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