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
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2014, 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/lov/lov_pool.c
  33  *
  34  * OST pool methods
  35  *
  36  * Author: Jacques-Charles LAFOUCRIERE <jc.lafoucriere@cea.fr>
  37  * Author: Alex Lyashkov <Alexey.Lyashkov@Sun.COM>
  38  * Author: Nathaniel Rutman <Nathan.Rutman@Sun.COM>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LOV
  42
  43 #include <libcfs/libcfs.h>
  44
  45 #include <obd.h>
  46 #include "lov_internal.h"
  47
  48 #define pool_tgt(_p, _i) \
  49                 _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
  50
  51 static void lov_pool_getref(struct pool_desc *pool)
  52 {
  53         CDEBUG(D_INFO, "pool %p\n", pool);
  54         atomic_inc(&pool->pool_refcount);
  55 }
  56
  57 static void lov_pool_putref(struct pool_desc *pool)
  58 {
  59         CDEBUG(D_INFO, "pool %p\n", pool);
  60         if (atomic_dec_and_test(&pool->pool_refcount)) {
  61                 LASSERT(hlist_unhashed(&pool->pool_hash));
  62                 LASSERT(list_empty(&pool->pool_list));
  63                 LASSERT(pool->pool_proc_entry == NULL);
  64                 lov_ost_pool_free(&(pool->pool_obds));
  65                 OBD_FREE_PTR(pool);
  66                 EXIT;
  67         }
  68 }
  69
  70 static void lov_pool_putref_locked(struct pool_desc *pool)
  71 {
  72         CDEBUG(D_INFO, "pool %p\n", pool);
  73         LASSERT(atomic_read(&pool->pool_refcount) > 1);
  74
  75         atomic_dec(&pool->pool_refcount);
  76 }
  77
  78 /*
  79  * hash function using a Rotating Hash algorithm
  80  * Knuth, D. The Art of Computer Programming,
  81  * Volume 3: Sorting and Searching,
  82  * Chapter 6.4.
  83  * Addison Wesley, 1973
  84  */
  85 static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key,
  86                          unsigned mask)
  87 {
  88         int i;
  89         __u32 result;
  90         char *poolname;
  91
  92         result = 0;
  93         poolname = (char *)key;
  94         for (i = 0; i < LOV_MAXPOOLNAME; i++) {
  95                 if (poolname[i] == '\0')
  96                         break;
  97                 result = (result << 4)^(result >> 28) ^  poolname[i];
  98         }
  99         return (result % mask);
 100 }
 101
 102 static void *pool_key(struct hlist_node *hnode)
 103 {
 104         struct pool_desc *pool;
 105
 106         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 107         return (pool->pool_name);
 108 }
 109
 110 static int
 111 pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
 112 {
 113         char *pool_name;
 114         struct pool_desc *pool;
 115
 116         pool_name = (char *)key;
 117         pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
 118         return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
 119 }
 120
 121 static void *pool_hashobject(struct hlist_node *hnode)
 122 {
 123         return hlist_entry(hnode, struct pool_desc, pool_hash);
 124 }
 125
 126 static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
 127 {
 128         struct pool_desc *pool;
 129
 130         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 131         lov_pool_getref(pool);
 132 }
 133
 134 static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
 135                                          struct hlist_node *hnode)
 136 {
 137         struct pool_desc *pool;
 138
 139         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 140         lov_pool_putref_locked(pool);
 141 }
 142
 143 struct cfs_hash_ops pool_hash_operations = {
 144         .hs_hash        = pool_hashfn,
 145         .hs_key         = pool_key,
 146         .hs_keycmp      = pool_hashkey_keycmp,
 147         .hs_object      = pool_hashobject,
 148         .hs_get         = pool_hashrefcount_get,
 149         .hs_put_locked  = pool_hashrefcount_put_locked,
 150
 151 };
 152
 153 #ifdef CONFIG_PROC_FS
 154 /*
 155  * pool /proc 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         iter->idx++;
 184         if (iter->idx >= pool_tgt_count(iter->pool)) {
 185                 iter->idx = prev_idx; /* we stay on the last entry */
 186                 return NULL;
 187         }
 188         (*pos)++;
 189         /* return != NULL to continue */
 190         return iter;
 191 }
 192
 193 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
 194 {
 195         struct pool_desc *pool = (struct pool_desc *)s->private;
 196         struct pool_iterator *iter;
 197
 198         lov_pool_getref(pool);
 199         if ((pool_tgt_count(pool) == 0) ||
 200             (*pos >= pool_tgt_count(pool))) {
 201                 /* iter is not created, so stop() has no way to
 202                  * find pool to dec ref */
 203                 lov_pool_putref(pool);
 204                 return NULL;
 205         }
 206
 207         OBD_ALLOC_PTR(iter);
 208         if (!iter)
 209                 return ERR_PTR(-ENOMEM);
 210         iter->magic = POOL_IT_MAGIC;
 211         iter->pool = pool;
 212         iter->idx = 0;
 213
 214         /* we use seq_file private field to memorized iterator so
 215          * we can free it at stop() */
 216         /* /!\ do not forget to restore it to pool before freeing it */
 217         s->private = iter;
 218         down_read(&pool_tgt_rw_sem(pool));
 219         if (*pos > 0) {
 220                 loff_t i;
 221                 void *ptr;
 222
 223                 i = 0;
 224                 do {
 225                      ptr = pool_proc_next(s, &iter, &i);
 226                 } while ((i < *pos) && (ptr != NULL));
 227                 return ptr;
 228         }
 229         return iter;
 230 }
 231
 232 static void pool_proc_stop(struct seq_file *s, void *v)
 233 {
 234         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 235
 236         /* in some cases stop() method is called 2 times, without
 237          * calling start() method (see seq_read() from fs/seq_file.c)
 238          * we have to free only if s->private is an iterator */
 239         if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
 240                 up_read(&pool_tgt_rw_sem(iter->pool));
 241                 /* we restore s->private so next call to pool_proc_start()
 242                  * will work */
 243                 s->private = iter->pool;
 244                 lov_pool_putref(iter->pool);
 245                 OBD_FREE_PTR(iter);
 246         }
 247         return;
 248 }
 249
 250 static int pool_proc_show(struct seq_file *s, void *v)
 251 {
 252         struct pool_iterator *iter = (struct pool_iterator *)v;
 253         struct lov_tgt_desc *tgt;
 254
 255         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
 256         LASSERT(iter->pool != NULL);
 257         LASSERT(iter->idx <= pool_tgt_count(iter->pool));
 258
 259         tgt = pool_tgt(iter->pool, iter->idx);
 260         if (tgt)
 261                 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
 262
 263         return 0;
 264 }
 265
 266 static struct seq_operations pool_proc_ops = {
 267         .start          = pool_proc_start,
 268         .next           = pool_proc_next,
 269         .stop           = pool_proc_stop,
 270         .show           = pool_proc_show,
 271 };
 272
 273 static int pool_proc_open(struct inode *inode, struct file *file)
 274 {
 275         int rc;
 276
 277         rc = seq_open(file, &pool_proc_ops);
 278         if (!rc) {
 279                 struct seq_file *s = file->private_data;
 280                 s->private = PDE_DATA(inode);
 281         }
 282         return rc;
 283 }
 284
 285 static struct file_operations pool_proc_operations = {
 286         .open           = pool_proc_open,
 287         .read           = seq_read,
 288         .llseek         = seq_lseek,
 289         .release        = seq_release,
 290 };
 291 #endif /* CONFIG_PROC_FS */
 292
 293 void lov_dump_pool(int level, struct pool_desc *pool)
 294 {
 295         int i;
 296
 297         lov_pool_getref(pool);
 298
 299         CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
 300                pool->pool_name, pool->pool_obds.op_count);
 301         down_read(&pool_tgt_rw_sem(pool));
 302
 303         for (i = 0; i < pool_tgt_count(pool) ; i++) {
 304                 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
 305                         continue;
 306                 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
 307                        pool->pool_name, i,
 308                        obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
 309         }
 310
 311         up_read(&pool_tgt_rw_sem(pool));
 312         lov_pool_putref(pool);
 313 }
 314
 315 #define LOV_POOL_INIT_COUNT 2
 316 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
 317 {
 318         ENTRY;
 319
 320         if (count == 0)
 321                 count = LOV_POOL_INIT_COUNT;
 322         op->op_array = NULL;
 323         op->op_count = 0;
 324         init_rwsem(&op->op_rw_sem);
 325         op->op_size = count * sizeof(op->op_array[0]);
 326         OBD_ALLOC(op->op_array, op->op_size);
 327         if (op->op_array == NULL) {
 328                 op->op_size = 0;
 329                 RETURN(-ENOMEM);
 330         }
 331         EXIT;
 332         return 0;
 333 }
 334
 335 /* Caller must hold write op_rwlock */
 336 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
 337 {
 338         __u32 *new;
 339         __u32 new_size;
 340
 341         LASSERT(min_count != 0);
 342
 343         if (op->op_count * sizeof(op->op_array[0]) < op->op_size)
 344                 return 0;
 345
 346         new_size = max_t(__u32, min_count * sizeof(op->op_array[0]),
 347                          2 * op->op_size);
 348         OBD_ALLOC(new, new_size);
 349         if (new == NULL)
 350                 return -ENOMEM;
 351
 352         /* copy old array to new one */
 353         memcpy(new, op->op_array, op->op_size);
 354         OBD_FREE(op->op_array, op->op_size);
 355         op->op_array = new;
 356         op->op_size = new_size;
 357         return 0;
 358 }
 359
 360 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
 361 {
 362         int rc = 0, i;
 363         ENTRY;
 364
 365         down_write(&op->op_rw_sem);
 366
 367         rc = lov_ost_pool_extend(op, min_count);
 368         if (rc)
 369                 GOTO(out, rc);
 370
 371         /* search ost in pool array */
 372         for (i = 0; i < op->op_count; i++) {
 373                 if (op->op_array[i] == idx)
 374                         GOTO(out, rc = -EEXIST);
 375         }
 376         /* ost not found we add it */
 377         op->op_array[op->op_count] = idx;
 378         op->op_count++;
 379         EXIT;
 380 out:
 381         up_write(&op->op_rw_sem);
 382         return rc;
 383 }
 384
 385 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
 386 {
 387         int i;
 388         ENTRY;
 389
 390         down_write(&op->op_rw_sem);
 391
 392         for (i = 0; i < op->op_count; i++) {
 393                 if (op->op_array[i] == idx) {
 394                         memmove(&op->op_array[i], &op->op_array[i + 1],
 395                                 (op->op_count - i - 1) * sizeof(op->op_array[0]));
 396                         op->op_count--;
 397                         up_write(&op->op_rw_sem);
 398                         EXIT;
 399                         return 0;
 400                 }
 401         }
 402
 403         up_write(&op->op_rw_sem);
 404         RETURN(-EINVAL);
 405 }
 406
 407 int lov_ost_pool_free(struct ost_pool *op)
 408 {
 409         ENTRY;
 410
 411         if (op->op_size == 0)
 412                 RETURN(0);
 413
 414         down_write(&op->op_rw_sem);
 415
 416         OBD_FREE(op->op_array, op->op_size);
 417         op->op_array = NULL;
 418         op->op_count = 0;
 419         op->op_size = 0;
 420
 421         up_write(&op->op_rw_sem);
 422         RETURN(0);
 423 }
 424
 425
 426 int lov_pool_new(struct obd_device *obd, char *poolname)
 427 {
 428         struct lov_obd *lov;
 429         struct pool_desc *new_pool;
 430         int rc;
 431         ENTRY;
 432
 433         lov = &(obd->u.lov);
 434
 435         if (strlen(poolname) > LOV_MAXPOOLNAME)
 436                 RETURN(-ENAMETOOLONG);
 437
 438         OBD_ALLOC_PTR(new_pool);
 439         if (new_pool == NULL)
 440                 RETURN(-ENOMEM);
 441
 442         strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
 443         new_pool->pool_lobd = obd;
 444         /* ref count init to 1 because when created a pool is always used
 445          * up to deletion
 446          */
 447         atomic_set(&new_pool->pool_refcount, 1);
 448         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
 449         if (rc)
 450                 GOTO(out_err, rc);
 451
 452         INIT_HLIST_NODE(&new_pool->pool_hash);
 453
 454 #ifdef CONFIG_PROC_FS
 455         /* get ref for /proc file */
 456         lov_pool_getref(new_pool);
 457         new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
 458                                                        poolname, new_pool,
 459                                                        &pool_proc_operations);
 460         if (IS_ERR(new_pool->pool_proc_entry)) {
 461                 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
 462                 new_pool->pool_proc_entry = NULL;
 463                 lov_pool_putref(new_pool);
 464         }
 465         CDEBUG(D_INFO, "pool %p - proc %p\n",
 466                new_pool, new_pool->pool_proc_entry);
 467 #endif
 468
 469         spin_lock(&obd->obd_dev_lock);
 470         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
 471         lov->lov_pool_count++;
 472         spin_unlock(&obd->obd_dev_lock);
 473
 474         /* add to find only when it fully ready  */
 475         rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
 476                                  &new_pool->pool_hash);
 477         if (rc)
 478                 GOTO(out_err, rc = -EEXIST);
 479
 480         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
 481                poolname, lov->lov_pool_count);
 482
 483         RETURN(0);
 484
 485 out_err:
 486         spin_lock(&obd->obd_dev_lock);
 487         list_del_init(&new_pool->pool_list);
 488         lov->lov_pool_count--;
 489         spin_unlock(&obd->obd_dev_lock);
 490         lprocfs_remove(&new_pool->pool_proc_entry);
 491         lov_ost_pool_free(&new_pool->pool_obds);
 492         OBD_FREE_PTR(new_pool);
 493
 494         return rc;
 495 }
 496
 497 int lov_pool_del(struct obd_device *obd, char *poolname)
 498 {
 499         struct lov_obd *lov;
 500         struct pool_desc *pool;
 501         ENTRY;
 502
 503         lov = &(obd->u.lov);
 504
 505         /* lookup and kill hash reference */
 506         pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
 507         if (pool == NULL)
 508                 RETURN(-ENOENT);
 509
 510         if (pool->pool_proc_entry != NULL) {
 511                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
 512                 lprocfs_remove(&pool->pool_proc_entry);
 513                 lov_pool_putref(pool);
 514         }
 515
 516         spin_lock(&obd->obd_dev_lock);
 517         list_del_init(&pool->pool_list);
 518         lov->lov_pool_count--;
 519         spin_unlock(&obd->obd_dev_lock);
 520
 521         /* release last reference */
 522         lov_pool_putref(pool);
 523
 524         RETURN(0);
 525 }
 526
 527
 528 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
 529 {
 530         struct obd_uuid ost_uuid;
 531         struct lov_obd *lov;
 532         struct pool_desc *pool;
 533         unsigned int lov_idx;
 534         int rc;
 535         ENTRY;
 536
 537         lov = &(obd->u.lov);
 538
 539         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 540         if (pool == NULL)
 541                 RETURN(-ENOENT);
 542
 543         obd_str2uuid(&ost_uuid, ostname);
 544
 545
 546         /* search ost in lov array */
 547         obd_getref(obd);
 548         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 549                 if (!lov->lov_tgts[lov_idx])
 550                         continue;
 551                 if (obd_uuid_equals(&ost_uuid,
 552                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 553                         break;
 554         }
 555         /* test if ost found in lov */
 556         if (lov_idx == lov->desc.ld_tgt_count)
 557                 GOTO(out, rc = -EINVAL);
 558
 559         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
 560         if (rc)
 561                 GOTO(out, rc);
 562
 563         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
 564                ostname, poolname,  pool_tgt_count(pool));
 565
 566         EXIT;
 567 out:
 568         obd_putref(obd);
 569         lov_pool_putref(pool);
 570         return rc;
 571 }
 572
 573 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
 574 {
 575         struct obd_uuid ost_uuid;
 576         struct lov_obd *lov;
 577         struct pool_desc *pool;
 578         unsigned int lov_idx;
 579         int rc = 0;
 580         ENTRY;
 581
 582         lov = &(obd->u.lov);
 583
 584         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 585         if (pool == NULL)
 586                 RETURN(-ENOENT);
 587
 588         obd_str2uuid(&ost_uuid, ostname);
 589
 590         obd_getref(obd);
 591         /* search ost in lov array, to get index */
 592         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 593                 if (!lov->lov_tgts[lov_idx])
 594                         continue;
 595
 596                 if (obd_uuid_equals(&ost_uuid,
 597                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 598                         break;
 599         }
 600
 601         /* test if ost found in lov */
 602         if (lov_idx == lov->desc.ld_tgt_count)
 603                 GOTO(out, rc = -EINVAL);
 604
 605         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
 606
 607         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
 608                poolname);
 609
 610         EXIT;
 611 out:
 612         obd_putref(obd);
 613         lov_pool_putref(pool);
 614         return rc;
 615 }