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, 2014, 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 <libcfs/libcfs.h>
  48
  49 #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 static 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_proc_entry == NULL);
  68                 lov_ost_pool_free(&(pool->pool_obds));
  69                 OBD_FREE_PTR(pool);
  70                 EXIT;
  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(cfs_hash_t *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
 114 pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
 115 {
 116         char *pool_name;
 117         struct pool_desc *pool;
 118
 119         pool_name = (char *)key;
 120         pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
 121         return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
 122 }
 123
 124 static void *pool_hashobject(struct hlist_node *hnode)
 125 {
 126         return hlist_entry(hnode, struct pool_desc, pool_hash);
 127 }
 128
 129 static void pool_hashrefcount_get(cfs_hash_t *hs, struct hlist_node *hnode)
 130 {
 131         struct pool_desc *pool;
 132
 133         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 134         lov_pool_getref(pool);
 135 }
 136
 137 static void pool_hashrefcount_put_locked(cfs_hash_t *hs,
 138                                          struct hlist_node *hnode)
 139 {
 140         struct pool_desc *pool;
 141
 142         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 143         lov_pool_putref_locked(pool);
 144 }
 145
 146 cfs_hash_ops_t pool_hash_operations = {
 147         .hs_hash        = pool_hashfn,
 148         .hs_key         = pool_key,
 149         .hs_keycmp      = pool_hashkey_keycmp,
 150         .hs_object      = pool_hashobject,
 151         .hs_get         = pool_hashrefcount_get,
 152         .hs_put_locked  = pool_hashrefcount_put_locked,
 153
 154 };
 155
 156 #ifdef CONFIG_PROC_FS
 157 /* ifdef needed for liblustre support */
 158 /*
 159  * pool /proc seq_file methods
 160  */
 161 /*
 162  * iterator is used to go through the target pool entries
 163  * index is the current entry index in the lp_array[] array
 164  * index >= pos returned to the seq_file interface
 165  * pos is from 0 to (pool->pool_obds.op_count - 1)
 166  */
 167 #define POOL_IT_MAGIC 0xB001CEA0
 168 struct pool_iterator {
 169         int magic;
 170         struct pool_desc *pool;
 171         int idx;        /* from 0 to pool_tgt_size - 1 */
 172 };
 173
 174 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
 175 {
 176         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 177         int prev_idx;
 178
 179         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
 180
 181         /* test if end of file */
 182         if (*pos >= pool_tgt_count(iter->pool))
 183                 return NULL;
 184
 185         /* iterate to find a non empty entry */
 186         prev_idx = iter->idx;
 187         down_read(&pool_tgt_rw_sem(iter->pool));
 188         iter->idx++;
 189         if (iter->idx == pool_tgt_count(iter->pool)) {
 190                 iter->idx = prev_idx; /* we stay on the last entry */
 191                 up_read(&pool_tgt_rw_sem(iter->pool));
 192                 return NULL;
 193         }
 194         up_read(&pool_tgt_rw_sem(iter->pool));
 195         (*pos)++;
 196         /* return != NULL to continue */
 197         return iter;
 198 }
 199
 200 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
 201 {
 202         struct pool_desc *pool = (struct pool_desc *)s->private;
 203         struct pool_iterator *iter;
 204
 205         lov_pool_getref(pool);
 206         if ((pool_tgt_count(pool) == 0) ||
 207             (*pos >= pool_tgt_count(pool))) {
 208                 /* iter is not created, so stop() has no way to
 209                  * find pool to dec ref */
 210                 lov_pool_putref(pool);
 211                 return NULL;
 212         }
 213
 214         OBD_ALLOC_PTR(iter);
 215         if (!iter)
 216                 return ERR_PTR(-ENOMEM);
 217         iter->magic = POOL_IT_MAGIC;
 218         iter->pool = pool;
 219         iter->idx = 0;
 220
 221         /* we use seq_file private field to memorized iterator so
 222          * we can free it at stop() */
 223         /* /!\ do not forget to restore it to pool before freeing it */
 224         s->private = iter;
 225         if (*pos > 0) {
 226                 loff_t i;
 227                 void *ptr;
 228
 229                 i = 0;
 230                 do {
 231                      ptr = pool_proc_next(s, &iter, &i);
 232                 } while ((i < *pos) && (ptr != NULL));
 233                 return ptr;
 234         }
 235         return iter;
 236 }
 237
 238 static void pool_proc_stop(struct seq_file *s, void *v)
 239 {
 240         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 241
 242         /* in some cases stop() method is called 2 times, without
 243          * calling start() method (see seq_read() from fs/seq_file.c)
 244          * we have to free only if s->private is an iterator */
 245         if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
 246                 /* we restore s->private so next call to pool_proc_start()
 247                  * will work */
 248                 s->private = iter->pool;
 249                 lov_pool_putref(iter->pool);
 250                 OBD_FREE_PTR(iter);
 251         }
 252         return;
 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 != NULL);
 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 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                 s->private = PDE_DATA(inode);
 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 #endif /* CONFIG_PROC_FS */
 299
 300 void lov_dump_pool(int level, struct pool_desc *pool)
 301 {
 302         int i;
 303
 304         lov_pool_getref(pool);
 305
 306         CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
 307                pool->pool_name, pool->pool_obds.op_count);
 308         down_read(&pool_tgt_rw_sem(pool));
 309
 310         for (i = 0; i < pool_tgt_count(pool) ; i++) {
 311                 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
 312                         continue;
 313                 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
 314                        pool->pool_name, i,
 315                        obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
 316         }
 317
 318         up_read(&pool_tgt_rw_sem(pool));
 319         lov_pool_putref(pool);
 320 }
 321
 322 #define LOV_POOL_INIT_COUNT 2
 323 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
 324 {
 325         ENTRY;
 326
 327         if (count == 0)
 328                 count = LOV_POOL_INIT_COUNT;
 329         op->op_array = NULL;
 330         op->op_count = 0;
 331         init_rwsem(&op->op_rw_sem);
 332         op->op_size = count;
 333         OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
 334         if (op->op_array == NULL) {
 335                 op->op_size = 0;
 336                 RETURN(-ENOMEM);
 337         }
 338         EXIT;
 339         return 0;
 340 }
 341
 342 /* Caller must hold write op_rwlock */
 343 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
 344 {
 345         __u32 *new;
 346         int new_size;
 347
 348         LASSERT(min_count != 0);
 349
 350         if (op->op_count < op->op_size)
 351                 return 0;
 352
 353         new_size = max(min_count, 2 * op->op_size);
 354         OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
 355         if (new == NULL)
 356                 return -ENOMEM;
 357
 358         /* copy old array to new one */
 359         memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
 360         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
 361         op->op_array = new;
 362         op->op_size = new_size;
 363         return 0;
 364 }
 365
 366 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
 367 {
 368         int rc = 0, i;
 369         ENTRY;
 370
 371         down_write(&op->op_rw_sem);
 372
 373         rc = lov_ost_pool_extend(op, min_count);
 374         if (rc)
 375                 GOTO(out, rc);
 376
 377         /* search ost in pool array */
 378         for (i = 0; i < op->op_count; i++) {
 379                 if (op->op_array[i] == idx)
 380                         GOTO(out, rc = -EEXIST);
 381         }
 382         /* ost not found we add it */
 383         op->op_array[op->op_count] = idx;
 384         op->op_count++;
 385         EXIT;
 386 out:
 387         up_write(&op->op_rw_sem);
 388         return rc;
 389 }
 390
 391 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
 392 {
 393         int i;
 394         ENTRY;
 395
 396         down_write(&op->op_rw_sem);
 397
 398         for (i = 0; i < op->op_count; i++) {
 399                 if (op->op_array[i] == idx) {
 400                         memmove(&op->op_array[i], &op->op_array[i + 1],
 401                                 (op->op_count - i - 1) * sizeof(op->op_array[0]));
 402                         op->op_count--;
 403                         up_write(&op->op_rw_sem);
 404                         EXIT;
 405                         return 0;
 406                 }
 407         }
 408
 409         up_write(&op->op_rw_sem);
 410         RETURN(-EINVAL);
 411 }
 412
 413 int lov_ost_pool_free(struct ost_pool *op)
 414 {
 415         ENTRY;
 416
 417         if (op->op_size == 0)
 418                 RETURN(0);
 419
 420         down_write(&op->op_rw_sem);
 421
 422         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
 423         op->op_array = NULL;
 424         op->op_count = 0;
 425         op->op_size = 0;
 426
 427         up_write(&op->op_rw_sem);
 428         RETURN(0);
 429 }
 430
 431
 432 int lov_pool_new(struct obd_device *obd, char *poolname)
 433 {
 434         struct lov_obd *lov;
 435         struct pool_desc *new_pool;
 436         int rc;
 437         ENTRY;
 438
 439         lov = &(obd->u.lov);
 440
 441         if (strlen(poolname) > LOV_MAXPOOLNAME)
 442                 RETURN(-ENAMETOOLONG);
 443
 444         OBD_ALLOC_PTR(new_pool);
 445         if (new_pool == NULL)
 446                 RETURN(-ENOMEM);
 447
 448         strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
 449         new_pool->pool_lobd = obd;
 450         /* ref count init to 1 because when created a pool is always used
 451          * up to deletion
 452          */
 453         atomic_set(&new_pool->pool_refcount, 1);
 454         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
 455         if (rc)
 456                 GOTO(out_err, rc);
 457
 458         INIT_HLIST_NODE(&new_pool->pool_hash);
 459
 460 #ifdef CONFIG_PROC_FS
 461         /* get ref for /proc file */
 462         lov_pool_getref(new_pool);
 463         new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
 464                                                        poolname, new_pool,
 465                                                        &pool_proc_operations);
 466         if (IS_ERR(new_pool->pool_proc_entry)) {
 467                 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
 468                 new_pool->pool_proc_entry = NULL;
 469                 lov_pool_putref(new_pool);
 470         }
 471         CDEBUG(D_INFO, "pool %p - proc %p\n",
 472                new_pool, new_pool->pool_proc_entry);
 473 #endif
 474
 475         spin_lock(&obd->obd_dev_lock);
 476         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
 477         lov->lov_pool_count++;
 478         spin_unlock(&obd->obd_dev_lock);
 479
 480         /* add to find only when it fully ready  */
 481         rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
 482                                  &new_pool->pool_hash);
 483         if (rc)
 484                 GOTO(out_err, rc = -EEXIST);
 485
 486         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
 487                poolname, lov->lov_pool_count);
 488
 489         RETURN(0);
 490
 491 out_err:
 492         spin_lock(&obd->obd_dev_lock);
 493         list_del_init(&new_pool->pool_list);
 494         lov->lov_pool_count--;
 495         spin_unlock(&obd->obd_dev_lock);
 496         lprocfs_remove(&new_pool->pool_proc_entry);
 497         lov_ost_pool_free(&new_pool->pool_obds);
 498         OBD_FREE_PTR(new_pool);
 499
 500         return rc;
 501 }
 502
 503 int lov_pool_del(struct obd_device *obd, char *poolname)
 504 {
 505         struct lov_obd *lov;
 506         struct pool_desc *pool;
 507         ENTRY;
 508
 509         lov = &(obd->u.lov);
 510
 511         /* lookup and kill hash reference */
 512         pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
 513         if (pool == NULL)
 514                 RETURN(-ENOENT);
 515
 516         if (pool->pool_proc_entry != NULL) {
 517                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
 518                 lprocfs_remove(&pool->pool_proc_entry);
 519                 lov_pool_putref(pool);
 520         }
 521
 522         spin_lock(&obd->obd_dev_lock);
 523         list_del_init(&pool->pool_list);
 524         lov->lov_pool_count--;
 525         spin_unlock(&obd->obd_dev_lock);
 526
 527         /* release last reference */
 528         lov_pool_putref(pool);
 529
 530         RETURN(0);
 531 }
 532
 533
 534 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
 535 {
 536         struct obd_uuid ost_uuid;
 537         struct lov_obd *lov;
 538         struct pool_desc *pool;
 539         unsigned int lov_idx;
 540         int rc;
 541         ENTRY;
 542
 543         lov = &(obd->u.lov);
 544
 545         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 546         if (pool == NULL)
 547                 RETURN(-ENOENT);
 548
 549         obd_str2uuid(&ost_uuid, ostname);
 550
 551
 552         /* search ost in lov array */
 553         obd_getref(obd);
 554         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 555                 if (!lov->lov_tgts[lov_idx])
 556                         continue;
 557                 if (obd_uuid_equals(&ost_uuid,
 558                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 559                         break;
 560         }
 561         /* test if ost found in lov */
 562         if (lov_idx == lov->desc.ld_tgt_count)
 563                 GOTO(out, rc = -EINVAL);
 564
 565         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
 566         if (rc)
 567                 GOTO(out, rc);
 568
 569         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
 570                ostname, poolname,  pool_tgt_count(pool));
 571
 572         EXIT;
 573 out:
 574         obd_putref(obd);
 575         lov_pool_putref(pool);
 576         return rc;
 577 }
 578
 579 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
 580 {
 581         struct obd_uuid ost_uuid;
 582         struct lov_obd *lov;
 583         struct pool_desc *pool;
 584         unsigned int lov_idx;
 585         int rc = 0;
 586         ENTRY;
 587
 588         lov = &(obd->u.lov);
 589
 590         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
 591         if (pool == NULL)
 592                 RETURN(-ENOENT);
 593
 594         obd_str2uuid(&ost_uuid, ostname);
 595
 596         obd_getref(obd);
 597         /* search ost in lov array, to get index */
 598         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
 599                 if (!lov->lov_tgts[lov_idx])
 600                         continue;
 601
 602                 if (obd_uuid_equals(&ost_uuid,
 603                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
 604                         break;
 605         }
 606
 607         /* test if ost found in lov */
 608         if (lov_idx == lov->desc.ld_tgt_count)
 609                 GOTO(out, rc = -EINVAL);
 610
 611         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
 612
 613         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
 614                poolname);
 615
 616         EXIT;
 617 out:
 618         obd_putref(obd);
 619         lov_pool_putref(pool);
 620         return rc;
 621 }