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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * Implementation of cl_page for OSC layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_OSC
  43
  44 #include "osc_cl_internal.h"
  45
  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);
  50
  51 /** \addtogroup osc
  52  *  @{
  53  */
  54
  55 /*
  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.
  61  */
  62 #if 0
  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)
  66 {
  67         struct cl_page         *page;
  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;
  73         ldlm_mode_t             dlmmode;
  74         __u64                   flags;
  75
  76         might_sleep();
  77
  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);
  84
  85         flags = LDLM_FL_TEST_LOCK | LDLM_FL_BLOCK_GRANTED;
  86         if (pending)
  87                 flags |= LDLM_FL_CBPENDING;
  88
  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);
  94 }
  95
  96 /**
  97  * Checks an invariant that a page in the cache is covered by a lock, as
  98  * needed.
  99  */
 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)
 103 {
 104         struct cl_object_header *hdr;
 105         struct cl_lock          *scan;
 106         struct cl_page          *page;
 107         struct cl_lock_descr    *descr;
 108         int result;
 109
 110         LINVRNT(!opg->ops_temp);
 111
 112         page = opg->ops_cl.cpl_page;
 113         if (page->cp_owner != NULL &&
 114             cl_io_top(page->cp_owner)->ci_lockreq == CILR_NEVER)
 115                 /*
 116                  * If IO is done without locks (liblustre, or lloop), lock is
 117                  * not required.
 118                  */
 119                 result = 1;
 120         else
 121                 /* otherwise check for a DLM lock */
 122         result = osc_page_is_dlocked(env, opg, mode, 1, unref);
 123         if (result == 0) {
 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) {
 132                         /*
 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().
 137                          */
 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;
 142
 143                                 olck = osc_lock_at(scan);
 144                                 result = osc_lock_is_lockless(olck);
 145                                 break;
 146                         }
 147                 }
 148                 spin_unlock(&hdr->coh_lock_guard);
 149         }
 150         return result;
 151 }
 152 #else
 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)
 156 {
 157         return 1;
 158 }
 159 #endif
 160
 161 /*****************************************************************************
 162  *
 163  * Page operations.
 164  *
 165  */
 166 static void osc_page_fini(const struct lu_env *env,
 167                           struct cl_page_slice *slice)
 168 {
 169         struct osc_page *opg = cl2osc_page(slice);
 170         CDEBUG(D_TRACE, "%p\n", opg);
 171         LASSERT(opg->ops_lock == NULL);
 172 }
 173
 174 static void osc_page_transfer_get(struct osc_page *opg, const char *label)
 175 {
 176         struct cl_page *page = cl_page_top(opg->ops_cl.cpl_page);
 177
 178         LASSERT(!opg->ops_transfer_pinned);
 179         cl_page_get(page);
 180         lu_ref_add_atomic(&page->cp_reference, label, page);
 181         opg->ops_transfer_pinned = 1;
 182 }
 183
 184 static void osc_page_transfer_put(const struct lu_env *env,
 185                                   struct osc_page *opg)
 186 {
 187         struct cl_page *page = cl_page_top(opg->ops_cl.cpl_page);
 188
 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);
 193         }
 194 }
 195
 196 /**
 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()).
 200  */
 201 static void osc_page_transfer_add(const struct lu_env *env,
 202                                   struct osc_page *opg, enum cl_req_type crt)
 203 {
 204         struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
 205
 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);
 209
 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);
 214 }
 215
 216 int osc_page_cache_add(const struct lu_env *env,
 217                         const struct cl_page_slice *slice, struct cl_io *io)
 218 {
 219         struct osc_page *opg = cl2osc_page(slice);
 220         int result;
 221         ENTRY;
 222
 223         LINVRNT(osc_page_protected(env, opg, CLM_WRITE, 0));
 224
 225         osc_page_transfer_get(opg, "transfer\0cache");
 226         result = osc_queue_async_io(env, io, opg);
 227         if (result != 0)
 228                 osc_page_transfer_put(env, opg);
 229         else
 230                 osc_page_transfer_add(env, opg, CRT_WRITE);
 231
 232         RETURN(result);
 233 }
 234
 235 void osc_index2policy(ldlm_policy_data_t *policy, const struct cl_object *obj,
 236                       pgoff_t start, pgoff_t end)
 237 {
 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;
 241 }
 242
 243 static int osc_page_addref_lock(const struct lu_env *env,
 244                                 struct osc_page *opg,
 245                                 struct cl_lock *lock)
 246 {
 247         struct osc_lock *olock;
 248         int              rc;
 249
 250         LASSERT(opg->ops_lock == NULL);
 251
 252         olock = osc_lock_at(lock);
 253         if (cfs_atomic_inc_return(&olock->ols_pageref) <= 0) {
 254                 cfs_atomic_dec(&olock->ols_pageref);
 255                 rc = -ENODATA;
 256         } else {
 257                 cl_lock_get(lock);
 258                 opg->ops_lock = lock;
 259                 rc = 0;
 260         }
 261         return rc;
 262 }
 263
 264 static void osc_page_putref_lock(const struct lu_env *env,
 265                                  struct osc_page *opg)
 266 {
 267         struct cl_lock  *lock = opg->ops_lock;
 268         struct osc_lock *olock;
 269
 270         LASSERT(lock != NULL);
 271         olock = osc_lock_at(lock);
 272
 273         cfs_atomic_dec(&olock->ols_pageref);
 274         opg->ops_lock = NULL;
 275
 276         cl_lock_put(env, lock);
 277 }
 278
 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)
 282 {
 283         struct cl_lock *lock;
 284         int             result = -ENODATA;
 285
 286         ENTRY;
 287         lock = cl_lock_at_page(env, slice->cpl_obj, slice->cpl_page,
 288                                NULL, 1, 0);
 289         if (lock != NULL) {
 290                 if (osc_page_addref_lock(env, cl2osc_page(slice), lock) == 0)
 291                         result = -EBUSY;
 292                 cl_lock_put(env, lock);
 293         }
 294         RETURN(result);
 295 }
 296
 297 static void osc_page_disown(const struct lu_env *env,
 298                             const struct cl_page_slice *slice,
 299                             struct cl_io *io)
 300 {
 301         struct osc_page *opg = cl2osc_page(slice);
 302
 303         if (unlikely(opg->ops_lock))
 304                 osc_page_putref_lock(env, opg);
 305 }
 306
 307 static void osc_page_completion_read(const struct lu_env *env,
 308                                      const struct cl_page_slice *slice,
 309                                      int ioret)
 310 {
 311         struct osc_page   *opg = cl2osc_page(slice);
 312
 313         if (likely(opg->ops_lock))
 314                 osc_page_putref_lock(env, opg);
 315 }
 316
 317 static void osc_page_completion_write(const struct lu_env *env,
 318                                       const struct cl_page_slice *slice,
 319                                       int ioret)
 320 {
 321 }
 322
 323 static const char *osc_list(cfs_list_t *head)
 324 {
 325         return cfs_list_empty(head) ? "-" : "+";
 326 }
 327
 328 static inline cfs_time_t osc_submit_duration(struct osc_page *opg)
 329 {
 330         if (opg->ops_submit_time == 0)
 331                 return 0;
 332
 333         return (cfs_time_current() - opg->ops_submit_time);
 334 }
 335
 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)
 339 {
 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;
 344
 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",
 351                           opg,
 352                           /* 1 */
 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),
 357                           /* 2 */
 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,
 361                           /* 3 */
 362                           osc_list(&opg->ops_inflight),
 363                           opg->ops_submitter, opg->ops_transfer_pinned,
 364                           osc_submit_duration(opg), opg->ops_srvlock,
 365                           /* 4 */
 366                           cli->cl_r_in_flight, cli->cl_w_in_flight,
 367                           cli->cl_max_rpcs_in_flight,
 368                           cli->cl_avail_grant,
 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),
 374                           /* 5 */
 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));
 384 }
 385
 386 static void osc_page_delete(const struct lu_env *env,
 387                             const struct cl_page_slice *slice)
 388 {
 389         struct osc_page   *opg = cl2osc_page(slice);
 390         struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
 391         int rc;
 392
 393         LINVRNT(opg->ops_temp || osc_page_protected(env, opg, CLM_READ, 1));
 394
 395         ENTRY;
 396         CDEBUG(D_TRACE, "%p\n", opg);
 397         osc_page_transfer_put(env, opg);
 398         rc = osc_teardown_async_page(env, obj, opg);
 399         if (rc) {
 400                 CL_PAGE_DEBUG(D_ERROR, env, cl_page_top(slice->cpl_page),
 401                               "Trying to teardown failed: %d\n", rc);
 402                 LASSERT(0);
 403         }
 404
 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;
 410         }
 411         spin_unlock(&obj->oo_seatbelt);
 412
 413         osc_lru_del(osc_cli(obj), opg);
 414
 415         if (slice->cpl_page->cp_type == CPT_CACHEABLE) {
 416                 void *value;
 417
 418                 spin_lock(&obj->oo_tree_lock);
 419                 value = radix_tree_delete(&obj->oo_tree, osc_index(opg));
 420                 if (value != NULL)
 421                         --obj->oo_npages;
 422                 spin_unlock(&obj->oo_tree_lock);
 423
 424                 LASSERT(ergo(value != NULL, value == opg));
 425         }
 426
 427         EXIT;
 428 }
 429
 430 void osc_page_clip(const struct lu_env *env, const struct cl_page_slice *slice,
 431                    int from, int to)
 432 {
 433         struct osc_page       *opg = cl2osc_page(slice);
 434         struct osc_async_page *oap = &opg->ops_oap;
 435
 436         LINVRNT(osc_page_protected(env, opg, CLM_READ, 0));
 437
 438         opg->ops_from = from;
 439         opg->ops_to   = to;
 440         spin_lock(&oap->oap_lock);
 441         oap->oap_async_flags |= ASYNC_COUNT_STABLE;
 442         spin_unlock(&oap->oap_lock);
 443 }
 444
 445 static int osc_page_cancel(const struct lu_env *env,
 446                            const struct cl_page_slice *slice)
 447 {
 448         struct osc_page *opg = cl2osc_page(slice);
 449         int rc = 0;
 450
 451         LINVRNT(osc_page_protected(env, opg, CLM_READ, 0));
 452
 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));
 459         return rc;
 460 }
 461
 462 static int osc_page_flush(const struct lu_env *env,
 463                           const struct cl_page_slice *slice,
 464                           struct cl_io *io)
 465 {
 466         struct osc_page *opg = cl2osc_page(slice);
 467         int rc = 0;
 468         ENTRY;
 469         rc = osc_flush_async_page(env, io, opg);
 470         RETURN(rc);
 471 }
 472
 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,
 479         .io = {
 480                 [CRT_READ] = {
 481                         .cpo_completion = osc_page_completion_read
 482                 },
 483                 [CRT_WRITE] = {
 484                         .cpo_completion = osc_page_completion_write
 485                 }
 486         },
 487         .cpo_clip           = osc_page_clip,
 488         .cpo_cancel         = osc_page_cancel,
 489         .cpo_flush          = osc_page_flush
 490 };
 491
 492 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
 493                   struct cl_page *page, struct page *vmpage)
 494 {
 495         struct osc_object *osc = cl2osc(obj);
 496         struct osc_page   *opg = cl_object_page_slice(obj, page);
 497         int result;
 498
 499         opg->ops_from = 0;
 500         opg->ops_to   = PAGE_CACHE_SIZE;
 501
 502         result = osc_prep_async_page(osc, opg, vmpage,
 503                                         cl_offset(obj, page->cp_index));
 504         if (result == 0) {
 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,
 508                                 &osc_page_ops);
 509         }
 510         /*
 511          * Cannot assert osc_page_protected() here as read-ahead
 512          * creates temporary pages outside of a lock.
 513          */
 514 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
 515         opg->ops_temp = !osc_page_protected(env, opg, CLM_READ, 1);
 516 #endif
 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);
 521
 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);
 525                 if (result == 0) {
 526                         spin_lock(&osc->oo_tree_lock);
 527                         result = radix_tree_insert(&osc->oo_tree,
 528                                                    page->cp_index, opg);
 529                         if (result == 0)
 530                                 ++osc->oo_npages;
 531                         spin_unlock(&osc->oo_tree_lock);
 532                         LASSERT(result == 0);
 533                 }
 534         }
 535
 536         return result;
 537 }
 538
 539 int osc_over_unstable_soft_limit(struct client_obd *cli)
 540 {
 541         long obd_upages, obd_dpages, osc_upages;
 542
 543         /* Can't check cli->cl_unstable_count, therefore, no soft limit */
 544         if (cli == NULL)
 545                 return 0;
 546
 547         obd_upages = cfs_atomic_read(&obd_unstable_pages);
 548         obd_dpages = cfs_atomic_read(&obd_dirty_pages);
 549
 550         osc_upages = cfs_atomic_read(&cli->cl_unstable_count);
 551
 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;
 561 }
 562
 563 /**
 564  * Helper function called by osc_io_submit() for every page in an immediate
 565  * transfer (i.e., transferred synchronously).
 566  */
 567 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
 568                      enum cl_req_type crt, int brw_flags)
 569 {
 570         struct osc_async_page *oap = &opg->ops_oap;
 571         struct osc_object     *obj = oap->oap_obj;
 572
 573         LINVRNT(osc_page_protected(env, opg,
 574                                    crt == CRT_WRITE ? CLM_WRITE : CLM_READ, 1));
 575
 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);
 580
 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;
 585
 586         if (osc_over_unstable_soft_limit(oap->oap_cli))
 587                 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
 588
 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;
 593         }
 594
 595         opg->ops_submit_time = cfs_time_current();
 596         osc_page_transfer_get(opg, "transfer\0imm");
 597         osc_page_transfer_add(env, opg, crt);
 598 }
 599
 600 /* --------------- LRU page management ------------------ */
 601
 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
 608  * at any time.
 609  */
 610
 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 */
 617
 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.
 621  *
 622  * Return how many LRU pages should be freed. */
 623 static int osc_cache_too_much(struct client_obd *cli)
 624 {
 625         struct cl_client_cache *cache = cli->cl_cache;
 626         int pages = cfs_atomic_read(&cli->cl_lru_in_list);
 627         unsigned long budget;
 628
 629         budget = cache->ccc_lru_max / cfs_atomic_read(&cache->ccc_users);
 630
 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) {
 634                 if (pages >= budget)
 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;
 640         return 0;
 641 }
 642
 643 int lru_queue_work(const struct lu_env *env, void *data)
 644 {
 645         struct client_obd *cli = data;
 646
 647         CDEBUG(D_CACHE, "Run LRU work for client obd %p.\n", cli);
 648
 649         if (osc_cache_too_much(cli))
 650                 osc_lru_shrink(env, cli, lru_shrink_max, true);
 651
 652         RETURN(0);
 653 }
 654
 655 void osc_lru_add_batch(struct client_obd *cli, cfs_list_t *plist)
 656 {
 657         CFS_LIST_HEAD(lru);
 658         struct osc_async_page *oap;
 659         int npages = 0;
 660
 661         cfs_list_for_each_entry(oap, plist, oap_pending_item) {
 662                 struct osc_page *opg = oap2osc_page(oap);
 663
 664                 if (!opg->ops_in_lru)
 665                         continue;
 666
 667                 ++npages;
 668                 LASSERT(cfs_list_empty(&opg->ops_lru));
 669                 cfs_list_add(&opg->ops_lru, &lru);
 670         }
 671
 672         if (npages > 0) {
 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);
 678
 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);
 682         }
 683 }
 684
 685 static void __osc_lru_del(struct client_obd *cli, struct osc_page *opg)
 686 {
 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);
 690 }
 691
 692 /**
 693  * Page is being destroyed. The page may be not in LRU list, if the transfer
 694  * has never finished(error occurred).
 695  */
 696 static void osc_lru_del(struct client_obd *cli, struct osc_page *opg)
 697 {
 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);
 702                 } else {
 703                         LASSERT(cfs_atomic_read(&cli->cl_lru_busy) > 0);
 704                         cfs_atomic_dec(&cli->cl_lru_busy);
 705                 }
 706                 client_obd_list_unlock(&cli->cl_lru_list_lock);
 707
 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);
 715         } else {
 716                 LASSERT(cfs_list_empty(&opg->ops_lru));
 717         }
 718 }
 719
 720 /**
 721  * Delete page from LRUlist for redirty.
 722  */
 723 static void osc_lru_use(struct client_obd *cli, struct osc_page *opg)
 724 {
 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);
 732         }
 733 }
 734
 735 static void discard_pagevec(const struct lu_env *env, struct cl_io *io,
 736                                 struct cl_page **pvec, int max_index)
 737 {
 738         int i;
 739
 740         for (i = 0; i < max_index; i++) {
 741                 struct cl_page *page = pvec[i];
 742
 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);
 747
 748                 pvec[i] = NULL;
 749         }
 750 }
 751
 752 /**
 753  * Drop @target of pages from LRU at most.
 754  */
 755 int osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
 756                    int target, bool force)
 757 {
 758         struct cl_io *io;
 759         struct cl_object *clobj = NULL;
 760         struct cl_page **pvec;
 761         struct osc_page *opg;
 762         int maxscan = 0;
 763         int count = 0;
 764         int index = 0;
 765         int rc = 0;
 766         ENTRY;
 767
 768         LASSERT(cfs_atomic_read(&cli->cl_lru_in_list) >= 0);
 769         if (cfs_atomic_read(&cli->cl_lru_in_list) == 0 || target <= 0)
 770                 RETURN(0);
 771
 772         if (!force) {
 773                 if (cfs_atomic_read(&cli->cl_lru_shrinkers) > 0)
 774                         RETURN(-EBUSY);
 775
 776                 if (cfs_atomic_inc_return(&cli->cl_lru_shrinkers) > 1) {
 777                         cfs_atomic_dec(&cli->cl_lru_shrinkers);
 778                         RETURN(-EBUSY);
 779                 }
 780         } else {
 781                 cfs_atomic_inc(&cli->cl_lru_shrinkers);
 782         }
 783
 784         pvec = (struct cl_page **)osc_env_info(env)->oti_pvec;
 785         io = &osc_env_info(env)->oti_io;
 786
 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;
 792
 793                 if (--maxscan < 0)
 794                         break;
 795
 796                 opg = cfs_list_entry(cli->cl_lru_list.next, struct osc_page,
 797                                      ops_lru);
 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);
 801                         continue;
 802                 }
 803
 804                 LASSERT(page->cp_obj != NULL);
 805                 if (clobj != page->cp_obj) {
 806                         struct cl_object *tmp = page->cp_obj;
 807
 808                         cl_object_get(tmp);
 809                         client_obd_list_unlock(&cli->cl_lru_list_lock);
 810
 811                         if (clobj != NULL) {
 812                                 discard_pagevec(env, io, pvec, index);
 813                                 index = 0;
 814
 815                                 cl_io_fini(env, io);
 816                                 cl_object_put(env, clobj);
 817                                 clobj = NULL;
 818                         }
 819
 820                         clobj = tmp;
 821                         io->ci_obj = clobj;
 822                         io->ci_ignore_layout = 1;
 823                         rc = cl_io_init(env, io, CIT_MISC, clobj);
 824
 825                         client_obd_list_lock(&cli->cl_lru_list_lock);
 826
 827                         if (rc != 0)
 828                                 break;
 829
 830                         ++maxscan;
 831                         continue;
 832                 }
 833
 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
 837                                  * lock contention */
 838                                 __osc_lru_del(cli, opg);
 839                                 opg->ops_in_lru = 0; /* will be discarded */
 840
 841                                 cl_page_get(page);
 842                                 will_free = true;
 843                         } else {
 844                                 cl_page_disown(env, io, page);
 845                         }
 846                 }
 847
 848                 if (!will_free) {
 849                         cfs_list_move_tail(&opg->ops_lru, &cli->cl_lru_list);
 850                         continue;
 851                 }
 852
 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);
 858                         index = 0;
 859
 860                         client_obd_list_lock(&cli->cl_lru_list_lock);
 861                 }
 862
 863                 if (++count >= target)
 864                         break;
 865         }
 866         client_obd_list_unlock(&cli->cl_lru_list_lock);
 867
 868         if (clobj != NULL) {
 869                 discard_pagevec(env, io, pvec, index);
 870
 871                 cl_io_fini(env, io);
 872                 cl_object_put(env, clobj);
 873         }
 874
 875         cfs_atomic_dec(&cli->cl_lru_shrinkers);
 876         if (count > 0) {
 877                 cfs_atomic_add(count, cli->cl_lru_left);
 878                 wake_up_all(&osc_lru_waitq);
 879         }
 880         RETURN(count > 0 ? count : rc);
 881 }
 882
 883 static inline int max_to_shrink(struct client_obd *cli)
 884 {
 885         return min(cfs_atomic_read(&cli->cl_lru_in_list) >> 1, lru_shrink_max);
 886 }
 887
 888 int osc_lru_reclaim(struct client_obd *cli)
 889 {
 890         struct cl_env_nest nest;
 891         struct lu_env *env;
 892         struct cl_client_cache *cache = cli->cl_cache;
 893         int max_scans;
 894         int rc = 0;
 895         ENTRY;
 896
 897         LASSERT(cache != NULL);
 898         LASSERT(!cfs_list_empty(&cache->ccc_lru));
 899
 900         env = cl_env_nested_get(&nest);
 901         if (IS_ERR(env))
 902                 RETURN(rc);
 903
 904         rc = osc_lru_shrink(env, cli, osc_cache_too_much(cli), false);
 905         if (rc != 0) {
 906                 if (rc == -EBUSY)
 907                         rc = 0;
 908
 909                 CDEBUG(D_CACHE, "%s: Free %d pages from own LRU: %p.\n",
 910                         cli->cl_import->imp_obd->obd_name, rc, cli);
 911                 GOTO(out, rc);
 912         }
 913
 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));
 918
 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);
 924
 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,
 928                                         cl_lru_osc);
 929
 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));
 934
 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);
 938
 939                         rc = osc_lru_shrink(env, cli, osc_cache_too_much(cli),
 940                                             true);
 941                         spin_lock(&cache->ccc_lru_lock);
 942                         if (rc != 0)
 943                                 break;
 944                 }
 945         }
 946         spin_unlock(&cache->ccc_lru_lock);
 947
 948 out:
 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);
 952         return rc;
 953 }
 954
 955 static int osc_lru_reserve(const struct lu_env *env, struct osc_object *obj,
 956                            struct osc_page *opg)
 957 {
 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);
 961         int rc = 0;
 962         ENTRY;
 963
 964         if (cli->cl_cache == NULL) /* shall not be in LRU */
 965                 RETURN(0);
 966
 967         if (oio->oi_lru_reserved > 0) {
 968                 --oio->oi_lru_reserved;
 969                 goto out;
 970         }
 971
 972         LASSERT(cfs_atomic_read(cli->cl_lru_left) >= 0);
 973         while (!cfs_atomic_add_unless(cli->cl_lru_left, -1, 0)) {
 974
 975                 /* run out of LRU spaces, try to drop some by itself */
 976                 rc = osc_lru_reclaim(cli);
 977                 if (rc < 0)
 978                         break;
 979                 if (rc > 0)
 980                         continue;
 981
 982                 cond_resched();
 983                 rc = l_wait_event(osc_lru_waitq,
 984                                 cfs_atomic_read(cli->cl_lru_left) > 0,
 985                                 &lwi);
 986                 if (rc < 0)
 987                         break;
 988         }
 989
 990 out:
 991         if (rc >= 0) {
 992                 cfs_atomic_inc(&cli->cl_lru_busy);
 993                 opg->ops_in_lru = 1;
 994                 rc = 0;
 995         }
 996
 997         RETURN(rc);
 998 }
 999
1000 /** @} osc */