4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Implementation of cl_page for OSC layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
42 #define DEBUG_SUBSYSTEM S_OSC
44 #include "osc_cl_internal.h"
46 static void osc_lru_del(struct client_obd *cli, struct osc_page *opg);
47 static void osc_lru_use(struct client_obd *cli, struct osc_page *opg);
48 static int osc_lru_reserve(const struct lu_env *env, struct osc_object *obj,
49 struct osc_page *opg);
56 * Comment out osc_page_protected because it may sleep inside the
57 * the client_obd_list_lock.
58 * client_obd_list_lock -> osc_ap_completion -> osc_completion ->
59 * -> osc_page_protected -> osc_page_is_dlocked -> osc_match_base
60 * -> ldlm_lock_match -> sptlrpc_import_check_ctx -> sleep.
63 static int osc_page_is_dlocked(const struct lu_env *env,
64 const struct osc_page *opg,
65 enum cl_lock_mode mode, int pending, int unref)
68 struct osc_object *obj;
69 struct osc_thread_info *info;
70 struct ldlm_res_id *resname;
71 struct lustre_handle *lockh;
72 ldlm_policy_data_t *policy;
78 info = osc_env_info(env);
79 resname = &info->oti_resname;
80 policy = &info->oti_policy;
81 lockh = &info->oti_handle;
82 page = opg->ops_cl.cpl_page;
83 obj = cl2osc(opg->ops_cl.cpl_obj);
85 flags = LDLM_FL_TEST_LOCK | LDLM_FL_BLOCK_GRANTED;
87 flags |= LDLM_FL_CBPENDING;
89 dlmmode = osc_cl_lock2ldlm(mode) | LCK_PW;
90 osc_lock_build_res(env, obj, resname);
91 osc_index2policy(policy, page->cp_obj, page->cp_index, page->cp_index);
92 return osc_match_base(osc_export(obj), resname, LDLM_EXTENT, policy,
93 dlmmode, &flags, NULL, lockh, unref);
97 * Checks an invariant that a page in the cache is covered by a lock, as
100 static int osc_page_protected(const struct lu_env *env,
101 const struct osc_page *opg,
102 enum cl_lock_mode mode, int unref)
104 struct cl_object_header *hdr;
105 struct cl_lock *scan;
106 struct cl_page *page;
107 struct cl_lock_descr *descr;
110 LINVRNT(!opg->ops_temp);
112 page = opg->ops_cl.cpl_page;
113 if (page->cp_owner != NULL &&
114 cl_io_top(page->cp_owner)->ci_lockreq == CILR_NEVER)
116 * If IO is done without locks (liblustre, or lloop), lock is
121 /* otherwise check for a DLM lock */
122 result = osc_page_is_dlocked(env, opg, mode, 1, unref);
124 /* maybe this page is a part of a lockless io? */
125 hdr = cl_object_header(opg->ops_cl.cpl_obj);
126 descr = &osc_env_info(env)->oti_descr;
127 descr->cld_mode = mode;
128 descr->cld_start = page->cp_index;
129 descr->cld_end = page->cp_index;
130 spin_lock(&hdr->coh_lock_guard);
131 cfs_list_for_each_entry(scan, &hdr->coh_locks, cll_linkage) {
133 * Lock-less sub-lock has to be either in HELD state
134 * (when io is actively going on), or in CACHED state,
135 * when top-lock is being unlocked:
136 * cl_io_unlock()->cl_unuse()->...->lov_lock_unuse().
138 if ((scan->cll_state == CLS_HELD ||
139 scan->cll_state == CLS_CACHED) &&
140 cl_lock_ext_match(&scan->cll_descr, descr)) {
141 struct osc_lock *olck;
143 olck = osc_lock_at(scan);
144 result = osc_lock_is_lockless(olck);
148 spin_unlock(&hdr->coh_lock_guard);
153 static int osc_page_protected(const struct lu_env *env,
154 const struct osc_page *opg,
155 enum cl_lock_mode mode, int unref)
161 /*****************************************************************************
166 static void osc_page_fini(const struct lu_env *env,
167 struct cl_page_slice *slice)
169 struct osc_page *opg = cl2osc_page(slice);
170 CDEBUG(D_TRACE, "%p\n", opg);
171 LASSERT(opg->ops_lock == NULL);
174 static void osc_page_transfer_get(struct osc_page *opg, const char *label)
176 struct cl_page *page = cl_page_top(opg->ops_cl.cpl_page);
178 LASSERT(!opg->ops_transfer_pinned);
180 lu_ref_add_atomic(&page->cp_reference, label, page);
181 opg->ops_transfer_pinned = 1;
184 static void osc_page_transfer_put(const struct lu_env *env,
185 struct osc_page *opg)
187 struct cl_page *page = cl_page_top(opg->ops_cl.cpl_page);
189 if (opg->ops_transfer_pinned) {
190 opg->ops_transfer_pinned = 0;
191 lu_ref_del(&page->cp_reference, "transfer", page);
192 cl_page_put(env, page);
197 * This is called once for every page when it is submitted for a transfer
198 * either opportunistic (osc_page_cache_add()), or immediate
199 * (osc_page_submit()).
201 static void osc_page_transfer_add(const struct lu_env *env,
202 struct osc_page *opg, enum cl_req_type crt)
204 struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
206 /* ops_lru and ops_inflight share the same field, so take it from LRU
207 * first and then use it as inflight. */
208 osc_lru_use(osc_cli(obj), opg);
210 spin_lock(&obj->oo_seatbelt);
211 cfs_list_add(&opg->ops_inflight, &obj->oo_inflight[crt]);
212 opg->ops_submitter = current;
213 spin_unlock(&obj->oo_seatbelt);
216 int osc_page_cache_add(const struct lu_env *env,
217 const struct cl_page_slice *slice, struct cl_io *io)
219 struct osc_page *opg = cl2osc_page(slice);
223 LINVRNT(osc_page_protected(env, opg, CLM_WRITE, 0));
225 osc_page_transfer_get(opg, "transfer\0cache");
226 result = osc_queue_async_io(env, io, opg);
228 osc_page_transfer_put(env, opg);
230 osc_page_transfer_add(env, opg, CRT_WRITE);
235 void osc_index2policy(ldlm_policy_data_t *policy, const struct cl_object *obj,
236 pgoff_t start, pgoff_t end)
238 memset(policy, 0, sizeof *policy);
239 policy->l_extent.start = cl_offset(obj, start);
240 policy->l_extent.end = cl_offset(obj, end + 1) - 1;
243 static int osc_page_addref_lock(const struct lu_env *env,
244 struct osc_page *opg,
245 struct cl_lock *lock)
247 struct osc_lock *olock;
250 LASSERT(opg->ops_lock == NULL);
252 olock = osc_lock_at(lock);
253 if (cfs_atomic_inc_return(&olock->ols_pageref) <= 0) {
254 cfs_atomic_dec(&olock->ols_pageref);
258 opg->ops_lock = lock;
264 static void osc_page_putref_lock(const struct lu_env *env,
265 struct osc_page *opg)
267 struct cl_lock *lock = opg->ops_lock;
268 struct osc_lock *olock;
270 LASSERT(lock != NULL);
271 olock = osc_lock_at(lock);
273 cfs_atomic_dec(&olock->ols_pageref);
274 opg->ops_lock = NULL;
276 cl_lock_put(env, lock);
279 static int osc_page_is_under_lock(const struct lu_env *env,
280 const struct cl_page_slice *slice,
281 struct cl_io *unused)
283 struct cl_lock *lock;
284 int result = -ENODATA;
287 lock = cl_lock_at_page(env, slice->cpl_obj, slice->cpl_page,
290 if (osc_page_addref_lock(env, cl2osc_page(slice), lock) == 0)
292 cl_lock_put(env, lock);
297 static void osc_page_disown(const struct lu_env *env,
298 const struct cl_page_slice *slice,
301 struct osc_page *opg = cl2osc_page(slice);
303 if (unlikely(opg->ops_lock))
304 osc_page_putref_lock(env, opg);
307 static void osc_page_completion_read(const struct lu_env *env,
308 const struct cl_page_slice *slice,
311 struct osc_page *opg = cl2osc_page(slice);
313 if (likely(opg->ops_lock))
314 osc_page_putref_lock(env, opg);
317 static void osc_page_completion_write(const struct lu_env *env,
318 const struct cl_page_slice *slice,
323 static const char *osc_list(cfs_list_t *head)
325 return cfs_list_empty(head) ? "-" : "+";
328 static inline cfs_time_t osc_submit_duration(struct osc_page *opg)
330 if (opg->ops_submit_time == 0)
333 return (cfs_time_current() - opg->ops_submit_time);
336 static int osc_page_print(const struct lu_env *env,
337 const struct cl_page_slice *slice,
338 void *cookie, lu_printer_t printer)
340 struct osc_page *opg = cl2osc_page(slice);
341 struct osc_async_page *oap = &opg->ops_oap;
342 struct osc_object *obj = cl2osc(slice->cpl_obj);
343 struct client_obd *cli = &osc_export(obj)->exp_obd->u.cli;
345 return (*printer)(env, cookie, LUSTRE_OSC_NAME"-page@%p: "
346 "1< %#x %d %u %s %s > "
347 "2< "LPU64" %u %u %#x %#x | %p %p %p > "
348 "3< %s %p %d %lu %d > "
349 "4< %d %d %d %lu %s | %s %s %s %s > "
350 "5< %s %s %s %s | %d %s | %d %s %s>\n",
353 oap->oap_magic, oap->oap_cmd,
354 oap->oap_interrupted,
355 osc_list(&oap->oap_pending_item),
356 osc_list(&oap->oap_rpc_item),
358 oap->oap_obj_off, oap->oap_page_off, oap->oap_count,
359 oap->oap_async_flags, oap->oap_brw_flags,
360 oap->oap_request, oap->oap_cli, obj,
362 osc_list(&opg->ops_inflight),
363 opg->ops_submitter, opg->ops_transfer_pinned,
364 osc_submit_duration(opg), opg->ops_srvlock,
366 cli->cl_r_in_flight, cli->cl_w_in_flight,
367 cli->cl_max_rpcs_in_flight,
369 osc_list(&cli->cl_cache_waiters),
370 osc_list(&cli->cl_loi_ready_list),
371 osc_list(&cli->cl_loi_hp_ready_list),
372 osc_list(&cli->cl_loi_write_list),
373 osc_list(&cli->cl_loi_read_list),
375 osc_list(&obj->oo_ready_item),
376 osc_list(&obj->oo_hp_ready_item),
377 osc_list(&obj->oo_write_item),
378 osc_list(&obj->oo_read_item),
379 cfs_atomic_read(&obj->oo_nr_reads),
380 osc_list(&obj->oo_reading_exts),
381 cfs_atomic_read(&obj->oo_nr_writes),
382 osc_list(&obj->oo_hp_exts),
383 osc_list(&obj->oo_urgent_exts));
386 static void osc_page_delete(const struct lu_env *env,
387 const struct cl_page_slice *slice)
389 struct osc_page *opg = cl2osc_page(slice);
390 struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
393 LINVRNT(opg->ops_temp || osc_page_protected(env, opg, CLM_READ, 1));
396 CDEBUG(D_TRACE, "%p\n", opg);
397 osc_page_transfer_put(env, opg);
398 rc = osc_teardown_async_page(env, obj, opg);
400 CL_PAGE_DEBUG(D_ERROR, env, cl_page_top(slice->cpl_page),
401 "Trying to teardown failed: %d\n", rc);
405 spin_lock(&obj->oo_seatbelt);
406 if (opg->ops_submitter != NULL) {
407 LASSERT(!cfs_list_empty(&opg->ops_inflight));
408 cfs_list_del_init(&opg->ops_inflight);
409 opg->ops_submitter = NULL;
411 spin_unlock(&obj->oo_seatbelt);
413 osc_lru_del(osc_cli(obj), opg);
415 if (slice->cpl_page->cp_type == CPT_CACHEABLE) {
418 spin_lock(&obj->oo_tree_lock);
419 value = radix_tree_delete(&obj->oo_tree, osc_index(opg));
422 spin_unlock(&obj->oo_tree_lock);
424 LASSERT(ergo(value != NULL, value == opg));
430 void osc_page_clip(const struct lu_env *env, const struct cl_page_slice *slice,
433 struct osc_page *opg = cl2osc_page(slice);
434 struct osc_async_page *oap = &opg->ops_oap;
436 LINVRNT(osc_page_protected(env, opg, CLM_READ, 0));
438 opg->ops_from = from;
440 spin_lock(&oap->oap_lock);
441 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
442 spin_unlock(&oap->oap_lock);
445 static int osc_page_cancel(const struct lu_env *env,
446 const struct cl_page_slice *slice)
448 struct osc_page *opg = cl2osc_page(slice);
451 LINVRNT(osc_page_protected(env, opg, CLM_READ, 0));
453 /* Check if the transferring against this page
454 * is completed, or not even queued. */
455 if (opg->ops_transfer_pinned)
456 /* FIXME: may not be interrupted.. */
457 rc = osc_cancel_async_page(env, opg);
458 LASSERT(ergo(rc == 0, opg->ops_transfer_pinned == 0));
462 static int osc_page_flush(const struct lu_env *env,
463 const struct cl_page_slice *slice,
466 struct osc_page *opg = cl2osc_page(slice);
469 rc = osc_flush_async_page(env, io, opg);
473 static const struct cl_page_operations osc_page_ops = {
474 .cpo_fini = osc_page_fini,
475 .cpo_print = osc_page_print,
476 .cpo_delete = osc_page_delete,
477 .cpo_is_under_lock = osc_page_is_under_lock,
478 .cpo_disown = osc_page_disown,
481 .cpo_completion = osc_page_completion_read
484 .cpo_completion = osc_page_completion_write
487 .cpo_clip = osc_page_clip,
488 .cpo_cancel = osc_page_cancel,
489 .cpo_flush = osc_page_flush
492 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
493 struct cl_page *page, struct page *vmpage)
495 struct osc_object *osc = cl2osc(obj);
496 struct osc_page *opg = cl_object_page_slice(obj, page);
500 opg->ops_to = PAGE_CACHE_SIZE;
502 result = osc_prep_async_page(osc, opg, vmpage,
503 cl_offset(obj, page->cp_index));
505 struct osc_io *oio = osc_env_io(env);
506 opg->ops_srvlock = osc_io_srvlock(oio);
507 cl_page_slice_add(page, &opg->ops_cl, obj,
511 * Cannot assert osc_page_protected() here as read-ahead
512 * creates temporary pages outside of a lock.
514 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
515 opg->ops_temp = !osc_page_protected(env, opg, CLM_READ, 1);
517 /* ops_inflight and ops_lru are the same field, but it doesn't
518 * hurt to initialize it twice :-) */
519 CFS_INIT_LIST_HEAD(&opg->ops_inflight);
520 CFS_INIT_LIST_HEAD(&opg->ops_lru);
522 /* reserve an LRU space for this page */
523 if (page->cp_type == CPT_CACHEABLE && result == 0) {
524 result = osc_lru_reserve(env, osc, opg);
526 spin_lock(&osc->oo_tree_lock);
527 result = radix_tree_insert(&osc->oo_tree,
528 page->cp_index, opg);
531 spin_unlock(&osc->oo_tree_lock);
532 LASSERT(result == 0);
539 int osc_over_unstable_soft_limit(struct client_obd *cli)
541 long obd_upages, obd_dpages, osc_upages;
543 /* Can't check cli->cl_unstable_count, therefore, no soft limit */
547 obd_upages = cfs_atomic_read(&obd_unstable_pages);
548 obd_dpages = cfs_atomic_read(&obd_dirty_pages);
550 osc_upages = cfs_atomic_read(&cli->cl_unstable_count);
552 /* obd_max_dirty_pages is the max number of (dirty + unstable)
553 * pages allowed at any given time. To simulate an unstable page
554 * only limit, we subtract the current number of dirty pages
555 * from this max. This difference is roughly the amount of pages
556 * currently available for unstable pages. Thus, the soft limit
557 * is half of that difference. Check osc_upages to ensure we don't
558 * set SOFT_SYNC for OSCs without any outstanding unstable pages. */
559 return osc_upages != 0 &&
560 obd_upages >= (obd_max_dirty_pages - obd_dpages) / 2;
564 * Helper function called by osc_io_submit() for every page in an immediate
565 * transfer (i.e., transferred synchronously).
567 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
568 enum cl_req_type crt, int brw_flags)
570 struct osc_async_page *oap = &opg->ops_oap;
571 struct osc_object *obj = oap->oap_obj;
573 LINVRNT(osc_page_protected(env, opg,
574 crt == CRT_WRITE ? CLM_WRITE : CLM_READ, 1));
576 LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, "
577 "magic 0x%x\n", oap, oap->oap_magic);
578 LASSERT(oap->oap_async_flags & ASYNC_READY);
579 LASSERT(oap->oap_async_flags & ASYNC_COUNT_STABLE);
581 oap->oap_cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
582 oap->oap_page_off = opg->ops_from;
583 oap->oap_count = opg->ops_to - opg->ops_from;
584 oap->oap_brw_flags = OBD_BRW_SYNC | brw_flags;
586 if (osc_over_unstable_soft_limit(oap->oap_cli))
587 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
589 if (!client_is_remote(osc_export(obj)) &&
590 cfs_capable(CFS_CAP_SYS_RESOURCE)) {
591 oap->oap_brw_flags |= OBD_BRW_NOQUOTA;
592 oap->oap_cmd |= OBD_BRW_NOQUOTA;
595 opg->ops_submit_time = cfs_time_current();
596 osc_page_transfer_get(opg, "transfer\0imm");
597 osc_page_transfer_add(env, opg, crt);
600 /* --------------- LRU page management ------------------ */
602 /* OSC is a natural place to manage LRU pages as applications are specialized
603 * to write OSC by OSC. Ideally, if one OSC is used more frequently it should
604 * occupy more LRU slots. On the other hand, we should avoid using up all LRU
605 * slots (client_obd::cl_lru_left) otherwise process has to be put into sleep
606 * for free LRU slots - this will be very bad so the algorithm requires each
607 * OSC to free slots voluntarily to maintain a reasonable number of free slots
611 static CFS_DECL_WAITQ(osc_lru_waitq);
612 /* LRU pages are freed in batch mode. OSC should at least free this
613 * number of pages to avoid running out of LRU budget, and.. */
614 static const int lru_shrink_min = 2 << (20 - PAGE_CACHE_SHIFT); /* 2M */
615 /* free this number at most otherwise it will take too long time to finsih. */
616 static const int lru_shrink_max = 8 << (20 - PAGE_CACHE_SHIFT); /* 8M */
618 /* Check if we can free LRU slots from this OSC. If there exists LRU waiters,
619 * we should free slots aggressively. In this way, slots are freed in a steady
620 * step to maintain fairness among OSCs.
622 * Return how many LRU pages should be freed. */
623 static int osc_cache_too_much(struct client_obd *cli)
625 struct cl_client_cache *cache = cli->cl_cache;
626 int pages = cfs_atomic_read(&cli->cl_lru_in_list);
627 unsigned long budget;
629 budget = cache->ccc_lru_max / cfs_atomic_read(&cache->ccc_users);
631 /* if it's going to run out LRU slots, we should free some, but not
632 * too much to maintain faireness among OSCs. */
633 if (cfs_atomic_read(cli->cl_lru_left) < cache->ccc_lru_max >> 4) {
635 return lru_shrink_max;
636 else if (pages >= budget / 2)
637 return lru_shrink_min;
638 } else if (pages >= budget * 2)
639 return lru_shrink_min;
643 int lru_queue_work(const struct lu_env *env, void *data)
645 struct client_obd *cli = data;
647 CDEBUG(D_CACHE, "Run LRU work for client obd %p.\n", cli);
649 if (osc_cache_too_much(cli))
650 osc_lru_shrink(env, cli, lru_shrink_max, true);
655 void osc_lru_add_batch(struct client_obd *cli, cfs_list_t *plist)
658 struct osc_async_page *oap;
661 cfs_list_for_each_entry(oap, plist, oap_pending_item) {
662 struct osc_page *opg = oap2osc_page(oap);
664 if (!opg->ops_in_lru)
668 LASSERT(cfs_list_empty(&opg->ops_lru));
669 cfs_list_add(&opg->ops_lru, &lru);
673 client_obd_list_lock(&cli->cl_lru_list_lock);
674 cfs_list_splice_tail(&lru, &cli->cl_lru_list);
675 cfs_atomic_sub(npages, &cli->cl_lru_busy);
676 cfs_atomic_add(npages, &cli->cl_lru_in_list);
677 client_obd_list_unlock(&cli->cl_lru_list_lock);
679 /* XXX: May set force to be true for better performance */
680 if (osc_cache_too_much(cli))
681 (void)ptlrpcd_queue_work(cli->cl_lru_work);
685 static void __osc_lru_del(struct client_obd *cli, struct osc_page *opg)
687 LASSERT(cfs_atomic_read(&cli->cl_lru_in_list) > 0);
688 cfs_list_del_init(&opg->ops_lru);
689 cfs_atomic_dec(&cli->cl_lru_in_list);
693 * Page is being destroyed. The page may be not in LRU list, if the transfer
694 * has never finished(error occurred).
696 static void osc_lru_del(struct client_obd *cli, struct osc_page *opg)
698 if (opg->ops_in_lru) {
699 client_obd_list_lock(&cli->cl_lru_list_lock);
700 if (!cfs_list_empty(&opg->ops_lru)) {
701 __osc_lru_del(cli, opg);
703 LASSERT(cfs_atomic_read(&cli->cl_lru_busy) > 0);
704 cfs_atomic_dec(&cli->cl_lru_busy);
706 client_obd_list_unlock(&cli->cl_lru_list_lock);
708 cfs_atomic_inc(cli->cl_lru_left);
709 /* this is a great place to release more LRU pages if
710 * this osc occupies too many LRU pages and kernel is
711 * stealing one of them. */
712 if (!memory_pressure_get())
713 (void)ptlrpcd_queue_work(cli->cl_lru_work);
714 wake_up(&osc_lru_waitq);
716 LASSERT(cfs_list_empty(&opg->ops_lru));
721 * Delete page from LRUlist for redirty.
723 static void osc_lru_use(struct client_obd *cli, struct osc_page *opg)
725 /* If page is being transfered for the first time,
726 * ops_lru should be empty */
727 if (opg->ops_in_lru && !cfs_list_empty(&opg->ops_lru)) {
728 client_obd_list_lock(&cli->cl_lru_list_lock);
729 __osc_lru_del(cli, opg);
730 client_obd_list_unlock(&cli->cl_lru_list_lock);
731 cfs_atomic_inc(&cli->cl_lru_busy);
735 static void discard_pagevec(const struct lu_env *env, struct cl_io *io,
736 struct cl_page **pvec, int max_index)
740 for (i = 0; i < max_index; i++) {
741 struct cl_page *page = pvec[i];
743 LASSERT(cl_page_is_owned(page, io));
744 cl_page_discard(env, io, page);
745 cl_page_disown(env, io, page);
746 cl_page_put(env, page);
753 * Drop @target of pages from LRU at most.
755 int osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
756 int target, bool force)
759 struct cl_object *clobj = NULL;
760 struct cl_page **pvec;
761 struct osc_page *opg;
768 LASSERT(cfs_atomic_read(&cli->cl_lru_in_list) >= 0);
769 if (cfs_atomic_read(&cli->cl_lru_in_list) == 0 || target <= 0)
773 if (cfs_atomic_read(&cli->cl_lru_shrinkers) > 0)
776 if (cfs_atomic_inc_return(&cli->cl_lru_shrinkers) > 1) {
777 cfs_atomic_dec(&cli->cl_lru_shrinkers);
781 cfs_atomic_inc(&cli->cl_lru_shrinkers);
784 pvec = (struct cl_page **)osc_env_info(env)->oti_pvec;
785 io = &osc_env_info(env)->oti_io;
787 client_obd_list_lock(&cli->cl_lru_list_lock);
788 maxscan = min(target << 1, cfs_atomic_read(&cli->cl_lru_in_list));
789 while (!cfs_list_empty(&cli->cl_lru_list)) {
790 struct cl_page *page;
791 bool will_free = false;
796 opg = cfs_list_entry(cli->cl_lru_list.next, struct osc_page,
798 page = cl_page_top(opg->ops_cl.cpl_page);
799 if (cl_page_in_use_noref(page)) {
800 cfs_list_move_tail(&opg->ops_lru, &cli->cl_lru_list);
804 LASSERT(page->cp_obj != NULL);
805 if (clobj != page->cp_obj) {
806 struct cl_object *tmp = page->cp_obj;
809 client_obd_list_unlock(&cli->cl_lru_list_lock);
812 discard_pagevec(env, io, pvec, index);
816 cl_object_put(env, clobj);
822 io->ci_ignore_layout = 1;
823 rc = cl_io_init(env, io, CIT_MISC, clobj);
825 client_obd_list_lock(&cli->cl_lru_list_lock);
834 if (cl_page_own_try(env, io, page) == 0) {
835 if (!cl_page_in_use_noref(page)) {
836 /* remove it from lru list earlier to avoid
838 __osc_lru_del(cli, opg);
839 opg->ops_in_lru = 0; /* will be discarded */
844 cl_page_disown(env, io, page);
849 cfs_list_move_tail(&opg->ops_lru, &cli->cl_lru_list);
853 /* Don't discard and free the page with cl_lru_list held */
854 pvec[index++] = page;
855 if (unlikely(index == OTI_PVEC_SIZE)) {
856 client_obd_list_unlock(&cli->cl_lru_list_lock);
857 discard_pagevec(env, io, pvec, index);
860 client_obd_list_lock(&cli->cl_lru_list_lock);
863 if (++count >= target)
866 client_obd_list_unlock(&cli->cl_lru_list_lock);
869 discard_pagevec(env, io, pvec, index);
872 cl_object_put(env, clobj);
875 cfs_atomic_dec(&cli->cl_lru_shrinkers);
877 cfs_atomic_add(count, cli->cl_lru_left);
878 wake_up_all(&osc_lru_waitq);
880 RETURN(count > 0 ? count : rc);
883 static inline int max_to_shrink(struct client_obd *cli)
885 return min(cfs_atomic_read(&cli->cl_lru_in_list) >> 1, lru_shrink_max);
888 int osc_lru_reclaim(struct client_obd *cli)
890 struct cl_env_nest nest;
892 struct cl_client_cache *cache = cli->cl_cache;
897 LASSERT(cache != NULL);
898 LASSERT(!cfs_list_empty(&cache->ccc_lru));
900 env = cl_env_nested_get(&nest);
904 rc = osc_lru_shrink(env, cli, osc_cache_too_much(cli), false);
909 CDEBUG(D_CACHE, "%s: Free %d pages from own LRU: %p.\n",
910 cli->cl_import->imp_obd->obd_name, rc, cli);
914 CDEBUG(D_CACHE, "%s: cli %p no free slots, pages: %d, busy: %d.\n",
915 cli->cl_import->imp_obd->obd_name, cli,
916 cfs_atomic_read(&cli->cl_lru_in_list),
917 cfs_atomic_read(&cli->cl_lru_busy));
919 /* Reclaim LRU slots from other client_obd as it can't free enough
920 * from its own. This should rarely happen. */
921 spin_lock(&cache->ccc_lru_lock);
922 cache->ccc_lru_shrinkers++;
923 cfs_list_move_tail(&cli->cl_lru_osc, &cache->ccc_lru);
925 max_scans = cfs_atomic_read(&cache->ccc_users);
926 while (--max_scans > 0 && !cfs_list_empty(&cache->ccc_lru)) {
927 cli = cfs_list_entry(cache->ccc_lru.next, struct client_obd,
930 CDEBUG(D_CACHE, "%s: cli %p LRU pages: %d, busy: %d.\n",
931 cli->cl_import->imp_obd->obd_name, cli,
932 cfs_atomic_read(&cli->cl_lru_in_list),
933 cfs_atomic_read(&cli->cl_lru_busy));
935 cfs_list_move_tail(&cli->cl_lru_osc, &cache->ccc_lru);
936 if (osc_cache_too_much(cli) > 0) {
937 spin_unlock(&cache->ccc_lru_lock);
939 rc = osc_lru_shrink(env, cli, osc_cache_too_much(cli),
941 spin_lock(&cache->ccc_lru_lock);
946 spin_unlock(&cache->ccc_lru_lock);
949 cl_env_nested_put(&nest, env);
950 CDEBUG(D_CACHE, "%s: cli %p freed %d pages.\n",
951 cli->cl_import->imp_obd->obd_name, cli, rc);
955 static int osc_lru_reserve(const struct lu_env *env, struct osc_object *obj,
956 struct osc_page *opg)
958 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
959 struct osc_io *oio = osc_env_io(env);
960 struct client_obd *cli = osc_cli(obj);
964 if (cli->cl_cache == NULL) /* shall not be in LRU */
967 if (oio->oi_lru_reserved > 0) {
968 --oio->oi_lru_reserved;
972 LASSERT(cfs_atomic_read(cli->cl_lru_left) >= 0);
973 while (!cfs_atomic_add_unless(cli->cl_lru_left, -1, 0)) {
975 /* run out of LRU spaces, try to drop some by itself */
976 rc = osc_lru_reclaim(cli);
983 rc = l_wait_event(osc_lru_waitq,
984 cfs_atomic_read(cli->cl_lru_left) > 0,
992 cfs_atomic_inc(&cli->cl_lru_busy);