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 <linux/libcfs/libcfs.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_proc_entry
== NULL
);
68 lov_ost_pool_free(&(pool
->pool_rr
.lqr_pool
));
69 lov_ost_pool_free(&(pool
->pool_obds
));
74 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(cfs_hash_t
*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(cfs_hash_t
*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(cfs_hash_t
*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 cfs_hash_ops_t 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
,
156 /* ifdef needed for liblustre support */
158 * pool /proc seq_file methods
161 * iterator is used to go through the target pool entries
162 * index is the current entry index in the lp_array[] array
163 * index >= pos returned to the seq_file interface
164 * pos is from 0 to (pool->pool_obds.op_count - 1)
166 #define POOL_IT_MAGIC 0xB001CEA0
167 struct pool_iterator
{
169 struct pool_desc
*pool
;
170 int idx
; /* from 0 to pool_tgt_size - 1 */
173 static void *pool_proc_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
175 struct pool_iterator
*iter
= (struct pool_iterator
*)s
->private;
178 LASSERTF(iter
->magic
== POOL_IT_MAGIC
, "%08X", iter
->magic
);
180 /* test if end of file */
181 if (*pos
>= pool_tgt_count(iter
->pool
))
184 /* iterate to find a non empty entry */
185 prev_idx
= iter
->idx
;
186 down_read(&pool_tgt_rw_sem(iter
->pool
));
188 if (iter
->idx
== pool_tgt_count(iter
->pool
)) {
189 iter
->idx
= prev_idx
; /* we stay on the last entry */
190 up_read(&pool_tgt_rw_sem(iter
->pool
));
193 up_read(&pool_tgt_rw_sem(iter
->pool
));
195 /* return != NULL to continue */
199 static void *pool_proc_start(struct seq_file
*s
, loff_t
*pos
)
201 struct pool_desc
*pool
= (struct pool_desc
*)s
->private;
202 struct pool_iterator
*iter
;
204 lov_pool_getref(pool
);
205 if ((pool_tgt_count(pool
) == 0) ||
206 (*pos
>= pool_tgt_count(pool
))) {
207 /* iter is not created, so stop() has no way to
208 * find pool to dec ref */
209 lov_pool_putref(pool
);
215 return ERR_PTR(-ENOMEM
);
216 iter
->magic
= POOL_IT_MAGIC
;
220 /* we use seq_file private field to memorized iterator so
221 * we can free it at stop() */
222 /* /!\ do not forget to restore it to pool before freeing it */
230 ptr
= pool_proc_next(s
, &iter
, &i
);
231 } while ((i
< *pos
) && (ptr
!= NULL
));
237 static void pool_proc_stop(struct seq_file
*s
, void *v
)
239 struct pool_iterator
*iter
= (struct pool_iterator
*)s
->private;
241 /* in some cases stop() method is called 2 times, without
242 * calling start() method (see seq_read() from fs/seq_file.c)
243 * 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()
247 s
->private = iter
->pool
;
248 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", iter
->magic
);
260 LASSERT(iter
->pool
!= NULL
);
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 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
;
286 s
->private = PDE_DATA(inode
);
291 static struct file_operations pool_proc_operations
= {
292 .open
= pool_proc_open
,
295 .release
= seq_release
,
299 void lov_dump_pool(int level
, struct pool_desc
*pool
)
303 lov_pool_getref(pool
);
305 CDEBUG(level
, "pool "LOV_POOLNAMEF
" has %d members\n",
306 pool
->pool_name
, pool
->pool_obds
.op_count
);
307 down_read(&pool_tgt_rw_sem(pool
));
309 for (i
= 0; i
< pool_tgt_count(pool
) ; i
++) {
310 if (!pool_tgt(pool
, i
) || !(pool_tgt(pool
, i
))->ltd_exp
)
312 CDEBUG(level
, "pool "LOV_POOLNAMEF
"[%d] = %s\n",
314 obd_uuid2str(&((pool_tgt(pool
, i
))->ltd_uuid
)));
317 up_read(&pool_tgt_rw_sem(pool
));
318 lov_pool_putref(pool
);
321 #define LOV_POOL_INIT_COUNT 2
322 int lov_ost_pool_init(struct ost_pool
*op
, unsigned int count
)
325 count
= LOV_POOL_INIT_COUNT
;
328 init_rwsem(&op
->op_rw_sem
);
330 OBD_ALLOC(op
->op_array
, op
->op_size
* sizeof(op
->op_array
[0]));
331 if (op
->op_array
== NULL
) {
338 /* Caller must hold write op_rwlock */
339 int lov_ost_pool_extend(struct ost_pool
*op
, unsigned int min_count
)
344 LASSERT(min_count
!= 0);
346 if (op
->op_count
< op
->op_size
)
349 new_size
= max(min_count
, 2 * op
->op_size
);
350 OBD_ALLOC(new, new_size
* sizeof(op
->op_array
[0]));
354 /* copy old array to new one */
355 memcpy(new, op
->op_array
, op
->op_size
* sizeof(op
->op_array
[0]));
356 OBD_FREE(op
->op_array
, op
->op_size
* sizeof(op
->op_array
[0]));
358 op
->op_size
= new_size
;
362 int lov_ost_pool_add(struct ost_pool
*op
, __u32 idx
, unsigned int min_count
)
366 down_write(&op
->op_rw_sem
);
368 rc
= lov_ost_pool_extend(op
, min_count
);
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
);
377 /* ost not found we add it */
378 op
->op_array
[op
->op_count
] = idx
;
381 up_write(&op
->op_rw_sem
);
385 int lov_ost_pool_remove(struct ost_pool
*op
, __u32 idx
)
389 down_write(&op
->op_rw_sem
);
391 for (i
= 0; i
< op
->op_count
; i
++) {
392 if (op
->op_array
[i
] == idx
) {
393 memmove(&op
->op_array
[i
], &op
->op_array
[i
+ 1],
394 (op
->op_count
- i
- 1) * sizeof(op
->op_array
[0]));
396 up_write(&op
->op_rw_sem
);
401 up_write(&op
->op_rw_sem
);
405 int lov_ost_pool_free(struct ost_pool
*op
)
407 if (op
->op_size
== 0)
410 down_write(&op
->op_rw_sem
);
412 OBD_FREE(op
->op_array
, op
->op_size
* sizeof(op
->op_array
[0]));
417 up_write(&op
->op_rw_sem
);
422 int lov_pool_new(struct obd_device
*obd
, char *poolname
)
425 struct pool_desc
*new_pool
;
430 if (strlen(poolname
) > LOV_MAXPOOLNAME
)
431 return -ENAMETOOLONG
;
433 OBD_ALLOC_PTR(new_pool
);
434 if (new_pool
== NULL
)
437 strncpy(new_pool
->pool_name
, poolname
, LOV_MAXPOOLNAME
);
438 new_pool
->pool_name
[LOV_MAXPOOLNAME
] = '\0';
439 new_pool
->pool_lobd
= obd
;
440 /* ref count init to 1 because when created a pool is always used
443 atomic_set(&new_pool
->pool_refcount
, 1);
444 rc
= lov_ost_pool_init(&new_pool
->pool_obds
, 0);
448 memset(&(new_pool
->pool_rr
), 0, sizeof(struct lov_qos_rr
));
449 rc
= lov_ost_pool_init(&new_pool
->pool_rr
.lqr_pool
, 0);
451 GOTO(out_free_pool_obds
, rc
);
453 INIT_HLIST_NODE(&new_pool
->pool_hash
);
456 /* we need this assert seq_file is not implementated for liblustre */
457 /* get ref for /proc file */
458 lov_pool_getref(new_pool
);
459 new_pool
->pool_proc_entry
= lprocfs_add_simple(lov
->lov_pool_proc_entry
,
461 &pool_proc_operations
);
462 if (IS_ERR(new_pool
->pool_proc_entry
)) {
463 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF
"\n", poolname
);
464 new_pool
->pool_proc_entry
= NULL
;
465 lov_pool_putref(new_pool
);
467 CDEBUG(D_INFO
, "pool %p - proc %p\n", new_pool
, new_pool
->pool_proc_entry
);
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
);
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
);
479 GOTO(out_err
, rc
= -EEXIST
);
481 CDEBUG(D_CONFIG
, LOV_POOLNAMEF
" is pool #%d\n",
482 poolname
, lov
->lov_pool_count
);
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
);
492 lprocfs_remove(&new_pool
->pool_proc_entry
);
494 lov_ost_pool_free(&new_pool
->pool_rr
.lqr_pool
);
496 lov_ost_pool_free(&new_pool
->pool_obds
);
497 OBD_FREE_PTR(new_pool
);
501 int lov_pool_del(struct obd_device
*obd
, char *poolname
)
504 struct pool_desc
*pool
;
508 /* lookup and kill hash reference */
509 pool
= cfs_hash_del_key(lov
->lov_pools_hash_body
, poolname
);
513 if (pool
->pool_proc_entry
!= NULL
) {
514 CDEBUG(D_INFO
, "proc entry %p\n", pool
->pool_proc_entry
);
515 lprocfs_remove(&pool
->pool_proc_entry
);
516 lov_pool_putref(pool
);
519 spin_lock(&obd
->obd_dev_lock
);
520 list_del_init(&pool
->pool_list
);
521 lov
->lov_pool_count
--;
522 spin_unlock(&obd
->obd_dev_lock
);
524 /* release last reference */
525 lov_pool_putref(pool
);
531 int lov_pool_add(struct obd_device
*obd
, char *poolname
, char *ostname
)
533 struct obd_uuid ost_uuid
;
535 struct pool_desc
*pool
;
536 unsigned int lov_idx
;
541 pool
= cfs_hash_lookup(lov
->lov_pools_hash_body
, poolname
);
545 obd_str2uuid(&ost_uuid
, ostname
);
548 /* search ost in lov array */
550 for (lov_idx
= 0; lov_idx
< lov
->desc
.ld_tgt_count
; lov_idx
++) {
551 if (!lov
->lov_tgts
[lov_idx
])
553 if (obd_uuid_equals(&ost_uuid
,
554 &(lov
->lov_tgts
[lov_idx
]->ltd_uuid
)))
557 /* test if ost found in lov */
558 if (lov_idx
== lov
->desc
.ld_tgt_count
)
559 GOTO(out
, rc
= -EINVAL
);
561 rc
= lov_ost_pool_add(&pool
->pool_obds
, lov_idx
, lov
->lov_tgt_size
);
565 pool
->pool_rr
.lqr_dirty
= 1;
567 CDEBUG(D_CONFIG
, "Added %s to "LOV_POOLNAMEF
" as member %d\n",
568 ostname
, poolname
, pool_tgt_count(pool
));
572 lov_pool_putref(pool
);
576 int lov_pool_remove(struct obd_device
*obd
, char *poolname
, char *ostname
)
578 struct obd_uuid ost_uuid
;
580 struct pool_desc
*pool
;
581 unsigned int lov_idx
;
586 pool
= cfs_hash_lookup(lov
->lov_pools_hash_body
, poolname
);
590 obd_str2uuid(&ost_uuid
, ostname
);
593 /* search ost in lov array, to get index */
594 for (lov_idx
= 0; lov_idx
< lov
->desc
.ld_tgt_count
; lov_idx
++) {
595 if (!lov
->lov_tgts
[lov_idx
])
598 if (obd_uuid_equals(&ost_uuid
,
599 &(lov
->lov_tgts
[lov_idx
]->ltd_uuid
)))
603 /* test if ost found in lov */
604 if (lov_idx
== lov
->desc
.ld_tgt_count
)
605 GOTO(out
, rc
= -EINVAL
);
607 lov_ost_pool_remove(&pool
->pool_obds
, lov_idx
);
609 pool
->pool_rr
.lqr_dirty
= 1;
611 CDEBUG(D_CONFIG
, "%s removed from "LOV_POOLNAMEF
"\n", ostname
,
616 lov_pool_putref(pool
);
620 int lov_check_index_in_pool(__u32 idx
, struct pool_desc
*pool
)
624 /* caller may no have a ref on pool if it got the pool
625 * without calling lov_find_pool() (e.g. go through the lov pool
628 lov_pool_getref(pool
);
630 down_read(&pool_tgt_rw_sem(pool
));
632 for (i
= 0; i
< pool_tgt_count(pool
); i
++) {
633 if (pool_tgt_array(pool
)[i
] == idx
)
638 up_read(&pool_tgt_rw_sem(pool
));
640 lov_pool_putref(pool
);
644 struct pool_desc
*lov_find_pool(struct lov_obd
*lov
, char *poolname
)
646 struct pool_desc
*pool
;
649 if (poolname
[0] != '\0') {
650 pool
= cfs_hash_lookup(lov
->lov_pools_hash_body
, poolname
);
652 CWARN("Request for an unknown pool ("LOV_POOLNAMEF
")\n",
654 if ((pool
!= NULL
) && (pool_tgt_count(pool
) == 0)) {
655 CWARN("Request for an empty pool ("LOV_POOLNAMEF
")\n",
657 /* pool is ignored, so we remove ref on it */
658 lov_pool_putref(pool
);