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