drivers/staging/lustre/lustre/lov/lov_pool.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 [sun.com URL with a
  18  * copy of GPLv2].
  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) 2012, 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  * lustre/lov/lov_pool.c
  37  *
  38  * OST pool methods
  39  *
  40  * Author: Jacques-Charles LAFOUCRIERE <jc.lafoucriere@cea.fr>
  41  * Author: Alex Lyashkov <Alexey.Lyashkov@Sun.COM>
  42  * Author: Nathaniel Rutman <Nathan.Rutman@Sun.COM>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_LOV
  46
  47 #include "../../include/linux/libcfs/libcfs.h"
  48
  49 #include "../include/obd.h"
  50 #include "lov_internal.h"
  51
  52 #define pool_tgt(_p, _i) \
  53                 _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
  54
  55 static void lov_pool_getref(struct pool_desc *pool)
  56 {
  57         CDEBUG(D_INFO, "pool %p\n", pool);
  58         atomic_inc(&pool->pool_refcount);
  59 }
  60
  61 void lov_pool_putref(struct pool_desc *pool)
  62 {
  63         CDEBUG(D_INFO, "pool %p\n", pool);
  64         if (atomic_dec_and_test(&pool->pool_refcount)) {
  65                 LASSERT(hlist_unhashed(&pool->pool_hash));
  66                 LASSERT(list_empty(&pool->pool_list));
  67                 LASSERT(!pool->pool_debugfs_entry);
  68                 lov_ost_pool_free(&(pool->pool_rr.lqr_pool));
  69                 lov_ost_pool_free(&(pool->pool_obds));
  70                 kfree(pool);
  71         }
  72 }
  73
  74 static void lov_pool_putref_locked(struct pool_desc *pool)
  75 {
  76         CDEBUG(D_INFO, "pool %p\n", pool);
  77         LASSERT(atomic_read(&pool->pool_refcount) > 1);
  78
  79         atomic_dec(&pool->pool_refcount);
  80 }
  81
  82 /*
  83  * hash function using a Rotating Hash algorithm
  84  * Knuth, D. The Art of Computer Programming,
  85  * Volume 3: Sorting and Searching,
  86  * Chapter 6.4.
  87  * Addison Wesley, 1973
  88  */
  89 static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key, unsigned mask)
  90 {
  91         int i;
  92         __u32 result;
  93         char *poolname;
  94
  95         result = 0;
  96         poolname = (char *)key;
  97         for (i = 0; i < LOV_MAXPOOLNAME; i++) {
  98                 if (poolname[i] == '\0')
  99                         break;
 100                 result = (result << 4)^(result >> 28) ^  poolname[i];
 101         }
 102         return (result % mask);
 103 }
 104
 105 static void *pool_key(struct hlist_node *hnode)
 106 {
 107         struct pool_desc *pool;
 108
 109         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 110         return pool->pool_name;
 111 }
 112
 113 static int pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
 114 {
 115         char *pool_name;
 116         struct pool_desc *pool;
 117
 118         pool_name = (char *)key;
 119         pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
 120         return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
 121 }
 122
 123 static void *pool_hashobject(struct hlist_node *hnode)
 124 {
 125         return hlist_entry(hnode, struct pool_desc, pool_hash);
 126 }
 127
 128 static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
 129 {
 130         struct pool_desc *pool;
 131
 132         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 133         lov_pool_getref(pool);
 134 }
 135
 136 static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
 137                                          struct hlist_node *hnode)
 138 {
 139         struct pool_desc *pool;
 140
 141         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 142         lov_pool_putref_locked(pool);
 143 }
 144
 145 struct cfs_hash_ops pool_hash_operations = {
 146         .hs_hash        = pool_hashfn,
 147         .hs_key         = pool_key,
 148         .hs_keycmp      = pool_hashkey_keycmp,
 149         .hs_object      = pool_hashobject,
 150         .hs_get         = pool_hashrefcount_get,
 151         .hs_put_locked  = pool_hashrefcount_put_locked,
 152
 153 };
 154
 155 /*
 156  * pool debugfs seq_file methods
 157  */
 158 /*
 159  * iterator is used to go through the target pool entries
 160  * index is the current entry index in the lp_array[] array
 161  * index >= pos returned to the seq_file interface
 162  * pos is from 0 to (pool->pool_obds.op_count - 1)
 163  */
 164 #define POOL_IT_MAGIC 0xB001CEA0
 165 struct pool_iterator {
 166         int magic;
 167         struct pool_desc *pool;
 168         int idx;        /* from 0 to pool_tgt_size - 1 */
 169 };
 170
 171 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
 172 {
 173         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 174         int prev_idx;
 175
 176         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
 177
 178         /* test if end of file */
 179         if (*pos >= pool_tgt_count(iter->pool))
 180                 return NULL;
 181
 182         /* iterate to find a non empty entry */
 183         prev_idx = iter->idx;
 184         down_read(&pool_tgt_rw_sem(iter->pool));
 185         iter->idx++;
 186         if (iter->idx == pool_tgt_count(iter->pool)) {
 187                 iter->idx = prev_idx; /* we stay on the last entry */
 188                 up_read(&pool_tgt_rw_sem(iter->pool));
 189                 return NULL;
 190         }
 191         up_read(&pool_tgt_rw_sem(iter->pool));
 192         (*pos)++;
 193         /* return != NULL to continue */
 194         return iter;
 195 }
 196
 197 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
 198 {
 199         struct pool_desc *pool = (struct pool_desc *)s->private;
 200         struct pool_iterator *iter;
 201
 202         lov_pool_getref(pool);
 203         if ((pool_tgt_count(pool) == 0) ||
 204             (*pos >= pool_tgt_count(pool))) {
 205                 /* iter is not created, so stop() has no way to
 206                  * find pool to dec ref
 207                  */
 208                 lov_pool_putref(pool);
 209                 return NULL;
 210         }
 211
 212         iter = kzalloc(sizeof(*iter), GFP_NOFS);
 213         if (!iter)
 214                 return ERR_PTR(-ENOMEM);
 215         iter->magic = POOL_IT_MAGIC;
 216         iter->pool = pool;
 217         iter->idx = 0;
 218
 219         /* we use seq_file private field to memorized iterator so
 220          * we can free it at stop()
 221          */
 222         /* /!\ do not forget to restore it to pool before freeing it */
 223         s->private = iter;
 224         if (*pos > 0) {
 225                 loff_t i;
 226                 void *ptr;
 227
 228                 i = 0;
 229                 do {
 230                         ptr = pool_proc_next(s, &iter, &i);
 231                 } while ((i < *pos) && ptr);
 232                 return ptr;
 233         }
 234         return iter;
 235 }
 236
 237 static void pool_proc_stop(struct seq_file *s, void *v)
 238 {
 239         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 240
 241         /* in some cases stop() method is called 2 times, without
 242          * calling start() method (see seq_read() from fs/seq_file.c)
 243          * we have to free only if s->private is an iterator
 244          */
 245         if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
 246                 /* we restore s->private so next call to pool_proc_start()
 247                  * will work
 248                  */
 249                 s->private = iter->pool;
 250                 lov_pool_putref(iter->pool);
 251                 kfree(iter);
 252         }
 253 }
 254
 255 static int pool_proc_show(struct seq_file *s, void *v)
 256 {
 257         struct pool_iterator *iter = (struct pool_iterator *)v;
 258         struct lov_tgt_desc *tgt;
 259
 260         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
 261         LASSERT(iter->pool);
 262         LASSERT(iter->idx <= pool_tgt_count(iter->pool));
 263
 264         down_read(&pool_tgt_rw_sem(iter->pool));
 265         tgt = pool_tgt(iter->pool, iter->idx);
 266         up_read(&pool_tgt_rw_sem(iter->pool));
 267         if (tgt)
 268                 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
 269
 270         return 0;
 271 }
 272
 273 static const struct seq_operations pool_proc_ops = {
 274         .start    = pool_proc_start,
 275         .next      = pool_proc_next,
 276         .stop      = pool_proc_stop,
 277         .show      = pool_proc_show,
 278 };
 279
 280 static int pool_proc_open(struct inode *inode, struct file *file)
 281 {
 282         int rc;
 283
 284         rc = seq_open(file, &pool_proc_ops);
 285         if (!rc) {
 286                 struct seq_file *s = file->private_data;
 287
 288                 s->private = inode->i_private;
 289         }
 290         return rc;
 291 }
 292
 293 static struct file_operations pool_proc_operations = {
 294         .open      = pool_proc_open,
 295         .read      = seq_read,
 296         .llseek  = seq_lseek,
 297         .release        = seq_release,
 298 };
 299
 300 #define LOV_POOL_INIT_COUNT 2
 301 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
 302 {
 303         if (count == 0)
 304                 count = LOV_POOL_INIT_COUNT;
 305         op->op_array = NULL;
 306         op->op_count = 0;
 307         init_rwsem(&op->op_rw_sem);
 308         op->op_size = count;
 309         op->op_array = kcalloc(op->op_size, sizeof(op->op_array[0]), GFP_NOFS);
 310         if (!op->op_array) {
 311                 op->op_size = 0;
 312                 return -ENOMEM;
 313         }
 314         return 0;
 315 }
 316
 317 /* Caller must hold write op_rwlock */
 318 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
 319 {
 320         __u32 *new;
 321         int new_size;
 322
 323         LASSERT(min_count != 0);
 324
 325         if (op->op_count < op->op_size)
 326                 return 0;
 327
 328         new_size = max(min_count, 2 * op->op_size);
 329         new = kcalloc(new_size, sizeof(op->op_array[0]), GFP_NOFS);
 330         if (!new)
 331                 return -ENOMEM;
 332
 333         /* copy old array to new one */
 334         memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
 335         kfree(op->op_array);
 336         op->op_array = new;
 337         op->op_size = new_size;
 338         return 0;
 339 }
 340
 341 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
 342 {
 343         int rc = 0, i;
 344
 345         down_write(&op->op_rw_sem);
 346
 347         rc = lov_ost_pool_extend(op, min_count);
 348         if (rc)
 349                 goto out;
 350
 351         /* search ost in pool array */
 352         for (i = 0; i < op->op_count; i++) {
 353                 if (op->op_array[i] == idx) {
 354                         rc = -EEXIST;
 355                         goto out;
 356                 }
 357         }
 358         /* ost not found we add it */
 359         op->op_array[op->op_count] = idx;
 360         op->op_count++;
 361 out:
 362         up_write(&op->op_rw_sem);
 363         return rc;
 364 }
 365
 366 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
 367 {
 368         int i;
 369
 370         down_write(&op->op_rw_sem);
 371
 372         for (i = 0; i < op->op_count; i++) {
 373                 if (op->op_array[i] == idx) {
 374                         memmove(&op->op_array[i], &op->op_array[i + 1],
 375                                 (op->op_count - i - 1) * sizeof(op->op_array[0]));
 376                         op->op_count--;
 377                         up_write(&op->op_rw_sem);
 378                         return 0;
 379                 }
 380         }
 381
 382         up_write(&op->op_rw_sem);
 383         return -EINVAL;
 384 }
 385
 386 int lov_ost_pool_free(struct ost_pool *op)
 387 {
 388         if (op->op_size == 0)
 389                 return 0;
 390
 391         down_write(&op->op_rw_sem);
 392
 393         kfree(op->op_array);
 394         op->op_array = NULL;
 395         op->op_count = 0;
 396         op->op_size = 0;
 397
 398         up_write(&op->op_rw_sem);
 399         return 0;
 400 }
 401
 402 int lov_pool_new(struct obd_device *obd, char *poolname)
 403 {
 404         struct lov_obd *lov;
 405         struct pool_desc *new_pool;
 406         int rc;
 407
 408         lov = &(obd->u.lov);
 409
 410         if (strlen(poolname) > LOV_MAXPOOLNAME)
 411                 return -ENAMETOOLONG;
 412
 413         new_pool = kzalloc(sizeof(*new_pool), GFP_NOFS);
 414         if (!new_pool)
 415                 return -ENOMEM;
 416
 417         strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
 418         new_pool->pool_lobd = obd;
 419         /* ref count init to 1 because when created a pool is always used
 420          * up to deletion
 421          */
 422         atomic_set(&new_pool->pool_refcount, 1);
 423         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
 424         if (rc)
 425                 goto out_err;
 426
 427         memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
 428         rc = lov_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
 429         if (rc)
 430                 goto out_free_pool_obds;
 431
 432         INIT_HLIST_NODE(&new_pool->pool_hash);
 433
 434         /* get ref for debugfs file */
 435         lov_pool_getref(new_pool);
 436         new_pool->pool_debugfs_entry = ldebugfs_add_simple(
 437                                                 lov->lov_pool_debugfs_entry,
 438                                                 poolname, new_pool,
 439                                                 &pool_proc_operations);
 440         if (IS_ERR_OR_NULL(new_pool->pool_debugfs_entry)) {
 441                 CWARN("Cannot add debugfs pool entry "LOV_POOLNAMEF"\n",
 442                       poolname);
 443                 new_pool->pool_debugfs_entry = NULL;
 444                 lov_pool_putref(new_pool);
 445         }
 446         CDEBUG(D_INFO, "pool %p - proc %p\n",
 447                new_pool, new_pool->pool_debugfs_entry);
 448
 449         spin_lock(&obd->obd_dev_lock);
 450         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
 451         lov->lov_pool_count++;
 452         spin_unlock(&obd->obd_dev_lock);
 453
 454         /* add to find only when it fully ready  */
 455         rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
 456                                  &new_pool->pool_hash);
 457         if (rc) {
 458                 rc = -EEXIST;
 459                 goto out_err;
 460         }
 461
 462         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
 463                poolname, lov->lov_pool_count);
 464
 465         return 0;
 466
 467 out_err:
 468         spin_lock(&obd->obd_dev_lock);
 469         list_del_init(&new_pool->pool_list);
 470         lov->lov_pool_count--;
 471         spin_unlock(&obd->obd_dev_lock);
 472
 473         ldebugfs_remove(&new_pool->pool_debugfs_entry);
 474
 475         lov_ost_pool_free(&new_pool->pool_rr.lqr_pool);
 476 out_free_pool_obds:
 477         lov_ost_pool_free(&new_pool->pool_obds);
 478         kfree(new_pool);
 479         return rc;
 480 }
 481
 482 int lov_pool_del(struct obd_device *obd, char *poolname)
 483 {
 484         struct lov_obd *lov;
 485         struct pool_desc *pool;
 486
 487         lov = &(obd->u.lov);
 488
 489         /* lookup and kill hash reference */
 490         pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
 491         if (!pool)
 492                 return -ENOENT;
 493
 494         if (!IS_ERR_OR_NULL(pool->pool_debugfs_entry)) {
 495                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_debugfs_entry);
 496                 ldebugfs_remove(&pool->pool_debugfs_entry);
 497                 lov_pool_putref(pool);
 498         }
 499
 500         spin_lock(&obd->obd_dev_lock);
 501         list_del_init(&pool->pool_list);
 502         lov->lov_pool_count--;
 503         spin_unlock(&obd->obd_dev_lock);
 504
 505         /* release last reference */
 506         lov_pool_putref(pool);
 507
 508         return 0;
 509 }
 510
 511 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
 512 {
 513         struct obd_uuid ost_uuid;
 514         struct lov_obd *lov;
 515         struct pool_desc *pool;
 516         unsigned int lov_idx;
 517         int rc;
 518
 519         lov = &(obd->u.lov);
 520
 521         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 522         if (!pool)
 523                 return -ENOENT;
 524
 525         obd_str2uuid(&ost_uuid, ostname);
 526
 527         /* search ost in lov array */
 528         obd_getref(obd);
 529         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 530                 if (!lov->lov_tgts[lov_idx])
 531                         continue;
 532                 if (obd_uuid_equals(&ost_uuid,
 533                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 534                         break;
 535         }
 536         /* test if ost found in lov */
 537         if (lov_idx == lov->desc.ld_tgt_count) {
 538                 rc = -EINVAL;
 539                 goto out;
 540         }
 541
 542         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
 543         if (rc)
 544                 goto out;
 545
 546         pool->pool_rr.lqr_dirty = 1;
 547
 548         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
 549                ostname, poolname,  pool_tgt_count(pool));
 550
 551 out:
 552         obd_putref(obd);
 553         lov_pool_putref(pool);
 554         return rc;
 555 }
 556
 557 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
 558 {
 559         struct obd_uuid ost_uuid;
 560         struct lov_obd *lov;
 561         struct pool_desc *pool;
 562         unsigned int lov_idx;
 563         int rc = 0;
 564
 565         lov = &(obd->u.lov);
 566
 567         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 568         if (!pool)
 569                 return -ENOENT;
 570
 571         obd_str2uuid(&ost_uuid, ostname);
 572
 573         obd_getref(obd);
 574         /* search ost in lov array, to get index */
 575         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 576                 if (!lov->lov_tgts[lov_idx])
 577                         continue;
 578
 579                 if (obd_uuid_equals(&ost_uuid,
 580                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 581                         break;
 582         }
 583
 584         /* test if ost found in lov */
 585         if (lov_idx == lov->desc.ld_tgt_count) {
 586                 rc = -EINVAL;
 587                 goto out;
 588         }
 589
 590         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
 591
 592         pool->pool_rr.lqr_dirty = 1;
 593
 594         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
 595                poolname);
 596
 597 out:
 598         obd_putref(obd);
 599         lov_pool_putref(pool);
 600         return rc;
 601 }
 602
 603 int lov_check_index_in_pool(__u32 idx, struct pool_desc *pool)
 604 {
 605         int i, rc;
 606
 607         /* caller may no have a ref on pool if it got the pool
 608          * without calling lov_find_pool() (e.g. go through the lov pool
 609          * list)
 610          */
 611         lov_pool_getref(pool);
 612
 613         down_read(&pool_tgt_rw_sem(pool));
 614
 615         for (i = 0; i < pool_tgt_count(pool); i++) {
 616                 if (pool_tgt_array(pool)[i] == idx) {
 617                         rc = 0;
 618                         goto out;
 619                 }
 620         }
 621         rc = -ENOENT;
 622 out:
 623         up_read(&pool_tgt_rw_sem(pool));
 624
 625         lov_pool_putref(pool);
 626         return rc;
 627 }
 628
 629 struct pool_desc *lov_find_pool(struct lov_obd *lov, char *poolname)
 630 {
 631         struct pool_desc *pool;
 632
 633         pool = NULL;
 634         if (poolname[0] != '\0') {
 635                 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 636                 if (!pool)
 637                         CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
 638                               poolname);
 639                 if (pool && (pool_tgt_count(pool) == 0)) {
 640                         CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
 641                               poolname);
 642                         /* pool is ignored, so we remove ref on it */
 643                         lov_pool_putref(pool);
 644                         pool = NULL;
 645                 }
 646         }
 647         return pool;
 648 }