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
47 #include "../../include/linux/libcfs/libcfs.h"
49 #include "../include/obd.h"
50 #include "lov_internal.h"
52 #define pool_tgt(_p, _i) \
53 _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
55 static void lov_pool_getref(struct pool_desc *pool)
57 CDEBUG(D_INFO, "pool %p\n", pool);
58 atomic_inc(&pool->pool_refcount);
61 void lov_pool_putref(struct pool_desc *pool)
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_debugfs_entry);
68 lov_ost_pool_free(&(pool->pool_obds));
73 static void lov_pool_putref_locked(struct pool_desc *pool)
75 CDEBUG(D_INFO, "pool %p\n", pool);
76 LASSERT(atomic_read(&pool->pool_refcount) > 1);
78 atomic_dec(&pool->pool_refcount);
82 * hash function using a Rotating Hash algorithm
83 * Knuth, D. The Art of Computer Programming,
84 * Volume 3: Sorting and Searching,
86 * Addison Wesley, 1973
88 static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key, unsigned mask)
95 poolname = (char *)key;
96 for (i = 0; i < LOV_MAXPOOLNAME; i++) {
97 if (poolname[i] == '\0')
99 result = (result << 4)^(result >> 28) ^ poolname[i];
101 return (result % mask);
104 static void *pool_key(struct hlist_node *hnode)
106 struct pool_desc *pool;
108 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
109 return pool->pool_name;
112 static int pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
115 struct pool_desc *pool;
117 pool_name = (char *)key;
118 pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
119 return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
122 static void *pool_hashobject(struct hlist_node *hnode)
124 return hlist_entry(hnode, struct pool_desc, pool_hash);
127 static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
129 struct pool_desc *pool;
131 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
132 lov_pool_getref(pool);
135 static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
136 struct hlist_node *hnode)
138 struct pool_desc *pool;
140 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
141 lov_pool_putref_locked(pool);
144 struct cfs_hash_ops pool_hash_operations = {
145 .hs_hash = pool_hashfn,
147 .hs_keycmp = pool_hashkey_keycmp,
148 .hs_object = pool_hashobject,
149 .hs_get = pool_hashrefcount_get,
150 .hs_put_locked = pool_hashrefcount_put_locked,
155 * pool debugfs seq_file methods
158 * iterator is used to go through the target pool entries
159 * index is the current entry index in the lp_array[] array
160 * index >= pos returned to the seq_file interface
161 * pos is from 0 to (pool->pool_obds.op_count - 1)
163 #define POOL_IT_MAGIC 0xB001CEA0
164 struct pool_iterator {
166 struct pool_desc *pool;
167 int idx; /* from 0 to pool_tgt_size - 1 */
170 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
172 struct pool_iterator *iter = (struct pool_iterator *)s->private;
175 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
177 /* test if end of file */
178 if (*pos >= pool_tgt_count(iter->pool))
181 /* iterate to find a non empty entry */
182 prev_idx = iter->idx;
183 down_read(&pool_tgt_rw_sem(iter->pool));
185 if (iter->idx == pool_tgt_count(iter->pool)) {
186 iter->idx = prev_idx; /* we stay on the last entry */
187 up_read(&pool_tgt_rw_sem(iter->pool));
190 up_read(&pool_tgt_rw_sem(iter->pool));
192 /* return != NULL to continue */
196 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
198 struct pool_desc *pool = (struct pool_desc *)s->private;
199 struct pool_iterator *iter;
201 lov_pool_getref(pool);
202 if ((pool_tgt_count(pool) == 0) ||
203 (*pos >= pool_tgt_count(pool))) {
204 /* iter is not created, so stop() has no way to
205 * find pool to dec ref
207 lov_pool_putref(pool);
211 iter = kzalloc(sizeof(*iter), GFP_NOFS);
213 return ERR_PTR(-ENOMEM);
214 iter->magic = POOL_IT_MAGIC;
218 /* we use seq_file private field to memorized iterator so
219 * we can free it at stop()
221 /* /!\ do not forget to restore it to pool before freeing it */
229 ptr = pool_proc_next(s, &iter, &i);
230 } while ((i < *pos) && ptr);
236 static void pool_proc_stop(struct seq_file *s, void *v)
238 struct pool_iterator *iter = (struct pool_iterator *)s->private;
240 /* in some cases stop() method is called 2 times, without
241 * calling start() method (see seq_read() from fs/seq_file.c)
242 * we have to free only if s->private is an iterator
244 if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
245 /* we restore s->private so next call to pool_proc_start()
248 s->private = iter->pool;
249 lov_pool_putref(iter->pool);
254 static int pool_proc_show(struct seq_file *s, void *v)
256 struct pool_iterator *iter = (struct pool_iterator *)v;
257 struct lov_tgt_desc *tgt;
259 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
261 LASSERT(iter->idx <= pool_tgt_count(iter->pool));
263 down_read(&pool_tgt_rw_sem(iter->pool));
264 tgt = pool_tgt(iter->pool, iter->idx);
265 up_read(&pool_tgt_rw_sem(iter->pool));
267 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
272 static const struct seq_operations pool_proc_ops = {
273 .start = pool_proc_start,
274 .next = pool_proc_next,
275 .stop = pool_proc_stop,
276 .show = pool_proc_show,
279 static int pool_proc_open(struct inode *inode, struct file *file)
283 rc = seq_open(file, &pool_proc_ops);
285 struct seq_file *s = file->private_data;
287 s->private = inode->i_private;
292 static struct file_operations pool_proc_operations = {
293 .open = pool_proc_open,
296 .release = seq_release,
299 #define LOV_POOL_INIT_COUNT 2
300 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
303 count = LOV_POOL_INIT_COUNT;
306 init_rwsem(&op->op_rw_sem);
308 op->op_array = kcalloc(op->op_size, sizeof(op->op_array[0]), GFP_NOFS);
316 /* Caller must hold write op_rwlock */
317 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
322 LASSERT(min_count != 0);
324 if (op->op_count < op->op_size)
327 new_size = max(min_count, 2 * op->op_size);
328 new = kcalloc(new_size, sizeof(op->op_array[0]), GFP_NOFS);
332 /* copy old array to new one */
333 memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
336 op->op_size = new_size;
340 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
344 down_write(&op->op_rw_sem);
346 rc = lov_ost_pool_extend(op, min_count);
350 /* search ost in pool array */
351 for (i = 0; i < op->op_count; i++) {
352 if (op->op_array[i] == idx) {
357 /* ost not found we add it */
358 op->op_array[op->op_count] = idx;
361 up_write(&op->op_rw_sem);
365 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
369 down_write(&op->op_rw_sem);
371 for (i = 0; i < op->op_count; i++) {
372 if (op->op_array[i] == idx) {
373 memmove(&op->op_array[i], &op->op_array[i + 1],
374 (op->op_count - i - 1) * sizeof(op->op_array[0]));
376 up_write(&op->op_rw_sem);
381 up_write(&op->op_rw_sem);
385 int lov_ost_pool_free(struct ost_pool *op)
387 if (op->op_size == 0)
390 down_write(&op->op_rw_sem);
397 up_write(&op->op_rw_sem);
401 int lov_pool_new(struct obd_device *obd, char *poolname)
404 struct pool_desc *new_pool;
409 if (strlen(poolname) > LOV_MAXPOOLNAME)
410 return -ENAMETOOLONG;
412 new_pool = kzalloc(sizeof(*new_pool), GFP_NOFS);
416 strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
417 new_pool->pool_lobd = obd;
418 /* ref count init to 1 because when created a pool is always used
421 atomic_set(&new_pool->pool_refcount, 1);
422 rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
426 INIT_HLIST_NODE(&new_pool->pool_hash);
428 /* get ref for debugfs file */
429 lov_pool_getref(new_pool);
430 new_pool->pool_debugfs_entry = ldebugfs_add_simple(
431 lov->lov_pool_debugfs_entry,
433 &pool_proc_operations);
434 if (IS_ERR_OR_NULL(new_pool->pool_debugfs_entry)) {
435 CWARN("Cannot add debugfs pool entry "LOV_POOLNAMEF"\n",
437 new_pool->pool_debugfs_entry = NULL;
438 lov_pool_putref(new_pool);
440 CDEBUG(D_INFO, "pool %p - proc %p\n",
441 new_pool, new_pool->pool_debugfs_entry);
443 spin_lock(&obd->obd_dev_lock);
444 list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
445 lov->lov_pool_count++;
446 spin_unlock(&obd->obd_dev_lock);
448 /* add to find only when it fully ready */
449 rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
450 &new_pool->pool_hash);
456 CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
457 poolname, lov->lov_pool_count);
462 spin_lock(&obd->obd_dev_lock);
463 list_del_init(&new_pool->pool_list);
464 lov->lov_pool_count--;
465 spin_unlock(&obd->obd_dev_lock);
466 ldebugfs_remove(&new_pool->pool_debugfs_entry);
467 lov_ost_pool_free(&new_pool->pool_obds);
473 int lov_pool_del(struct obd_device *obd, char *poolname)
476 struct pool_desc *pool;
480 /* lookup and kill hash reference */
481 pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
485 if (!IS_ERR_OR_NULL(pool->pool_debugfs_entry)) {
486 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_debugfs_entry);
487 ldebugfs_remove(&pool->pool_debugfs_entry);
488 lov_pool_putref(pool);
491 spin_lock(&obd->obd_dev_lock);
492 list_del_init(&pool->pool_list);
493 lov->lov_pool_count--;
494 spin_unlock(&obd->obd_dev_lock);
496 /* release last reference */
497 lov_pool_putref(pool);
502 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
504 struct obd_uuid ost_uuid;
506 struct pool_desc *pool;
507 unsigned int lov_idx;
512 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
516 obd_str2uuid(&ost_uuid, ostname);
518 /* search ost in lov array */
520 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
521 if (!lov->lov_tgts[lov_idx])
523 if (obd_uuid_equals(&ost_uuid,
524 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
527 /* test if ost found in lov */
528 if (lov_idx == lov->desc.ld_tgt_count) {
533 rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
537 CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
538 ostname, poolname, pool_tgt_count(pool));
542 lov_pool_putref(pool);
546 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
548 struct obd_uuid ost_uuid;
550 struct pool_desc *pool;
551 unsigned int lov_idx;
556 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
560 obd_str2uuid(&ost_uuid, ostname);
563 /* search ost in lov array, to get index */
564 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
565 if (!lov->lov_tgts[lov_idx])
568 if (obd_uuid_equals(&ost_uuid,
569 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
573 /* test if ost found in lov */
574 if (lov_idx == lov->desc.ld_tgt_count) {
579 lov_ost_pool_remove(&pool->pool_obds, lov_idx);
581 CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
586 lov_pool_putref(pool);