drivers/staging/lustre/lustre/osc/osc_page.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) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * Implementation of cl_page for OSC layer.
  33  *
  34  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  35  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_OSC
  39
  40 #include "osc_cl_internal.h"
  41
  42 static void osc_lru_del(struct client_obd *cli, struct osc_page *opg);
  43 static void osc_lru_use(struct client_obd *cli, struct osc_page *opg);
  44 static int osc_lru_reserve(const struct lu_env *env, struct osc_object *obj,
  45                            struct osc_page *opg);
  46
  47 /** \addtogroup osc
  48  *  @{
  49  */
  50
  51 /*****************************************************************************
  52  *
  53  * Page operations.
  54  *
  55  */
  56 static void osc_page_transfer_get(struct osc_page *opg, const char *label)
  57 {
  58         struct cl_page *page = opg->ops_cl.cpl_page;
  59
  60         LASSERT(!opg->ops_transfer_pinned);
  61         cl_page_get(page);
  62         lu_ref_add_atomic(&page->cp_reference, label, page);
  63         opg->ops_transfer_pinned = 1;
  64 }
  65
  66 static void osc_page_transfer_put(const struct lu_env *env,
  67                                   struct osc_page *opg)
  68 {
  69         struct cl_page *page = opg->ops_cl.cpl_page;
  70
  71         if (opg->ops_transfer_pinned) {
  72                 opg->ops_transfer_pinned = 0;
  73                 lu_ref_del(&page->cp_reference, "transfer", page);
  74                 cl_page_put(env, page);
  75         }
  76 }
  77
  78 /**
  79  * This is called once for every page when it is submitted for a transfer
  80  * either opportunistic (osc_page_cache_add()), or immediate
  81  * (osc_page_submit()).
  82  */
  83 static void osc_page_transfer_add(const struct lu_env *env,
  84                                   struct osc_page *opg, enum cl_req_type crt)
  85 {
  86         struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
  87
  88         osc_lru_use(osc_cli(obj), opg);
  89
  90         spin_lock(&obj->oo_seatbelt);
  91         list_add(&opg->ops_inflight, &obj->oo_inflight[crt]);
  92         opg->ops_submitter = current;
  93         spin_unlock(&obj->oo_seatbelt);
  94 }
  95
  96 int osc_page_cache_add(const struct lu_env *env,
  97                        const struct cl_page_slice *slice, struct cl_io *io)
  98 {
  99         struct osc_page *opg = cl2osc_page(slice);
 100         int result;
 101
 102         osc_page_transfer_get(opg, "transfer\0cache");
 103         result = osc_queue_async_io(env, io, opg);
 104         if (result != 0)
 105                 osc_page_transfer_put(env, opg);
 106         else
 107                 osc_page_transfer_add(env, opg, CRT_WRITE);
 108
 109         return result;
 110 }
 111
 112 void osc_index2policy(ldlm_policy_data_t *policy, const struct cl_object *obj,
 113                       pgoff_t start, pgoff_t end)
 114 {
 115         memset(policy, 0, sizeof(*policy));
 116         policy->l_extent.start = cl_offset(obj, start);
 117         policy->l_extent.end = cl_offset(obj, end + 1) - 1;
 118 }
 119
 120 static int osc_page_is_under_lock(const struct lu_env *env,
 121                                   const struct cl_page_slice *slice,
 122                                   struct cl_io *unused, pgoff_t *max_index)
 123 {
 124         struct osc_page *opg = cl2osc_page(slice);
 125         struct ldlm_lock *dlmlock;
 126         int result = -ENODATA;
 127
 128         dlmlock = osc_dlmlock_at_pgoff(env, cl2osc(slice->cpl_obj),
 129                                        osc_index(opg), 1, 0);
 130         if (dlmlock) {
 131                 *max_index = cl_index(slice->cpl_obj,
 132                                       dlmlock->l_policy_data.l_extent.end);
 133                 LDLM_LOCK_PUT(dlmlock);
 134                 result = 0;
 135         }
 136         return result;
 137 }
 138
 139 static const char *osc_list(struct list_head *head)
 140 {
 141         return list_empty(head) ? "-" : "+";
 142 }
 143
 144 static inline unsigned long osc_submit_duration(struct osc_page *opg)
 145 {
 146         if (opg->ops_submit_time == 0)
 147                 return 0;
 148
 149         return (cfs_time_current() - opg->ops_submit_time);
 150 }
 151
 152 static int osc_page_print(const struct lu_env *env,
 153                           const struct cl_page_slice *slice,
 154                           void *cookie, lu_printer_t printer)
 155 {
 156         struct osc_page *opg = cl2osc_page(slice);
 157         struct osc_async_page *oap = &opg->ops_oap;
 158         struct osc_object *obj = cl2osc(slice->cpl_obj);
 159         struct client_obd *cli = &osc_export(obj)->exp_obd->u.cli;
 160
 161         return (*printer)(env, cookie, LUSTRE_OSC_NAME "-page@%p %lu: 1< %#x %d %u %s %s > 2< %llu %u %u %#x %#x | %p %p %p > 3< %s %p %d %lu %d > 4< %d %d %d %lu %s | %s %s %s %s > 5< %s %s %s %s | %d %s | %d %s %s>\n",
 162                           opg, osc_index(opg),
 163                           /* 1 */
 164                           oap->oap_magic, oap->oap_cmd,
 165                           oap->oap_interrupted,
 166                           osc_list(&oap->oap_pending_item),
 167                           osc_list(&oap->oap_rpc_item),
 168                           /* 2 */
 169                           oap->oap_obj_off, oap->oap_page_off, oap->oap_count,
 170                           oap->oap_async_flags, oap->oap_brw_flags,
 171                           oap->oap_request, oap->oap_cli, obj,
 172                           /* 3 */
 173                           osc_list(&opg->ops_inflight),
 174                           opg->ops_submitter, opg->ops_transfer_pinned,
 175                           osc_submit_duration(opg), opg->ops_srvlock,
 176                           /* 4 */
 177                           cli->cl_r_in_flight, cli->cl_w_in_flight,
 178                           cli->cl_max_rpcs_in_flight,
 179                           cli->cl_avail_grant,
 180                           osc_list(&cli->cl_cache_waiters),
 181                           osc_list(&cli->cl_loi_ready_list),
 182                           osc_list(&cli->cl_loi_hp_ready_list),
 183                           osc_list(&cli->cl_loi_write_list),
 184                           osc_list(&cli->cl_loi_read_list),
 185                           /* 5 */
 186                           osc_list(&obj->oo_ready_item),
 187                           osc_list(&obj->oo_hp_ready_item),
 188                           osc_list(&obj->oo_write_item),
 189                           osc_list(&obj->oo_read_item),
 190                           atomic_read(&obj->oo_nr_reads),
 191                           osc_list(&obj->oo_reading_exts),
 192                           atomic_read(&obj->oo_nr_writes),
 193                           osc_list(&obj->oo_hp_exts),
 194                           osc_list(&obj->oo_urgent_exts));
 195 }
 196
 197 static void osc_page_delete(const struct lu_env *env,
 198                             const struct cl_page_slice *slice)
 199 {
 200         struct osc_page *opg = cl2osc_page(slice);
 201         struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
 202         int rc;
 203
 204         CDEBUG(D_TRACE, "%p\n", opg);
 205         osc_page_transfer_put(env, opg);
 206         rc = osc_teardown_async_page(env, obj, opg);
 207         if (rc) {
 208                 CL_PAGE_DEBUG(D_ERROR, env, slice->cpl_page,
 209                               "Trying to teardown failed: %d\n", rc);
 210                 LASSERT(0);
 211         }
 212
 213         spin_lock(&obj->oo_seatbelt);
 214         if (opg->ops_submitter) {
 215                 LASSERT(!list_empty(&opg->ops_inflight));
 216                 list_del_init(&opg->ops_inflight);
 217                 opg->ops_submitter = NULL;
 218         }
 219         spin_unlock(&obj->oo_seatbelt);
 220
 221         osc_lru_del(osc_cli(obj), opg);
 222
 223         if (slice->cpl_page->cp_type == CPT_CACHEABLE) {
 224                 void *value;
 225
 226                 spin_lock(&obj->oo_tree_lock);
 227                 value = radix_tree_delete(&obj->oo_tree, osc_index(opg));
 228                 if (value)
 229                         --obj->oo_npages;
 230                 spin_unlock(&obj->oo_tree_lock);
 231
 232                 LASSERT(ergo(value, value == opg));
 233         }
 234 }
 235
 236 static void osc_page_clip(const struct lu_env *env,
 237                           const struct cl_page_slice *slice, int from, int to)
 238 {
 239         struct osc_page *opg = cl2osc_page(slice);
 240         struct osc_async_page *oap = &opg->ops_oap;
 241
 242         opg->ops_from = from;
 243         opg->ops_to = to;
 244         spin_lock(&oap->oap_lock);
 245         oap->oap_async_flags |= ASYNC_COUNT_STABLE;
 246         spin_unlock(&oap->oap_lock);
 247 }
 248
 249 static int osc_page_cancel(const struct lu_env *env,
 250                            const struct cl_page_slice *slice)
 251 {
 252         struct osc_page *opg = cl2osc_page(slice);
 253         int rc = 0;
 254
 255         /* Check if the transferring against this page
 256          * is completed, or not even queued.
 257          */
 258         if (opg->ops_transfer_pinned)
 259                 /* FIXME: may not be interrupted.. */
 260                 rc = osc_cancel_async_page(env, opg);
 261         LASSERT(ergo(rc == 0, opg->ops_transfer_pinned == 0));
 262         return rc;
 263 }
 264
 265 static int osc_page_flush(const struct lu_env *env,
 266                           const struct cl_page_slice *slice,
 267                           struct cl_io *io)
 268 {
 269         struct osc_page *opg = cl2osc_page(slice);
 270         int rc;
 271
 272         rc = osc_flush_async_page(env, io, opg);
 273         return rc;
 274 }
 275
 276 static const struct cl_page_operations osc_page_ops = {
 277         .cpo_print       = osc_page_print,
 278         .cpo_delete     = osc_page_delete,
 279         .cpo_is_under_lock = osc_page_is_under_lock,
 280         .cpo_clip          = osc_page_clip,
 281         .cpo_cancel      = osc_page_cancel,
 282         .cpo_flush        = osc_page_flush
 283 };
 284
 285 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
 286                   struct cl_page *page, pgoff_t index)
 287 {
 288         struct osc_object *osc = cl2osc(obj);
 289         struct osc_page *opg = cl_object_page_slice(obj, page);
 290         int result;
 291
 292         opg->ops_from = 0;
 293         opg->ops_to = PAGE_SIZE;
 294
 295         result = osc_prep_async_page(osc, opg, page->cp_vmpage,
 296                                      cl_offset(obj, index));
 297         if (result == 0) {
 298                 struct osc_io *oio = osc_env_io(env);
 299
 300                 opg->ops_srvlock = osc_io_srvlock(oio);
 301                 cl_page_slice_add(page, &opg->ops_cl, obj, index,
 302                                   &osc_page_ops);
 303         }
 304         /* ops_inflight and ops_lru are the same field, but it doesn't
 305          * hurt to initialize it twice :-)
 306          */
 307         INIT_LIST_HEAD(&opg->ops_inflight);
 308         INIT_LIST_HEAD(&opg->ops_lru);
 309
 310         /* reserve an LRU space for this page */
 311         if (page->cp_type == CPT_CACHEABLE && result == 0) {
 312                 result = osc_lru_reserve(env, osc, opg);
 313                 if (result == 0) {
 314                         spin_lock(&osc->oo_tree_lock);
 315                         result = radix_tree_insert(&osc->oo_tree, index, opg);
 316                         if (result == 0)
 317                                 ++osc->oo_npages;
 318                         spin_unlock(&osc->oo_tree_lock);
 319                         LASSERT(result == 0);
 320                 }
 321         }
 322
 323         return result;
 324 }
 325
 326 int osc_over_unstable_soft_limit(struct client_obd *cli)
 327 {
 328         long obd_upages, obd_dpages, osc_upages;
 329
 330         /* Can't check cli->cl_unstable_count, therefore, no soft limit */
 331         if (!cli)
 332                 return 0;
 333
 334         obd_upages = atomic_read(&obd_unstable_pages);
 335         obd_dpages = atomic_read(&obd_dirty_pages);
 336
 337         osc_upages = atomic_read(&cli->cl_unstable_count);
 338
 339         /*
 340          * obd_max_dirty_pages is the max number of (dirty + unstable)
 341          * pages allowed at any given time. To simulate an unstable page
 342          * only limit, we subtract the current number of dirty pages
 343          * from this max. This difference is roughly the amount of pages
 344          * currently available for unstable pages. Thus, the soft limit
 345          * is half of that difference. Check osc_upages to ensure we don't
 346          * set SOFT_SYNC for OSCs without any outstanding unstable pages.
 347          */
 348         return osc_upages &&
 349                obd_upages >= (obd_max_dirty_pages - obd_dpages) / 2;
 350 }
 351
 352 /**
 353  * Helper function called by osc_io_submit() for every page in an immediate
 354  * transfer (i.e., transferred synchronously).
 355  */
 356 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
 357                      enum cl_req_type crt, int brw_flags)
 358 {
 359         struct osc_async_page *oap = &opg->ops_oap;
 360
 361         LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, magic 0x%x\n",
 362                  oap, oap->oap_magic);
 363         LASSERT(oap->oap_async_flags & ASYNC_READY);
 364         LASSERT(oap->oap_async_flags & ASYNC_COUNT_STABLE);
 365
 366         oap->oap_cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
 367         oap->oap_page_off = opg->ops_from;
 368         oap->oap_count = opg->ops_to - opg->ops_from;
 369         oap->oap_brw_flags = brw_flags | OBD_BRW_SYNC;
 370
 371         if (osc_over_unstable_soft_limit(oap->oap_cli))
 372                 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
 373
 374         if (capable(CFS_CAP_SYS_RESOURCE)) {
 375                 oap->oap_brw_flags |= OBD_BRW_NOQUOTA;
 376                 oap->oap_cmd |= OBD_BRW_NOQUOTA;
 377         }
 378
 379         opg->ops_submit_time = cfs_time_current();
 380         osc_page_transfer_get(opg, "transfer\0imm");
 381         osc_page_transfer_add(env, opg, crt);
 382 }
 383
 384 /* --------------- LRU page management ------------------ */
 385
 386 /* OSC is a natural place to manage LRU pages as applications are specialized
 387  * to write OSC by OSC. Ideally, if one OSC is used more frequently it should
 388  * occupy more LRU slots. On the other hand, we should avoid using up all LRU
 389  * slots (client_obd::cl_lru_left) otherwise process has to be put into sleep
 390  * for free LRU slots - this will be very bad so the algorithm requires each
 391  * OSC to free slots voluntarily to maintain a reasonable number of free slots
 392  * at any time.
 393  */
 394
 395 static DECLARE_WAIT_QUEUE_HEAD(osc_lru_waitq);
 396 /* LRU pages are freed in batch mode. OSC should at least free this
 397  * number of pages to avoid running out of LRU budget, and..
 398  */
 399 static const int lru_shrink_min = 2 << (20 - PAGE_SHIFT);  /* 2M */
 400 /* free this number at most otherwise it will take too long time to finish. */
 401 static const int lru_shrink_max = 8 << (20 - PAGE_SHIFT); /* 8M */
 402
 403 /* Check if we can free LRU slots from this OSC. If there exists LRU waiters,
 404  * we should free slots aggressively. In this way, slots are freed in a steady
 405  * step to maintain fairness among OSCs.
 406  *
 407  * Return how many LRU pages should be freed.
 408  */
 409 static int osc_cache_too_much(struct client_obd *cli)
 410 {
 411         struct cl_client_cache *cache = cli->cl_cache;
 412         int pages = atomic_read(&cli->cl_lru_in_list);
 413         unsigned long budget;
 414
 415         budget = cache->ccc_lru_max / (atomic_read(&cache->ccc_users) - 2);
 416
 417         /* if it's going to run out LRU slots, we should free some, but not
 418          * too much to maintain fairness among OSCs.
 419          */
 420         if (atomic_read(cli->cl_lru_left) < cache->ccc_lru_max >> 4) {
 421                 if (pages >= budget)
 422                         return lru_shrink_max;
 423                 else if (pages >= budget / 2)
 424                         return lru_shrink_min;
 425         } else if (pages >= budget * 2) {
 426                 return lru_shrink_min;
 427         }
 428         return 0;
 429 }
 430
 431 int lru_queue_work(const struct lu_env *env, void *data)
 432 {
 433         struct client_obd *cli = data;
 434
 435         CDEBUG(D_CACHE, "Run LRU work for client obd %p.\n", cli);
 436
 437         if (osc_cache_too_much(cli))
 438                 osc_lru_shrink(env, cli, lru_shrink_max, true);
 439
 440         return 0;
 441 }
 442
 443 void osc_lru_add_batch(struct client_obd *cli, struct list_head *plist)
 444 {
 445         LIST_HEAD(lru);
 446         struct osc_async_page *oap;
 447         int npages = 0;
 448
 449         list_for_each_entry(oap, plist, oap_pending_item) {
 450                 struct osc_page *opg = oap2osc_page(oap);
 451
 452                 if (!opg->ops_in_lru)
 453                         continue;
 454
 455                 ++npages;
 456                 LASSERT(list_empty(&opg->ops_lru));
 457                 list_add(&opg->ops_lru, &lru);
 458         }
 459
 460         if (npages > 0) {
 461                 spin_lock(&cli->cl_lru_list_lock);
 462                 list_splice_tail(&lru, &cli->cl_lru_list);
 463                 atomic_sub(npages, &cli->cl_lru_busy);
 464                 atomic_add(npages, &cli->cl_lru_in_list);
 465                 spin_unlock(&cli->cl_lru_list_lock);
 466
 467                 /* XXX: May set force to be true for better performance */
 468                 if (osc_cache_too_much(cli))
 469                         (void)ptlrpcd_queue_work(cli->cl_lru_work);
 470         }
 471 }
 472
 473 static void __osc_lru_del(struct client_obd *cli, struct osc_page *opg)
 474 {
 475         LASSERT(atomic_read(&cli->cl_lru_in_list) > 0);
 476         list_del_init(&opg->ops_lru);
 477         atomic_dec(&cli->cl_lru_in_list);
 478 }
 479
 480 /**
 481  * Page is being destroyed. The page may be not in LRU list, if the transfer
 482  * has never finished(error occurred).
 483  */
 484 static void osc_lru_del(struct client_obd *cli, struct osc_page *opg)
 485 {
 486         if (opg->ops_in_lru) {
 487                 spin_lock(&cli->cl_lru_list_lock);
 488                 if (!list_empty(&opg->ops_lru)) {
 489                         __osc_lru_del(cli, opg);
 490                 } else {
 491                         LASSERT(atomic_read(&cli->cl_lru_busy) > 0);
 492                         atomic_dec(&cli->cl_lru_busy);
 493                 }
 494                 spin_unlock(&cli->cl_lru_list_lock);
 495
 496                 atomic_inc(cli->cl_lru_left);
 497                 /* this is a great place to release more LRU pages if
 498                  * this osc occupies too many LRU pages and kernel is
 499                  * stealing one of them.
 500                  */
 501                 if (!memory_pressure_get())
 502                         (void)ptlrpcd_queue_work(cli->cl_lru_work);
 503                 wake_up(&osc_lru_waitq);
 504         } else {
 505                 LASSERT(list_empty(&opg->ops_lru));
 506         }
 507 }
 508
 509 /**
 510  * Delete page from LRUlist for redirty.
 511  */
 512 static void osc_lru_use(struct client_obd *cli, struct osc_page *opg)
 513 {
 514         /* If page is being transferred for the first time,
 515          * ops_lru should be empty
 516          */
 517         if (opg->ops_in_lru && !list_empty(&opg->ops_lru)) {
 518                 spin_lock(&cli->cl_lru_list_lock);
 519                 __osc_lru_del(cli, opg);
 520                 spin_unlock(&cli->cl_lru_list_lock);
 521                 atomic_inc(&cli->cl_lru_busy);
 522         }
 523 }
 524
 525 static void discard_pagevec(const struct lu_env *env, struct cl_io *io,
 526                             struct cl_page **pvec, int max_index)
 527 {
 528         int i;
 529
 530         for (i = 0; i < max_index; i++) {
 531                 struct cl_page *page = pvec[i];
 532
 533                 LASSERT(cl_page_is_owned(page, io));
 534                 cl_page_discard(env, io, page);
 535                 cl_page_disown(env, io, page);
 536                 cl_page_put(env, page);
 537
 538                 pvec[i] = NULL;
 539         }
 540 }
 541
 542 /**
 543  * Drop @target of pages from LRU at most.
 544  */
 545 int osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
 546                    int target, bool force)
 547 {
 548         struct cl_io *io;
 549         struct cl_object *clobj = NULL;
 550         struct cl_page **pvec;
 551         struct osc_page *opg;
 552         struct osc_page *temp;
 553         int maxscan = 0;
 554         int count = 0;
 555         int index = 0;
 556         int rc = 0;
 557
 558         LASSERT(atomic_read(&cli->cl_lru_in_list) >= 0);
 559         if (atomic_read(&cli->cl_lru_in_list) == 0 || target <= 0)
 560                 return 0;
 561
 562         if (!force) {
 563                 if (atomic_read(&cli->cl_lru_shrinkers) > 0)
 564                         return -EBUSY;
 565
 566                 if (atomic_inc_return(&cli->cl_lru_shrinkers) > 1) {
 567                         atomic_dec(&cli->cl_lru_shrinkers);
 568                         return -EBUSY;
 569                 }
 570         } else {
 571                 atomic_inc(&cli->cl_lru_shrinkers);
 572         }
 573
 574         pvec = (struct cl_page **)osc_env_info(env)->oti_pvec;
 575         io = &osc_env_info(env)->oti_io;
 576
 577         spin_lock(&cli->cl_lru_list_lock);
 578         maxscan = min(target << 1, atomic_read(&cli->cl_lru_in_list));
 579         list_for_each_entry_safe(opg, temp, &cli->cl_lru_list, ops_lru) {
 580                 struct cl_page *page;
 581                 bool will_free = false;
 582
 583                 if (--maxscan < 0)
 584                         break;
 585
 586                 page = opg->ops_cl.cpl_page;
 587                 if (cl_page_in_use_noref(page)) {
 588                         list_move_tail(&opg->ops_lru, &cli->cl_lru_list);
 589                         continue;
 590                 }
 591
 592                 LASSERT(page->cp_obj);
 593                 if (clobj != page->cp_obj) {
 594                         struct cl_object *tmp = page->cp_obj;
 595
 596                         cl_object_get(tmp);
 597                         spin_unlock(&cli->cl_lru_list_lock);
 598
 599                         if (clobj) {
 600                                 discard_pagevec(env, io, pvec, index);
 601                                 index = 0;
 602
 603                                 cl_io_fini(env, io);
 604                                 cl_object_put(env, clobj);
 605                                 clobj = NULL;
 606                         }
 607
 608                         clobj = tmp;
 609                         io->ci_obj = clobj;
 610                         io->ci_ignore_layout = 1;
 611                         rc = cl_io_init(env, io, CIT_MISC, clobj);
 612
 613                         spin_lock(&cli->cl_lru_list_lock);
 614
 615                         if (rc != 0)
 616                                 break;
 617
 618                         ++maxscan;
 619                         continue;
 620                 }
 621
 622                 if (cl_page_own_try(env, io, page) == 0) {
 623                         if (!cl_page_in_use_noref(page)) {
 624                                 /* remove it from lru list earlier to avoid
 625                                  * lock contention
 626                                  */
 627                                 __osc_lru_del(cli, opg);
 628                                 opg->ops_in_lru = 0; /* will be discarded */
 629
 630                                 cl_page_get(page);
 631                                 will_free = true;
 632                         } else {
 633                                 cl_page_disown(env, io, page);
 634                         }
 635                 }
 636
 637                 if (!will_free) {
 638                         list_move_tail(&opg->ops_lru, &cli->cl_lru_list);
 639                         continue;
 640                 }
 641
 642                 /* Don't discard and free the page with cl_lru_list held */
 643                 pvec[index++] = page;
 644                 if (unlikely(index == OTI_PVEC_SIZE)) {
 645                         spin_unlock(&cli->cl_lru_list_lock);
 646                         discard_pagevec(env, io, pvec, index);
 647                         index = 0;
 648
 649                         spin_lock(&cli->cl_lru_list_lock);
 650                 }
 651
 652                 if (++count >= target)
 653                         break;
 654         }
 655         spin_unlock(&cli->cl_lru_list_lock);
 656
 657         if (clobj) {
 658                 discard_pagevec(env, io, pvec, index);
 659
 660                 cl_io_fini(env, io);
 661                 cl_object_put(env, clobj);
 662         }
 663
 664         atomic_dec(&cli->cl_lru_shrinkers);
 665         if (count > 0) {
 666                 atomic_add(count, cli->cl_lru_left);
 667                 wake_up_all(&osc_lru_waitq);
 668         }
 669         return count > 0 ? count : rc;
 670 }
 671
 672 static inline int max_to_shrink(struct client_obd *cli)
 673 {
 674         return min(atomic_read(&cli->cl_lru_in_list) >> 1, lru_shrink_max);
 675 }
 676
 677 int osc_lru_reclaim(struct client_obd *cli)
 678 {
 679         struct cl_env_nest nest;
 680         struct lu_env *env;
 681         struct cl_client_cache *cache = cli->cl_cache;
 682         int max_scans;
 683         int rc = 0;
 684
 685         LASSERT(cache);
 686
 687         env = cl_env_nested_get(&nest);
 688         if (IS_ERR(env))
 689                 return 0;
 690
 691         rc = osc_lru_shrink(env, cli, osc_cache_too_much(cli), false);
 692         if (rc != 0) {
 693                 if (rc == -EBUSY)
 694                         rc = 0;
 695
 696                 CDEBUG(D_CACHE, "%s: Free %d pages from own LRU: %p.\n",
 697                        cli->cl_import->imp_obd->obd_name, rc, cli);
 698                 goto out;
 699         }
 700
 701         CDEBUG(D_CACHE, "%s: cli %p no free slots, pages: %d, busy: %d.\n",
 702                cli->cl_import->imp_obd->obd_name, cli,
 703                atomic_read(&cli->cl_lru_in_list),
 704                atomic_read(&cli->cl_lru_busy));
 705
 706         /* Reclaim LRU slots from other client_obd as it can't free enough
 707          * from its own. This should rarely happen.
 708          */
 709         spin_lock(&cache->ccc_lru_lock);
 710         LASSERT(!list_empty(&cache->ccc_lru));
 711
 712         cache->ccc_lru_shrinkers++;
 713         list_move_tail(&cli->cl_lru_osc, &cache->ccc_lru);
 714
 715         max_scans = atomic_read(&cache->ccc_users) - 2;
 716         while (--max_scans > 0 && !list_empty(&cache->ccc_lru)) {
 717                 cli = list_entry(cache->ccc_lru.next, struct client_obd,
 718                                  cl_lru_osc);
 719
 720                 CDEBUG(D_CACHE, "%s: cli %p LRU pages: %d, busy: %d.\n",
 721                        cli->cl_import->imp_obd->obd_name, cli,
 722                        atomic_read(&cli->cl_lru_in_list),
 723                        atomic_read(&cli->cl_lru_busy));
 724
 725                 list_move_tail(&cli->cl_lru_osc, &cache->ccc_lru);
 726                 if (osc_cache_too_much(cli) > 0) {
 727                         spin_unlock(&cache->ccc_lru_lock);
 728
 729                         rc = osc_lru_shrink(env, cli, osc_cache_too_much(cli),
 730                                             true);
 731                         spin_lock(&cache->ccc_lru_lock);
 732                         if (rc != 0)
 733                                 break;
 734                 }
 735         }
 736         spin_unlock(&cache->ccc_lru_lock);
 737
 738 out:
 739         cl_env_nested_put(&nest, env);
 740         CDEBUG(D_CACHE, "%s: cli %p freed %d pages.\n",
 741                cli->cl_import->imp_obd->obd_name, cli, rc);
 742         return rc;
 743 }
 744
 745 static int osc_lru_reserve(const struct lu_env *env, struct osc_object *obj,
 746                            struct osc_page *opg)
 747 {
 748         struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
 749         struct osc_io *oio = osc_env_io(env);
 750         struct client_obd *cli = osc_cli(obj);
 751         int rc = 0;
 752
 753         if (!cli->cl_cache) /* shall not be in LRU */
 754                 return 0;
 755
 756         if (oio->oi_lru_reserved > 0) {
 757                 --oio->oi_lru_reserved;
 758                 goto out;
 759         }
 760
 761         LASSERT(atomic_read(cli->cl_lru_left) >= 0);
 762         while (!atomic_add_unless(cli->cl_lru_left, -1, 0)) {
 763                 /* run out of LRU spaces, try to drop some by itself */
 764                 rc = osc_lru_reclaim(cli);
 765                 if (rc < 0)
 766                         break;
 767                 if (rc > 0)
 768                         continue;
 769
 770                 cond_resched();
 771
 772                 rc = l_wait_event(osc_lru_waitq,
 773                                   atomic_read(cli->cl_lru_left) > 0,
 774                                   &lwi);
 775
 776                 if (rc < 0)
 777                         break;
 778         }
 779
 780 out:
 781         if (rc >= 0) {
 782                 atomic_inc(&cli->cl_lru_busy);
 783                 opg->ops_in_lru = 1;
 784                 rc = 0;
 785         }
 786
 787         return rc;
 788 }
 789
 790 /** @} osc */