lustre/lov/lov_pool.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see [sun.com URL with a
  20  * copy of GPLv2].
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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  */
  42
  43 #define DEBUG_SUBSYSTEM S_LOV
  44
  45 #ifdef __KERNEL__
  46 #include <libcfs/libcfs.h>
  47 #else
  48 #include <liblustre.h>
  49 #endif
  50
  51 #include <obd.h>
  52 #include "lov_internal.h"
  53
  54 /*
  55  * hash function using a Rotating Hash algorithm
  56  * Knuth, D. The Art of Computer Programming,
  57  * Volume 3: Sorting and Searching,
  58  * Chapter 6.4.
  59  * Addison Wesley, 1973
  60  */
  61 static __u32 pool_hashfn(lustre_hash_t *hash_body, void *key, unsigned mask)
  62 {
  63         int i;
  64         __u32 result;
  65         char *poolname;
  66
  67         result = 0;
  68         poolname = (char *)key;
  69         for (i = 0; i < MAXPOOLNAME; i++) {
  70                 if (poolname[i] == '\0')
  71                         break;
  72                 result = (result << 4)^(result >> 28) ^  poolname[i];
  73         }
  74         return (result % mask);
  75 }
  76
  77 static void *pool_key(struct hlist_node *hnode)
  78 {
  79         struct pool_desc *pool;
  80
  81         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
  82         return (pool->pool_name);
  83 }
  84
  85 static int pool_hashkey_compare(void *key, struct hlist_node *compared_hnode)
  86 {
  87         char *pool_name;
  88         struct pool_desc *pool;
  89         int rc;
  90
  91         pool_name = (char *)key;
  92         pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
  93         rc = strncmp(pool_name, pool->pool_name, MAXPOOLNAME);
  94         return (!rc);
  95 }
  96
  97 static void *pool_hashrefcount_get(struct hlist_node *hnode)
  98 {
  99         struct pool_desc *pool;
 100
 101         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 102         return (pool);
 103 }
 104
 105 static void *pool_hashrefcount_put(struct hlist_node *hnode)
 106 {
 107         struct pool_desc *pool;
 108
 109         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
 110         return (pool);
 111 }
 112
 113 lustre_hash_ops_t pool_hash_operations = {
 114         .lh_hash        = pool_hashfn,
 115         .lh_key         = pool_key,
 116         .lh_compare     = pool_hashkey_compare,
 117         .lh_get         = pool_hashrefcount_get,
 118         .lh_put         = pool_hashrefcount_put,
 119 };
 120
 121 #ifdef LPROCFS
 122 /* ifdef needed for liblustre support */
 123 /*
 124  * pool /proc seq_file methods
 125  */
 126 /*
 127  * iterator is used to go through the target pool entries
 128  * index is the current entry index in the lp_array[] array
 129  * index >= pos returned to the seq_file interface
 130  * pos is from 0 to (pool->pool_obds.op_count - 1)
 131  */
 132 #define POOL_IT_MAGIC 0xB001CEA0
 133 struct pool_iterator {
 134         int magic;
 135         struct pool_desc *pool;
 136         int idx;        /* from 0 to pool_tgt_size - 1 */
 137 };
 138
 139 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
 140 {
 141         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 142         int prev_idx;
 143
 144         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
 145
 146         /* test if end of file */
 147         if (*pos >= pool_tgt_count(iter->pool))
 148                 return NULL;
 149
 150         /* iterate to find a non empty entry */
 151         prev_idx = iter->idx;
 152         read_lock(&pool_tgt_rwlock(iter->pool));
 153         iter->idx++;
 154         if (iter->idx == pool_tgt_count(iter->pool)) {
 155                 iter->idx = prev_idx; /* we stay on the last entry */
 156                 read_unlock(&pool_tgt_rwlock(iter->pool));
 157                 return NULL;
 158         }
 159         read_unlock(&pool_tgt_rwlock(iter->pool));
 160         (*pos)++;
 161         /* return != NULL to continue */
 162         return iter;
 163 }
 164
 165 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
 166 {
 167         struct pool_desc *pool = (struct pool_desc *)s->private;
 168         struct pool_iterator *iter;
 169
 170         if ((pool_tgt_count(pool) == 0) ||
 171             (*pos >= pool_tgt_count(pool)))
 172                 return NULL;
 173
 174         OBD_ALLOC(iter, sizeof(struct pool_iterator));
 175         if (!iter)
 176                 return ERR_PTR(-ENOMEM);
 177         iter->magic = POOL_IT_MAGIC;
 178         iter->pool = pool;
 179         iter->idx = 0;
 180
 181         /* we use seq_file private field to memorized iterator so
 182          * we can free it at stop() */
 183         /* /!\ do not forget to restore it to pool before freeing it */
 184         s->private = iter;
 185         if (*pos > 0) {
 186                 loff_t i;
 187                 void *ptr;
 188
 189                 i = 0;
 190                 do {
 191                      ptr = pool_proc_next(s, &iter, &i);
 192                 } while ((i < *pos) && (ptr != NULL));
 193                 return ptr;
 194         }
 195         return iter;
 196 }
 197
 198 static void pool_proc_stop(struct seq_file *s, void *v)
 199 {
 200         struct pool_iterator *iter = (struct pool_iterator *)s->private;
 201
 202         /* in some cases stop() method is called 2 times, without
 203          * calling start() method (see seq_read() from fs/seq_file.c)
 204          * we have to free only if s->private is an iterator */
 205         if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
 206                 /* we restore s->private so next call to pool_proc_start()
 207                  * will work */
 208                 s->private = iter->pool;
 209                 OBD_FREE(iter, sizeof(struct pool_iterator));
 210         }
 211         return;
 212 }
 213
 214 static int pool_proc_show(struct seq_file *s, void *v)
 215 {
 216         struct pool_iterator *iter = (struct pool_iterator *)v;
 217         struct lov_tgt_desc *tgt;
 218
 219         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
 220         LASSERT(iter->pool != NULL);
 221         LASSERT(iter->idx <= pool_tgt_count(iter->pool));
 222
 223         read_lock(&pool_tgt_rwlock(iter->pool));
 224         tgt = pool_tgt(iter->pool, iter->idx);
 225         read_unlock(&pool_tgt_rwlock(iter->pool));
 226         if (tgt)
 227                 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
 228
 229         return 0;
 230 }
 231
 232 static struct seq_operations pool_proc_ops = {
 233         .start          = pool_proc_start,
 234         .next           = pool_proc_next,
 235         .stop           = pool_proc_stop,
 236         .show           = pool_proc_show,
 237 };
 238
 239 static int pool_proc_open(struct inode *inode, struct file *file)
 240 {
 241         int rc;
 242
 243         rc = seq_open(file, &pool_proc_ops);
 244         if (!rc) {
 245                 struct seq_file *s = file->private_data;
 246                 s->private = PROC_I(inode)->pde->data;
 247         }
 248         return rc;
 249 }
 250
 251 static struct file_operations pool_proc_operations = {
 252         .open           = pool_proc_open,
 253         .read           = seq_read,
 254         .llseek         = seq_lseek,
 255         .release        = seq_release,
 256 };
 257 #endif /* LPROCFS */
 258
 259 void lov_dump_pool(int level, struct pool_desc *pool)
 260 {
 261         int i;
 262
 263         CDEBUG(level, "pool "POOLNAMEF" has %d members\n",
 264                pool->pool_name, pool->pool_obds.op_count);
 265         read_lock(&pool_tgt_rwlock(pool));
 266         for (i = 0; i < pool_tgt_count(pool) ; i++) {
 267                 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
 268                         continue;
 269                 CDEBUG(level, "pool "POOLNAMEF"[%d] = %s\n", pool->pool_name,
 270                        i, obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
 271         }
 272         read_unlock(&pool_tgt_rwlock(pool));
 273 }
 274
 275 #define LOV_POOL_INIT_COUNT 2
 276 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
 277 {
 278         if (count == 0)
 279                 count = LOV_POOL_INIT_COUNT;
 280         op->op_array = NULL;
 281         op->op_count = 0;
 282         op->op_rwlock = RW_LOCK_UNLOCKED;
 283         op->op_size = count;
 284         OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
 285         if (op->op_array == NULL) {
 286                 op->op_size = 0;
 287                 return -ENOMEM;
 288         }
 289         return 0;
 290 }
 291
 292 /* Caller must hold write op_rwlock */
 293 int lov_ost_pool_extend(struct ost_pool *op, unsigned int max_count)
 294 {
 295         __u32 *new;
 296         int new_size;
 297
 298         LASSERT(max_count != 0);
 299
 300         if (op->op_count < op->op_size)
 301                 return 0;
 302
 303         new_size = min(max_count, 2 * op->op_size);
 304         OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
 305         if (new == NULL)
 306                 return -ENOMEM;
 307
 308         /* copy old array to new one */
 309         memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
 310         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
 311         op->op_array = new;
 312         op->op_size = new_size;
 313         return 0;
 314 }
 315
 316 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int max_count)
 317 {
 318         int rc = 0, i;
 319         ENTRY;
 320
 321         write_lock(&op->op_rwlock);
 322
 323         rc = lov_ost_pool_extend(op, max_count);
 324         if (rc)
 325                 GOTO(out, rc);
 326
 327         /* search ost in pool array */
 328         for (i = 0; i < op->op_count; i++) {
 329                 if (op->op_array[i] == idx)
 330                         GOTO(out, rc = -EEXIST);
 331         }
 332         /* ost not found we add it */
 333         op->op_array[op->op_count] = idx;
 334         op->op_count++;
 335 out:
 336         write_unlock(&op->op_rwlock);
 337         return rc;
 338 }
 339
 340 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
 341 {
 342         int i;
 343
 344         write_lock(&op->op_rwlock);
 345         for (i = 0; i < op->op_count; i++) {
 346                 if (op->op_array[i] == idx) {
 347                         memmove(&op->op_array[i], &op->op_array[i + 1],
 348                                 (op->op_count - i - 1) * sizeof(op->op_array[0]));
 349                         op->op_count--;
 350                         write_unlock(&op->op_rwlock);
 351                         return 0;
 352                 }
 353         }
 354         write_unlock(&op->op_rwlock);
 355         return -EINVAL;
 356 }
 357
 358 int lov_ost_pool_free(struct ost_pool *op)
 359 {
 360         if (op->op_size == 0)
 361                 return 0;
 362
 363         write_lock(&op->op_rwlock);
 364         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
 365         op->op_array = NULL;
 366         op->op_count = 0;
 367         op->op_size = 0;
 368         write_unlock(&op->op_rwlock);
 369         return 0;
 370 }
 371
 372
 373 int lov_pool_new(struct obd_device *obd, char *poolname)
 374 {
 375         struct lov_obd *lov;
 376         struct pool_desc *new_pool;
 377         int rc;
 378
 379         lov = &(obd->u.lov);
 380
 381         if (strlen(poolname) > MAXPOOLNAME)
 382                 return -ENAMETOOLONG;
 383
 384         OBD_ALLOC_PTR(new_pool);
 385         if (new_pool == NULL)
 386                 return -ENOMEM;
 387
 388         strncpy(new_pool->pool_name, poolname, MAXPOOLNAME);
 389         new_pool->pool_name[MAXPOOLNAME] = '\0';
 390         new_pool->pool_lov = lov;
 391         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
 392         if (rc)
 393                GOTO(out_err, rc);
 394
 395         memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
 396         rc = lov_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
 397         if (rc) {
 398                 lov_ost_pool_free(&new_pool->pool_obds);
 399                 GOTO(out_err, rc);
 400         }
 401
 402         spin_lock(&obd->obd_dev_lock);
 403         /* check if pool already exists */
 404         if (lustre_hash_lookup(lov->lov_pools_hash_body, poolname) != NULL) {
 405                 spin_unlock(&obd->obd_dev_lock);
 406                 lov_ost_pool_free(&new_pool->pool_rr.lqr_pool);
 407                 lov_ost_pool_free(&new_pool->pool_obds);
 408                 GOTO(out_err, rc = -EEXIST);
 409         }
 410
 411         INIT_HLIST_NODE(&new_pool->pool_hash);
 412         lustre_hash_add_unique(lov->lov_pools_hash_body, poolname,
 413                                &new_pool->pool_hash);
 414         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
 415         lov->lov_pool_count++;
 416         spin_unlock(&obd->obd_dev_lock);
 417
 418         CDEBUG(D_CONFIG, POOLNAMEF" is pool #%d\n",
 419                poolname, lov->lov_pool_count);
 420
 421 #ifdef LPROCFS
 422         /* ifdef needed for liblustre */
 423         new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
 424                                                        poolname, NULL, NULL,
 425                                                        new_pool,
 426                                                        &pool_proc_operations);
 427 #endif
 428
 429         if (IS_ERR(new_pool->pool_proc_entry)) {
 430                 CWARN("Cannot add proc pool entry "POOLNAMEF"\n", poolname);
 431                 new_pool->pool_proc_entry = NULL;
 432         }
 433
 434         return 0;
 435
 436 out_err:
 437         OBD_FREE_PTR(new_pool);
 438         return rc;
 439 }
 440
 441 int lov_pool_del(struct obd_device *obd, char *poolname)
 442 {
 443         struct lov_obd *lov;
 444         struct pool_desc *pool;
 445
 446         lov = &(obd->u.lov);
 447
 448         spin_lock(&obd->obd_dev_lock);
 449         pool = lustre_hash_lookup(lov->lov_pools_hash_body,
 450                                              poolname);
 451         if (pool == NULL) {
 452                 spin_unlock(&obd->obd_dev_lock);
 453                 return -ENOENT;
 454         }
 455
 456 #ifdef LPROCFS
 457         if (pool->pool_proc_entry != NULL)
 458                 remove_proc_entry(pool->pool_proc_entry->name,
 459                                   pool->pool_proc_entry->parent);
 460 #endif
 461
 462         lustre_hash_del_key(lov->lov_pools_hash_body, poolname);
 463
 464         lov->lov_pool_count--;
 465
 466         spin_unlock(&obd->obd_dev_lock);
 467
 468         /* pool struct is not freed because it may be used by
 469          * some open in /proc. The struct is freed at lov_cleanup()
 470         */
 471         return 0;
 472 }
 473
 474
 475 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
 476 {
 477         struct obd_uuid ost_uuid;
 478         struct lov_obd *lov;
 479         struct pool_desc *pool;
 480         unsigned int i, lov_idx;
 481         int rc;
 482
 483         lov = &(obd->u.lov);
 484
 485         pool = lustre_hash_lookup(lov->lov_pools_hash_body, poolname);
 486         if (pool == NULL)
 487                 return -ENOENT;
 488
 489         obd_str2uuid(&ost_uuid, ostname);
 490
 491
 492         /* search ost in lov array */
 493         mutex_down(&lov->lov_lock);
 494         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 495                 if (!lov->lov_tgts[i])
 496                         continue;
 497                 if (obd_uuid_equals(&ost_uuid, &(lov->lov_tgts[i]->ltd_uuid)))
 498                         break;
 499         }
 500
 501         /* test if ost found in lov */
 502         if (i == lov->desc.ld_tgt_count) {
 503                 mutex_up(&lov->lov_lock);
 504                 return -EINVAL;
 505         }
 506         mutex_up(&lov->lov_lock);
 507
 508         lov_idx = i;
 509
 510         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
 511         if (rc)
 512                 return rc;
 513
 514         pool->pool_rr.lqr_dirty = 1;
 515
 516         CDEBUG(D_CONFIG, "Added %s to "POOLNAMEF" as member %d\n",
 517                ostname, poolname,  pool_tgt_count(pool));
 518         return 0;
 519 }
 520
 521 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
 522 {
 523         struct obd_uuid ost_uuid;
 524         struct lov_obd *lov;
 525         struct pool_desc *pool;
 526         unsigned int i, lov_idx;
 527
 528         lov = &(obd->u.lov);
 529
 530         spin_lock(&obd->obd_dev_lock);
 531         pool = lustre_hash_lookup(lov->lov_pools_hash_body, poolname);
 532         if (pool == NULL) {
 533                 spin_unlock(&obd->obd_dev_lock);
 534                 return -ENOENT;
 535         }
 536
 537         obd_str2uuid(&ost_uuid, ostname);
 538
 539         /* search ost in lov array, to get index */
 540         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 541                 if (!lov->lov_tgts[i])
 542                         continue;
 543
 544                 if (obd_uuid_equals(&ost_uuid, &(lov->lov_tgts[i]->ltd_uuid)))
 545                         break;
 546         }
 547
 548         /* test if ost found in lov */
 549         if (i == lov->desc.ld_tgt_count) {
 550                 spin_unlock(&obd->obd_dev_lock);
 551                 return -EINVAL;
 552         }
 553
 554         spin_unlock(&obd->obd_dev_lock);
 555
 556         lov_idx = i;
 557
 558         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
 559
 560         pool->pool_rr.lqr_dirty = 1;
 561
 562         CDEBUG(D_CONFIG, "%s removed from "POOLNAMEF"\n", ostname, poolname);
 563
 564         return 0;
 565 }
 566
 567 int lov_check_index_in_pool(__u32 idx, struct pool_desc *pool)
 568 {
 569         int i;
 570
 571         read_lock(&pool_tgt_rwlock(pool));
 572         for (i = 0; i < pool_tgt_count(pool); i++) {
 573                 if (pool_tgt_array(pool)[i] == idx) {
 574                         read_unlock(&pool_tgt_rwlock(pool));
 575                         return 0;
 576                 }
 577         }
 578         read_unlock(&pool_tgt_rwlock(pool));
 579         return -ENOENT;
 580 }
 581
 582 struct pool_desc *lov_find_pool(struct lov_obd *lov, char *poolname)
 583 {
 584         struct pool_desc *pool;
 585
 586         pool = NULL;
 587         if (poolname[0] != '\0') {
 588                 pool = lustre_hash_lookup(lov->lov_pools_hash_body, poolname);
 589                 if (pool == NULL)
 590                         CWARN("Request for an unknown pool ("POOLNAMEF")\n",
 591                               poolname);
 592                 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
 593                         CWARN("Request for an empty pool ("POOLNAMEF")\n",
 594                                poolname);
 595                         pool = NULL;
 596                 }
 597         }
 598         return pool;
 599 }
 600