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 /* ifdef needed for liblustre support */
 155 /*
 156  * pool /proc 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                 lov_pool_putref(pool);
 208                 return NULL;
 209         }
 210
 211         OBD_ALLOC_PTR(iter);
 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         /* /!\ do not forget to restore it to pool before freeing it */
 221         s->private = iter;
 222         if (*pos > 0) {
 223                 loff_t i;
 224                 void *ptr;
 225
 226                 i = 0;
 227                 do {
 228                      ptr = pool_proc_next(s, &iter, &i);
 229                 } while ((i < *pos) && (ptr != NULL));
 230                 return ptr;
 231         }
 232         return iter;
 233 }
 234
 235 static void pool_proc_stop(struct seq_file *s, void *v)
 236 {
 237         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 238
 239         /* in some cases stop() method is called 2 times, without
 240          * calling start() method (see seq_read() from fs/seq_file.c)
 241          * we have to free only if s->private is an iterator */
 242         if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
 243                 /* we restore s->private so next call to pool_proc_start()
 244                  * will work */
 245                 s->private = iter->pool;
 246                 lov_pool_putref(iter->pool);
 247                 OBD_FREE_PTR(iter);
 248         }
 249         return;
 250 }
 251
 252 static int pool_proc_show(struct seq_file *s, void *v)
 253 {
 254         struct pool_iterator *iter = (struct pool_iterator *)v;
 255         struct lov_tgt_desc *tgt;
 256
 257         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
 258         LASSERT(iter->pool != NULL);
 259         LASSERT(iter->idx <= pool_tgt_count(iter->pool));
 260
 261         down_read(&pool_tgt_rw_sem(iter->pool));
 262         tgt = pool_tgt(iter->pool, iter->idx);
 263         up_read(&pool_tgt_rw_sem(iter->pool));
 264         if (tgt)
 265                 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
 266
 267         return 0;
 268 }
 269
 270 static struct seq_operations pool_proc_ops = {
 271         .start          = pool_proc_start,
 272         .next           = pool_proc_next,
 273         .stop           = pool_proc_stop,
 274         .show           = pool_proc_show,
 275 };
 276
 277 static int pool_proc_open(struct inode *inode, struct file *file)
 278 {
 279         int rc;
 280
 281         rc = seq_open(file, &pool_proc_ops);
 282         if (!rc) {
 283                 struct seq_file *s = file->private_data;
 284                 s->private = PDE_DATA(inode);
 285         }
 286         return rc;
 287 }
 288
 289 static struct file_operations pool_proc_operations = {
 290         .open           = pool_proc_open,
 291         .read           = seq_read,
 292         .llseek         = seq_lseek,
 293         .release        = seq_release,
 294 };
 295 #endif /* CONFIG_PROC_FS */
 296
 297 void lov_dump_pool(int level, struct pool_desc *pool)
 298 {
 299         int i;
 300
 301         lov_pool_getref(pool);
 302
 303         CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
 304                pool->pool_name, pool->pool_obds.op_count);
 305         down_read(&pool_tgt_rw_sem(pool));
 306
 307         for (i = 0; i < pool_tgt_count(pool) ; i++) {
 308                 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
 309                         continue;
 310                 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
 311                        pool->pool_name, i,
 312                        obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
 313         }
 314
 315         up_read(&pool_tgt_rw_sem(pool));
 316         lov_pool_putref(pool);
 317 }
 318
 319 #define LOV_POOL_INIT_COUNT 2
 320 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
 321 {
 322         ENTRY;
 323
 324         if (count == 0)
 325                 count = LOV_POOL_INIT_COUNT;
 326         op->op_array = NULL;
 327         op->op_count = 0;
 328         init_rwsem(&op->op_rw_sem);
 329         op->op_size = count * sizeof(op->op_array[0]);
 330         OBD_ALLOC(op->op_array, op->op_size);
 331         if (op->op_array == NULL) {
 332                 op->op_size = 0;
 333                 RETURN(-ENOMEM);
 334         }
 335         EXIT;
 336         return 0;
 337 }
 338
 339 /* Caller must hold write op_rwlock */
 340 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
 341 {
 342         __u32 *new;
 343         __u32 new_size;
 344
 345         LASSERT(min_count != 0);
 346
 347         if (op->op_count * sizeof(op->op_array[0]) < op->op_size)
 348                 return 0;
 349
 350         new_size = max_t(__u32, min_count * sizeof(op->op_array[0]),
 351                          2 * op->op_size);
 352         OBD_ALLOC(new, new_size);
 353         if (new == NULL)
 354                 return -ENOMEM;
 355
 356         /* copy old array to new one */
 357         memcpy(new, op->op_array, op->op_size);
 358         OBD_FREE(op->op_array, op->op_size);
 359         op->op_array = new;
 360         op->op_size = new_size;
 361         return 0;
 362 }
 363
 364 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
 365 {
 366         int rc = 0, i;
 367         ENTRY;
 368
 369         down_write(&op->op_rw_sem);
 370
 371         rc = lov_ost_pool_extend(op, min_count);
 372         if (rc)
 373                 GOTO(out, rc);
 374
 375         /* search ost in pool array */
 376         for (i = 0; i < op->op_count; i++) {
 377                 if (op->op_array[i] == idx)
 378                         GOTO(out, rc = -EEXIST);
 379         }
 380         /* ost not found we add it */
 381         op->op_array[op->op_count] = idx;
 382         op->op_count++;
 383         EXIT;
 384 out:
 385         up_write(&op->op_rw_sem);
 386         return rc;
 387 }
 388
 389 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
 390 {
 391         int i;
 392         ENTRY;
 393
 394         down_write(&op->op_rw_sem);
 395
 396         for (i = 0; i < op->op_count; i++) {
 397                 if (op->op_array[i] == idx) {
 398                         memmove(&op->op_array[i], &op->op_array[i + 1],
 399                                 (op->op_count - i - 1) * sizeof(op->op_array[0]));
 400                         op->op_count--;
 401                         up_write(&op->op_rw_sem);
 402                         EXIT;
 403                         return 0;
 404                 }
 405         }
 406
 407         up_write(&op->op_rw_sem);
 408         RETURN(-EINVAL);
 409 }
 410
 411 int lov_ost_pool_free(struct ost_pool *op)
 412 {
 413         ENTRY;
 414
 415         if (op->op_size == 0)
 416                 RETURN(0);
 417
 418         down_write(&op->op_rw_sem);
 419
 420         OBD_FREE(op->op_array, op->op_size);
 421         op->op_array = NULL;
 422         op->op_count = 0;
 423         op->op_size = 0;
 424
 425         up_write(&op->op_rw_sem);
 426         RETURN(0);
 427 }
 428
 429
 430 int lov_pool_new(struct obd_device *obd, char *poolname)
 431 {
 432         struct lov_obd *lov;
 433         struct pool_desc *new_pool;
 434         int rc;
 435         ENTRY;
 436
 437         lov = &(obd->u.lov);
 438
 439         if (strlen(poolname) > LOV_MAXPOOLNAME)
 440                 RETURN(-ENAMETOOLONG);
 441
 442         OBD_ALLOC_PTR(new_pool);
 443         if (new_pool == NULL)
 444                 RETURN(-ENOMEM);
 445
 446         strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
 447         new_pool->pool_lobd = obd;
 448         /* ref count init to 1 because when created a pool is always used
 449          * up to deletion
 450          */
 451         atomic_set(&new_pool->pool_refcount, 1);
 452         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
 453         if (rc)
 454                 GOTO(out_err, rc);
 455
 456         INIT_HLIST_NODE(&new_pool->pool_hash);
 457
 458 #ifdef CONFIG_PROC_FS
 459         /* get ref for /proc file */
 460         lov_pool_getref(new_pool);
 461         new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
 462                                                        poolname, new_pool,
 463                                                        &pool_proc_operations);
 464         if (IS_ERR(new_pool->pool_proc_entry)) {
 465                 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
 466                 new_pool->pool_proc_entry = NULL;
 467                 lov_pool_putref(new_pool);
 468         }
 469         CDEBUG(D_INFO, "pool %p - proc %p\n",
 470                new_pool, new_pool->pool_proc_entry);
 471 #endif
 472
 473         spin_lock(&obd->obd_dev_lock);
 474         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
 475         lov->lov_pool_count++;
 476         spin_unlock(&obd->obd_dev_lock);
 477
 478         /* add to find only when it fully ready  */
 479         rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
 480                                  &new_pool->pool_hash);
 481         if (rc)
 482                 GOTO(out_err, rc = -EEXIST);
 483
 484         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
 485                poolname, lov->lov_pool_count);
 486
 487         RETURN(0);
 488
 489 out_err:
 490         spin_lock(&obd->obd_dev_lock);
 491         list_del_init(&new_pool->pool_list);
 492         lov->lov_pool_count--;
 493         spin_unlock(&obd->obd_dev_lock);
 494         lprocfs_remove(&new_pool->pool_proc_entry);
 495         lov_ost_pool_free(&new_pool->pool_obds);
 496         OBD_FREE_PTR(new_pool);
 497
 498         return rc;
 499 }
 500
 501 int lov_pool_del(struct obd_device *obd, char *poolname)
 502 {
 503         struct lov_obd *lov;
 504         struct pool_desc *pool;
 505         ENTRY;
 506
 507         lov = &(obd->u.lov);
 508
 509         /* lookup and kill hash reference */
 510         pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
 511         if (pool == NULL)
 512                 RETURN(-ENOENT);
 513
 514         if (pool->pool_proc_entry != NULL) {
 515                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
 516                 lprocfs_remove(&pool->pool_proc_entry);
 517                 lov_pool_putref(pool);
 518         }
 519
 520         spin_lock(&obd->obd_dev_lock);
 521         list_del_init(&pool->pool_list);
 522         lov->lov_pool_count--;
 523         spin_unlock(&obd->obd_dev_lock);
 524
 525         /* release last reference */
 526         lov_pool_putref(pool);
 527
 528         RETURN(0);
 529 }
 530
 531
 532 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
 533 {
 534         struct obd_uuid ost_uuid;
 535         struct lov_obd *lov;
 536         struct pool_desc *pool;
 537         unsigned int lov_idx;
 538         int rc;
 539         ENTRY;
 540
 541         lov = &(obd->u.lov);
 542
 543         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 544         if (pool == NULL)
 545                 RETURN(-ENOENT);
 546
 547         obd_str2uuid(&ost_uuid, ostname);
 548
 549
 550         /* search ost in lov array */
 551         obd_getref(obd);
 552         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 553                 if (!lov->lov_tgts[lov_idx])
 554                         continue;
 555                 if (obd_uuid_equals(&ost_uuid,
 556                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 557                         break;
 558         }
 559         /* test if ost found in lov */
 560         if (lov_idx == lov->desc.ld_tgt_count)
 561                 GOTO(out, rc = -EINVAL);
 562
 563         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
 564         if (rc)
 565                 GOTO(out, rc);
 566
 567         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
 568                ostname, poolname,  pool_tgt_count(pool));
 569
 570         EXIT;
 571 out:
 572         obd_putref(obd);
 573         lov_pool_putref(pool);
 574         return rc;
 575 }
 576
 577 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
 578 {
 579         struct obd_uuid ost_uuid;
 580         struct lov_obd *lov;
 581         struct pool_desc *pool;
 582         unsigned int lov_idx;
 583         int rc = 0;
 584         ENTRY;
 585
 586         lov = &(obd->u.lov);
 587
 588         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 589         if (pool == NULL)
 590                 RETURN(-ENOENT);
 591
 592         obd_str2uuid(&ost_uuid, ostname);
 593
 594         obd_getref(obd);
 595         /* search ost in lov array, to get index */
 596         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 597                 if (!lov->lov_tgts[lov_idx])
 598                         continue;
 599
 600                 if (obd_uuid_equals(&ost_uuid,
 601                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 602                         break;
 603         }
 604
 605         /* test if ost found in lov */
 606         if (lov_idx == lov->desc.ld_tgt_count)
 607                 GOTO(out, rc = -EINVAL);
 608
 609         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
 610
 611         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
 612                poolname);
 613
 614         EXIT;
 615 out:
 616         obd_putref(obd);
 617         lov_pool_putref(pool);
 618         return rc;
 619 }