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 == NULL);
68 lov_ost_pool_free(&(pool->pool_rr.lqr_pool));
69 lov_ost_pool_free(&(pool->pool_obds));
74 static void lov_pool_putref_locked(struct pool_desc *pool)
76 CDEBUG(D_INFO, "pool %p\n", pool);
77 LASSERT(atomic_read(&pool->pool_refcount) > 1);
79 atomic_dec(&pool->pool_refcount);
83 * hash function using a Rotating Hash algorithm
84 * Knuth, D. The Art of Computer Programming,
85 * Volume 3: Sorting and Searching,
87 * Addison Wesley, 1973
89 static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key, unsigned mask)
96 poolname = (char *)key;
97 for (i = 0; i < LOV_MAXPOOLNAME; i++) {
98 if (poolname[i] == '\0')
100 result = (result << 4)^(result >> 28) ^ poolname[i];
102 return (result % mask);
105 static void *pool_key(struct hlist_node *hnode)
107 struct pool_desc *pool;
109 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
110 return pool->pool_name;
113 static int pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
116 struct pool_desc *pool;
118 pool_name = (char *)key;
119 pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
120 return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
123 static void *pool_hashobject(struct hlist_node *hnode)
125 return hlist_entry(hnode, struct pool_desc, pool_hash);
128 static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
130 struct pool_desc *pool;
132 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
133 lov_pool_getref(pool);
136 static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
137 struct hlist_node *hnode)
139 struct pool_desc *pool;
141 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
142 lov_pool_putref_locked(pool);
145 struct cfs_hash_ops pool_hash_operations = {
146 .hs_hash = pool_hashfn,
148 .hs_keycmp = pool_hashkey_keycmp,
149 .hs_object = pool_hashobject,
150 .hs_get = pool_hashrefcount_get,
151 .hs_put_locked = pool_hashrefcount_put_locked,
155 /* ifdef needed for liblustre support */
157 * pool /proc seq_file methods
160 * iterator is used to go through the target pool entries
161 * index is the current entry index in the lp_array[] array
162 * index >= pos returned to the seq_file interface
163 * pos is from 0 to (pool->pool_obds.op_count - 1)
165 #define POOL_IT_MAGIC 0xB001CEA0
166 struct pool_iterator {
168 struct pool_desc *pool;
169 int idx; /* from 0 to pool_tgt_size - 1 */
172 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
174 struct pool_iterator *iter = (struct pool_iterator *)s->private;
177 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
179 /* test if end of file */
180 if (*pos >= pool_tgt_count(iter->pool))
183 /* iterate to find a non empty entry */
184 prev_idx = iter->idx;
185 down_read(&pool_tgt_rw_sem(iter->pool));
187 if (iter->idx == pool_tgt_count(iter->pool)) {
188 iter->idx = prev_idx; /* we stay on the last entry */
189 up_read(&pool_tgt_rw_sem(iter->pool));
192 up_read(&pool_tgt_rw_sem(iter->pool));
194 /* return != NULL to continue */
198 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
200 struct pool_desc *pool = (struct pool_desc *)s->private;
201 struct pool_iterator *iter;
203 lov_pool_getref(pool);
204 if ((pool_tgt_count(pool) == 0) ||
205 (*pos >= pool_tgt_count(pool))) {
206 /* iter is not created, so stop() has no way to
207 * find pool to dec ref */
208 lov_pool_putref(pool);
212 iter = kzalloc(sizeof(*iter), GFP_NOFS);
214 return ERR_PTR(-ENOMEM);
215 iter->magic = POOL_IT_MAGIC;
219 /* we use seq_file private field to memorized iterator so
220 * 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 != NULL));
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 */
243 if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
244 /* we restore s->private so next call to pool_proc_start()
246 s->private = iter->pool;
247 lov_pool_putref(iter->pool);
253 static int pool_proc_show(struct seq_file *s, void *v)
255 struct pool_iterator *iter = (struct pool_iterator *)v;
256 struct lov_tgt_desc *tgt;
258 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
259 LASSERT(iter->pool != NULL);
260 LASSERT(iter->idx <= pool_tgt_count(iter->pool));
262 down_read(&pool_tgt_rw_sem(iter->pool));
263 tgt = pool_tgt(iter->pool, iter->idx);
264 up_read(&pool_tgt_rw_sem(iter->pool));
266 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
271 static const struct seq_operations pool_proc_ops = {
272 .start = pool_proc_start,
273 .next = pool_proc_next,
274 .stop = pool_proc_stop,
275 .show = pool_proc_show,
278 static int pool_proc_open(struct inode *inode, struct file *file)
282 rc = seq_open(file, &pool_proc_ops);
284 struct seq_file *s = file->private_data;
286 s->private = inode->i_private;
291 static struct file_operations pool_proc_operations = {
292 .open = pool_proc_open,
295 .release = seq_release,
298 #define LOV_POOL_INIT_COUNT 2
299 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
302 count = LOV_POOL_INIT_COUNT;
305 init_rwsem(&op->op_rw_sem);
307 op->op_array = kcalloc(op->op_size, sizeof(op->op_array[0]), GFP_NOFS);
308 if (op->op_array == NULL) {
315 /* Caller must hold write op_rwlock */
316 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
321 LASSERT(min_count != 0);
323 if (op->op_count < op->op_size)
326 new_size = max(min_count, 2 * op->op_size);
327 new = kcalloc(new_size, sizeof(op->op_array[0]), GFP_NOFS);
331 /* copy old array to new one */
332 memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
335 op->op_size = new_size;
339 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
343 down_write(&op->op_rw_sem);
345 rc = lov_ost_pool_extend(op, min_count);
349 /* search ost in pool array */
350 for (i = 0; i < op->op_count; i++) {
351 if (op->op_array[i] == idx) {
356 /* ost not found we add it */
357 op->op_array[op->op_count] = idx;
360 up_write(&op->op_rw_sem);
364 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
368 down_write(&op->op_rw_sem);
370 for (i = 0; i < op->op_count; i++) {
371 if (op->op_array[i] == idx) {
372 memmove(&op->op_array[i], &op->op_array[i + 1],
373 (op->op_count - i - 1) * sizeof(op->op_array[0]));
375 up_write(&op->op_rw_sem);
380 up_write(&op->op_rw_sem);
384 int lov_ost_pool_free(struct ost_pool *op)
386 if (op->op_size == 0)
389 down_write(&op->op_rw_sem);
396 up_write(&op->op_rw_sem);
400 int lov_pool_new(struct obd_device *obd, char *poolname)
403 struct pool_desc *new_pool;
408 if (strlen(poolname) > LOV_MAXPOOLNAME)
409 return -ENAMETOOLONG;
411 new_pool = kzalloc(sizeof(*new_pool), GFP_NOFS);
415 strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
416 new_pool->pool_lobd = obd;
417 /* ref count init to 1 because when created a pool is always used
420 atomic_set(&new_pool->pool_refcount, 1);
421 rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
425 memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
426 rc = lov_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
428 goto out_free_pool_obds;
430 INIT_HLIST_NODE(&new_pool->pool_hash);
432 /* we need this assert seq_file is not implemented for liblustre */
433 /* get ref for /proc file */
434 lov_pool_getref(new_pool);
435 new_pool->pool_debugfs_entry = ldebugfs_add_simple(
436 lov->lov_pool_debugfs_entry,
438 &pool_proc_operations);
439 if (IS_ERR_OR_NULL(new_pool->pool_debugfs_entry)) {
440 CWARN("Cannot add debugfs pool entry "LOV_POOLNAMEF"\n",
442 new_pool->pool_debugfs_entry = NULL;
443 lov_pool_putref(new_pool);
445 CDEBUG(D_INFO, "pool %p - proc %p\n",
446 new_pool, new_pool->pool_debugfs_entry);
448 spin_lock(&obd->obd_dev_lock);
449 list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
450 lov->lov_pool_count++;
451 spin_unlock(&obd->obd_dev_lock);
453 /* add to find only when it fully ready */
454 rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
455 &new_pool->pool_hash);
461 CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
462 poolname, lov->lov_pool_count);
467 spin_lock(&obd->obd_dev_lock);
468 list_del_init(&new_pool->pool_list);
469 lov->lov_pool_count--;
470 spin_unlock(&obd->obd_dev_lock);
472 ldebugfs_remove(&new_pool->pool_debugfs_entry);
474 lov_ost_pool_free(&new_pool->pool_rr.lqr_pool);
476 lov_ost_pool_free(&new_pool->pool_obds);
481 int lov_pool_del(struct obd_device *obd, char *poolname)
484 struct pool_desc *pool;
488 /* lookup and kill hash reference */
489 pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
493 if (!IS_ERR_OR_NULL(pool->pool_debugfs_entry)) {
494 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_debugfs_entry);
495 ldebugfs_remove(&pool->pool_debugfs_entry);
496 lov_pool_putref(pool);
499 spin_lock(&obd->obd_dev_lock);
500 list_del_init(&pool->pool_list);
501 lov->lov_pool_count--;
502 spin_unlock(&obd->obd_dev_lock);
504 /* release last reference */
505 lov_pool_putref(pool);
510 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
512 struct obd_uuid ost_uuid;
514 struct pool_desc *pool;
515 unsigned int lov_idx;
520 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
524 obd_str2uuid(&ost_uuid, ostname);
526 /* search ost in lov array */
528 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
529 if (!lov->lov_tgts[lov_idx])
531 if (obd_uuid_equals(&ost_uuid,
532 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
535 /* test if ost found in lov */
536 if (lov_idx == lov->desc.ld_tgt_count) {
541 rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
545 pool->pool_rr.lqr_dirty = 1;
547 CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
548 ostname, poolname, pool_tgt_count(pool));
552 lov_pool_putref(pool);
556 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
558 struct obd_uuid ost_uuid;
560 struct pool_desc *pool;
561 unsigned int lov_idx;
566 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
570 obd_str2uuid(&ost_uuid, ostname);
573 /* search ost in lov array, to get index */
574 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
575 if (!lov->lov_tgts[lov_idx])
578 if (obd_uuid_equals(&ost_uuid,
579 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
583 /* test if ost found in lov */
584 if (lov_idx == lov->desc.ld_tgt_count) {
589 lov_ost_pool_remove(&pool->pool_obds, lov_idx);
591 pool->pool_rr.lqr_dirty = 1;
593 CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
598 lov_pool_putref(pool);
602 int lov_check_index_in_pool(__u32 idx, struct pool_desc *pool)
606 /* caller may no have a ref on pool if it got the pool
607 * without calling lov_find_pool() (e.g. go through the lov pool
610 lov_pool_getref(pool);
612 down_read(&pool_tgt_rw_sem(pool));
614 for (i = 0; i < pool_tgt_count(pool); i++) {
615 if (pool_tgt_array(pool)[i] == idx) {
622 up_read(&pool_tgt_rw_sem(pool));
624 lov_pool_putref(pool);
628 struct pool_desc *lov_find_pool(struct lov_obd *lov, char *poolname)
630 struct pool_desc *pool;
633 if (poolname[0] != '\0') {
634 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
636 CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
638 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
639 CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
641 /* pool is ignored, so we remove ref on it */
642 lov_pool_putref(pool);