Whamcloud - gitweb
LU-13004 ptlrpc: Allow BULK_BUF_KIOV to accept a kvec
[fs/lustre-release.git] / lustre / lov / lov_pool.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/lov/lov_pool.c
33  *
34  * OST pool methods
35  *
36  * Author: Jacques-Charles LAFOUCRIERE <jc.lafoucriere@cea.fr>
37  * Author: Alex Lyashkov <Alexey.Lyashkov@Sun.COM>
38  * Author: Nathaniel Rutman <Nathan.Rutman@Sun.COM>
39  */
40
41 #define DEBUG_SUBSYSTEM S_LOV
42
43 #include <libcfs/libcfs.h>
44
45 #include <obd.h>
46 #include "lov_internal.h"
47
48 #define pool_tgt(_p, _i) \
49                 _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
50
51 static void lov_pool_getref(struct pool_desc *pool)
52 {
53         CDEBUG(D_INFO, "pool %p\n", pool);
54         atomic_inc(&pool->pool_refcount);
55 }
56
57 static void lov_pool_putref(struct pool_desc *pool)
58 {
59         CDEBUG(D_INFO, "pool %p\n", pool);
60         if (atomic_dec_and_test(&pool->pool_refcount)) {
61                 LASSERT(hlist_unhashed(&pool->pool_hash));
62                 LASSERT(list_empty(&pool->pool_list));
63                 LASSERT(pool->pool_proc_entry == NULL);
64                 lov_ost_pool_free(&(pool->pool_obds));
65                 OBD_FREE_PTR(pool);
66                 EXIT;
67         }
68 }
69
70 static void lov_pool_putref_locked(struct pool_desc *pool)
71 {
72         CDEBUG(D_INFO, "pool %p\n", pool);
73         LASSERT(atomic_read(&pool->pool_refcount) > 1);
74
75         atomic_dec(&pool->pool_refcount);
76 }
77
78 /*
79  * hash function using a Rotating Hash algorithm
80  * Knuth, D. The Art of Computer Programming,
81  * Volume 3: Sorting and Searching,
82  * Chapter 6.4.
83  * Addison Wesley, 1973
84  */
85 static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key,
86                          unsigned mask)
87 {
88         int i;
89         __u32 result;
90         char *poolname;
91
92         result = 0;
93         poolname = (char *)key;
94         for (i = 0; i < LOV_MAXPOOLNAME; i++) {
95                 if (poolname[i] == '\0')
96                         break;
97                 result = (result << 4)^(result >> 28) ^  poolname[i];
98         }
99         return (result % mask);
100 }
101
102 static void *pool_key(struct hlist_node *hnode)
103 {
104         struct pool_desc *pool;
105
106         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
107         return (pool->pool_name);
108 }
109
110 static int
111 pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
112 {
113         char *pool_name;
114         struct pool_desc *pool;
115
116         pool_name = (char *)key;
117         pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
118         return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
119 }
120
121 static void *pool_hashobject(struct hlist_node *hnode)
122 {
123         return hlist_entry(hnode, struct pool_desc, pool_hash);
124 }
125
126 static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
127 {
128         struct pool_desc *pool;
129
130         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
131         lov_pool_getref(pool);
132 }
133
134 static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
135                                          struct hlist_node *hnode)
136 {
137         struct pool_desc *pool;
138
139         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
140         lov_pool_putref_locked(pool);
141 }
142
143 struct cfs_hash_ops pool_hash_operations = {
144         .hs_hash        = pool_hashfn,
145         .hs_key         = pool_key,
146         .hs_keycmp      = pool_hashkey_keycmp,
147         .hs_object      = pool_hashobject,
148         .hs_get         = pool_hashrefcount_get,
149         .hs_put_locked  = pool_hashrefcount_put_locked,
150
151 };
152
153 #ifdef CONFIG_PROC_FS
154 /*
155  * pool /proc seq_file methods
156  */
157 /*
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)
162  */
163 #define POOL_IT_MAGIC 0xB001CEA0
164 struct pool_iterator {
165         int magic;
166         struct pool_desc *pool;
167         int idx;        /* from 0 to pool_tgt_size - 1 */
168 };
169
170 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
171 {
172         struct pool_iterator *iter = (struct pool_iterator *)s->private;
173         int prev_idx;
174
175         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
176
177         /* test if end of file */
178         if (*pos >= pool_tgt_count(iter->pool))
179                 return NULL;
180
181         /* iterate to find a non empty entry */
182         prev_idx = iter->idx;
183         iter->idx++;
184         if (iter->idx >= pool_tgt_count(iter->pool)) {
185                 iter->idx = prev_idx; /* we stay on the last entry */
186                 return NULL;
187         }
188         (*pos)++;
189         /* return != NULL to continue */
190         return iter;
191 }
192
193 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
194 {
195         struct pool_desc *pool = (struct pool_desc *)s->private;
196         struct pool_iterator *iter;
197
198         lov_pool_getref(pool);
199         if ((pool_tgt_count(pool) == 0) ||
200             (*pos >= pool_tgt_count(pool))) {
201                 /* iter is not created, so stop() has no way to
202                  * find pool to dec ref */
203                 lov_pool_putref(pool);
204                 return NULL;
205         }
206
207         OBD_ALLOC_PTR(iter);
208         if (!iter)
209                 return ERR_PTR(-ENOMEM);
210         iter->magic = POOL_IT_MAGIC;
211         iter->pool = pool;
212         iter->idx = 0;
213
214         /* we use seq_file private field to memorized iterator so
215          * we can free it at stop() */
216         /* /!\ do not forget to restore it to pool before freeing it */
217         s->private = iter;
218         down_read(&pool_tgt_rw_sem(pool));
219         if (*pos > 0) {
220                 loff_t i;
221                 void *ptr;
222
223                 i = 0;
224                 do {
225                      ptr = pool_proc_next(s, &iter, &i);
226                 } while ((i < *pos) && (ptr != NULL));
227                 return ptr;
228         }
229         return iter;
230 }
231
232 static void pool_proc_stop(struct seq_file *s, void *v)
233 {
234         struct pool_iterator *iter = (struct pool_iterator *)s->private;
235
236         /* in some cases stop() method is called 2 times, without
237          * calling start() method (see seq_read() from fs/seq_file.c)
238          * we have to free only if s->private is an iterator */
239         if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
240                 up_read(&pool_tgt_rw_sem(iter->pool));
241                 /* we restore s->private so next call to pool_proc_start()
242                  * will work */
243                 s->private = iter->pool;
244                 lov_pool_putref(iter->pool);
245                 OBD_FREE_PTR(iter);
246         }
247 }
248
249 static int pool_proc_show(struct seq_file *s, void *v)
250 {
251         struct pool_iterator *iter = (struct pool_iterator *)v;
252         struct lov_tgt_desc *tgt;
253
254         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X\n", iter->magic);
255         LASSERT(iter->pool != NULL);
256         LASSERT(iter->idx <= pool_tgt_count(iter->pool));
257
258         tgt = pool_tgt(iter->pool, iter->idx);
259         if (tgt)
260                 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
261
262         return 0;
263 }
264
265 static struct seq_operations pool_proc_ops = {
266         .start          = pool_proc_start,
267         .next           = pool_proc_next,
268         .stop           = pool_proc_stop,
269         .show           = pool_proc_show,
270 };
271
272 static int pool_proc_open(struct inode *inode, struct file *file)
273 {
274         int rc;
275
276         rc = seq_open(file, &pool_proc_ops);
277         if (!rc) {
278                 struct seq_file *s = file->private_data;
279                 s->private = PDE_DATA(inode);
280         }
281         return rc;
282 }
283
284 static struct file_operations pool_proc_operations = {
285         .open           = pool_proc_open,
286         .read           = seq_read,
287         .llseek         = seq_lseek,
288         .release        = seq_release,
289 };
290 #endif /* CONFIG_PROC_FS */
291
292 void lov_dump_pool(int level, struct pool_desc *pool)
293 {
294         int i;
295
296         lov_pool_getref(pool);
297
298         CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
299                pool->pool_name, pool->pool_obds.op_count);
300         down_read(&pool_tgt_rw_sem(pool));
301
302         for (i = 0; i < pool_tgt_count(pool) ; i++) {
303                 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
304                         continue;
305                 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
306                        pool->pool_name, i,
307                        obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
308         }
309
310         up_read(&pool_tgt_rw_sem(pool));
311         lov_pool_putref(pool);
312 }
313
314 #define LOV_POOL_INIT_COUNT 2
315 int lov_ost_pool_init(struct lu_tgt_pool *op, unsigned int count)
316 {
317         ENTRY;
318
319         if (count == 0)
320                 count = LOV_POOL_INIT_COUNT;
321         op->op_array = NULL;
322         op->op_count = 0;
323         init_rwsem(&op->op_rw_sem);
324         op->op_size = count * sizeof(op->op_array[0]);
325         OBD_ALLOC(op->op_array, op->op_size);
326         if (op->op_array == NULL) {
327                 op->op_size = 0;
328                 RETURN(-ENOMEM);
329         }
330         EXIT;
331         return 0;
332 }
333
334 /* Caller must hold write op_rwlock */
335 int lov_ost_pool_extend(struct lu_tgt_pool *op, unsigned int min_count)
336 {
337         __u32 *new;
338         __u32 new_size;
339
340         LASSERT(min_count != 0);
341
342         if (op->op_count * sizeof(op->op_array[0]) < op->op_size)
343                 return 0;
344
345         new_size = max_t(__u32, min_count * sizeof(op->op_array[0]),
346                          2 * op->op_size);
347         OBD_ALLOC(new, new_size);
348         if (new == NULL)
349                 return -ENOMEM;
350
351         /* copy old array to new one */
352         memcpy(new, op->op_array, op->op_size);
353         OBD_FREE(op->op_array, op->op_size);
354         op->op_array = new;
355         op->op_size = new_size;
356         return 0;
357 }
358
359 int lov_ost_pool_add(struct lu_tgt_pool *op, __u32 idx, unsigned int min_count)
360 {
361         int rc = 0, i;
362         ENTRY;
363
364         down_write(&op->op_rw_sem);
365
366         rc = lov_ost_pool_extend(op, min_count);
367         if (rc)
368                 GOTO(out, rc);
369
370         /* search ost in pool array */
371         for (i = 0; i < op->op_count; i++) {
372                 if (op->op_array[i] == idx)
373                         GOTO(out, rc = -EEXIST);
374         }
375         /* ost not found we add it */
376         op->op_array[op->op_count] = idx;
377         op->op_count++;
378         EXIT;
379 out:
380         up_write(&op->op_rw_sem);
381         return rc;
382 }
383
384 int lov_ost_pool_remove(struct lu_tgt_pool *op, __u32 idx)
385 {
386         int i;
387         ENTRY;
388
389         down_write(&op->op_rw_sem);
390
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]));
395                         op->op_count--;
396                         up_write(&op->op_rw_sem);
397                         EXIT;
398                         return 0;
399                 }
400         }
401
402         up_write(&op->op_rw_sem);
403         RETURN(-EINVAL);
404 }
405
406 int lov_ost_pool_free(struct lu_tgt_pool *op)
407 {
408         ENTRY;
409
410         if (op->op_size == 0)
411                 RETURN(0);
412
413         down_write(&op->op_rw_sem);
414
415         OBD_FREE(op->op_array, op->op_size);
416         op->op_array = NULL;
417         op->op_count = 0;
418         op->op_size = 0;
419
420         up_write(&op->op_rw_sem);
421         RETURN(0);
422 }
423
424
425 int lov_pool_new(struct obd_device *obd, char *poolname)
426 {
427         struct lov_obd *lov;
428         struct pool_desc *new_pool;
429         int rc;
430         ENTRY;
431
432         lov = &(obd->u.lov);
433
434         if (strlen(poolname) > LOV_MAXPOOLNAME)
435                 RETURN(-ENAMETOOLONG);
436
437         OBD_ALLOC_PTR(new_pool);
438         if (new_pool == NULL)
439                 RETURN(-ENOMEM);
440
441         strlcpy(new_pool->pool_name, poolname, sizeof(new_pool->pool_name));
442         new_pool->pool_lobd = obd;
443         /* ref count init to 1 because when created a pool is always used
444          * up to deletion
445          */
446         atomic_set(&new_pool->pool_refcount, 1);
447         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
448         if (rc)
449                 GOTO(out_err, rc);
450
451         INIT_HLIST_NODE(&new_pool->pool_hash);
452
453 #ifdef CONFIG_PROC_FS
454         /* get ref for /proc file */
455         lov_pool_getref(new_pool);
456         new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
457                                                        poolname, new_pool,
458                                                        &pool_proc_operations);
459         if (IS_ERR(new_pool->pool_proc_entry)) {
460                 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
461                 new_pool->pool_proc_entry = NULL;
462                 lov_pool_putref(new_pool);
463         }
464         CDEBUG(D_INFO, "pool %p - proc %p\n",
465                new_pool, new_pool->pool_proc_entry);
466 #endif
467
468         spin_lock(&obd->obd_dev_lock);
469         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
470         lov->lov_pool_count++;
471         spin_unlock(&obd->obd_dev_lock);
472
473         /* add to find only when it fully ready  */
474         rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
475                                  &new_pool->pool_hash);
476         if (rc)
477                 GOTO(out_err, rc = -EEXIST);
478
479         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
480                poolname, lov->lov_pool_count);
481
482         RETURN(0);
483
484 out_err:
485         spin_lock(&obd->obd_dev_lock);
486         list_del_init(&new_pool->pool_list);
487         lov->lov_pool_count--;
488         spin_unlock(&obd->obd_dev_lock);
489         lprocfs_remove(&new_pool->pool_proc_entry);
490         lov_ost_pool_free(&new_pool->pool_obds);
491         OBD_FREE_PTR(new_pool);
492
493         return rc;
494 }
495
496 int lov_pool_del(struct obd_device *obd, char *poolname)
497 {
498         struct lov_obd *lov;
499         struct pool_desc *pool;
500         ENTRY;
501
502         lov = &(obd->u.lov);
503
504         /* lookup and kill hash reference */
505         pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
506         if (pool == NULL)
507                 RETURN(-ENOENT);
508
509         if (pool->pool_proc_entry != NULL) {
510                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
511                 lprocfs_remove(&pool->pool_proc_entry);
512                 lov_pool_putref(pool);
513         }
514
515         spin_lock(&obd->obd_dev_lock);
516         list_del_init(&pool->pool_list);
517         lov->lov_pool_count--;
518         spin_unlock(&obd->obd_dev_lock);
519
520         /* release last reference */
521         lov_pool_putref(pool);
522
523         RETURN(0);
524 }
525
526
527 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
528 {
529         struct obd_uuid ost_uuid;
530         struct lov_obd *lov;
531         struct pool_desc *pool;
532         unsigned int lov_idx;
533         int rc;
534         ENTRY;
535
536         lov = &(obd->u.lov);
537
538         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
539         if (pool == NULL)
540                 RETURN(-ENOENT);
541
542         obd_str2uuid(&ost_uuid, ostname);
543
544
545         /* search ost in lov array */
546         lov_tgts_getref(obd);
547         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
548                 if (!lov->lov_tgts[lov_idx])
549                         continue;
550                 if (obd_uuid_equals(&ost_uuid,
551                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
552                         break;
553         }
554         /* test if ost found in lov */
555         if (lov_idx == lov->desc.ld_tgt_count)
556                 GOTO(out, rc = -EINVAL);
557
558         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
559         if (rc)
560                 GOTO(out, rc);
561
562         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
563                ostname, poolname,  pool_tgt_count(pool));
564
565         EXIT;
566 out:
567         lov_tgts_putref(obd);
568         lov_pool_putref(pool);
569
570         return rc;
571 }
572
573 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
574 {
575         struct obd_uuid ost_uuid;
576         struct lov_obd *lov;
577         struct pool_desc *pool;
578         unsigned int lov_idx;
579         int rc = 0;
580         ENTRY;
581
582         lov = &(obd->u.lov);
583
584         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
585         if (pool == NULL)
586                 RETURN(-ENOENT);
587
588         obd_str2uuid(&ost_uuid, ostname);
589
590         lov_tgts_getref(obd);
591         /* search ost in lov array, to get index */
592         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
593                 if (!lov->lov_tgts[lov_idx])
594                         continue;
595
596                 if (obd_uuid_equals(&ost_uuid,
597                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
598                         break;
599         }
600
601         /* test if ost found in lov */
602         if (lov_idx == lov->desc.ld_tgt_count)
603                 GOTO(out, rc = -EINVAL);
604
605         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
606
607         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
608                poolname);
609
610         EXIT;
611 out:
612         lov_tgts_putref(obd);
613         lov_pool_putref(pool);
614
615         return rc;
616 }