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