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 atomic_inc(&pool->pool_refcount);
65 void lov_pool_putref(struct pool_desc *pool)
67 CDEBUG(D_INFO, "pool %p\n", pool);
68 if (atomic_dec_and_test(&pool->pool_refcount)) {
69 LASSERT(hlist_unhashed(&pool->pool_hash));
70 LASSERT(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(atomic_read(&pool->pool_refcount) > 1);
83 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(struct hlist_node *hnode)
111 struct pool_desc *pool;
113 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
114 return (pool->pool_name);
118 pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
121 struct pool_desc *pool;
123 pool_name = (char *)key;
124 pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
125 return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
128 static void *pool_hashobject(struct hlist_node *hnode)
130 return hlist_entry(hnode, struct pool_desc, pool_hash);
133 static void pool_hashrefcount_get(cfs_hash_t *hs, struct hlist_node *hnode)
135 struct pool_desc *pool;
137 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
138 lov_pool_getref(pool);
141 static void pool_hashrefcount_put_locked(cfs_hash_t *hs,
142 struct hlist_node *hnode)
144 struct pool_desc *pool;
146 pool = hlist_entry(hnode, struct pool_desc, pool_hash);
147 lov_pool_putref_locked(pool);
150 cfs_hash_ops_t pool_hash_operations = {
151 .hs_hash = pool_hashfn,
153 .hs_keycmp = pool_hashkey_keycmp,
154 .hs_object = pool_hashobject,
155 .hs_get = pool_hashrefcount_get,
156 .hs_put_locked = pool_hashrefcount_put_locked,
161 /* ifdef needed for liblustre support */
163 * pool /proc seq_file methods
166 * iterator is used to go through the target pool entries
167 * index is the current entry index in the lp_array[] array
168 * index >= pos returned to the seq_file interface
169 * pos is from 0 to (pool->pool_obds.op_count - 1)
171 #define POOL_IT_MAGIC 0xB001CEA0
172 struct pool_iterator {
174 struct pool_desc *pool;
175 int idx; /* from 0 to pool_tgt_size - 1 */
178 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
180 struct pool_iterator *iter = (struct pool_iterator *)s->private;
183 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
185 /* test if end of file */
186 if (*pos >= pool_tgt_count(iter->pool))
189 /* iterate to find a non empty entry */
190 prev_idx = iter->idx;
191 down_read(&pool_tgt_rw_sem(iter->pool));
193 if (iter->idx == pool_tgt_count(iter->pool)) {
194 iter->idx = prev_idx; /* we stay on the last entry */
195 up_read(&pool_tgt_rw_sem(iter->pool));
198 up_read(&pool_tgt_rw_sem(iter->pool));
200 /* return != NULL to continue */
204 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
206 struct pool_desc *pool = (struct pool_desc *)s->private;
207 struct pool_iterator *iter;
209 lov_pool_getref(pool);
210 if ((pool_tgt_count(pool) == 0) ||
211 (*pos >= pool_tgt_count(pool))) {
212 /* iter is not created, so stop() has no way to
213 * find pool to dec ref */
214 lov_pool_putref(pool);
220 return ERR_PTR(-ENOMEM);
221 iter->magic = POOL_IT_MAGIC;
225 /* we use seq_file private field to memorized iterator so
226 * we can free it at stop() */
227 /* /!\ do not forget to restore it to pool before freeing it */
235 ptr = pool_proc_next(s, &iter, &i);
236 } while ((i < *pos) && (ptr != NULL));
242 static void pool_proc_stop(struct seq_file *s, void *v)
244 struct pool_iterator *iter = (struct pool_iterator *)s->private;
246 /* in some cases stop() method is called 2 times, without
247 * calling start() method (see seq_read() from fs/seq_file.c)
248 * we have to free only if s->private is an iterator */
249 if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
250 /* we restore s->private so next call to pool_proc_start()
252 s->private = iter->pool;
253 lov_pool_putref(iter->pool);
259 static int pool_proc_show(struct seq_file *s, void *v)
261 struct pool_iterator *iter = (struct pool_iterator *)v;
262 struct lov_tgt_desc *tgt;
264 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
265 LASSERT(iter->pool != NULL);
266 LASSERT(iter->idx <= pool_tgt_count(iter->pool));
268 down_read(&pool_tgt_rw_sem(iter->pool));
269 tgt = pool_tgt(iter->pool, iter->idx);
270 up_read(&pool_tgt_rw_sem(iter->pool));
272 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
277 static struct seq_operations pool_proc_ops = {
278 .start = pool_proc_start,
279 .next = pool_proc_next,
280 .stop = pool_proc_stop,
281 .show = pool_proc_show,
284 static int pool_proc_open(struct inode *inode, struct file *file)
288 rc = seq_open(file, &pool_proc_ops);
290 struct seq_file *s = file->private_data;
291 s->private = PDE_DATA(inode);
296 static struct file_operations pool_proc_operations = {
297 .open = pool_proc_open,
300 .release = seq_release,
304 void lov_dump_pool(int level, struct pool_desc *pool)
308 lov_pool_getref(pool);
310 CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
311 pool->pool_name, pool->pool_obds.op_count);
312 down_read(&pool_tgt_rw_sem(pool));
314 for (i = 0; i < pool_tgt_count(pool) ; i++) {
315 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
317 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
319 obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
322 up_read(&pool_tgt_rw_sem(pool));
323 lov_pool_putref(pool);
326 #define LOV_POOL_INIT_COUNT 2
327 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
332 count = LOV_POOL_INIT_COUNT;
335 init_rwsem(&op->op_rw_sem);
337 OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
338 if (op->op_array == NULL) {
346 /* Caller must hold write op_rwlock */
347 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
352 LASSERT(min_count != 0);
354 if (op->op_count < op->op_size)
357 new_size = max(min_count, 2 * op->op_size);
358 OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
362 /* copy old array to new one */
363 memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
364 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
366 op->op_size = new_size;
370 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
375 down_write(&op->op_rw_sem);
377 rc = lov_ost_pool_extend(op, min_count);
381 /* search ost in pool array */
382 for (i = 0; i < op->op_count; i++) {
383 if (op->op_array[i] == idx)
384 GOTO(out, rc = -EEXIST);
386 /* ost not found we add it */
387 op->op_array[op->op_count] = idx;
391 up_write(&op->op_rw_sem);
395 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
400 down_write(&op->op_rw_sem);
402 for (i = 0; i < op->op_count; i++) {
403 if (op->op_array[i] == idx) {
404 memmove(&op->op_array[i], &op->op_array[i + 1],
405 (op->op_count - i - 1) * sizeof(op->op_array[0]));
407 up_write(&op->op_rw_sem);
413 up_write(&op->op_rw_sem);
417 int lov_ost_pool_free(struct ost_pool *op)
421 if (op->op_size == 0)
424 down_write(&op->op_rw_sem);
426 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
431 up_write(&op->op_rw_sem);
436 int lov_pool_new(struct obd_device *obd, char *poolname)
439 struct pool_desc *new_pool;
445 if (strlen(poolname) > LOV_MAXPOOLNAME)
446 RETURN(-ENAMETOOLONG);
448 OBD_ALLOC_PTR(new_pool);
449 if (new_pool == NULL)
452 strncpy(new_pool->pool_name, poolname, LOV_MAXPOOLNAME);
453 new_pool->pool_name[LOV_MAXPOOLNAME] = '\0';
454 new_pool->pool_lobd = obd;
455 /* ref count init to 1 because when created a pool is always used
458 atomic_set(&new_pool->pool_refcount, 1);
459 rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
463 INIT_HLIST_NODE(&new_pool->pool_hash);
466 /* we need this assert seq_file is not implementated for liblustre */
467 /* get ref for /proc file */
468 lov_pool_getref(new_pool);
469 new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
471 #ifndef HAVE_ONLY_PROCFS_SEQ
475 &pool_proc_operations);
476 if (IS_ERR(new_pool->pool_proc_entry)) {
477 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
478 new_pool->pool_proc_entry = NULL;
479 lov_pool_putref(new_pool);
481 CDEBUG(D_INFO, "pool %p - proc %p\n", new_pool, new_pool->pool_proc_entry);
484 spin_lock(&obd->obd_dev_lock);
485 list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
486 lov->lov_pool_count++;
487 spin_unlock(&obd->obd_dev_lock);
489 /* add to find only when it fully ready */
490 rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
491 &new_pool->pool_hash);
493 GOTO(out_err, rc = -EEXIST);
495 CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
496 poolname, lov->lov_pool_count);
501 spin_lock(&obd->obd_dev_lock);
502 list_del_init(&new_pool->pool_list);
503 lov->lov_pool_count--;
504 spin_unlock(&obd->obd_dev_lock);
505 lprocfs_remove(&new_pool->pool_proc_entry);
506 lov_ost_pool_free(&new_pool->pool_obds);
507 OBD_FREE_PTR(new_pool);
512 int lov_pool_del(struct obd_device *obd, char *poolname)
515 struct pool_desc *pool;
520 /* lookup and kill hash reference */
521 pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
525 if (pool->pool_proc_entry != NULL) {
526 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
527 lprocfs_remove(&pool->pool_proc_entry);
528 lov_pool_putref(pool);
531 spin_lock(&obd->obd_dev_lock);
532 list_del_init(&pool->pool_list);
533 lov->lov_pool_count--;
534 spin_unlock(&obd->obd_dev_lock);
536 /* release last reference */
537 lov_pool_putref(pool);
543 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
545 struct obd_uuid ost_uuid;
547 struct pool_desc *pool;
548 unsigned int lov_idx;
554 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
558 obd_str2uuid(&ost_uuid, ostname);
561 /* search ost in lov array */
563 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
564 if (!lov->lov_tgts[lov_idx])
566 if (obd_uuid_equals(&ost_uuid,
567 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
570 /* test if ost found in lov */
571 if (lov_idx == lov->desc.ld_tgt_count)
572 GOTO(out, rc = -EINVAL);
574 rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
578 CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
579 ostname, poolname, pool_tgt_count(pool));
584 lov_pool_putref(pool);
588 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
590 struct obd_uuid ost_uuid;
592 struct pool_desc *pool;
593 unsigned int lov_idx;
599 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
603 obd_str2uuid(&ost_uuid, ostname);
606 /* search ost in lov array, to get index */
607 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
608 if (!lov->lov_tgts[lov_idx])
611 if (obd_uuid_equals(&ost_uuid,
612 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
616 /* test if ost found in lov */
617 if (lov_idx == lov->desc.ld_tgt_count)
618 GOTO(out, rc = -EINVAL);
620 lov_ost_pool_remove(&pool->pool_obds, lov_idx);
622 CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
628 lov_pool_putref(pool);