drivers/staging/lustre/lustre/fld/fld_cache.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2013, 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  * lustre/fld/fld_cache.c
  33  *
  34  * FLD (Fids Location Database)
  35  *
  36  * Author: Pravin Shelar <pravin.shelar@sun.com>
  37  * Author: Yury Umanets <umka@clusterfs.com>
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_FLD
  41
  42 #include "../../include/linux/libcfs/libcfs.h"
  43 #include <linux/module.h>
  44 #include <asm/div64.h>
  45
  46 #include "../include/obd.h"
  47 #include "../include/obd_class.h"
  48 #include "../include/lustre_ver.h"
  49 #include "../include/obd_support.h"
  50 #include "../include/lprocfs_status.h"
  51
  52 #include "../include/lustre_req_layout.h"
  53 #include "../include/lustre_fld.h"
  54 #include "fld_internal.h"
  55
  56 /**
  57  * create fld cache.
  58  */
  59 struct fld_cache *fld_cache_init(const char *name,
  60                                  int cache_size, int cache_threshold)
  61 {
  62         struct fld_cache *cache;
  63
  64         LASSERT(name);
  65         LASSERT(cache_threshold < cache_size);
  66
  67         cache = kzalloc(sizeof(*cache), GFP_NOFS);
  68         if (!cache)
  69                 return ERR_PTR(-ENOMEM);
  70
  71         INIT_LIST_HEAD(&cache->fci_entries_head);
  72         INIT_LIST_HEAD(&cache->fci_lru);
  73
  74         cache->fci_cache_count = 0;
  75         rwlock_init(&cache->fci_lock);
  76
  77         strlcpy(cache->fci_name, name,
  78                 sizeof(cache->fci_name));
  79
  80         cache->fci_cache_size = cache_size;
  81         cache->fci_threshold = cache_threshold;
  82
  83         /* Init fld cache info. */
  84         memset(&cache->fci_stat, 0, sizeof(cache->fci_stat));
  85
  86         CDEBUG(D_INFO, "%s: FLD cache - Size: %d, Threshold: %d\n",
  87                cache->fci_name, cache_size, cache_threshold);
  88
  89         return cache;
  90 }
  91
  92 /**
  93  * destroy fld cache.
  94  */
  95 void fld_cache_fini(struct fld_cache *cache)
  96 {
  97         __u64 pct;
  98
  99         LASSERT(cache);
 100         fld_cache_flush(cache);
 101
 102         if (cache->fci_stat.fst_count > 0) {
 103                 pct = cache->fci_stat.fst_cache * 100;
 104                 do_div(pct, cache->fci_stat.fst_count);
 105         } else {
 106                 pct = 0;
 107         }
 108
 109         CDEBUG(D_INFO, "FLD cache statistics (%s):\n", cache->fci_name);
 110         CDEBUG(D_INFO, "  Total reqs: %llu\n", cache->fci_stat.fst_count);
 111         CDEBUG(D_INFO, "  Cache reqs: %llu\n", cache->fci_stat.fst_cache);
 112         CDEBUG(D_INFO, "  Cache hits: %llu%%\n", pct);
 113
 114         kfree(cache);
 115 }
 116
 117 /**
 118  * delete given node from list.
 119  */
 120 static void fld_cache_entry_delete(struct fld_cache *cache,
 121                                    struct fld_cache_entry *node)
 122 {
 123         list_del(&node->fce_list);
 124         list_del(&node->fce_lru);
 125         cache->fci_cache_count--;
 126         kfree(node);
 127 }
 128
 129 /**
 130  * fix list by checking new entry with NEXT entry in order.
 131  */
 132 static void fld_fix_new_list(struct fld_cache *cache)
 133 {
 134         struct fld_cache_entry *f_curr;
 135         struct fld_cache_entry *f_next;
 136         struct lu_seq_range *c_range;
 137         struct lu_seq_range *n_range;
 138         struct list_head *head = &cache->fci_entries_head;
 139
 140 restart_fixup:
 141
 142         list_for_each_entry_safe(f_curr, f_next, head, fce_list) {
 143                 c_range = &f_curr->fce_range;
 144                 n_range = &f_next->fce_range;
 145
 146                 LASSERT(range_is_sane(c_range));
 147                 if (&f_next->fce_list == head)
 148                         break;
 149
 150                 if (c_range->lsr_flags != n_range->lsr_flags)
 151                         continue;
 152
 153                 LASSERTF(c_range->lsr_start <= n_range->lsr_start,
 154                          "cur lsr_start "DRANGE" next lsr_start "DRANGE"\n",
 155                          PRANGE(c_range), PRANGE(n_range));
 156
 157                 /* check merge possibility with next range */
 158                 if (c_range->lsr_end == n_range->lsr_start) {
 159                         if (c_range->lsr_index != n_range->lsr_index)
 160                                 continue;
 161                         n_range->lsr_start = c_range->lsr_start;
 162                         fld_cache_entry_delete(cache, f_curr);
 163                         continue;
 164                 }
 165
 166                 /* check if current range overlaps with next range. */
 167                 if (n_range->lsr_start < c_range->lsr_end) {
 168                         if (c_range->lsr_index == n_range->lsr_index) {
 169                                 n_range->lsr_start = c_range->lsr_start;
 170                                 n_range->lsr_end = max(c_range->lsr_end,
 171                                                        n_range->lsr_end);
 172                                 fld_cache_entry_delete(cache, f_curr);
 173                         } else {
 174                                 if (n_range->lsr_end <= c_range->lsr_end) {
 175                                         *n_range = *c_range;
 176                                         fld_cache_entry_delete(cache, f_curr);
 177                                 } else {
 178                                         n_range->lsr_start = c_range->lsr_end;
 179                                 }
 180                         }
 181
 182                         /* we could have overlap over next
 183                          * range too. better restart.
 184                          */
 185                         goto restart_fixup;
 186                 }
 187
 188                 /* kill duplicates */
 189                 if (c_range->lsr_start == n_range->lsr_start &&
 190                     c_range->lsr_end == n_range->lsr_end)
 191                         fld_cache_entry_delete(cache, f_curr);
 192         }
 193 }
 194
 195 /**
 196  * add node to fld cache
 197  */
 198 static inline void fld_cache_entry_add(struct fld_cache *cache,
 199                                        struct fld_cache_entry *f_new,
 200                                        struct list_head *pos)
 201 {
 202         list_add(&f_new->fce_list, pos);
 203         list_add(&f_new->fce_lru, &cache->fci_lru);
 204
 205         cache->fci_cache_count++;
 206         fld_fix_new_list(cache);
 207 }
 208
 209 /**
 210  * Check if cache needs to be shrunk. If so - do it.
 211  * Remove one entry in list and so on until cache is shrunk enough.
 212  */
 213 static int fld_cache_shrink(struct fld_cache *cache)
 214 {
 215         struct fld_cache_entry *flde;
 216         struct list_head *curr;
 217         int num = 0;
 218
 219         if (cache->fci_cache_count < cache->fci_cache_size)
 220                 return 0;
 221
 222         curr = cache->fci_lru.prev;
 223
 224         while (cache->fci_cache_count + cache->fci_threshold >
 225                cache->fci_cache_size && curr != &cache->fci_lru) {
 226                 flde = list_entry(curr, struct fld_cache_entry, fce_lru);
 227                 curr = curr->prev;
 228                 fld_cache_entry_delete(cache, flde);
 229                 num++;
 230         }
 231
 232         CDEBUG(D_INFO, "%s: FLD cache - Shrunk by %d entries\n",
 233                cache->fci_name, num);
 234
 235         return 0;
 236 }
 237
 238 /**
 239  * kill all fld cache entries.
 240  */
 241 void fld_cache_flush(struct fld_cache *cache)
 242 {
 243         write_lock(&cache->fci_lock);
 244         cache->fci_cache_size = 0;
 245         fld_cache_shrink(cache);
 246         write_unlock(&cache->fci_lock);
 247 }
 248
 249 /**
 250  * punch hole in existing range. divide this range and add new
 251  * entry accordingly.
 252  */
 253
 254 static void fld_cache_punch_hole(struct fld_cache *cache,
 255                                  struct fld_cache_entry *f_curr,
 256                                  struct fld_cache_entry *f_new)
 257 {
 258         const struct lu_seq_range *range = &f_new->fce_range;
 259         const u64 new_start  = range->lsr_start;
 260         const u64 new_end  = range->lsr_end;
 261         struct fld_cache_entry *fldt;
 262
 263         fldt = kzalloc(sizeof(*fldt), GFP_ATOMIC);
 264         if (!fldt) {
 265                 kfree(f_new);
 266                 /* overlap is not allowed, so dont mess up list. */
 267                 return;
 268         }
 269         /*  break f_curr RANGE into three RANGES:
 270          *      f_curr, f_new , fldt
 271          */
 272
 273         /* f_new = *range */
 274
 275         /* fldt */
 276         fldt->fce_range.lsr_start = new_end;
 277         fldt->fce_range.lsr_end = f_curr->fce_range.lsr_end;
 278         fldt->fce_range.lsr_index = f_curr->fce_range.lsr_index;
 279
 280         /* f_curr */
 281         f_curr->fce_range.lsr_end = new_start;
 282
 283         /* add these two entries to list */
 284         fld_cache_entry_add(cache, f_new, &f_curr->fce_list);
 285         fld_cache_entry_add(cache, fldt, &f_new->fce_list);
 286
 287         /* no need to fixup */
 288 }
 289
 290 /**
 291  * handle range overlap in fld cache.
 292  */
 293 static void fld_cache_overlap_handle(struct fld_cache *cache,
 294                                      struct fld_cache_entry *f_curr,
 295                                      struct fld_cache_entry *f_new)
 296 {
 297         const struct lu_seq_range *range = &f_new->fce_range;
 298         const u64 new_start  = range->lsr_start;
 299         const u64 new_end  = range->lsr_end;
 300         const u32 mdt = range->lsr_index;
 301
 302         /* this is overlap case, these case are checking overlapping with
 303          * prev range only. fixup will handle overlapping with next range.
 304          */
 305
 306         if (f_curr->fce_range.lsr_index == mdt) {
 307                 f_curr->fce_range.lsr_start = min(f_curr->fce_range.lsr_start,
 308                                                   new_start);
 309
 310                 f_curr->fce_range.lsr_end = max(f_curr->fce_range.lsr_end,
 311                                                 new_end);
 312
 313                 kfree(f_new);
 314                 fld_fix_new_list(cache);
 315
 316         } else if (new_start <= f_curr->fce_range.lsr_start &&
 317                         f_curr->fce_range.lsr_end <= new_end) {
 318                 /* case 1: new range completely overshadowed existing range.
 319                  *       e.g. whole range migrated. update fld cache entry
 320                  */
 321
 322                 f_curr->fce_range = *range;
 323                 kfree(f_new);
 324                 fld_fix_new_list(cache);
 325
 326         } else if (f_curr->fce_range.lsr_start < new_start &&
 327                         new_end < f_curr->fce_range.lsr_end) {
 328                 /* case 2: new range fit within existing range. */
 329
 330                 fld_cache_punch_hole(cache, f_curr, f_new);
 331
 332         } else  if (new_end <= f_curr->fce_range.lsr_end) {
 333                 /* case 3: overlap:
 334                  *       [new_start [c_start  new_end)  c_end)
 335                  */
 336
 337                 LASSERT(new_start <= f_curr->fce_range.lsr_start);
 338
 339                 f_curr->fce_range.lsr_start = new_end;
 340                 fld_cache_entry_add(cache, f_new, f_curr->fce_list.prev);
 341
 342         } else if (f_curr->fce_range.lsr_start <= new_start) {
 343                 /* case 4: overlap:
 344                  *       [c_start [new_start c_end) new_end)
 345                  */
 346
 347                 LASSERT(f_curr->fce_range.lsr_end <= new_end);
 348
 349                 f_curr->fce_range.lsr_end = new_start;
 350                 fld_cache_entry_add(cache, f_new, &f_curr->fce_list);
 351         } else
 352                 CERROR("NEW range ="DRANGE" curr = "DRANGE"\n",
 353                        PRANGE(range), PRANGE(&f_curr->fce_range));
 354 }
 355
 356 struct fld_cache_entry
 357 *fld_cache_entry_create(const struct lu_seq_range *range)
 358 {
 359         struct fld_cache_entry *f_new;
 360
 361         LASSERT(range_is_sane(range));
 362
 363         f_new = kzalloc(sizeof(*f_new), GFP_NOFS);
 364         if (!f_new)
 365                 return ERR_PTR(-ENOMEM);
 366
 367         f_new->fce_range = *range;
 368         return f_new;
 369 }
 370
 371 /**
 372  * Insert FLD entry in FLD cache.
 373  *
 374  * This function handles all cases of merging and breaking up of
 375  * ranges.
 376  */
 377 static int fld_cache_insert_nolock(struct fld_cache *cache,
 378                                    struct fld_cache_entry *f_new)
 379 {
 380         struct fld_cache_entry *f_curr;
 381         struct fld_cache_entry *n;
 382         struct list_head *head;
 383         struct list_head *prev = NULL;
 384         const u64 new_start  = f_new->fce_range.lsr_start;
 385         const u64 new_end  = f_new->fce_range.lsr_end;
 386         __u32 new_flags  = f_new->fce_range.lsr_flags;
 387
 388         /*
 389          * Duplicate entries are eliminated in insert op.
 390          * So we don't need to search new entry before starting
 391          * insertion loop.
 392          */
 393
 394         if (!cache->fci_no_shrink)
 395                 fld_cache_shrink(cache);
 396
 397         head = &cache->fci_entries_head;
 398
 399         list_for_each_entry_safe(f_curr, n, head, fce_list) {
 400                 /* add list if next is end of list */
 401                 if (new_end < f_curr->fce_range.lsr_start ||
 402                     (new_end == f_curr->fce_range.lsr_start &&
 403                      new_flags != f_curr->fce_range.lsr_flags))
 404                         break;
 405
 406                 prev = &f_curr->fce_list;
 407                 /* check if this range is to left of new range. */
 408                 if (new_start < f_curr->fce_range.lsr_end &&
 409                     new_flags == f_curr->fce_range.lsr_flags) {
 410                         fld_cache_overlap_handle(cache, f_curr, f_new);
 411                         goto out;
 412                 }
 413         }
 414
 415         if (!prev)
 416                 prev = head;
 417
 418         CDEBUG(D_INFO, "insert range "DRANGE"\n", PRANGE(&f_new->fce_range));
 419         /* Add new entry to cache and lru list. */
 420         fld_cache_entry_add(cache, f_new, prev);
 421 out:
 422         return 0;
 423 }
 424
 425 int fld_cache_insert(struct fld_cache *cache,
 426                      const struct lu_seq_range *range)
 427 {
 428         struct fld_cache_entry  *flde;
 429         int rc;
 430
 431         flde = fld_cache_entry_create(range);
 432         if (IS_ERR(flde))
 433                 return PTR_ERR(flde);
 434
 435         write_lock(&cache->fci_lock);
 436         rc = fld_cache_insert_nolock(cache, flde);
 437         write_unlock(&cache->fci_lock);
 438         if (rc)
 439                 kfree(flde);
 440
 441         return rc;
 442 }
 443
 444 /**
 445  * Delete FLD entry in FLD cache.
 446  *
 447  */
 448
 449 struct fld_cache_entry
 450 *fld_cache_entry_lookup_nolock(struct fld_cache *cache,
 451                               struct lu_seq_range *range)
 452 {
 453         struct fld_cache_entry *flde;
 454         struct fld_cache_entry *got = NULL;
 455         struct list_head *head;
 456
 457         head = &cache->fci_entries_head;
 458         list_for_each_entry(flde, head, fce_list) {
 459                 if (range->lsr_start == flde->fce_range.lsr_start ||
 460                     (range->lsr_end == flde->fce_range.lsr_end &&
 461                      range->lsr_flags == flde->fce_range.lsr_flags)) {
 462                         got = flde;
 463                         break;
 464                 }
 465         }
 466
 467         return got;
 468 }
 469
 470 /**
 471  * lookup \a seq sequence for range in fld cache.
 472  */
 473 struct fld_cache_entry
 474 *fld_cache_entry_lookup(struct fld_cache *cache, struct lu_seq_range *range)
 475 {
 476         struct fld_cache_entry *got = NULL;
 477
 478         read_lock(&cache->fci_lock);
 479         got = fld_cache_entry_lookup_nolock(cache, range);
 480         read_unlock(&cache->fci_lock);
 481         return got;
 482 }
 483
 484 /**
 485  * lookup \a seq sequence for range in fld cache.
 486  */
 487 int fld_cache_lookup(struct fld_cache *cache,
 488                      const u64 seq, struct lu_seq_range *range)
 489 {
 490         struct fld_cache_entry *flde;
 491         struct fld_cache_entry *prev = NULL;
 492         struct list_head *head;
 493
 494         read_lock(&cache->fci_lock);
 495         head = &cache->fci_entries_head;
 496
 497         cache->fci_stat.fst_count++;
 498         list_for_each_entry(flde, head, fce_list) {
 499                 if (flde->fce_range.lsr_start > seq) {
 500                         if (prev)
 501                                 *range = prev->fce_range;
 502                         break;
 503                 }
 504
 505                 prev = flde;
 506                 if (range_within(&flde->fce_range, seq)) {
 507                         *range = flde->fce_range;
 508
 509                         cache->fci_stat.fst_cache++;
 510                         read_unlock(&cache->fci_lock);
 511                         return 0;
 512                 }
 513         }
 514         read_unlock(&cache->fci_lock);
 515         return -ENOENT;
 516 }