0)
300 OSC_EXTENT_DUMP(D_ERROR, ext,
301 "%s:%d sanity check %p failed with rc = %d\n",
302 func, line, ext, rc);
303 return rc;
304 }
305
306 #define sanity_check_nolock(ext) \
307 osc_extent_sanity_check0(ext, __func__, __LINE__)
308
309 #define sanity_check(ext) ({ \
310 int __res; \
311 osc_object_lock((ext)->oe_obj); \
312 __res = sanity_check_nolock(ext); \
313 osc_object_unlock((ext)->oe_obj); \
314 __res; \
315 })
316
317 /**
318 * sanity check - to make sure there is no overlapped extent in the tree.
319 */
320 static int osc_extent_is_overlapped(struct osc_object *obj,
321 struct osc_extent *ext)
322 {
323 struct osc_extent *tmp;
324
325 LASSERT(osc_object_is_locked(obj));
326
327 if (!extent_debug)
328 return 0;
329
330 for (tmp = first_extent(obj); tmp; tmp = next_extent(tmp)) {
331 if (tmp == ext)
332 continue;
333 if (tmp->oe_end >= ext->oe_start &&
334 tmp->oe_start <= ext->oe_end)
335 return 1;
336 }
337 return 0;
338 }
339
340 static void osc_extent_state_set(struct osc_extent *ext, int state)
341 {
342 LASSERT(osc_object_is_locked(ext->oe_obj));
343 LASSERT(state >= OES_INV && state < OES_STATE_MAX);
344
345 /* Never try to sanity check a state changing extent :-) */
346 /* LASSERT(sanity_check_nolock(ext) == 0); */
347
348 /* TODO: validate the state machine */
349 ext->oe_state = state;
350 wake_up_all(&ext->oe_waitq);
351 }
352
353 static struct osc_extent *osc_extent_alloc(struct osc_object *obj)
354 {
355 struct osc_extent *ext;
356
357 ext = kmem_cache_zalloc(osc_extent_kmem, GFP_NOFS);
358 if (!ext)
359 return NULL;
360
361 RB_CLEAR_NODE(&ext->oe_node);
362 ext->oe_obj = obj;
363 atomic_set(&ext->oe_refc, 1);
364 atomic_set(&ext->oe_users, 0);
365 INIT_LIST_HEAD(&ext->oe_link);
366 ext->oe_state = OES_INV;
367 INIT_LIST_HEAD(&ext->oe_pages);
368 init_waitqueue_head(&ext->oe_waitq);
369 ext->oe_dlmlock = NULL;
370
371 return ext;
372 }
373
374 static void osc_extent_free(struct osc_extent *ext)
375 {
376 kmem_cache_free(osc_extent_kmem, ext);
377 }
378
379 static struct osc_extent *osc_extent_get(struct osc_extent *ext)
380 {
381 LASSERT(atomic_read(&ext->oe_refc) >= 0);
382 atomic_inc(&ext->oe_refc);
383 return ext;
384 }
385
386 static void osc_extent_put(const struct lu_env *env, struct osc_extent *ext)
387 {
388 LASSERT(atomic_read(&ext->oe_refc) > 0);
389 if (atomic_dec_and_test(&ext->oe_refc)) {
390 LASSERT(list_empty(&ext->oe_link));
391 LASSERT(atomic_read(&ext->oe_users) == 0);
392 LASSERT(ext->oe_state == OES_INV);
393 LASSERT(!ext->oe_intree);
394
395 if (ext->oe_dlmlock) {
396 lu_ref_add(&ext->oe_dlmlock->l_reference,
397 "osc_extent", ext);
398 LDLM_LOCK_PUT(ext->oe_dlmlock);
399 ext->oe_dlmlock = NULL;
400 }
401 osc_extent_free(ext);
402 }
403 }
404
405 /**
406 * osc_extent_put_trust() is a special version of osc_extent_put() when
407 * it's known that the caller is not the last user. This is to address the
408 * problem of lacking of lu_env ;-).
409 */
410 static void osc_extent_put_trust(struct osc_extent *ext)
411 {
412 LASSERT(atomic_read(&ext->oe_refc) > 1);
413 LASSERT(osc_object_is_locked(ext->oe_obj));
414 atomic_dec(&ext->oe_refc);
415 }
416
417 /**
418 * Return the extent which includes pgoff @index, or return the greatest
419 * previous extent in the tree.
420 */
421 static struct osc_extent *osc_extent_search(struct osc_object *obj,
422 pgoff_t index)
423 {
424 struct rb_node *n = obj->oo_root.rb_node;
425 struct osc_extent *tmp, *p = NULL;
426
427 LASSERT(osc_object_is_locked(obj));
428 while (n) {
429 tmp = rb_extent(n);
430 if (index < tmp->oe_start) {
431 n = n->rb_left;
432 } else if (index > tmp->oe_end) {
433 p = rb_extent(n);
434 n = n->rb_right;
435 } else {
436 return tmp;
437 }
438 }
439 return p;
440 }
441
442 /*
443 * Return the extent covering @index, otherwise return NULL.
444 * caller must have held object lock.
445 */
446 static struct osc_extent *osc_extent_lookup(struct osc_object *obj,
447 pgoff_t index)
448 {
449 struct osc_extent *ext;
450
451 ext = osc_extent_search(obj, index);
452 if (ext && ext->oe_start <= index && index <= ext->oe_end)
453 return osc_extent_get(ext);
454 return NULL;
455 }
456
457 /* caller must have held object lock. */
458 static void osc_extent_insert(struct osc_object *obj, struct osc_extent *ext)
459 {
460 struct rb_node **n = &obj->oo_root.rb_node;
461 struct rb_node *parent = NULL;
462 struct osc_extent *tmp;
463
464 LASSERT(ext->oe_intree == 0);
465 LASSERT(ext->oe_obj == obj);
466 LASSERT(osc_object_is_locked(obj));
467 while (*n) {
468 tmp = rb_extent(*n);
469 parent = *n;
470
471 if (ext->oe_end < tmp->oe_start)
472 n = &(*n)->rb_left;
473 else if (ext->oe_start > tmp->oe_end)
474 n = &(*n)->rb_right;
475 else
476 EASSERTF(0, tmp, EXTSTR"\n", EXTPARA(ext));
477 }
478 rb_link_node(&ext->oe_node, parent, n);
479 rb_insert_color(&ext->oe_node, &obj->oo_root);
480 osc_extent_get(ext);
481 ext->oe_intree = 1;
482 }
483
484 /* caller must have held object lock. */
485 static void osc_extent_erase(struct osc_extent *ext)
486 {
487 struct osc_object *obj = ext->oe_obj;
488
489 LASSERT(osc_object_is_locked(obj));
490 if (ext->oe_intree) {
491 rb_erase(&ext->oe_node, &obj->oo_root);
492 ext->oe_intree = 0;
493 /* rbtree held a refcount */
494 osc_extent_put_trust(ext);
495 }
496 }
497
498 static struct osc_extent *osc_extent_hold(struct osc_extent *ext)
499 {
500 struct osc_object *obj = ext->oe_obj;
501
502 LASSERT(osc_object_is_locked(obj));
503 LASSERT(ext->oe_state == OES_ACTIVE || ext->oe_state == OES_CACHE);
504 if (ext->oe_state == OES_CACHE) {
505 osc_extent_state_set(ext, OES_ACTIVE);
506 osc_update_pending(obj, OBD_BRW_WRITE, -ext->oe_nr_pages);
507 }
508 atomic_inc(&ext->oe_users);
509 list_del_init(&ext->oe_link);
510 return osc_extent_get(ext);
511 }
512
513 static void __osc_extent_remove(struct osc_extent *ext)
514 {
515 LASSERT(osc_object_is_locked(ext->oe_obj));
516 LASSERT(list_empty(&ext->oe_pages));
517 osc_extent_erase(ext);
518 list_del_init(&ext->oe_link);
519 osc_extent_state_set(ext, OES_INV);
520 OSC_EXTENT_DUMP(D_CACHE, ext, "destroyed.\n");
521 }
522
523 static void osc_extent_remove(struct osc_extent *ext)
524 {
525 struct osc_object *obj = ext->oe_obj;
526
527 osc_object_lock(obj);
528 __osc_extent_remove(ext);
529 osc_object_unlock(obj);
530 }
531
532 /**
533 * This function is used to merge extents to get better performance. It checks
534 * if @cur and @victim are contiguous at chunk level.
535 */
536 static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur,
537 struct osc_extent *victim)
538 {
539 struct osc_object *obj = cur->oe_obj;
540 pgoff_t chunk_start;
541 pgoff_t chunk_end;
542 int ppc_bits;
543
544 LASSERT(cur->oe_state == OES_CACHE);
545 LASSERT(osc_object_is_locked(obj));
546 if (!victim)
547 return -EINVAL;
548
549 if (victim->oe_state != OES_CACHE || victim->oe_fsync_wait)
550 return -EBUSY;
551
552 if (cur->oe_max_end != victim->oe_max_end)
553 return -ERANGE;
554
555 LASSERT(cur->oe_dlmlock == victim->oe_dlmlock);
556 ppc_bits = osc_cli(obj)->cl_chunkbits - PAGE_SHIFT;
557 chunk_start = cur->oe_start >> ppc_bits;
558 chunk_end = cur->oe_end >> ppc_bits;
559 if (chunk_start != (victim->oe_end >> ppc_bits) + 1 &&
560 chunk_end + 1 != victim->oe_start >> ppc_bits)
561 return -ERANGE;
562
563 OSC_EXTENT_DUMP(D_CACHE, victim, "will be merged by %p.\n", cur);
564
565 cur->oe_start = min(cur->oe_start, victim->oe_start);
566 cur->oe_end = max(cur->oe_end, victim->oe_end);
567 cur->oe_grants += victim->oe_grants;
568 cur->oe_nr_pages += victim->oe_nr_pages;
569 /* only the following bits are needed to merge */
570 cur->oe_urgent |= victim->oe_urgent;
571 cur->oe_memalloc |= victim->oe_memalloc;
572 list_splice_init(&victim->oe_pages, &cur->oe_pages);
573 list_del_init(&victim->oe_link);
574 victim->oe_nr_pages = 0;
575
576 osc_extent_get(victim);
577 __osc_extent_remove(victim);
578 osc_extent_put(env, victim);
579
580 OSC_EXTENT_DUMP(D_CACHE, cur, "after merging %p.\n", victim);
581 return 0;
582 }
583
584 /**
585 * Drop user count of osc_extent, and unplug IO asynchronously.
586 */
587 void osc_extent_release(const struct lu_env *env, struct osc_extent *ext)
588 {
589 struct osc_object *obj = ext->oe_obj;
590
591 LASSERT(atomic_read(&ext->oe_users) > 0);
592 LASSERT(sanity_check(ext) == 0);
593 LASSERT(ext->oe_grants > 0);
594
595 if (atomic_dec_and_lock(&ext->oe_users, &obj->oo_lock)) {
596 LASSERT(ext->oe_state == OES_ACTIVE);
597 if (ext->oe_trunc_pending) {
598 /* a truncate process is waiting for this extent.
599 * This may happen due to a race, check
600 * osc_cache_truncate_start().
601 */
602 osc_extent_state_set(ext, OES_TRUNC);
603 ext->oe_trunc_pending = 0;
604 } else {
605 osc_extent_state_set(ext, OES_CACHE);
606 osc_update_pending(obj, OBD_BRW_WRITE,
607 ext->oe_nr_pages);
608
609 /* try to merge the previous and next extent. */
610 osc_extent_merge(env, ext, prev_extent(ext));
611 osc_extent_merge(env, ext, next_extent(ext));
612
613 if (ext->oe_urgent)
614 list_move_tail(&ext->oe_link,
615 &obj->oo_urgent_exts);
616 }
617 osc_object_unlock(obj);
618
619 osc_io_unplug_async(env, osc_cli(obj), obj);
620 }
621 osc_extent_put(env, ext);
622 }
623
624 static inline int overlapped(struct osc_extent *ex1, struct osc_extent *ex2)
625 {
626 return !(ex1->oe_end < ex2->oe_start || ex2->oe_end < ex1->oe_start);
627 }
628
629 /**
630 * Find or create an extent which includes @index, core function to manage
631 * extent tree.
632 */
633 static struct osc_extent *osc_extent_find(const struct lu_env *env,
634 struct osc_object *obj, pgoff_t index,
635 int *grants)
636 {
637 struct client_obd *cli = osc_cli(obj);
638 struct osc_lock *olck;
639 struct cl_lock_descr *descr;
640 struct osc_extent *cur;
641 struct osc_extent *ext;
642 struct osc_extent *conflict = NULL;
643 struct osc_extent *found = NULL;
644 pgoff_t chunk;
645 pgoff_t max_end;
646 int max_pages; /* max_pages_per_rpc */
647 int chunksize;
648 int ppc_bits; /* pages per chunk bits */
649 int chunk_mask;
650 int rc;
651
652 cur = osc_extent_alloc(obj);
653 if (!cur)
654 return ERR_PTR(-ENOMEM);
655
656 olck = osc_env_io(env)->oi_write_osclock;
657 LASSERTF(olck, "page %lu is not covered by lock\n", index);
658 LASSERT(olck->ols_state == OLS_GRANTED);
659
660 descr = &olck->ols_cl.cls_lock->cll_descr;
661 LASSERT(descr->cld_mode >= CLM_WRITE);
662
663 LASSERT(cli->cl_chunkbits >= PAGE_SHIFT);
664 ppc_bits = cli->cl_chunkbits - PAGE_SHIFT;
665 chunk_mask = ~((1 << ppc_bits) - 1);
666 chunksize = 1 << cli->cl_chunkbits;
667 chunk = index >> ppc_bits;
668
669 /* align end to rpc edge, rpc size may not be a power 2 integer. */
670 max_pages = cli->cl_max_pages_per_rpc;
671 LASSERT((max_pages & ~chunk_mask) == 0);
672 max_end = index - (index % max_pages) + max_pages - 1;
673 max_end = min_t(pgoff_t, max_end, descr->cld_end);
674
675 /* initialize new extent by parameters so far */
676 cur->oe_max_end = max_end;
677 cur->oe_start = index & chunk_mask;
678 cur->oe_end = ((index + ~chunk_mask + 1) & chunk_mask) - 1;
679 if (cur->oe_start < descr->cld_start)
680 cur->oe_start = descr->cld_start;
681 if (cur->oe_end > max_end)
682 cur->oe_end = max_end;
683 cur->oe_grants = 0;
684 cur->oe_mppr = max_pages;
685 if (olck->ols_dlmlock) {
686 LASSERT(olck->ols_hold);
687 cur->oe_dlmlock = LDLM_LOCK_GET(olck->ols_dlmlock);
688 lu_ref_add(&olck->ols_dlmlock->l_reference, "osc_extent", cur);
689 }
690
691 /* grants has been allocated by caller */
692 LASSERTF(*grants >= chunksize + cli->cl_extent_tax,
693 "%u/%u/%u.\n", *grants, chunksize, cli->cl_extent_tax);
694 LASSERTF((max_end - cur->oe_start) < max_pages, EXTSTR"\n",
695 EXTPARA(cur));
696
697 restart:
698 osc_object_lock(obj);
699 ext = osc_extent_search(obj, cur->oe_start);
700 if (!ext)
701 ext = first_extent(obj);
702 while (ext) {
703 loff_t ext_chk_start = ext->oe_start >> ppc_bits;
704 loff_t ext_chk_end = ext->oe_end >> ppc_bits;
705
706 LASSERT(sanity_check_nolock(ext) == 0);
707 if (chunk > ext_chk_end + 1)
708 break;
709
710 /* if covering by different locks, no chance to match */
711 if (olck->ols_dlmlock != ext->oe_dlmlock) {
712 EASSERTF(!overlapped(ext, cur), ext,
713 EXTSTR"\n", EXTPARA(cur));
714
715 ext = next_extent(ext);
716 continue;
717 }
718
719 /* discontiguous chunks? */
720 if (chunk + 1 < ext_chk_start) {
721 ext = next_extent(ext);
722 continue;
723 }
724
725 /* ok, from now on, ext and cur have these attrs:
726 * 1. covered by the same lock
727 * 2. contiguous at chunk level or overlapping.
728 */
729
730 if (overlapped(ext, cur)) {
731 /* cur is the minimum unit, so overlapping means
732 * full contain.
733 */
734 EASSERTF((ext->oe_start <= cur->oe_start &&
735 ext->oe_end >= cur->oe_end),
736 ext, EXTSTR"\n", EXTPARA(cur));
737
738 if (ext->oe_state > OES_CACHE || ext->oe_fsync_wait) {
739 /* for simplicity, we wait for this extent to
740 * finish before going forward.
741 */
742 conflict = osc_extent_get(ext);
743 break;
744 }
745
746 found = osc_extent_hold(ext);
747 break;
748 }
749
750 /* non-overlapped extent */
751 if (ext->oe_state != OES_CACHE || ext->oe_fsync_wait) {
752 /* we can't do anything for a non OES_CACHE extent, or
753 * if there is someone waiting for this extent to be
754 * flushed, try next one.
755 */
756 ext = next_extent(ext);
757 continue;
758 }
759
760 /* check if they belong to the same rpc slot before trying to
761 * merge. the extents are not overlapped and contiguous at
762 * chunk level to get here.
763 */
764 if (ext->oe_max_end != max_end) {
765 /* if they don't belong to the same RPC slot or
766 * max_pages_per_rpc has ever changed, do not merge.
767 */
768 ext = next_extent(ext);
769 continue;
770 }
771
772 /* it's required that an extent must be contiguous at chunk
773 * level so that we know the whole extent is covered by grant
774 * (the pages in the extent are NOT required to be contiguous).
775 * Otherwise, it will be too much difficult to know which
776 * chunks have grants allocated.
777 */
778
779 /* try to do front merge - extend ext's start */
780 if (chunk + 1 == ext_chk_start) {
781 /* ext must be chunk size aligned */
782 EASSERT((ext->oe_start & ~chunk_mask) == 0, ext);
783
784 /* pull ext's start back to cover cur */
785 ext->oe_start = cur->oe_start;
786 ext->oe_grants += chunksize;
787 *grants -= chunksize;
788
789 found = osc_extent_hold(ext);
790 } else if (chunk == ext_chk_end + 1) {
791 /* rear merge */
792 ext->oe_end = cur->oe_end;
793 ext->oe_grants += chunksize;
794 *grants -= chunksize;
795
796 /* try to merge with the next one because we just fill
797 * in a gap
798 */
799 if (osc_extent_merge(env, ext, next_extent(ext)) == 0)
800 /* we can save extent tax from next extent */
801 *grants += cli->cl_extent_tax;
802
803 found = osc_extent_hold(ext);
804 }
805 if (found)
806 break;
807
808 ext = next_extent(ext);
809 }
810
811 osc_extent_tree_dump(D_CACHE, obj);
812 if (found) {
813 LASSERT(!conflict);
814 if (!IS_ERR(found)) {
815 LASSERT(found->oe_dlmlock == cur->oe_dlmlock);
816 OSC_EXTENT_DUMP(D_CACHE, found,
817 "found caching ext for %lu.\n", index);
818 }
819 } else if (!conflict) {
820 /* create a new extent */
821 EASSERT(osc_extent_is_overlapped(obj, cur) == 0, cur);
822 cur->oe_grants = chunksize + cli->cl_extent_tax;
823 *grants -= cur->oe_grants;
824 LASSERT(*grants >= 0);
825
826 cur->oe_state = OES_CACHE;
827 found = osc_extent_hold(cur);
828 osc_extent_insert(obj, cur);
829 OSC_EXTENT_DUMP(D_CACHE, cur, "add into tree %lu/%lu.\n",
830 index, descr->cld_end);
831 }
832 osc_object_unlock(obj);
833
834 if (conflict) {
835 LASSERT(!found);
836
837 /* waiting for IO to finish. Please notice that it's impossible
838 * to be an OES_TRUNC extent.
839 */
840 rc = osc_extent_wait(env, conflict, OES_INV);
841 osc_extent_put(env, conflict);
842 conflict = NULL;
843 if (rc < 0) {
844 found = ERR_PTR(rc);
845 goto out;
846 }
847
848 goto restart;
849 }
850
851 out:
852 osc_extent_put(env, cur);
853 LASSERT(*grants >= 0);
854 return found;
855 }
856
857 /**
858 * Called when IO is finished to an extent.
859 */
860 int osc_extent_finish(const struct lu_env *env, struct osc_extent *ext,
861 int sent, int rc)
862 {
863 struct client_obd *cli = osc_cli(ext->oe_obj);
864 struct osc_async_page *oap;
865 struct osc_async_page *tmp;
866 int nr_pages = ext->oe_nr_pages;
867 int lost_grant = 0;
868 int blocksize = cli->cl_import->imp_obd->obd_osfs.os_bsize ? : 4096;
869 __u64 last_off = 0;
870 int last_count = -1;
871
872 OSC_EXTENT_DUMP(D_CACHE, ext, "extent finished.\n");
873
874 ext->oe_rc = rc ?: ext->oe_nr_pages;
875 EASSERT(ergo(rc == 0, ext->oe_state == OES_RPC), ext);
876
877 osc_lru_add_batch(cli, &ext->oe_pages);
878 list_for_each_entry_safe(oap, tmp, &ext->oe_pages, oap_pending_item) {
879 list_del_init(&oap->oap_rpc_item);
880 list_del_init(&oap->oap_pending_item);
881 if (last_off <= oap->oap_obj_off) {
882 last_off = oap->oap_obj_off;
883 last_count = oap->oap_count;
884 }
885
886 --ext->oe_nr_pages;
887 osc_ap_completion(env, cli, oap, sent, rc);
888 }
889 EASSERT(ext->oe_nr_pages == 0, ext);
890
891 if (!sent) {
892 lost_grant = ext->oe_grants;
893 } else if (blocksize < PAGE_SIZE &&
894 last_count != PAGE_SIZE) {
895 /* For short writes we shouldn't count parts of pages that
896 * span a whole chunk on the OST side, or our accounting goes
897 * wrong. Should match the code in filter_grant_check.
898 */
899 int offset = last_off & ~PAGE_MASK;
900 int count = last_count + (offset & (blocksize - 1));
901 int end = (offset + last_count) & (blocksize - 1);
902 if (end)
903 count += blocksize - end;
904
905 lost_grant = PAGE_SIZE - count;
906 }
907 if (ext->oe_grants > 0)
908 osc_free_grant(cli, nr_pages, lost_grant);
909
910 osc_extent_remove(ext);
911 /* put the refcount for RPC */
912 osc_extent_put(env, ext);
913 return 0;
914 }
915
916 static int extent_wait_cb(struct osc_extent *ext, int state)
917 {
918 int ret;
919
920 osc_object_lock(ext->oe_obj);
921 ret = ext->oe_state == state;
922 osc_object_unlock(ext->oe_obj);
923
924 return ret;
925 }
926
927 /**
928 * Wait for the extent's state to become @state.
929 */
930 static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
931 int state)
932 {
933 struct osc_object *obj = ext->oe_obj;
934 struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(600), NULL,
935 LWI_ON_SIGNAL_NOOP, NULL);
936 int rc = 0;
937
938 osc_object_lock(obj);
939 LASSERT(sanity_check_nolock(ext) == 0);
940 /* `Kick' this extent only if the caller is waiting for it to be
941 * written out.
942 */
943 if (state == OES_INV && !ext->oe_urgent && !ext->oe_hp &&
944 !ext->oe_trunc_pending) {
945 if (ext->oe_state == OES_ACTIVE) {
946 ext->oe_urgent = 1;
947 } else if (ext->oe_state == OES_CACHE) {
948 ext->oe_urgent = 1;
949 osc_extent_hold(ext);
950 rc = 1;
951 }
952 }
953 osc_object_unlock(obj);
954 if (rc == 1)
955 osc_extent_release(env, ext);
956
957 /* wait for the extent until its state becomes @state */
958 rc = l_wait_event(ext->oe_waitq, extent_wait_cb(ext, state), &lwi);
959 if (rc == -ETIMEDOUT) {
960 OSC_EXTENT_DUMP(D_ERROR, ext,
961 "%s: wait ext to %d timedout, recovery in progress?\n",
962 osc_export(obj)->exp_obd->obd_name, state);
963
964 lwi = LWI_INTR(NULL, NULL);
965 rc = l_wait_event(ext->oe_waitq, extent_wait_cb(ext, state),
966 &lwi);
967 }
968 if (rc == 0 && ext->oe_rc < 0)
969 rc = ext->oe_rc;
970 return rc;
971 }
972
973 /**
974 * Discard pages with index greater than @size. If @ext is overlapped with
975 * @size, then partial truncate happens.
976 */
977 static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index,
978 bool partial)
979 {
980 struct cl_env_nest nest;
981 struct lu_env *env;
982 struct cl_io *io;
983 struct osc_object *obj = ext->oe_obj;
984 struct client_obd *cli = osc_cli(obj);
985 struct osc_async_page *oap;
986 struct osc_async_page *tmp;
987 int pages_in_chunk = 0;
988 int ppc_bits = cli->cl_chunkbits - PAGE_SHIFT;
989 __u64 trunc_chunk = trunc_index >> ppc_bits;
990 int grants = 0;
991 int nr_pages = 0;
992 int rc = 0;
993
994 LASSERT(sanity_check(ext) == 0);
995 EASSERT(ext->oe_state == OES_TRUNC, ext);
996 EASSERT(!ext->oe_urgent, ext);
997
998 /* Request new lu_env.
999 * We can't use that env from osc_cache_truncate_start() because
1000 * it's from lov_io_sub and not fully initialized.
1001 */
1002 env = cl_env_nested_get(&nest);
1003 io = &osc_env_info(env)->oti_io;
1004 io->ci_obj = cl_object_top(osc2cl(obj));
1005 rc = cl_io_init(env, io, CIT_MISC, io->ci_obj);
1006 if (rc < 0)
1007 goto out;
1008
1009 /* discard all pages with index greater then trunc_index */
1010 list_for_each_entry_safe(oap, tmp, &ext->oe_pages, oap_pending_item) {
1011 pgoff_t index = osc_index(oap2osc(oap));
1012 struct cl_page *page = oap2cl_page(oap);
1013
1014 LASSERT(list_empty(&oap->oap_rpc_item));
1015
1016 /* only discard the pages with their index greater than
1017 * trunc_index, and ...
1018 */
1019 if (index < trunc_index ||
1020 (index == trunc_index && partial)) {
1021 /* accounting how many pages remaining in the chunk
1022 * so that we can calculate grants correctly. */
1023 if (index >> ppc_bits == trunc_chunk)
1024 ++pages_in_chunk;
1025 continue;
1026 }
1027
1028 list_del_init(&oap->oap_pending_item);
1029
1030 cl_page_get(page);
1031 lu_ref_add(&page->cp_reference, "truncate", current);
1032
1033 if (cl_page_own(env, io, page) == 0) {
1034 cl_page_discard(env, io, page);
1035 cl_page_disown(env, io, page);
1036 } else {
1037 LASSERT(page->cp_state == CPS_FREEING);
1038 LASSERT(0);
1039 }
1040
1041 lu_ref_del(&page->cp_reference, "truncate", current);
1042 cl_page_put(env, page);
1043
1044 --ext->oe_nr_pages;
1045 ++nr_pages;
1046 }
1047 EASSERTF(ergo(ext->oe_start >= trunc_index + !!partial,
1048 ext->oe_nr_pages == 0),
1049 ext, "trunc_index %lu, partial %d\n", trunc_index, partial);
1050
1051 osc_object_lock(obj);
1052 if (ext->oe_nr_pages == 0) {
1053 LASSERT(pages_in_chunk == 0);
1054 grants = ext->oe_grants;
1055 ext->oe_grants = 0;
1056 } else { /* calculate how many grants we can free */
1057 int chunks = (ext->oe_end >> ppc_bits) - trunc_chunk;
1058 pgoff_t last_index;
1059
1060 /* if there is no pages in this chunk, we can also free grants
1061 * for the last chunk
1062 */
1063 if (pages_in_chunk == 0) {
1064 /* if this is the 1st chunk and no pages in this chunk,
1065 * ext->oe_nr_pages must be zero, so we should be in
1066 * the other if-clause.
1067 */
1068 LASSERT(trunc_chunk > 0);
1069 --trunc_chunk;
1070 ++chunks;
1071 }
1072
1073 /* this is what we can free from this extent */
1074 grants = chunks << cli->cl_chunkbits;
1075 ext->oe_grants -= grants;
1076 last_index = ((trunc_chunk + 1) << ppc_bits) - 1;
1077 ext->oe_end = min(last_index, ext->oe_max_end);
1078 LASSERT(ext->oe_end >= ext->oe_start);
1079 LASSERT(ext->oe_grants > 0);
1080 }
1081 osc_object_unlock(obj);
1082
1083 if (grants > 0 || nr_pages > 0)
1084 osc_free_grant(cli, nr_pages, grants);
1085
1086 out:
1087 cl_io_fini(env, io);
1088 cl_env_nested_put(&nest, env);
1089 return rc;
1090 }
1091
1092 /**
1093 * This function is used to make the extent prepared for transfer.
1094 * A race with flushing page - ll_writepage() has to be handled cautiously.
1095 */
1096 static int osc_extent_make_ready(const struct lu_env *env,
1097 struct osc_extent *ext)
1098 {
1099 struct osc_async_page *oap;
1100 struct osc_async_page *last = NULL;
1101 struct osc_object *obj = ext->oe_obj;
1102 int page_count = 0;
1103 int rc;
1104
1105 /* we're going to grab page lock, so object lock must not be taken. */
1106 LASSERT(sanity_check(ext) == 0);
1107 /* in locking state, any process should not touch this extent. */
1108 EASSERT(ext->oe_state == OES_LOCKING, ext);
1109 EASSERT(ext->oe_owner, ext);
1110
1111 OSC_EXTENT_DUMP(D_CACHE, ext, "make ready\n");
1112
1113 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
1114 ++page_count;
1115 if (!last || last->oap_obj_off < oap->oap_obj_off)
1116 last = oap;
1117
1118 /* checking ASYNC_READY is race safe */
1119 if ((oap->oap_async_flags & ASYNC_READY) != 0)
1120 continue;
1121
1122 rc = osc_make_ready(env, oap, OBD_BRW_WRITE);
1123 switch (rc) {
1124 case 0:
1125 spin_lock(&oap->oap_lock);
1126 oap->oap_async_flags |= ASYNC_READY;
1127 spin_unlock(&oap->oap_lock);
1128 break;
1129 case -EALREADY:
1130 LASSERT((oap->oap_async_flags & ASYNC_READY) != 0);
1131 break;
1132 default:
1133 LASSERTF(0, "unknown return code: %d\n", rc);
1134 }
1135 }
1136
1137 LASSERT(page_count == ext->oe_nr_pages);
1138 LASSERT(last);
1139 /* the last page is the only one we need to refresh its count by
1140 * the size of file.
1141 */
1142 if (!(last->oap_async_flags & ASYNC_COUNT_STABLE)) {
1143 last->oap_count = osc_refresh_count(env, last, OBD_BRW_WRITE);
1144 LASSERT(last->oap_count > 0);
1145 LASSERT(last->oap_page_off + last->oap_count <= PAGE_SIZE);
1146 spin_lock(&last->oap_lock);
1147 last->oap_async_flags |= ASYNC_COUNT_STABLE;
1148 spin_unlock(&last->oap_lock);
1149 }
1150
1151 /* for the rest of pages, we don't need to call osf_refresh_count()
1152 * because it's known they are not the last page
1153 */
1154 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
1155 if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
1156 oap->oap_count = PAGE_SIZE - oap->oap_page_off;
1157 spin_lock(&last->oap_lock);
1158 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
1159 spin_unlock(&last->oap_lock);
1160 }
1161 }
1162
1163 osc_object_lock(obj);
1164 osc_extent_state_set(ext, OES_RPC);
1165 osc_object_unlock(obj);
1166 /* get a refcount for RPC. */
1167 osc_extent_get(ext);
1168
1169 return 0;
1170 }
1171
1172 /**
1173 * Quick and simple version of osc_extent_find(). This function is frequently
1174 * called to expand the extent for the same IO. To expand the extent, the
1175 * page index must be in the same or next chunk of ext->oe_end.
1176 */
1177 static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, int *grants)
1178 {
1179 struct osc_object *obj = ext->oe_obj;
1180 struct client_obd *cli = osc_cli(obj);
1181 struct osc_extent *next;
1182 int ppc_bits = cli->cl_chunkbits - PAGE_SHIFT;
1183 pgoff_t chunk = index >> ppc_bits;
1184 pgoff_t end_chunk;
1185 pgoff_t end_index;
1186 int chunksize = 1 << cli->cl_chunkbits;
1187 int rc = 0;
1188
1189 LASSERT(ext->oe_max_end >= index && ext->oe_start <= index);
1190 osc_object_lock(obj);
1191 LASSERT(sanity_check_nolock(ext) == 0);
1192 end_chunk = ext->oe_end >> ppc_bits;
1193 if (chunk > end_chunk + 1) {
1194 rc = -ERANGE;
1195 goto out;
1196 }
1197
1198 if (end_chunk >= chunk) {
1199 rc = 0;
1200 goto out;
1201 }
1202
1203 LASSERT(end_chunk + 1 == chunk);
1204 /* try to expand this extent to cover @index */
1205 end_index = min(ext->oe_max_end, ((chunk + 1) << ppc_bits) - 1);
1206
1207 next = next_extent(ext);
1208 if (next && next->oe_start <= end_index) {
1209 /* complex mode - overlapped with the next extent,
1210 * this case will be handled by osc_extent_find()
1211 */
1212 rc = -EAGAIN;
1213 goto out;
1214 }
1215
1216 ext->oe_end = end_index;
1217 ext->oe_grants += chunksize;
1218 *grants -= chunksize;
1219 LASSERT(*grants >= 0);
1220 EASSERTF(osc_extent_is_overlapped(obj, ext) == 0, ext,
1221 "overlapped after expanding for %lu.\n", index);
1222
1223 out:
1224 osc_object_unlock(obj);
1225 return rc;
1226 }
1227
1228 static void osc_extent_tree_dump0(int level, struct osc_object *obj,
1229 const char *func, int line)
1230 {
1231 struct osc_extent *ext;
1232 int cnt;
1233
1234 CDEBUG(level, "Dump object %p extents at %s:%d, mppr: %u.\n",
1235 obj, func, line, osc_cli(obj)->cl_max_pages_per_rpc);
1236
1237 /* osc_object_lock(obj); */
1238 cnt = 1;
1239 for (ext = first_extent(obj); ext; ext = next_extent(ext))
1240 OSC_EXTENT_DUMP(level, ext, "in tree %d.\n", cnt++);
1241
1242 cnt = 1;
1243 list_for_each_entry(ext, &obj->oo_hp_exts, oe_link)
1244 OSC_EXTENT_DUMP(level, ext, "hp %d.\n", cnt++);
1245
1246 cnt = 1;
1247 list_for_each_entry(ext, &obj->oo_urgent_exts, oe_link)
1248 OSC_EXTENT_DUMP(level, ext, "urgent %d.\n", cnt++);
1249
1250 cnt = 1;
1251 list_for_each_entry(ext, &obj->oo_reading_exts, oe_link)
1252 OSC_EXTENT_DUMP(level, ext, "reading %d.\n", cnt++);
1253 /* osc_object_unlock(obj); */
1254 }
1255
1256 /* ------------------ osc extent end ------------------ */
1257
1258 static inline int osc_is_ready(struct osc_object *osc)
1259 {
1260 return !list_empty(&osc->oo_ready_item) ||
1261 !list_empty(&osc->oo_hp_ready_item);
1262 }
1263
1264 #define OSC_IO_DEBUG(OSC, STR, args...) \
1265 CDEBUG(D_CACHE, "obj %p ready %d|%c|%c wr %d|%c|%c rd %d|%c " STR, \
1266 (OSC), osc_is_ready(OSC), \
1267 list_empty_marker(&(OSC)->oo_hp_ready_item), \
1268 list_empty_marker(&(OSC)->oo_ready_item), \
1269 atomic_read(&(OSC)->oo_nr_writes), \
1270 list_empty_marker(&(OSC)->oo_hp_exts), \
1271 list_empty_marker(&(OSC)->oo_urgent_exts), \
1272 atomic_read(&(OSC)->oo_nr_reads), \
1273 list_empty_marker(&(OSC)->oo_reading_exts), \
1274 ##args)
1275
1276 static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap,
1277 int cmd)
1278 {
1279 struct osc_page *opg = oap2osc_page(oap);
1280 struct cl_page *page = oap2cl_page(oap);
1281 int result;
1282
1283 LASSERT(cmd == OBD_BRW_WRITE); /* no cached reads */
1284
1285 result = cl_page_make_ready(env, page, CRT_WRITE);
1286 if (result == 0)
1287 opg->ops_submit_time = cfs_time_current();
1288 return result;
1289 }
1290
1291 static int osc_refresh_count(const struct lu_env *env,
1292 struct osc_async_page *oap, int cmd)
1293 {
1294 struct osc_page *opg = oap2osc_page(oap);
1295 pgoff_t index = osc_index(oap2osc(oap));
1296 struct cl_object *obj;
1297 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1298
1299 int result;
1300 loff_t kms;
1301
1302 /* readpage queues with _COUNT_STABLE, shouldn't get here. */
1303 LASSERT(!(cmd & OBD_BRW_READ));
1304 obj = opg->ops_cl.cpl_obj;
1305
1306 cl_object_attr_lock(obj);
1307 result = cl_object_attr_get(env, obj, attr);
1308 cl_object_attr_unlock(obj);
1309 if (result < 0)
1310 return result;
1311 kms = attr->cat_kms;
1312 if (cl_offset(obj, index) >= kms)
1313 /* catch race with truncate */
1314 return 0;
1315 else if (cl_offset(obj, index + 1) > kms)
1316 /* catch sub-page write at end of file */
1317 return kms % PAGE_SIZE;
1318 else
1319 return PAGE_SIZE;
1320 }
1321
1322 static int osc_completion(const struct lu_env *env, struct osc_async_page *oap,
1323 int cmd, int rc)
1324 {
1325 struct osc_page *opg = oap2osc_page(oap);
1326 struct cl_page *page = oap2cl_page(oap);
1327 struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
1328 enum cl_req_type crt;
1329 int srvlock;
1330
1331 cmd &= ~OBD_BRW_NOQUOTA;
1332 LASSERTF(equi(page->cp_state == CPS_PAGEIN, cmd == OBD_BRW_READ),
1333 "cp_state:%u, cmd:%d\n", page->cp_state, cmd);
1334 LASSERTF(equi(page->cp_state == CPS_PAGEOUT, cmd == OBD_BRW_WRITE),
1335 "cp_state:%u, cmd:%d\n", page->cp_state, cmd);
1336 LASSERT(opg->ops_transfer_pinned);
1337
1338 /*
1339 * page->cp_req can be NULL if io submission failed before
1340 * cl_req was allocated.
1341 */
1342 if (page->cp_req)
1343 cl_req_page_done(env, page);
1344 LASSERT(!page->cp_req);
1345
1346 crt = cmd == OBD_BRW_READ ? CRT_READ : CRT_WRITE;
1347 /* Clear opg->ops_transfer_pinned before VM lock is released. */
1348 opg->ops_transfer_pinned = 0;
1349
1350 spin_lock(&obj->oo_seatbelt);
1351 LASSERT(opg->ops_submitter);
1352 LASSERT(!list_empty(&opg->ops_inflight));
1353 list_del_init(&opg->ops_inflight);
1354 opg->ops_submitter = NULL;
1355 spin_unlock(&obj->oo_seatbelt);
1356
1357 opg->ops_submit_time = 0;
1358 srvlock = oap->oap_brw_flags & OBD_BRW_SRVLOCK;
1359
1360 /* statistic */
1361 if (rc == 0 && srvlock) {
1362 struct lu_device *ld = opg->ops_cl.cpl_obj->co_lu.lo_dev;
1363 struct osc_stats *stats = &lu2osc_dev(ld)->od_stats;
1364 int bytes = oap->oap_count;
1365
1366 if (crt == CRT_READ)
1367 stats->os_lockless_reads += bytes;
1368 else
1369 stats->os_lockless_writes += bytes;
1370 }
1371
1372 /*
1373 * This has to be the last operation with the page, as locks are
1374 * released in cl_page_completion() and nothing except for the
1375 * reference counter protects page from concurrent reclaim.
1376 */
1377 lu_ref_del(&page->cp_reference, "transfer", page);
1378
1379 cl_page_completion(env, page, crt, rc);
1380
1381 return 0;
1382 }
1383
1384 #define OSC_DUMP_GRANT(lvl, cli, fmt, args...) do { \
1385 struct client_obd *__tmp = (cli); \
1386 CDEBUG(lvl, "%s: grant { dirty: %ld/%ld dirty_pages: %d/%d " \
1387 "unstable_pages: %d/%d dropped: %ld avail: %ld, " \
1388 "reserved: %ld, flight: %d } lru {in list: %d, " \
1389 "left: %d, waiters: %d }" fmt, \
1390 __tmp->cl_import->imp_obd->obd_name, \
1391 __tmp->cl_dirty, __tmp->cl_dirty_max, \
1392 atomic_read(&obd_dirty_pages), obd_max_dirty_pages, \
1393 atomic_read(&obd_unstable_pages), obd_max_dirty_pages, \
1394 __tmp->cl_lost_grant, __tmp->cl_avail_grant, \
1395 __tmp->cl_reserved_grant, __tmp->cl_w_in_flight, \
1396 atomic_read(&__tmp->cl_lru_in_list), \
1397 atomic_read(&__tmp->cl_lru_busy), \
1398 atomic_read(&__tmp->cl_lru_shrinkers), ##args); \
1399 } while (0)
1400
1401 /* caller must hold loi_list_lock */
1402 static void osc_consume_write_grant(struct client_obd *cli,
1403 struct brw_page *pga)
1404 {
1405 assert_spin_locked(&cli->cl_loi_list_lock);
1406 LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT));
1407 atomic_inc(&obd_dirty_pages);
1408 cli->cl_dirty += PAGE_SIZE;
1409 pga->flag |= OBD_BRW_FROM_GRANT;
1410 CDEBUG(D_CACHE, "using %lu grant credits for brw %p page %p\n",
1411 PAGE_SIZE, pga, pga->pg);
1412 osc_update_next_shrink(cli);
1413 }
1414
1415 /* the companion to osc_consume_write_grant, called when a brw has completed.
1416 * must be called with the loi lock held.
1417 */
1418 static void osc_release_write_grant(struct client_obd *cli,
1419 struct brw_page *pga)
1420 {
1421 assert_spin_locked(&cli->cl_loi_list_lock);
1422 if (!(pga->flag & OBD_BRW_FROM_GRANT)) {
1423 return;
1424 }
1425
1426 pga->flag &= ~OBD_BRW_FROM_GRANT;
1427 atomic_dec(&obd_dirty_pages);
1428 cli->cl_dirty -= PAGE_SIZE;
1429 if (pga->flag & OBD_BRW_NOCACHE) {
1430 pga->flag &= ~OBD_BRW_NOCACHE;
1431 atomic_dec(&obd_dirty_transit_pages);
1432 cli->cl_dirty_transit -= PAGE_SIZE;
1433 }
1434 }
1435
1436 /**
1437 * To avoid sleeping with object lock held, it's good for us allocate enough
1438 * grants before entering into critical section.
1439 *
1440 * spin_lock held by caller
1441 */
1442 static int osc_reserve_grant(struct client_obd *cli, unsigned int bytes)
1443 {
1444 int rc = -EDQUOT;
1445
1446 if (cli->cl_avail_grant >= bytes) {
1447 cli->cl_avail_grant -= bytes;
1448 cli->cl_reserved_grant += bytes;
1449 rc = 0;
1450 }
1451 return rc;
1452 }
1453
1454 static void __osc_unreserve_grant(struct client_obd *cli,
1455 unsigned int reserved, unsigned int unused)
1456 {
1457 /* it's quite normal for us to get more grant than reserved.
1458 * Thinking about a case that two extents merged by adding a new
1459 * chunk, we can save one extent tax. If extent tax is greater than
1460 * one chunk, we can save more grant by adding a new chunk
1461 */
1462 cli->cl_reserved_grant -= reserved;
1463 if (unused > reserved) {
1464 cli->cl_avail_grant += reserved;
1465 cli->cl_lost_grant += unused - reserved;
1466 } else {
1467 cli->cl_avail_grant += unused;
1468 }
1469 }
1470
1471 static void osc_unreserve_grant(struct client_obd *cli,
1472 unsigned int reserved, unsigned int unused)
1473 {
1474 spin_lock(&cli->cl_loi_list_lock);
1475 __osc_unreserve_grant(cli, reserved, unused);
1476 if (unused > 0)
1477 osc_wake_cache_waiters(cli);
1478 spin_unlock(&cli->cl_loi_list_lock);
1479 }
1480
1481 /**
1482 * Free grant after IO is finished or canceled.
1483 *
1484 * @lost_grant is used to remember how many grants we have allocated but not
1485 * used, we should return these grants to OST. There're two cases where grants
1486 * can be lost:
1487 * 1. truncate;
1488 * 2. blocksize at OST is less than PAGE_SIZE and a partial page was
1489 * written. In this case OST may use less chunks to serve this partial
1490 * write. OSTs don't actually know the page size on the client side. so
1491 * clients have to calculate lost grant by the blocksize on the OST.
1492 * See filter_grant_check() for details.
1493 */
1494 static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages,
1495 unsigned int lost_grant)
1496 {
1497 int grant = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
1498
1499 spin_lock(&cli->cl_loi_list_lock);
1500 atomic_sub(nr_pages, &obd_dirty_pages);
1501 cli->cl_dirty -= nr_pages << PAGE_SHIFT;
1502 cli->cl_lost_grant += lost_grant;
1503 if (cli->cl_avail_grant < grant && cli->cl_lost_grant >= grant) {
1504 /* borrow some grant from truncate to avoid the case that
1505 * truncate uses up all avail grant
1506 */
1507 cli->cl_lost_grant -= grant;
1508 cli->cl_avail_grant += grant;
1509 }
1510 osc_wake_cache_waiters(cli);
1511 spin_unlock(&cli->cl_loi_list_lock);
1512 CDEBUG(D_CACHE, "lost %u grant: %lu avail: %lu dirty: %lu\n",
1513 lost_grant, cli->cl_lost_grant,
1514 cli->cl_avail_grant, cli->cl_dirty);
1515 }
1516
1517 /**
1518 * The companion to osc_enter_cache(), called when @oap is no longer part of
1519 * the dirty accounting due to error.
1520 */
1521 static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap)
1522 {
1523 spin_lock(&cli->cl_loi_list_lock);
1524 osc_release_write_grant(cli, &oap->oap_brw_page);
1525 spin_unlock(&cli->cl_loi_list_lock);
1526 }
1527
1528 /**
1529 * Non-blocking version of osc_enter_cache() that consumes grant only when it
1530 * is available.
1531 */
1532 static int osc_enter_cache_try(struct client_obd *cli,
1533 struct osc_async_page *oap,
1534 int bytes, int transient)
1535 {
1536 int rc;
1537
1538 OSC_DUMP_GRANT(D_CACHE, cli, "need:%d.\n", bytes);
1539
1540 rc = osc_reserve_grant(cli, bytes);
1541 if (rc < 0)
1542 return 0;
1543
1544 if (cli->cl_dirty + PAGE_SIZE <= cli->cl_dirty_max &&
1545 atomic_read(&obd_unstable_pages) + 1 +
1546 atomic_read(&obd_dirty_pages) <= obd_max_dirty_pages) {
1547 osc_consume_write_grant(cli, &oap->oap_brw_page);
1548 if (transient) {
1549 cli->cl_dirty_transit += PAGE_SIZE;
1550 atomic_inc(&obd_dirty_transit_pages);
1551 oap->oap_brw_flags |= OBD_BRW_NOCACHE;
1552 }
1553 rc = 1;
1554 } else {
1555 __osc_unreserve_grant(cli, bytes, bytes);
1556 rc = 0;
1557 }
1558 return rc;
1559 }
1560
1561 static int ocw_granted(struct client_obd *cli, struct osc_cache_waiter *ocw)
1562 {
1563 int rc;
1564
1565 spin_lock(&cli->cl_loi_list_lock);
1566 rc = list_empty(&ocw->ocw_entry);
1567 spin_unlock(&cli->cl_loi_list_lock);
1568 return rc;
1569 }
1570
1571 /**
1572 * The main entry to reserve dirty page accounting. Usually the grant reserved
1573 * in this function will be freed in bulk in osc_free_grant() unless it fails
1574 * to add osc cache, in that case, it will be freed in osc_exit_cache().
1575 *
1576 * The process will be put into sleep if it's already run out of grant.
1577 */
1578 static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
1579 struct osc_async_page *oap, int bytes)
1580 {
1581 struct osc_object *osc = oap->oap_obj;
1582 struct lov_oinfo *loi = osc->oo_oinfo;
1583 struct osc_cache_waiter ocw;
1584 struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(600), NULL,
1585 LWI_ON_SIGNAL_NOOP, NULL);
1586 int rc = -EDQUOT;
1587
1588 OSC_DUMP_GRANT(D_CACHE, cli, "need:%d.\n", bytes);
1589
1590 spin_lock(&cli->cl_loi_list_lock);
1591
1592 /* force the caller to try sync io. this can jump the list
1593 * of queued writes and create a discontiguous rpc stream
1594 */
1595 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_NO_GRANT) ||
1596 cli->cl_dirty_max < PAGE_SIZE ||
1597 cli->cl_ar.ar_force_sync || loi->loi_ar.ar_force_sync) {
1598 rc = -EDQUOT;
1599 goto out;
1600 }
1601
1602 /* Hopefully normal case - cache space and write credits available */
1603 if (osc_enter_cache_try(cli, oap, bytes, 0)) {
1604 rc = 0;
1605 goto out;
1606 }
1607
1608 /* We can get here for two reasons: too many dirty pages in cache, or
1609 * run out of grants. In both cases we should write dirty pages out.
1610 * Adding a cache waiter will trigger urgent write-out no matter what
1611 * RPC size will be.
1612 * The exiting condition is no avail grants and no dirty pages caching,
1613 * that really means there is no space on the OST.
1614 */
1615 init_waitqueue_head(&ocw.ocw_waitq);
1616 ocw.ocw_oap = oap;
1617 ocw.ocw_grant = bytes;
1618 while (cli->cl_dirty > 0 || cli->cl_w_in_flight > 0) {
1619 list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters);
1620 ocw.ocw_rc = 0;
1621 spin_unlock(&cli->cl_loi_list_lock);
1622
1623 osc_io_unplug_async(env, cli, NULL);
1624
1625 CDEBUG(D_CACHE, "%s: sleeping for cache space @ %p for %p\n",
1626 cli->cl_import->imp_obd->obd_name, &ocw, oap);
1627
1628 rc = l_wait_event(ocw.ocw_waitq, ocw_granted(cli, &ocw), &lwi);
1629
1630 spin_lock(&cli->cl_loi_list_lock);
1631
1632 /* l_wait_event is interrupted by signal, or timed out */
1633 if (rc < 0) {
1634 if (rc == -ETIMEDOUT) {
1635 OSC_DUMP_GRANT(D_ERROR, cli,
1636 "try to reserve %d.\n", bytes);
1637 osc_extent_tree_dump(D_ERROR, osc);
1638 rc = -EDQUOT;
1639 }
1640
1641 list_del_init(&ocw.ocw_entry);
1642 goto out;
1643 }
1644
1645 LASSERT(list_empty(&ocw.ocw_entry));
1646 rc = ocw.ocw_rc;
1647
1648 if (rc != -EDQUOT)
1649 goto out;
1650 if (osc_enter_cache_try(cli, oap, bytes, 0)) {
1651 rc = 0;
1652 goto out;
1653 }
1654 }
1655 out:
1656 spin_unlock(&cli->cl_loi_list_lock);
1657 OSC_DUMP_GRANT(D_CACHE, cli, "returned %d.\n", rc);
1658 return rc;
1659 }
1660
1661 /* caller must hold loi_list_lock */
1662 void osc_wake_cache_waiters(struct client_obd *cli)
1663 {
1664 struct list_head *l, *tmp;
1665 struct osc_cache_waiter *ocw;
1666
1667 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
1668 ocw = list_entry(l, struct osc_cache_waiter, ocw_entry);
1669 list_del_init(&ocw->ocw_entry);
1670
1671 ocw->ocw_rc = -EDQUOT;
1672 /* we can't dirty more */
1673 if ((cli->cl_dirty + PAGE_SIZE > cli->cl_dirty_max) ||
1674 (atomic_read(&obd_unstable_pages) + 1 +
1675 atomic_read(&obd_dirty_pages) > obd_max_dirty_pages)) {
1676 CDEBUG(D_CACHE, "no dirty room: dirty: %ld osc max %ld, sys max %d\n",
1677 cli->cl_dirty,
1678 cli->cl_dirty_max, obd_max_dirty_pages);
1679 goto wakeup;
1680 }
1681
1682 ocw->ocw_rc = 0;
1683 if (!osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant, 0))
1684 ocw->ocw_rc = -EDQUOT;
1685
1686 wakeup:
1687 CDEBUG(D_CACHE, "wake up %p for oap %p, avail grant %ld, %d\n",
1688 ocw, ocw->ocw_oap, cli->cl_avail_grant, ocw->ocw_rc);
1689
1690 wake_up(&ocw->ocw_waitq);
1691 }
1692 }
1693
1694 static int osc_max_rpc_in_flight(struct client_obd *cli, struct osc_object *osc)
1695 {
1696 int hprpc = !!list_empty(&osc->oo_hp_exts);
1697
1698 return rpcs_in_flight(cli) >= cli->cl_max_rpcs_in_flight + hprpc;
1699 }
1700
1701 /* This maintains the lists of pending pages to read/write for a given object
1702 * (lop). This is used by osc_check_rpcs->osc_next_obj() and osc_list_maint()
1703 * to quickly find objects that are ready to send an RPC.
1704 */
1705 static int osc_makes_rpc(struct client_obd *cli, struct osc_object *osc,
1706 int cmd)
1707 {
1708 int invalid_import = 0;
1709
1710 /* if we have an invalid import we want to drain the queued pages
1711 * by forcing them through rpcs that immediately fail and complete
1712 * the pages. recovery relies on this to empty the queued pages
1713 * before canceling the locks and evicting down the llite pages
1714 */
1715 if (!cli->cl_import || cli->cl_import->imp_invalid)
1716 invalid_import = 1;
1717
1718 if (cmd & OBD_BRW_WRITE) {
1719 if (atomic_read(&osc->oo_nr_writes) == 0)
1720 return 0;
1721 if (invalid_import) {
1722 CDEBUG(D_CACHE, "invalid import forcing RPC\n");
1723 return 1;
1724 }
1725 if (!list_empty(&osc->oo_hp_exts)) {
1726 CDEBUG(D_CACHE, "high prio request forcing RPC\n");
1727 return 1;
1728 }
1729 if (!list_empty(&osc->oo_urgent_exts)) {
1730 CDEBUG(D_CACHE, "urgent request forcing RPC\n");
1731 return 1;
1732 }
1733 /* trigger a write rpc stream as long as there are dirtiers
1734 * waiting for space. as they're waiting, they're not going to
1735 * create more pages to coalesce with what's waiting..
1736 */
1737 if (!list_empty(&cli->cl_cache_waiters)) {
1738 CDEBUG(D_CACHE, "cache waiters forcing RPC\n");
1739 return 1;
1740 }
1741 if (atomic_read(&osc->oo_nr_writes) >=
1742 cli->cl_max_pages_per_rpc)
1743 return 1;
1744 } else {
1745 if (atomic_read(&osc->oo_nr_reads) == 0)
1746 return 0;
1747 if (invalid_import) {
1748 CDEBUG(D_CACHE, "invalid import forcing RPC\n");
1749 return 1;
1750 }
1751 /* all read are urgent. */
1752 if (!list_empty(&osc->oo_reading_exts))
1753 return 1;
1754 }
1755
1756 return 0;
1757 }
1758
1759 static void osc_update_pending(struct osc_object *obj, int cmd, int delta)
1760 {
1761 struct client_obd *cli = osc_cli(obj);
1762
1763 if (cmd & OBD_BRW_WRITE) {
1764 atomic_add(delta, &obj->oo_nr_writes);
1765 atomic_add(delta, &cli->cl_pending_w_pages);
1766 LASSERT(atomic_read(&obj->oo_nr_writes) >= 0);
1767 } else {
1768 atomic_add(delta, &obj->oo_nr_reads);
1769 atomic_add(delta, &cli->cl_pending_r_pages);
1770 LASSERT(atomic_read(&obj->oo_nr_reads) >= 0);
1771 }
1772 OSC_IO_DEBUG(obj, "update pending cmd %d delta %d.\n", cmd, delta);
1773 }
1774
1775 static int osc_makes_hprpc(struct osc_object *obj)
1776 {
1777 return !list_empty(&obj->oo_hp_exts);
1778 }
1779
1780 static void on_list(struct list_head *item, struct list_head *list, int should_be_on)
1781 {
1782 if (list_empty(item) && should_be_on)
1783 list_add_tail(item, list);
1784 else if (!list_empty(item) && !should_be_on)
1785 list_del_init(item);
1786 }
1787
1788 /* maintain the osc's cli list membership invariants so that osc_send_oap_rpc
1789 * can find pages to build into rpcs quickly
1790 */
1791 static int __osc_list_maint(struct client_obd *cli, struct osc_object *osc)
1792 {
1793 if (osc_makes_hprpc(osc)) {
1794 /* HP rpc */
1795 on_list(&osc->oo_ready_item, &cli->cl_loi_ready_list, 0);
1796 on_list(&osc->oo_hp_ready_item, &cli->cl_loi_hp_ready_list, 1);
1797 } else {
1798 on_list(&osc->oo_hp_ready_item, &cli->cl_loi_hp_ready_list, 0);
1799 on_list(&osc->oo_ready_item, &cli->cl_loi_ready_list,
1800 osc_makes_rpc(cli, osc, OBD_BRW_WRITE) ||
1801 osc_makes_rpc(cli, osc, OBD_BRW_READ));
1802 }
1803
1804 on_list(&osc->oo_write_item, &cli->cl_loi_write_list,
1805 atomic_read(&osc->oo_nr_writes) > 0);
1806
1807 on_list(&osc->oo_read_item, &cli->cl_loi_read_list,
1808 atomic_read(&osc->oo_nr_reads) > 0);
1809
1810 return osc_is_ready(osc);
1811 }
1812
1813 static int osc_list_maint(struct client_obd *cli, struct osc_object *osc)
1814 {
1815 int is_ready;
1816
1817 spin_lock(&cli->cl_loi_list_lock);
1818 is_ready = __osc_list_maint(cli, osc);
1819 spin_unlock(&cli->cl_loi_list_lock);
1820
1821 return is_ready;
1822 }
1823
1824 /* this is trying to propagate async writeback errors back up to the
1825 * application. As an async write fails we record the error code for later if
1826 * the app does an fsync. As long as errors persist we force future rpcs to be
1827 * sync so that the app can get a sync error and break the cycle of queueing
1828 * pages for which writeback will fail.
1829 */
1830 static void osc_process_ar(struct osc_async_rc *ar, __u64 xid,
1831 int rc)
1832 {
1833 if (rc) {
1834 if (!ar->ar_rc)
1835 ar->ar_rc = rc;
1836
1837 ar->ar_force_sync = 1;
1838 ar->ar_min_xid = ptlrpc_sample_next_xid();
1839 return;
1840 }
1841
1842 if (ar->ar_force_sync && (xid >= ar->ar_min_xid))
1843 ar->ar_force_sync = 0;
1844 }
1845
1846 /**
1847 * Performs "unstable" page accounting. This function balances the
1848 * increment operations performed in osc_inc_unstable_pages. It is
1849 * registered as the RPC request callback, and is executed when the
1850 * bulk RPC is committed on the server. Thus at this point, the pages
1851 * involved in the bulk transfer are no longer considered unstable.
1852 */
1853 void osc_dec_unstable_pages(struct ptlrpc_request *req)
1854 {
1855 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
1856 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
1857 int page_count = desc->bd_iov_count;
1858 int i;
1859
1860 /* No unstable page tracking */
1861 if (!cli->cl_cache)
1862 return;
1863
1864 LASSERT(page_count >= 0);
1865
1866 for (i = 0; i < page_count; i++)
1867 dec_zone_page_state(desc->bd_iov[i].kiov_page, NR_UNSTABLE_NFS);
1868
1869 atomic_sub(page_count, &cli->cl_cache->ccc_unstable_nr);
1870 LASSERT(atomic_read(&cli->cl_cache->ccc_unstable_nr) >= 0);
1871
1872 atomic_sub(page_count, &cli->cl_unstable_count);
1873 LASSERT(atomic_read(&cli->cl_unstable_count) >= 0);
1874
1875 atomic_sub(page_count, &obd_unstable_pages);
1876 LASSERT(atomic_read(&obd_unstable_pages) >= 0);
1877
1878 spin_lock(&req->rq_lock);
1879 req->rq_committed = 1;
1880 req->rq_unstable = 0;
1881 spin_unlock(&req->rq_lock);
1882
1883 wake_up_all(&cli->cl_cache->ccc_unstable_waitq);
1884 }
1885
1886 /* "unstable" page accounting. See: osc_dec_unstable_pages. */
1887 void osc_inc_unstable_pages(struct ptlrpc_request *req)
1888 {
1889 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
1890 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
1891 long page_count = desc->bd_iov_count;
1892 int i;
1893
1894 /* No unstable page tracking */
1895 if (!cli->cl_cache)
1896 return;
1897
1898 LASSERT(page_count >= 0);
1899
1900 for (i = 0; i < page_count; i++)
1901 inc_zone_page_state(desc->bd_iov[i].kiov_page, NR_UNSTABLE_NFS);
1902
1903 LASSERT(atomic_read(&cli->cl_cache->ccc_unstable_nr) >= 0);
1904 atomic_add(page_count, &cli->cl_cache->ccc_unstable_nr);
1905
1906 LASSERT(atomic_read(&cli->cl_unstable_count) >= 0);
1907 atomic_add(page_count, &cli->cl_unstable_count);
1908
1909 LASSERT(atomic_read(&obd_unstable_pages) >= 0);
1910 atomic_add(page_count, &obd_unstable_pages);
1911
1912 spin_lock(&req->rq_lock);
1913
1914 /*
1915 * If the request has already been committed (i.e. brw_commit
1916 * called via rq_commit_cb), we need to undo the unstable page
1917 * increments we just performed because rq_commit_cb wont be
1918 * called again. Otherwise, just set the commit callback so the
1919 * unstable page accounting is properly updated when the request
1920 * is committed
1921 */
1922 if (req->rq_committed) {
1923 /* Drop lock before calling osc_dec_unstable_pages */
1924 spin_unlock(&req->rq_lock);
1925 osc_dec_unstable_pages(req);
1926 spin_lock(&req->rq_lock);
1927 } else {
1928 req->rq_unstable = 1;
1929 req->rq_commit_cb = osc_dec_unstable_pages;
1930 }
1931
1932 spin_unlock(&req->rq_lock);
1933 }
1934
1935 /* this must be called holding the loi list lock to give coverage to exit_cache,
1936 * async_flag maintenance, and oap_request
1937 */
1938 static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
1939 struct osc_async_page *oap, int sent, int rc)
1940 {
1941 struct osc_object *osc = oap->oap_obj;
1942 struct lov_oinfo *loi = osc->oo_oinfo;
1943 __u64 xid = 0;
1944
1945 if (oap->oap_request) {
1946 if (!rc)
1947 osc_inc_unstable_pages(oap->oap_request);
1948
1949 xid = ptlrpc_req_xid(oap->oap_request);
1950 ptlrpc_req_finished(oap->oap_request);
1951 oap->oap_request = NULL;
1952 }
1953
1954 /* As the transfer for this page is being done, clear the flags */
1955 spin_lock(&oap->oap_lock);
1956 oap->oap_async_flags = 0;
1957 spin_unlock(&oap->oap_lock);
1958 oap->oap_interrupted = 0;
1959
1960 if (oap->oap_cmd & OBD_BRW_WRITE && xid > 0) {
1961 spin_lock(&cli->cl_loi_list_lock);
1962 osc_process_ar(&cli->cl_ar, xid, rc);
1963 osc_process_ar(&loi->loi_ar, xid, rc);
1964 spin_unlock(&cli->cl_loi_list_lock);
1965 }
1966
1967 rc = osc_completion(env, oap, oap->oap_cmd, rc);
1968 if (rc)
1969 CERROR("completion on oap %p obj %p returns %d.\n",
1970 oap, osc, rc);
1971 }
1972
1973 /**
1974 * Try to add extent to one RPC. We need to think about the following things:
1975 * - # of pages must not be over max_pages_per_rpc
1976 * - extent must be compatible with previous ones
1977 */
1978 static int try_to_add_extent_for_io(struct client_obd *cli,
1979 struct osc_extent *ext, struct list_head *rpclist,
1980 int *pc, unsigned int *max_pages)
1981 {
1982 struct osc_extent *tmp;
1983 struct osc_async_page *oap = list_first_entry(&ext->oe_pages,
1984 struct osc_async_page,
1985 oap_pending_item);
1986
1987 EASSERT((ext->oe_state == OES_CACHE || ext->oe_state == OES_LOCK_DONE),
1988 ext);
1989
1990 *max_pages = max(ext->oe_mppr, *max_pages);
1991 if (*pc + ext->oe_nr_pages > *max_pages)
1992 return 0;
1993
1994 list_for_each_entry(tmp, rpclist, oe_link) {
1995 struct osc_async_page *oap2;
1996
1997 oap2 = list_first_entry(&tmp->oe_pages, struct osc_async_page,
1998 oap_pending_item);
1999 EASSERT(tmp->oe_owner == current, tmp);
2000 if (oap2cl_page(oap)->cp_type != oap2cl_page(oap2)->cp_type) {
2001 CDEBUG(D_CACHE, "Do not permit different type of IO"
2002 " for a same RPC\n");
2003 return 0;
2004 }
2005
2006 if (tmp->oe_srvlock != ext->oe_srvlock ||
2007 !tmp->oe_grants != !ext->oe_grants)
2008 return 0;
2009
2010 /* remove break for strict check */
2011 break;
2012 }
2013
2014 *pc += ext->oe_nr_pages;
2015 list_move_tail(&ext->oe_link, rpclist);
2016 ext->oe_owner = current;
2017 return 1;
2018 }
2019
2020 /**
2021 * In order to prevent multiple ptlrpcd from breaking contiguous extents,
2022 * get_write_extent() takes all appropriate extents in atomic.
2023 *
2024 * The following policy is used to collect extents for IO:
2025 * 1. Add as many HP extents as possible;
2026 * 2. Add the first urgent extent in urgent extent list and take it out of
2027 * urgent list;
2028 * 3. Add subsequent extents of this urgent extent;
2029 * 4. If urgent list is not empty, goto 2;
2030 * 5. Traverse the extent tree from the 1st extent;
2031 * 6. Above steps exit if there is no space in this RPC.
2032 */
2033 static int get_write_extents(struct osc_object *obj, struct list_head *rpclist)
2034 {
2035 struct client_obd *cli = osc_cli(obj);
2036 struct osc_extent *ext;
2037 struct osc_extent *temp;
2038 int page_count = 0;
2039 unsigned int max_pages = cli->cl_max_pages_per_rpc;
2040
2041 LASSERT(osc_object_is_locked(obj));
2042 list_for_each_entry_safe(ext, temp, &obj->oo_hp_exts, oe_link) {
2043 LASSERT(ext->oe_state == OES_CACHE);
2044 if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count,
2045 &max_pages))
2046 return page_count;
2047 EASSERT(ext->oe_nr_pages <= max_pages, ext);
2048 }
2049 if (page_count == max_pages)
2050 return page_count;
2051
2052 while (!list_empty(&obj->oo_urgent_exts)) {
2053 ext = list_entry(obj->oo_urgent_exts.next,
2054 struct osc_extent, oe_link);
2055 if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count,
2056 &max_pages))
2057 return page_count;
2058
2059 if (!ext->oe_intree)
2060 continue;
2061
2062 while ((ext = next_extent(ext)) != NULL) {
2063 if ((ext->oe_state != OES_CACHE) ||
2064 (!list_empty(&ext->oe_link) &&
2065 ext->oe_owner))
2066 continue;
2067
2068 if (!try_to_add_extent_for_io(cli, ext, rpclist,
2069 &page_count, &max_pages))
2070 return page_count;
2071 }
2072 }
2073 if (page_count == max_pages)
2074 return page_count;
2075
2076 ext = first_extent(obj);
2077 while (ext) {
2078 if ((ext->oe_state != OES_CACHE) ||
2079 /* this extent may be already in current rpclist */
2080 (!list_empty(&ext->oe_link) && ext->oe_owner)) {
2081 ext = next_extent(ext);
2082 continue;
2083 }
2084
2085 if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count,
2086 &max_pages))
2087 return page_count;
2088
2089 ext = next_extent(ext);
2090 }
2091 return page_count;
2092 }
2093
2094 static int
2095 osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli,
2096 struct osc_object *osc)
2097 __must_hold(osc)
2098 {
2099 LIST_HEAD(rpclist);
2100 struct osc_extent *ext;
2101 struct osc_extent *tmp;
2102 struct osc_extent *first = NULL;
2103 u32 page_count = 0;
2104 int srvlock = 0;
2105 int rc = 0;
2106
2107 LASSERT(osc_object_is_locked(osc));
2108
2109 page_count = get_write_extents(osc, &rpclist);
2110 LASSERT(equi(page_count == 0, list_empty(&rpclist)));
2111
2112 if (list_empty(&rpclist))
2113 return 0;
2114
2115 osc_update_pending(osc, OBD_BRW_WRITE, -page_count);
2116
2117 list_for_each_entry(ext, &rpclist, oe_link) {
2118 LASSERT(ext->oe_state == OES_CACHE ||
2119 ext->oe_state == OES_LOCK_DONE);
2120 if (ext->oe_state == OES_CACHE)
2121 osc_extent_state_set(ext, OES_LOCKING);
2122 else
2123 osc_extent_state_set(ext, OES_RPC);
2124 }
2125
2126 /* we're going to grab page lock, so release object lock because
2127 * lock order is page lock -> object lock.
2128 */
2129 osc_object_unlock(osc);
2130
2131 list_for_each_entry_safe(ext, tmp, &rpclist, oe_link) {
2132 if (ext->oe_state == OES_LOCKING) {
2133 rc = osc_extent_make_ready(env, ext);
2134 if (unlikely(rc < 0)) {
2135 list_del_init(&ext->oe_link);
2136 osc_extent_finish(env, ext, 0, rc);
2137 continue;
2138 }
2139 }
2140 if (!first) {
2141 first = ext;
2142 srvlock = ext->oe_srvlock;
2143 } else {
2144 LASSERT(srvlock == ext->oe_srvlock);
2145 }
2146 }
2147
2148 if (!list_empty(&rpclist)) {
2149 LASSERT(page_count > 0);
2150 rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_WRITE);
2151 LASSERT(list_empty(&rpclist));
2152 }
2153
2154 osc_object_lock(osc);
2155 return rc;
2156 }
2157
2158 /**
2159 * prepare pages for ASYNC io and put pages in send queue.
2160 *
2161 * \param cmd OBD_BRW_* macroses
2162 * \param lop pending pages
2163 *
2164 * \return zero if no page added to send queue.
2165 * \return 1 if pages successfully added to send queue.
2166 * \return negative on errors.
2167 */
2168 static int
2169 osc_send_read_rpc(const struct lu_env *env, struct client_obd *cli,
2170 struct osc_object *osc)
2171 __must_hold(osc)
2172 {
2173 struct osc_extent *ext;
2174 struct osc_extent *next;
2175 LIST_HEAD(rpclist);
2176 int page_count = 0;
2177 unsigned int max_pages = cli->cl_max_pages_per_rpc;
2178 int rc = 0;
2179
2180 LASSERT(osc_object_is_locked(osc));
2181 list_for_each_entry_safe(ext, next, &osc->oo_reading_exts, oe_link) {
2182 EASSERT(ext->oe_state == OES_LOCK_DONE, ext);
2183 if (!try_to_add_extent_for_io(cli, ext, &rpclist, &page_count,
2184 &max_pages))
2185 break;
2186 osc_extent_state_set(ext, OES_RPC);
2187 EASSERT(ext->oe_nr_pages <= max_pages, ext);
2188 }
2189 LASSERT(page_count <= max_pages);
2190
2191 osc_update_pending(osc, OBD_BRW_READ, -page_count);
2192
2193 if (!list_empty(&rpclist)) {
2194 osc_object_unlock(osc);
2195
2196 LASSERT(page_count > 0);
2197 rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_READ);
2198 LASSERT(list_empty(&rpclist));
2199
2200 osc_object_lock(osc);
2201 }
2202 return rc;
2203 }
2204
2205 #define list_to_obj(list, item) ({ \
2206 struct list_head *__tmp = (list)->next; \
2207 list_del_init(__tmp); \
2208 list_entry(__tmp, struct osc_object, oo_##item); \
2209 })
2210
2211 /* This is called by osc_check_rpcs() to find which objects have pages that
2212 * we could be sending. These lists are maintained by osc_makes_rpc().
2213 */
2214 static struct osc_object *osc_next_obj(struct client_obd *cli)
2215 {
2216 /* First return objects that have blocked locks so that they
2217 * will be flushed quickly and other clients can get the lock,
2218 * then objects which have pages ready to be stuffed into RPCs
2219 */
2220 if (!list_empty(&cli->cl_loi_hp_ready_list))
2221 return list_to_obj(&cli->cl_loi_hp_ready_list, hp_ready_item);
2222 if (!list_empty(&cli->cl_loi_ready_list))
2223 return list_to_obj(&cli->cl_loi_ready_list, ready_item);
2224
2225 /* then if we have cache waiters, return all objects with queued
2226 * writes. This is especially important when many small files
2227 * have filled up the cache and not been fired into rpcs because
2228 * they don't pass the nr_pending/object threshold
2229 */
2230 if (!list_empty(&cli->cl_cache_waiters) &&
2231 !list_empty(&cli->cl_loi_write_list))
2232 return list_to_obj(&cli->cl_loi_write_list, write_item);
2233
2234 /* then return all queued objects when we have an invalid import
2235 * so that they get flushed
2236 */
2237 if (!cli->cl_import || cli->cl_import->imp_invalid) {
2238 if (!list_empty(&cli->cl_loi_write_list))
2239 return list_to_obj(&cli->cl_loi_write_list, write_item);
2240 if (!list_empty(&cli->cl_loi_read_list))
2241 return list_to_obj(&cli->cl_loi_read_list, read_item);
2242 }
2243 return NULL;
2244 }
2245
2246 /* called with the loi list lock held */
2247 static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli)
2248 __must_hold(&cli->cl_loi_list_lock)
2249 {
2250 struct osc_object *osc;
2251 int rc = 0;
2252
2253 while ((osc = osc_next_obj(cli)) != NULL) {
2254 struct cl_object *obj = osc2cl(osc);
2255 struct lu_ref_link link;
2256
2257 OSC_IO_DEBUG(osc, "%lu in flight\n", rpcs_in_flight(cli));
2258
2259 if (osc_max_rpc_in_flight(cli, osc)) {
2260 __osc_list_maint(cli, osc);
2261 break;
2262 }
2263
2264 cl_object_get(obj);
2265 spin_unlock(&cli->cl_loi_list_lock);
2266 lu_object_ref_add_at(&obj->co_lu, &link, "check", current);
2267
2268 /* attempt some read/write balancing by alternating between
2269 * reads and writes in an object. The makes_rpc checks here
2270 * would be redundant if we were getting read/write work items
2271 * instead of objects. we don't want send_oap_rpc to drain a
2272 * partial read pending queue when we're given this object to
2273 * do io on writes while there are cache waiters
2274 */
2275 osc_object_lock(osc);
2276 if (osc_makes_rpc(cli, osc, OBD_BRW_WRITE)) {
2277 rc = osc_send_write_rpc(env, cli, osc);
2278 if (rc < 0) {
2279 CERROR("Write request failed with %d\n", rc);
2280
2281 /* osc_send_write_rpc failed, mostly because of
2282 * memory pressure.
2283 *
2284 * It can't break here, because if:
2285 * - a page was submitted by osc_io_submit, so
2286 * page locked;
2287 * - no request in flight
2288 * - no subsequent request
2289 * The system will be in live-lock state,
2290 * because there is no chance to call
2291 * osc_io_unplug() and osc_check_rpcs() any
2292 * more. pdflush can't help in this case,
2293 * because it might be blocked at grabbing
2294 * the page lock as we mentioned.
2295 *
2296 * Anyway, continue to drain pages.
2297 */
2298 /* break; */
2299 }
2300 }
2301 if (osc_makes_rpc(cli, osc, OBD_BRW_READ)) {
2302 rc = osc_send_read_rpc(env, cli, osc);
2303 if (rc < 0)
2304 CERROR("Read request failed with %d\n", rc);
2305 }
2306 osc_object_unlock(osc);
2307
2308 osc_list_maint(cli, osc);
2309 lu_object_ref_del_at(&obj->co_lu, &link, "check", current);
2310 cl_object_put(env, obj);
2311
2312 spin_lock(&cli->cl_loi_list_lock);
2313 }
2314 }
2315
2316 static int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
2317 struct osc_object *osc, int async)
2318 {
2319 int rc = 0;
2320
2321 if (osc && osc_list_maint(cli, osc) == 0)
2322 return 0;
2323
2324 if (!async) {
2325 /* disable osc_lru_shrink() temporarily to avoid
2326 * potential stack overrun problem. LU-2859
2327 */
2328 atomic_inc(&cli->cl_lru_shrinkers);
2329 spin_lock(&cli->cl_loi_list_lock);
2330 osc_check_rpcs(env, cli);
2331 spin_unlock(&cli->cl_loi_list_lock);
2332 atomic_dec(&cli->cl_lru_shrinkers);
2333 } else {
2334 CDEBUG(D_CACHE, "Queue writeback work for client %p.\n", cli);
2335 LASSERT(cli->cl_writeback_work);
2336 rc = ptlrpcd_queue_work(cli->cl_writeback_work);
2337 }
2338 return rc;
2339 }
2340
2341 static int osc_io_unplug_async(const struct lu_env *env,
2342 struct client_obd *cli, struct osc_object *osc)
2343 {
2344 return osc_io_unplug0(env, cli, osc, 1);
2345 }
2346
2347 void osc_io_unplug(const struct lu_env *env, struct client_obd *cli,
2348 struct osc_object *osc)
2349 {
2350 (void)osc_io_unplug0(env, cli, osc, 0);
2351 }
2352
2353 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
2354 struct page *page, loff_t offset)
2355 {
2356 struct obd_export *exp = osc_export(osc);
2357 struct osc_async_page *oap = &ops->ops_oap;
2358
2359 if (!page)
2360 return cfs_size_round(sizeof(*oap));
2361
2362 oap->oap_magic = OAP_MAGIC;
2363 oap->oap_cli = &exp->exp_obd->u.cli;
2364 oap->oap_obj = osc;
2365
2366 oap->oap_page = page;
2367 oap->oap_obj_off = offset;
2368 LASSERT(!(offset & ~PAGE_MASK));
2369
2370 if (!client_is_remote(exp) && capable(CFS_CAP_SYS_RESOURCE))
2371 oap->oap_brw_flags = OBD_BRW_NOQUOTA;
2372
2373 INIT_LIST_HEAD(&oap->oap_pending_item);
2374 INIT_LIST_HEAD(&oap->oap_rpc_item);
2375
2376 spin_lock_init(&oap->oap_lock);
2377 CDEBUG(D_INFO, "oap %p page %p obj off %llu\n",
2378 oap, page, oap->oap_obj_off);
2379 return 0;
2380 }
2381
2382 int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
2383 struct osc_page *ops)
2384 {
2385 struct osc_io *oio = osc_env_io(env);
2386 struct osc_extent *ext = NULL;
2387 struct osc_async_page *oap = &ops->ops_oap;
2388 struct client_obd *cli = oap->oap_cli;
2389 struct osc_object *osc = oap->oap_obj;
2390 pgoff_t index;
2391 int grants = 0;
2392 int brw_flags = OBD_BRW_ASYNC;
2393 int cmd = OBD_BRW_WRITE;
2394 int need_release = 0;
2395 int rc = 0;
2396
2397 if (oap->oap_magic != OAP_MAGIC)
2398 return -EINVAL;
2399
2400 if (!cli->cl_import || cli->cl_import->imp_invalid)
2401 return -EIO;
2402
2403 if (!list_empty(&oap->oap_pending_item) ||
2404 !list_empty(&oap->oap_rpc_item))
2405 return -EBUSY;
2406
2407 /* Set the OBD_BRW_SRVLOCK before the page is queued. */
2408 brw_flags |= ops->ops_srvlock ? OBD_BRW_SRVLOCK : 0;
2409 if (!client_is_remote(osc_export(osc)) &&
2410 capable(CFS_CAP_SYS_RESOURCE)) {
2411 brw_flags |= OBD_BRW_NOQUOTA;
2412 cmd |= OBD_BRW_NOQUOTA;
2413 }
2414
2415 /* check if the file's owner/group is over quota */
2416 if (!(cmd & OBD_BRW_NOQUOTA)) {
2417 struct cl_object *obj;
2418 struct cl_attr *attr;
2419 unsigned int qid[MAXQUOTAS];
2420
2421 obj = cl_object_top(&osc->oo_cl);
2422 attr = &osc_env_info(env)->oti_attr;
2423
2424 cl_object_attr_lock(obj);
2425 rc = cl_object_attr_get(env, obj, attr);
2426 cl_object_attr_unlock(obj);
2427
2428 qid[USRQUOTA] = attr->cat_uid;
2429 qid[GRPQUOTA] = attr->cat_gid;
2430 if (rc == 0 && osc_quota_chkdq(cli, qid) == NO_QUOTA)
2431 rc = -EDQUOT;
2432 if (rc)
2433 return rc;
2434 }
2435
2436 if (osc_over_unstable_soft_limit(cli))
2437 brw_flags |= OBD_BRW_SOFT_SYNC;
2438
2439 oap->oap_cmd = cmd;
2440 oap->oap_page_off = ops->ops_from;
2441 oap->oap_count = ops->ops_to - ops->ops_from;
2442 /*
2443 * No need to hold a lock here,
2444 * since this page is not in any list yet.
2445 */
2446 oap->oap_async_flags = 0;
2447 oap->oap_brw_flags = brw_flags;
2448
2449 OSC_IO_DEBUG(osc, "oap %p page %p added for cmd %d\n",
2450 oap, oap->oap_page, oap->oap_cmd & OBD_BRW_RWMASK);
2451
2452 index = osc_index(oap2osc(oap));
2453
2454 /* Add this page into extent by the following steps:
2455 * 1. if there exists an active extent for this IO, mostly this page
2456 * can be added to the active extent and sometimes we need to
2457 * expand extent to accommodate this page;
2458 * 2. otherwise, a new extent will be allocated.
2459 */
2460
2461 ext = oio->oi_active;
2462 if (ext && ext->oe_start <= index && ext->oe_max_end >= index) {
2463 /* one chunk plus extent overhead must be enough to write this
2464 * page
2465 */
2466 grants = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
2467 if (ext->oe_end >= index)
2468 grants = 0;
2469
2470 /* it doesn't need any grant to dirty this page */
2471 spin_lock(&cli->cl_loi_list_lock);
2472 rc = osc_enter_cache_try(cli, oap, grants, 0);
2473 spin_unlock(&cli->cl_loi_list_lock);
2474 if (rc == 0) { /* try failed */
2475 grants = 0;
2476 need_release = 1;
2477 } else if (ext->oe_end < index) {
2478 int tmp = grants;
2479 /* try to expand this extent */
2480 rc = osc_extent_expand(ext, index, &tmp);
2481 if (rc < 0) {
2482 need_release = 1;
2483 /* don't free reserved grant */
2484 } else {
2485 OSC_EXTENT_DUMP(D_CACHE, ext,
2486 "expanded for %lu.\n", index);
2487 osc_unreserve_grant(cli, grants, tmp);
2488 grants = 0;
2489 }
2490 }
2491 rc = 0;
2492 } else if (ext) {
2493 /* index is located outside of active extent */
2494 need_release = 1;
2495 }
2496 if (need_release) {
2497 osc_extent_release(env, ext);
2498 oio->oi_active = NULL;
2499 ext = NULL;
2500 }
2501
2502 if (!ext) {
2503 int tmp = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
2504
2505 /* try to find new extent to cover this page */
2506 LASSERT(!oio->oi_active);
2507 /* we may have allocated grant for this page if we failed
2508 * to expand the previous active extent.
2509 */
2510 LASSERT(ergo(grants > 0, grants >= tmp));
2511
2512 rc = 0;
2513 if (grants == 0) {
2514 /* we haven't allocated grant for this page. */
2515 rc = osc_enter_cache(env, cli, oap, tmp);
2516 if (rc == 0)
2517 grants = tmp;
2518 }
2519
2520 tmp = grants;
2521 if (rc == 0) {
2522 ext = osc_extent_find(env, osc, index, &tmp);
2523 if (IS_ERR(ext)) {
2524 LASSERT(tmp == grants);
2525 osc_exit_cache(cli, oap);
2526 rc = PTR_ERR(ext);
2527 ext = NULL;
2528 } else {
2529 oio->oi_active = ext;
2530 }
2531 }
2532 if (grants > 0)
2533 osc_unreserve_grant(cli, grants, tmp);
2534 }
2535
2536 LASSERT(ergo(rc == 0, ext));
2537 if (ext) {
2538 EASSERTF(ext->oe_end >= index && ext->oe_start <= index,
2539 ext, "index = %lu.\n", index);
2540 LASSERT((oap->oap_brw_flags & OBD_BRW_FROM_GRANT) != 0);
2541
2542 osc_object_lock(osc);
2543 if (ext->oe_nr_pages == 0)
2544 ext->oe_srvlock = ops->ops_srvlock;
2545 else
2546 LASSERT(ext->oe_srvlock == ops->ops_srvlock);
2547 ++ext->oe_nr_pages;
2548 list_add_tail(&oap->oap_pending_item, &ext->oe_pages);
2549 osc_object_unlock(osc);
2550 }
2551 return rc;
2552 }
2553
2554 int osc_teardown_async_page(const struct lu_env *env,
2555 struct osc_object *obj, struct osc_page *ops)
2556 {
2557 struct osc_async_page *oap = &ops->ops_oap;
2558 struct osc_extent *ext = NULL;
2559 int rc = 0;
2560
2561 LASSERT(oap->oap_magic == OAP_MAGIC);
2562
2563 CDEBUG(D_INFO, "teardown oap %p page %p at index %lu.\n",
2564 oap, ops, osc_index(oap2osc(oap)));
2565
2566 osc_object_lock(obj);
2567 if (!list_empty(&oap->oap_rpc_item)) {
2568 CDEBUG(D_CACHE, "oap %p is not in cache.\n", oap);
2569 rc = -EBUSY;
2570 } else if (!list_empty(&oap->oap_pending_item)) {
2571 ext = osc_extent_lookup(obj, osc_index(oap2osc(oap)));
2572 /* only truncated pages are allowed to be taken out.
2573 * See osc_extent_truncate() and osc_cache_truncate_start()
2574 * for details.
2575 */
2576 if (ext && ext->oe_state != OES_TRUNC) {
2577 OSC_EXTENT_DUMP(D_ERROR, ext, "trunc at %lu.\n",
2578 osc_index(oap2osc(oap)));
2579 rc = -EBUSY;
2580 }
2581 }
2582 osc_object_unlock(obj);
2583 if (ext)
2584 osc_extent_put(env, ext);
2585 return rc;
2586 }
2587
2588 /**
2589 * This is called when a page is picked up by kernel to write out.
2590 *
2591 * We should find out the corresponding extent and add the whole extent
2592 * into urgent list. The extent may be being truncated or used, handle it
2593 * carefully.
2594 */
2595 int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
2596 struct osc_page *ops)
2597 {
2598 struct osc_extent *ext = NULL;
2599 struct osc_object *obj = cl2osc(ops->ops_cl.cpl_obj);
2600 struct cl_page *cp = ops->ops_cl.cpl_page;
2601 pgoff_t index = osc_index(ops);
2602 struct osc_async_page *oap = &ops->ops_oap;
2603 bool unplug = false;
2604 int rc = 0;
2605
2606 osc_object_lock(obj);
2607 ext = osc_extent_lookup(obj, index);
2608 if (!ext) {
2609 osc_extent_tree_dump(D_ERROR, obj);
2610 LASSERTF(0, "page index %lu is NOT covered.\n", index);
2611 }
2612
2613 switch (ext->oe_state) {
2614 case OES_RPC:
2615 case OES_LOCK_DONE:
2616 CL_PAGE_DEBUG(D_ERROR, env, cp, "flush an in-rpc page?\n");
2617 LASSERT(0);
2618 break;
2619 case OES_LOCKING:
2620 /* If we know this extent is being written out, we should abort
2621 * so that the writer can make this page ready. Otherwise, there
2622 * exists a deadlock problem because other process can wait for
2623 * page writeback bit holding page lock; and meanwhile in
2624 * vvp_page_make_ready(), we need to grab page lock before
2625 * really sending the RPC.
2626 */
2627 case OES_TRUNC:
2628 /* race with truncate, page will be redirtied */
2629 case OES_ACTIVE:
2630 /* The extent is active so we need to abort and let the caller
2631 * re-dirty the page. If we continued on here, and we were the
2632 * one making the extent active, we could deadlock waiting for
2633 * the page writeback to clear but it won't because the extent
2634 * is active and won't be written out.
2635 */
2636 rc = -EAGAIN;
2637 goto out;
2638 default:
2639 break;
2640 }
2641
2642 rc = cl_page_prep(env, io, cp, CRT_WRITE);
2643 if (rc)
2644 goto out;
2645
2646 spin_lock(&oap->oap_lock);
2647 oap->oap_async_flags |= ASYNC_READY|ASYNC_URGENT;
2648 spin_unlock(&oap->oap_lock);
2649
2650 if (memory_pressure_get())
2651 ext->oe_memalloc = 1;
2652
2653 ext->oe_urgent = 1;
2654 if (ext->oe_state == OES_CACHE) {
2655 OSC_EXTENT_DUMP(D_CACHE, ext,
2656 "flush page %p make it urgent.\n", oap);
2657 if (list_empty(&ext->oe_link))
2658 list_add_tail(&ext->oe_link, &obj->oo_urgent_exts);
2659 unplug = true;
2660 }
2661 rc = 0;
2662
2663 out:
2664 osc_object_unlock(obj);
2665 osc_extent_put(env, ext);
2666 if (unplug)
2667 osc_io_unplug_async(env, osc_cli(obj), obj);
2668 return rc;
2669 }
2670
2671 /**
2672 * this is called when a sync waiter receives an interruption. Its job is to
2673 * get the caller woken as soon as possible. If its page hasn't been put in an
2674 * rpc yet it can dequeue immediately. Otherwise it has to mark the rpc as
2675 * desiring interruption which will forcefully complete the rpc once the rpc
2676 * has timed out.
2677 */
2678 int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops)
2679 {
2680 struct osc_async_page *oap = &ops->ops_oap;
2681 struct osc_object *obj = oap->oap_obj;
2682 struct client_obd *cli = osc_cli(obj);
2683 struct osc_extent *ext;
2684 struct osc_extent *found = NULL;
2685 struct list_head *plist;
2686 pgoff_t index = osc_index(ops);
2687 int rc = -EBUSY;
2688 int cmd;
2689
2690 LASSERT(!oap->oap_interrupted);
2691 oap->oap_interrupted = 1;
2692
2693 /* Find out the caching extent */
2694 osc_object_lock(obj);
2695 if (oap->oap_cmd & OBD_BRW_WRITE) {
2696 plist = &obj->oo_urgent_exts;
2697 cmd = OBD_BRW_WRITE;
2698 } else {
2699 plist = &obj->oo_reading_exts;
2700 cmd = OBD_BRW_READ;
2701 }
2702 list_for_each_entry(ext, plist, oe_link) {
2703 if (ext->oe_start <= index && ext->oe_end >= index) {
2704 LASSERT(ext->oe_state == OES_LOCK_DONE);
2705 /* For OES_LOCK_DONE state extent, it has already held
2706 * a refcount for RPC.
2707 */
2708 found = osc_extent_get(ext);
2709 break;
2710 }
2711 }
2712 if (found) {
2713 list_del_init(&found->oe_link);
2714 osc_update_pending(obj, cmd, -found->oe_nr_pages);
2715 osc_object_unlock(obj);
2716
2717 osc_extent_finish(env, found, 0, -EINTR);
2718 osc_extent_put(env, found);
2719 rc = 0;
2720 } else {
2721 osc_object_unlock(obj);
2722 /* ok, it's been put in an rpc. only one oap gets a request
2723 * reference
2724 */
2725 if (oap->oap_request) {
2726 ptlrpc_mark_interrupted(oap->oap_request);
2727 ptlrpcd_wake(oap->oap_request);
2728 ptlrpc_req_finished(oap->oap_request);
2729 oap->oap_request = NULL;
2730 }
2731 }
2732
2733 osc_list_maint(cli, obj);
2734 return rc;
2735 }
2736
2737 int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj,
2738 struct list_head *list, int cmd, int brw_flags)
2739 {
2740 struct client_obd *cli = osc_cli(obj);
2741 struct osc_extent *ext;
2742 struct osc_async_page *oap, *tmp;
2743 int page_count = 0;
2744 int mppr = cli->cl_max_pages_per_rpc;
2745 pgoff_t start = CL_PAGE_EOF;
2746 pgoff_t end = 0;
2747
2748 list_for_each_entry(oap, list, oap_pending_item) {
2749 pgoff_t index = osc_index(oap2osc(oap));
2750
2751 if (index > end)
2752 end = index;
2753 if (index < start)
2754 start = index;
2755 ++page_count;
2756 mppr <<= (page_count > mppr);
2757 }
2758
2759 ext = osc_extent_alloc(obj);
2760 if (!ext) {
2761 list_for_each_entry_safe(oap, tmp, list, oap_pending_item) {
2762 list_del_init(&oap->oap_pending_item);
2763 osc_ap_completion(env, cli, oap, 0, -ENOMEM);
2764 }
2765 return -ENOMEM;
2766 }
2767
2768 ext->oe_rw = !!(cmd & OBD_BRW_READ);
2769 ext->oe_sync = 1;
2770 ext->oe_urgent = 1;
2771 ext->oe_start = start;
2772 ext->oe_end = end;
2773 ext->oe_max_end = end;
2774 ext->oe_obj = obj;
2775 ext->oe_srvlock = !!(brw_flags & OBD_BRW_SRVLOCK);
2776 ext->oe_nr_pages = page_count;
2777 ext->oe_mppr = mppr;
2778 list_splice_init(list, &ext->oe_pages);
2779
2780 osc_object_lock(obj);
2781 /* Reuse the initial refcount for RPC, don't drop it */
2782 osc_extent_state_set(ext, OES_LOCK_DONE);
2783 if (cmd & OBD_BRW_WRITE) {
2784 list_add_tail(&ext->oe_link, &obj->oo_urgent_exts);
2785 osc_update_pending(obj, OBD_BRW_WRITE, page_count);
2786 } else {
2787 list_add_tail(&ext->oe_link, &obj->oo_reading_exts);
2788 osc_update_pending(obj, OBD_BRW_READ, page_count);
2789 }
2790 osc_object_unlock(obj);
2791
2792 osc_io_unplug_async(env, cli, obj);
2793 return 0;
2794 }
2795
2796 /**
2797 * Called by osc_io_setattr_start() to freeze and destroy covering extents.
2798 */
2799 int osc_cache_truncate_start(const struct lu_env *env, struct osc_io *oio,
2800 struct osc_object *obj, __u64 size)
2801 {
2802 struct client_obd *cli = osc_cli(obj);
2803 struct osc_extent *ext;
2804 struct osc_extent *temp;
2805 struct osc_extent *waiting = NULL;
2806 pgoff_t index;
2807 LIST_HEAD(list);
2808 int result = 0;
2809 bool partial;
2810
2811 /* pages with index greater or equal to index will be truncated. */
2812 index = cl_index(osc2cl(obj), size);
2813 partial = size > cl_offset(osc2cl(obj), index);
2814
2815 again:
2816 osc_object_lock(obj);
2817 ext = osc_extent_search(obj, index);
2818 if (!ext)
2819 ext = first_extent(obj);
2820 else if (ext->oe_end < index)
2821 ext = next_extent(ext);
2822 while (ext) {
2823 EASSERT(ext->oe_state != OES_TRUNC, ext);
2824
2825 if (ext->oe_state > OES_CACHE || ext->oe_urgent) {
2826 /* if ext is in urgent state, it means there must exist
2827 * a page already having been flushed by write_page().
2828 * We have to wait for this extent because we can't
2829 * truncate that page.
2830 */
2831 LASSERT(!ext->oe_hp);
2832 OSC_EXTENT_DUMP(D_CACHE, ext,
2833 "waiting for busy extent\n");
2834 waiting = osc_extent_get(ext);
2835 break;
2836 }
2837
2838 OSC_EXTENT_DUMP(D_CACHE, ext, "try to trunc:%llu.\n", size);
2839
2840 osc_extent_get(ext);
2841 if (ext->oe_state == OES_ACTIVE) {
2842 /* though we grab inode mutex for write path, but we
2843 * release it before releasing extent(in osc_io_end()),
2844 * so there is a race window that an extent is still
2845 * in OES_ACTIVE when truncate starts.
2846 */
2847 LASSERT(!ext->oe_trunc_pending);
2848 ext->oe_trunc_pending = 1;
2849 } else {
2850 EASSERT(ext->oe_state == OES_CACHE, ext);
2851 osc_extent_state_set(ext, OES_TRUNC);
2852 osc_update_pending(obj, OBD_BRW_WRITE,
2853 -ext->oe_nr_pages);
2854 }
2855 EASSERT(list_empty(&ext->oe_link), ext);
2856 list_add_tail(&ext->oe_link, &list);
2857
2858 ext = next_extent(ext);
2859 }
2860 osc_object_unlock(obj);
2861
2862 osc_list_maint(cli, obj);
2863
2864 list_for_each_entry_safe(ext, temp, &list, oe_link) {
2865 int rc;
2866
2867 list_del_init(&ext->oe_link);
2868
2869 /* extent may be in OES_ACTIVE state because inode mutex
2870 * is released before osc_io_end() in file write case
2871 */
2872 if (ext->oe_state != OES_TRUNC)
2873 osc_extent_wait(env, ext, OES_TRUNC);
2874
2875 rc = osc_extent_truncate(ext, index, partial);
2876 if (rc < 0) {
2877 if (result == 0)
2878 result = rc;
2879
2880 OSC_EXTENT_DUMP(D_ERROR, ext,
2881 "truncate error %d\n", rc);
2882 } else if (ext->oe_nr_pages == 0) {
2883 osc_extent_remove(ext);
2884 } else {
2885 /* this must be an overlapped extent which means only
2886 * part of pages in this extent have been truncated.
2887 */
2888 EASSERTF(ext->oe_start <= index, ext,
2889 "trunc index = %lu/%d.\n", index, partial);
2890 /* fix index to skip this partially truncated extent */
2891 index = ext->oe_end + 1;
2892 partial = false;
2893
2894 /* we need to hold this extent in OES_TRUNC state so
2895 * that no writeback will happen. This is to avoid
2896 * BUG 17397.
2897 */
2898 LASSERT(!oio->oi_trunc);
2899 oio->oi_trunc = osc_extent_get(ext);
2900 OSC_EXTENT_DUMP(D_CACHE, ext,
2901 "trunc at %llu\n", size);
2902 }
2903 osc_extent_put(env, ext);
2904 }
2905 if (waiting) {
2906 int rc;
2907
2908 /* ignore the result of osc_extent_wait the write initiator
2909 * should take care of it.
2910 */
2911 rc = osc_extent_wait(env, waiting, OES_INV);
2912 if (rc < 0)
2913 OSC_EXTENT_DUMP(D_CACHE, waiting, "error: %d.\n", rc);
2914
2915 osc_extent_put(env, waiting);
2916 waiting = NULL;
2917 goto again;
2918 }
2919 return result;
2920 }
2921
2922 /**
2923 * Called after osc_io_setattr_end to add oio->oi_trunc back to cache.
2924 */
2925 void osc_cache_truncate_end(const struct lu_env *env, struct osc_io *oio,
2926 struct osc_object *obj)
2927 {
2928 struct osc_extent *ext = oio->oi_trunc;
2929
2930 oio->oi_trunc = NULL;
2931 if (ext) {
2932 bool unplug = false;
2933
2934 EASSERT(ext->oe_nr_pages > 0, ext);
2935 EASSERT(ext->oe_state == OES_TRUNC, ext);
2936 EASSERT(!ext->oe_urgent, ext);
2937
2938 OSC_EXTENT_DUMP(D_CACHE, ext, "trunc -> cache.\n");
2939 osc_object_lock(obj);
2940 osc_extent_state_set(ext, OES_CACHE);
2941 if (ext->oe_fsync_wait && !ext->oe_urgent) {
2942 ext->oe_urgent = 1;
2943 list_move_tail(&ext->oe_link, &obj->oo_urgent_exts);
2944 unplug = true;
2945 }
2946 osc_update_pending(obj, OBD_BRW_WRITE, ext->oe_nr_pages);
2947 osc_object_unlock(obj);
2948 osc_extent_put(env, ext);
2949
2950 if (unplug)
2951 osc_io_unplug_async(env, osc_cli(obj), obj);
2952 }
2953 }
2954
2955 /**
2956 * Wait for extents in a specific range to be written out.
2957 * The caller must have called osc_cache_writeback_range() to issue IO
2958 * otherwise it will take a long time for this function to finish.
2959 *
2960 * Caller must hold inode_mutex , or cancel exclusive dlm lock so that
2961 * nobody else can dirty this range of file while we're waiting for
2962 * extents to be written.
2963 */
2964 int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
2965 pgoff_t start, pgoff_t end)
2966 {
2967 struct osc_extent *ext;
2968 pgoff_t index = start;
2969 int result = 0;
2970
2971 again:
2972 osc_object_lock(obj);
2973 ext = osc_extent_search(obj, index);
2974 if (!ext)
2975 ext = first_extent(obj);
2976 else if (ext->oe_end < index)
2977 ext = next_extent(ext);
2978 while (ext) {
2979 int rc;
2980
2981 if (ext->oe_start > end)
2982 break;
2983
2984 if (!ext->oe_fsync_wait) {
2985 ext = next_extent(ext);
2986 continue;
2987 }
2988
2989 EASSERT(ergo(ext->oe_state == OES_CACHE,
2990 ext->oe_hp || ext->oe_urgent), ext);
2991 EASSERT(ergo(ext->oe_state == OES_ACTIVE,
2992 !ext->oe_hp && ext->oe_urgent), ext);
2993
2994 index = ext->oe_end + 1;
2995 osc_extent_get(ext);
2996 osc_object_unlock(obj);
2997
2998 rc = osc_extent_wait(env, ext, OES_INV);
2999 if (result == 0)
3000 result = rc;
3001 osc_extent_put(env, ext);
3002 goto again;
3003 }
3004 osc_object_unlock(obj);
3005
3006 OSC_IO_DEBUG(obj, "sync file range.\n");
3007 return result;
3008 }
3009
3010 /**
3011 * Called to write out a range of osc object.
3012 *
3013 * @hp : should be set this is caused by lock cancel;
3014 * @discard: is set if dirty pages should be dropped - file will be deleted or
3015 * truncated, this implies there is no partially discarding extents.
3016 *
3017 * Return how many pages will be issued, or error code if error occurred.
3018 */
3019 int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
3020 pgoff_t start, pgoff_t end, int hp, int discard)
3021 {
3022 struct osc_extent *ext;
3023 LIST_HEAD(discard_list);
3024 bool unplug = false;
3025 int result = 0;
3026
3027 osc_object_lock(obj);
3028 ext = osc_extent_search(obj, start);
3029 if (!ext)
3030 ext = first_extent(obj);
3031 else if (ext->oe_end < start)
3032 ext = next_extent(ext);
3033 while (ext) {
3034 if (ext->oe_start > end)
3035 break;
3036
3037 ext->oe_fsync_wait = 1;
3038 switch (ext->oe_state) {
3039 case OES_CACHE:
3040 result += ext->oe_nr_pages;
3041 if (!discard) {
3042 struct list_head *list = NULL;
3043
3044 if (hp) {
3045 EASSERT(!ext->oe_hp, ext);
3046 ext->oe_hp = 1;
3047 list = &obj->oo_hp_exts;
3048 } else if (!ext->oe_urgent) {
3049 ext->oe_urgent = 1;
3050 list = &obj->oo_urgent_exts;
3051 }
3052 if (list)
3053 list_move_tail(&ext->oe_link, list);
3054 unplug = true;
3055 } else {
3056 /* the only discarder is lock cancelling, so
3057 * [start, end] must contain this extent
3058 */
3059 EASSERT(ext->oe_start >= start &&
3060 ext->oe_max_end <= end, ext);
3061 osc_extent_state_set(ext, OES_LOCKING);
3062 ext->oe_owner = current;
3063 list_move_tail(&ext->oe_link, &discard_list);
3064 osc_update_pending(obj, OBD_BRW_WRITE,
3065 -ext->oe_nr_pages);
3066 }
3067 break;
3068 case OES_ACTIVE:
3069 /* It's pretty bad to wait for ACTIVE extents, because
3070 * we don't know how long we will wait for it to be
3071 * flushed since it may be blocked at awaiting more
3072 * grants. We do this for the correctness of fsync.
3073 */
3074 LASSERT(hp == 0 && discard == 0);
3075 ext->oe_urgent = 1;
3076 break;
3077 case OES_TRUNC:
3078 /* this extent is being truncated, can't do anything
3079 * for it now. it will be set to urgent after truncate
3080 * is finished in osc_cache_truncate_end().
3081 */
3082 default:
3083 break;
3084 }
3085 ext = next_extent(ext);
3086 }
3087 osc_object_unlock(obj);
3088
3089 LASSERT(ergo(!discard, list_empty(&discard_list)));
3090 if (!list_empty(&discard_list)) {
3091 struct osc_extent *tmp;
3092 int rc;
3093
3094 osc_list_maint(osc_cli(obj), obj);
3095 list_for_each_entry_safe(ext, tmp, &discard_list, oe_link) {
3096 list_del_init(&ext->oe_link);
3097 EASSERT(ext->oe_state == OES_LOCKING, ext);
3098
3099 /* Discard caching pages. We don't actually write this
3100 * extent out but we complete it as if we did.
3101 */
3102 rc = osc_extent_make_ready(env, ext);
3103 if (unlikely(rc < 0)) {
3104 OSC_EXTENT_DUMP(D_ERROR, ext,
3105 "make_ready returned %d\n", rc);
3106 if (result >= 0)
3107 result = rc;
3108 }
3109
3110 /* finish the extent as if the pages were sent */
3111 osc_extent_finish(env, ext, 0, 0);
3112 }
3113 }
3114
3115 if (unplug)
3116 osc_io_unplug(env, osc_cli(obj), obj);
3117
3118 if (hp || discard) {
3119 int rc;
3120
3121 rc = osc_cache_wait_range(env, obj, start, end);
3122 if (result >= 0 && rc < 0)
3123 result = rc;
3124 }
3125
3126 OSC_IO_DEBUG(obj, "pageout [%lu, %lu], %d.\n", start, end, result);
3127 return result;
3128 }
3129
3130 /**
3131 * Returns a list of pages by a given [start, end] of \a obj.
3132 *
3133 * \param resched If not NULL, then we give up before hogging CPU for too
3134 * long and set *resched = 1, in that case caller should implement a retry
3135 * logic.
3136 *
3137 * Gang tree lookup (radix_tree_gang_lookup()) optimization is absolutely
3138 * crucial in the face of [offset, EOF] locks.
3139 *
3140 * Return at least one page in @queue unless there is no covered page.
3141 */
3142 int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io,
3143 struct osc_object *osc, pgoff_t start, pgoff_t end,
3144 osc_page_gang_cbt cb, void *cbdata)
3145 {
3146 struct osc_page *ops;
3147 void **pvec;
3148 pgoff_t idx;
3149 unsigned int nr;
3150 unsigned int i;
3151 unsigned int j;
3152 int res = CLP_GANG_OKAY;
3153 bool tree_lock = true;
3154
3155 idx = start;
3156 pvec = osc_env_info(env)->oti_pvec;
3157 spin_lock(&osc->oo_tree_lock);
3158 while ((nr = radix_tree_gang_lookup(&osc->oo_tree, pvec,
3159 idx, OTI_PVEC_SIZE)) > 0) {
3160 struct cl_page *page;
3161 bool end_of_region = false;
3162
3163 for (i = 0, j = 0; i < nr; ++i) {
3164 ops = pvec[i];
3165 pvec[i] = NULL;
3166
3167 idx = osc_index(ops);
3168 if (idx > end) {
3169 end_of_region = true;
3170 break;
3171 }
3172
3173 page = ops->ops_cl.cpl_page;
3174 LASSERT(page->cp_type == CPT_CACHEABLE);
3175 if (page->cp_state == CPS_FREEING)
3176 continue;
3177
3178 cl_page_get(page);
3179 lu_ref_add_atomic(&page->cp_reference,
3180 "gang_lookup", current);
3181 pvec[j++] = ops;
3182 }
3183 ++idx;
3184
3185 /*
3186 * Here a delicate locking dance is performed. Current thread
3187 * holds a reference to a page, but has to own it before it
3188 * can be placed into queue. Owning implies waiting, so
3189 * radix-tree lock is to be released. After a wait one has to
3190 * check that pages weren't truncated (cl_page_own() returns
3191 * error in the latter case).
3192 */
3193 spin_unlock(&osc->oo_tree_lock);
3194 tree_lock = false;
3195
3196 for (i = 0; i < j; ++i) {
3197 ops = pvec[i];
3198 if (res == CLP_GANG_OKAY)
3199 res = (*cb)(env, io, ops, cbdata);
3200
3201 page = ops->ops_cl.cpl_page;
3202 lu_ref_del(&page->cp_reference, "gang_lookup", current);
3203 cl_page_put(env, page);
3204 }
3205 if (nr < OTI_PVEC_SIZE || end_of_region)
3206 break;
3207
3208 if (res == CLP_GANG_OKAY && need_resched())
3209 res = CLP_GANG_RESCHED;
3210 if (res != CLP_GANG_OKAY)
3211 break;
3212
3213 spin_lock(&osc->oo_tree_lock);
3214 tree_lock = true;
3215 }
3216 if (tree_lock)
3217 spin_unlock(&osc->oo_tree_lock);
3218 return res;
3219 }
3220
3221 /**
3222 * Check if page @page is covered by an extra lock or discard it.
3223 */
3224 static int check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
3225 struct osc_page *ops, void *cbdata)
3226 {
3227 struct osc_thread_info *info = osc_env_info(env);
3228 struct osc_object *osc = cbdata;
3229 pgoff_t index;
3230
3231 index = osc_index(ops);
3232 if (index >= info->oti_fn_index) {
3233 struct ldlm_lock *tmp;
3234 struct cl_page *page = ops->ops_cl.cpl_page;
3235
3236 /* refresh non-overlapped index */
3237 tmp = osc_dlmlock_at_pgoff(env, osc, index, 0, 0);
3238 if (tmp) {
3239 __u64 end = tmp->l_policy_data.l_extent.end;
3240 /* Cache the first-non-overlapped index so as to skip
3241 * all pages within [index, oti_fn_index). This is safe
3242 * because if tmp lock is canceled, it will discard
3243 * these pages.
3244 */
3245 info->oti_fn_index = cl_index(osc2cl(osc), end + 1);
3246 if (end == OBD_OBJECT_EOF)
3247 info->oti_fn_index = CL_PAGE_EOF;
3248 LDLM_LOCK_PUT(tmp);
3249 } else if (cl_page_own(env, io, page) == 0) {
3250 /* discard the page */
3251 cl_page_discard(env, io, page);
3252 cl_page_disown(env, io, page);
3253 } else {
3254 LASSERT(page->cp_state == CPS_FREEING);
3255 }
3256 }
3257
3258 info->oti_next_index = index + 1;
3259 return CLP_GANG_OKAY;
3260 }
3261
3262 static int discard_cb(const struct lu_env *env, struct cl_io *io,
3263 struct osc_page *ops, void *cbdata)
3264 {
3265 struct osc_thread_info *info = osc_env_info(env);
3266 struct cl_page *page = ops->ops_cl.cpl_page;
3267
3268 /* page is top page. */
3269 info->oti_next_index = osc_index(ops) + 1;
3270 if (cl_page_own(env, io, page) == 0) {
3271 KLASSERT(ergo(page->cp_type == CPT_CACHEABLE,
3272 !PageDirty(cl_page_vmpage(page))));
3273
3274 /* discard the page */
3275 cl_page_discard(env, io, page);
3276 cl_page_disown(env, io, page);
3277 } else {
3278 LASSERT(page->cp_state == CPS_FREEING);
3279 }
3280
3281 return CLP_GANG_OKAY;
3282 }
3283
3284 /**
3285 * Discard pages protected by the given lock. This function traverses radix
3286 * tree to find all covering pages and discard them. If a page is being covered
3287 * by other locks, it should remain in cache.
3288 *
3289 * If error happens on any step, the process continues anyway (the reasoning
3290 * behind this being that lock cancellation cannot be delayed indefinitely).
3291 */
3292 int osc_lock_discard_pages(const struct lu_env *env, struct osc_object *osc,
3293 pgoff_t start, pgoff_t end, enum cl_lock_mode mode)
3294 {
3295 struct osc_thread_info *info = osc_env_info(env);
3296 struct cl_io *io = &info->oti_io;
3297 osc_page_gang_cbt cb;
3298 int res;
3299 int result;
3300
3301 io->ci_obj = cl_object_top(osc2cl(osc));
3302 io->ci_ignore_layout = 1;
3303 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3304 if (result != 0)
3305 goto out;
3306
3307 cb = mode == CLM_READ ? check_and_discard_cb : discard_cb;
3308 info->oti_fn_index = start;
3309 info->oti_next_index = start;
3310 do {
3311 res = osc_page_gang_lookup(env, io, osc,
3312 info->oti_next_index, end, cb, osc);
3313 if (info->oti_next_index > end)
3314 break;
3315
3316 if (res == CLP_GANG_RESCHED)
3317 cond_resched();
3318 } while (res != CLP_GANG_OKAY);
3319 out:
3320 cl_io_fini(env, io);
3321 return result;
3322 }
3323
3324 /** @} osc */
Linux kernel - Deliverables
This page took 0.18555 seconds and 5 git commands to generate.