4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_pool.c
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>
45 #define DEBUG_SUBSYSTEM S_LOV
48 #include <libcfs/libcfs.h>
50 #include <liblustre.h>
54 #include "lov_internal.h"
56 #define pool_tgt(_p, _i) \
57 _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
59 static void lov_pool_getref(struct pool_desc *pool)
61 CDEBUG(D_INFO, "pool %p\n", pool);
62 cfs_atomic_inc(&pool->pool_refcount);
65 void lov_pool_putref(struct pool_desc *pool)
67 CDEBUG(D_INFO, "pool %p\n", pool);
68 if (cfs_atomic_dec_and_test(&pool->pool_refcount)) {
69 LASSERT(cfs_hlist_unhashed(&pool->pool_hash));
70 LASSERT(cfs_list_empty(&pool->pool_list));
71 LASSERT(pool->pool_proc_entry == NULL);
72 lov_ost_pool_free(&(pool->pool_obds));
78 void lov_pool_putref_locked(struct pool_desc *pool)
80 CDEBUG(D_INFO, "pool %p\n", pool);
81 LASSERT(cfs_atomic_read(&pool->pool_refcount) > 1);
83 cfs_atomic_dec(&pool->pool_refcount);
87 * hash function using a Rotating Hash algorithm
88 * Knuth, D. The Art of Computer Programming,
89 * Volume 3: Sorting and Searching,
91 * Addison Wesley, 1973
93 static __u32 pool_hashfn(cfs_hash_t *hash_body, const void *key, unsigned mask)
100 poolname = (char *)key;
101 for (i = 0; i < LOV_MAXPOOLNAME; i++) {
102 if (poolname[i] == '\0')
104 result = (result << 4)^(result >> 28) ^ poolname[i];
106 return (result % mask);
109 static void *pool_key(cfs_hlist_node_t *hnode)
111 struct pool_desc *pool;
113 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
114 return (pool->pool_name);
117 static int pool_hashkey_keycmp(const void *key, cfs_hlist_node_t *compared_hnode)
120 struct pool_desc *pool;
122 pool_name = (char *)key;
123 pool = cfs_hlist_entry(compared_hnode, struct pool_desc, pool_hash);
124 return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
127 static void *pool_hashobject(cfs_hlist_node_t *hnode)
129 return cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
132 static void pool_hashrefcount_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
134 struct pool_desc *pool;
136 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
137 lov_pool_getref(pool);
140 static void pool_hashrefcount_put_locked(cfs_hash_t *hs,
141 cfs_hlist_node_t *hnode)
143 struct pool_desc *pool;
145 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
146 lov_pool_putref_locked(pool);
149 cfs_hash_ops_t pool_hash_operations = {
150 .hs_hash = pool_hashfn,
152 .hs_keycmp = pool_hashkey_keycmp,
153 .hs_object = pool_hashobject,
154 .hs_get = pool_hashrefcount_get,
155 .hs_put_locked = pool_hashrefcount_put_locked,
160 /* ifdef needed for liblustre support */
162 * pool /proc seq_file methods
165 * iterator is used to go through the target pool entries
166 * index is the current entry index in the lp_array[] array
167 * index >= pos returned to the seq_file interface
168 * pos is from 0 to (pool->pool_obds.op_count - 1)
170 #define POOL_IT_MAGIC 0xB001CEA0
171 struct pool_iterator {
173 struct pool_desc *pool;
174 int idx; /* from 0 to pool_tgt_size - 1 */
177 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
179 struct pool_iterator *iter = (struct pool_iterator *)s->private;
182 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
184 /* test if end of file */
185 if (*pos >= pool_tgt_count(iter->pool))
188 /* iterate to find a non empty entry */
189 prev_idx = iter->idx;
190 down_read(&pool_tgt_rw_sem(iter->pool));
192 if (iter->idx == pool_tgt_count(iter->pool)) {
193 iter->idx = prev_idx; /* we stay on the last entry */
194 up_read(&pool_tgt_rw_sem(iter->pool));
197 up_read(&pool_tgt_rw_sem(iter->pool));
199 /* return != NULL to continue */
203 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
205 struct pool_desc *pool = (struct pool_desc *)s->private;
206 struct pool_iterator *iter;
208 lov_pool_getref(pool);
209 if ((pool_tgt_count(pool) == 0) ||
210 (*pos >= pool_tgt_count(pool))) {
211 /* iter is not created, so stop() has no way to
212 * find pool to dec ref */
213 lov_pool_putref(pool);
219 return ERR_PTR(-ENOMEM);
220 iter->magic = POOL_IT_MAGIC;
224 /* we use seq_file private field to memorized iterator so
225 * we can free it at stop() */
226 /* /!\ do not forget to restore it to pool before freeing it */
234 ptr = pool_proc_next(s, &iter, &i);
235 } while ((i < *pos) && (ptr != NULL));
241 static void pool_proc_stop(struct seq_file *s, void *v)
243 struct pool_iterator *iter = (struct pool_iterator *)s->private;
245 /* in some cases stop() method is called 2 times, without
246 * calling start() method (see seq_read() from fs/seq_file.c)
247 * we have to free only if s->private is an iterator */
248 if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
249 /* we restore s->private so next call to pool_proc_start()
251 s->private = iter->pool;
252 lov_pool_putref(iter->pool);
258 static int pool_proc_show(struct seq_file *s, void *v)
260 struct pool_iterator *iter = (struct pool_iterator *)v;
261 struct lov_tgt_desc *tgt;
263 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
264 LASSERT(iter->pool != NULL);
265 LASSERT(iter->idx <= pool_tgt_count(iter->pool));
267 down_read(&pool_tgt_rw_sem(iter->pool));
268 tgt = pool_tgt(iter->pool, iter->idx);
269 up_read(&pool_tgt_rw_sem(iter->pool));
271 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
276 static struct seq_operations pool_proc_ops = {
277 .start = pool_proc_start,
278 .next = pool_proc_next,
279 .stop = pool_proc_stop,
280 .show = pool_proc_show,
283 static int pool_proc_open(struct inode *inode, struct file *file)
287 rc = seq_open(file, &pool_proc_ops);
289 struct seq_file *s = file->private_data;
290 s->private = PDE_DATA(inode);
295 static struct file_operations pool_proc_operations = {
296 .open = pool_proc_open,
299 .release = seq_release,
303 void lov_dump_pool(int level, struct pool_desc *pool)
307 lov_pool_getref(pool);
309 CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
310 pool->pool_name, pool->pool_obds.op_count);
311 down_read(&pool_tgt_rw_sem(pool));
313 for (i = 0; i < pool_tgt_count(pool) ; i++) {
314 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
316 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
318 obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
321 up_read(&pool_tgt_rw_sem(pool));
322 lov_pool_putref(pool);
325 #define LOV_POOL_INIT_COUNT 2
326 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
331 count = LOV_POOL_INIT_COUNT;
334 init_rwsem(&op->op_rw_sem);
336 OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
337 if (op->op_array == NULL) {
345 /* Caller must hold write op_rwlock */
346 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
351 LASSERT(min_count != 0);
353 if (op->op_count < op->op_size)
356 new_size = max(min_count, 2 * op->op_size);
357 OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
361 /* copy old array to new one */
362 memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
363 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
365 op->op_size = new_size;
369 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
374 down_write(&op->op_rw_sem);
376 rc = lov_ost_pool_extend(op, min_count);
380 /* search ost in pool array */
381 for (i = 0; i < op->op_count; i++) {
382 if (op->op_array[i] == idx)
383 GOTO(out, rc = -EEXIST);
385 /* ost not found we add it */
386 op->op_array[op->op_count] = idx;
390 up_write(&op->op_rw_sem);
394 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
399 down_write(&op->op_rw_sem);
401 for (i = 0; i < op->op_count; i++) {
402 if (op->op_array[i] == idx) {
403 memmove(&op->op_array[i], &op->op_array[i + 1],
404 (op->op_count - i - 1) * sizeof(op->op_array[0]));
406 up_write(&op->op_rw_sem);
412 up_write(&op->op_rw_sem);
416 int lov_ost_pool_free(struct ost_pool *op)
420 if (op->op_size == 0)
423 down_write(&op->op_rw_sem);
425 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
430 up_write(&op->op_rw_sem);
435 int lov_pool_new(struct obd_device *obd, char *poolname)
438 struct pool_desc *new_pool;
444 if (strlen(poolname) > LOV_MAXPOOLNAME)
445 RETURN(-ENAMETOOLONG);
447 OBD_ALLOC_PTR(new_pool);
448 if (new_pool == NULL)
451 strncpy(new_pool->pool_name, poolname, LOV_MAXPOOLNAME);
452 new_pool->pool_name[LOV_MAXPOOLNAME] = '\0';
453 new_pool->pool_lobd = obd;
454 /* ref count init to 1 because when created a pool is always used
457 cfs_atomic_set(&new_pool->pool_refcount, 1);
458 rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
462 CFS_INIT_HLIST_NODE(&new_pool->pool_hash);
465 /* we need this assert seq_file is not implementated for liblustre */
466 /* get ref for /proc file */
467 lov_pool_getref(new_pool);
468 new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
470 #ifndef HAVE_ONLY_PROCFS_SEQ
474 &pool_proc_operations);
475 if (IS_ERR(new_pool->pool_proc_entry)) {
476 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
477 new_pool->pool_proc_entry = NULL;
478 lov_pool_putref(new_pool);
480 CDEBUG(D_INFO, "pool %p - proc %p\n", new_pool, new_pool->pool_proc_entry);
483 spin_lock(&obd->obd_dev_lock);
484 cfs_list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
485 lov->lov_pool_count++;
486 spin_unlock(&obd->obd_dev_lock);
488 /* add to find only when it fully ready */
489 rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
490 &new_pool->pool_hash);
492 GOTO(out_err, rc = -EEXIST);
494 CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
495 poolname, lov->lov_pool_count);
500 spin_lock(&obd->obd_dev_lock);
501 cfs_list_del_init(&new_pool->pool_list);
502 lov->lov_pool_count--;
503 spin_unlock(&obd->obd_dev_lock);
504 lprocfs_remove(&new_pool->pool_proc_entry);
505 lov_ost_pool_free(&new_pool->pool_obds);
506 OBD_FREE_PTR(new_pool);
511 int lov_pool_del(struct obd_device *obd, char *poolname)
514 struct pool_desc *pool;
519 /* lookup and kill hash reference */
520 pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
524 if (pool->pool_proc_entry != NULL) {
525 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
526 lprocfs_remove(&pool->pool_proc_entry);
527 lov_pool_putref(pool);
530 spin_lock(&obd->obd_dev_lock);
531 cfs_list_del_init(&pool->pool_list);
532 lov->lov_pool_count--;
533 spin_unlock(&obd->obd_dev_lock);
535 /* release last reference */
536 lov_pool_putref(pool);
542 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
544 struct obd_uuid ost_uuid;
546 struct pool_desc *pool;
547 unsigned int lov_idx;
553 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
557 obd_str2uuid(&ost_uuid, ostname);
560 /* search ost in lov array */
562 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
563 if (!lov->lov_tgts[lov_idx])
565 if (obd_uuid_equals(&ost_uuid,
566 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
569 /* test if ost found in lov */
570 if (lov_idx == lov->desc.ld_tgt_count)
571 GOTO(out, rc = -EINVAL);
573 rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
577 CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
578 ostname, poolname, pool_tgt_count(pool));
583 lov_pool_putref(pool);
587 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
589 struct obd_uuid ost_uuid;
591 struct pool_desc *pool;
592 unsigned int lov_idx;
598 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
602 obd_str2uuid(&ost_uuid, ostname);
605 /* search ost in lov array, to get index */
606 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
607 if (!lov->lov_tgts[lov_idx])
610 if (obd_uuid_equals(&ost_uuid,
611 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
615 /* test if ost found in lov */
616 if (lov_idx == lov->desc.ld_tgt_count)
617 GOTO(out, rc = -EINVAL);
619 lov_ost_pool_remove(&pool->pool_obds, lov_idx);
621 CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
627 lov_pool_putref(pool);
631 int lov_check_index_in_pool(__u32 idx, struct pool_desc *pool)
636 /* caller may no have a ref on pool if it got the pool
637 * without calling lov_find_pool() (e.g. go through the lov pool
640 lov_pool_getref(pool);
642 down_read(&pool_tgt_rw_sem(pool));
644 for (i = 0; i < pool_tgt_count(pool); i++) {
645 if (pool_tgt_array(pool)[i] == idx)
651 up_read(&pool_tgt_rw_sem(pool));
653 lov_pool_putref(pool);
657 struct pool_desc *lov_find_pool(struct lov_obd *lov, char *poolname)
659 struct pool_desc *pool;
662 if (poolname[0] != '\0') {
663 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
665 CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
667 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
668 CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
670 /* pool is ignored, so we remove ref on it */
671 lov_pool_putref(pool);