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