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, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2012, Intel Corporation.
25 * Use is subject to license terms.
27 * Author: Johann Lombardi <johann.lombardi@intel.com>
28 * Author: Niu Yawei <yawei.niu@intel.com>
32 * A Quota Master Target has a hash table where it stores qmt_pool_info
33 * structures. There is one such structure for each pool managed by the QMT.
35 * Each pool can have different quota types enforced (typically user & group
36 * quota). A pool is in charge of managing lquota_entry structures for each
37 * quota type. This is done by creating one lquota_entry site per quota
38 * type. A site stores entries in a hash table and read quota settings from disk
39 * when a given ID isn't present in the hash.
41 * The pool API exported here is the following:
42 * - qmt_pool_init(): initializes the general QMT structures used to manage
44 * - qmt_pool_fini(): frees the structures allocated by qmt_pool_fini().
45 * - qmt_pool_prepare(): sets up the on-disk indexes associated with each pool.
46 * - qmt_pool_new_conn(): is used to create a new slave index file.
47 * - qmt_pool_lqe_lookup(): returns an up-to-date lquota entry associated with
51 #define DEBUG_SUBSYSTEM S_LQUOTA
53 #include <obd_class.h>
54 #include <lprocfs_status.h>
55 #include "qmt_internal.h"
57 static void qmt_pool_free(const struct lu_env *, struct qmt_pool_info *);
60 * Static helper functions not used outside the scope of this file
64 * Reference counter management for qmt_pool_info structures
66 static inline void qpi_getref(struct qmt_pool_info *pool)
68 cfs_atomic_inc(&pool->qpi_ref);
71 static inline void qpi_putref(const struct lu_env *env,
72 struct qmt_pool_info *pool)
74 LASSERT(atomic_read(&pool->qpi_ref) > 0);
75 if (cfs_atomic_dec_and_test(&pool->qpi_ref))
76 qmt_pool_free(env, pool);
79 static inline void qpi_putref_locked(struct qmt_pool_info *pool)
81 LASSERT(cfs_atomic_read(&pool->qpi_ref) > 1);
82 cfs_atomic_dec(&pool->qpi_ref);
86 * Hash functions for qmt_pool_info management
89 static unsigned qpi_hash_hash(cfs_hash_t *hs, const void *key, unsigned mask)
91 return cfs_hash_u32_hash(*((__u32 *)key), mask);
94 static void *qpi_hash_key(cfs_hlist_node_t *hnode)
96 struct qmt_pool_info *pool;
97 pool = cfs_hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
98 return &pool->qpi_key;
101 static int qpi_hash_keycmp(const void *key, cfs_hlist_node_t *hnode)
103 struct qmt_pool_info *pool;
104 pool = cfs_hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
105 return pool->qpi_key == *((__u32 *)key);
108 static void *qpi_hash_object(cfs_hlist_node_t *hnode)
110 return cfs_hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
113 static void qpi_hash_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
115 struct qmt_pool_info *pool;
116 pool = cfs_hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
120 static void qpi_hash_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
122 struct qmt_pool_info *pool;
123 pool = cfs_hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
124 qpi_putref_locked(pool);
127 static void qpi_hash_exit(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
129 CERROR("Should not have any item left!\n");
132 /* vector of hash operations */
133 static cfs_hash_ops_t qpi_hash_ops = {
134 .hs_hash = qpi_hash_hash,
135 .hs_key = qpi_hash_key,
136 .hs_keycmp = qpi_hash_keycmp,
137 .hs_object = qpi_hash_object,
138 .hs_get = qpi_hash_get,
139 .hs_put_locked = qpi_hash_put_locked,
140 .hs_exit = qpi_hash_exit
143 /* some procfs helpers */
144 static int qpi_state_seq_show(struct seq_file *m, void *data)
146 struct qmt_pool_info *pool = m->private;
149 LASSERT(pool != NULL);
151 seq_printf(m, "pool:\n"
155 " least qunit: %lu\n",
156 pool->qpi_key & 0x0000ffff,
157 RES_NAME(pool->qpi_key >> 16),
158 cfs_atomic_read(&pool->qpi_ref),
159 pool->qpi_least_qunit);
161 for (type = 0; type < MAXQUOTAS; type++)
162 seq_printf(m, " %s:\n"
166 pool->qpi_slv_nr[type],
167 cfs_atomic_read(&pool->qpi_site[type]->lqs_hash->hs_count));
171 LPROC_SEQ_FOPS_RO(qpi_state);
173 static struct lprocfs_seq_vars lprocfs_quota_qpi_vars[] = {
174 { "info", &qpi_state_fops },
179 * Allocate a new qmt_pool_info structure and add it to the pool hash table
182 * \param env - is the environment passed by the caller
183 * \param qmt - is the quota master target
184 * \param pool_id - is the 16-bit pool identifier of the new pool to add
185 * \param pool_type - is the resource type of this pool instance, either
186 * LQUOTA_RES_MD or LQUOTA_RES_DT.
188 * \retval - 0 on success, appropriate error on failure
190 static int qmt_pool_alloc(const struct lu_env *env, struct qmt_device *qmt,
191 int pool_id, int pool_type)
193 struct qmt_thread_info *qti = qmt_info(env);
194 struct qmt_pool_info *pool;
201 CFS_INIT_LIST_HEAD(&pool->qpi_linkage);
203 /* assign key used by hash functions */
204 pool->qpi_key = pool_id + (pool_type << 16);
206 /* initialize refcount to 1, hash table will then grab an additional
208 atomic_set(&pool->qpi_ref, 1);
210 /* set up least qunit size to use for this pool */
211 pool->qpi_least_qunit = LQUOTA_LEAST_QUNIT(pool_type);
213 /* create pool proc directory */
214 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
215 pool->qpi_proc = lprocfs_seq_register(qti->qti_buf, qmt->qmt_proc,
216 lprocfs_quota_qpi_vars, pool);
217 if (IS_ERR(pool->qpi_proc)) {
218 rc = PTR_ERR(pool->qpi_proc);
219 CERROR("%s: failed to create proc entry for pool %s (%d)\n",
220 qmt->qmt_svname, qti->qti_buf, rc);
221 pool->qpi_proc = NULL;
225 /* grab reference on master target that this pool belongs to */
226 lu_device_get(qmt2lu_dev(qmt));
227 lu_ref_add(&qmt2lu_dev(qmt)->ld_reference, "pool", pool);
230 /* add to qmt hash table */
231 rc = cfs_hash_add_unique(qmt->qmt_pool_hash, &pool->qpi_key,
234 CERROR("%s: failed to add pool %s to qmt hash (%d)\n",
235 qmt->qmt_svname, qti->qti_buf, rc);
239 /* add to qmt pool list */
240 cfs_list_add_tail(&pool->qpi_linkage, &qmt->qmt_pool_list);
244 /* this frees the pool structure since refcount is equal to 1 */
245 qpi_putref(env, pool);
250 * Delete a qmt_pool_info instance and all structures associated.
252 * \param env - is the environment passed by the caller
253 * \param pool - is the qmt_pool_info structure to free
255 static void qmt_pool_free(const struct lu_env *env, struct qmt_pool_info *pool)
260 /* release proc entry */
261 if (pool->qpi_proc) {
262 lprocfs_remove(&pool->qpi_proc);
263 pool->qpi_proc = NULL;
266 /* release per-quota type site used to manage quota entries as well as
267 * references to global index files */
268 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
269 /* release lqe storing grace time */
270 if (pool->qpi_grace_lqe[qtype] != NULL)
271 lqe_putref(pool->qpi_grace_lqe[qtype]);
274 if (pool->qpi_site[qtype] != NULL &&
275 !IS_ERR(pool->qpi_site[qtype]))
276 lquota_site_free(env, pool->qpi_site[qtype]);
277 /* release reference to global index */
278 if (pool->qpi_glb_obj[qtype] != NULL &&
279 !IS_ERR(pool->qpi_glb_obj[qtype]))
280 lu_object_put(env, &pool->qpi_glb_obj[qtype]->do_lu);
283 /* release reference on pool directory */
284 if (pool->qpi_root != NULL && !IS_ERR(pool->qpi_root))
285 lu_object_put(env, &pool->qpi_root->do_lu);
287 /* release reference on the master target */
288 if (pool->qpi_qmt != NULL) {
289 struct lu_device *ld = qmt2lu_dev(pool->qpi_qmt);
292 lu_ref_del(&ld->ld_reference, "pool", pool);
293 pool->qpi_qmt = NULL;
296 LASSERT(cfs_list_empty(&pool->qpi_linkage));
301 * Look-up a pool in the hash table based on the pool ID and type.
303 * \param env - is the environment passed by the caller
304 * \param qmt - is the quota master target
305 * \param pool_id - is the 16-bit identifier of the pool to look up
306 * \param pool_type - is the type of this pool, either LQUOTA_RES_MD or
309 static struct qmt_pool_info *qmt_pool_lookup(const struct lu_env *env,
310 struct qmt_device *qmt,
311 int pool_id, int pool_type)
313 struct qmt_pool_info *pool;
317 LASSERT(qmt->qmt_pool_hash != NULL);
319 /* look-up pool in hash table */
320 key = pool_id + (pool_type << 16);
321 pool = cfs_hash_lookup(qmt->qmt_pool_hash, (void *)&key);
323 /* this qmt isn't managing this pool! */
324 CERROR("%s: looking up quota entry for a pool (0x%x/%d) which "
325 "isn't managed by this quota master target\n",
326 qmt->qmt_svname, pool_id, pool_type);
327 RETURN(ERR_PTR(-ENOENT));
333 * Functions implementing the pool API, used by the qmt handlers
337 * Destroy all pools which are still in the hash table and free the pool
340 * \param env - is the environment passed by the caller
341 * \param qmt - is the quota master target
344 void qmt_pool_fini(const struct lu_env *env, struct qmt_device *qmt)
346 struct qmt_pool_info *pool;
350 if (qmt->qmt_pool_hash == NULL)
353 /* parse list of pool and destroy each element */
354 cfs_list_for_each_safe(pos, n, &qmt->qmt_pool_list) {
355 pool = cfs_list_entry(pos, struct qmt_pool_info,
357 /* remove from hash */
358 cfs_hash_del(qmt->qmt_pool_hash, &pool->qpi_key,
361 /* remove from list */
362 cfs_list_del_init(&pool->qpi_linkage);
364 /* release extra reference taken in qmt_pool_alloc */
365 qpi_putref(env, pool);
367 LASSERT(cfs_list_empty(&qmt->qmt_pool_list));
369 cfs_hash_putref(qmt->qmt_pool_hash);
370 qmt->qmt_pool_hash = NULL;
375 * Initialize pool configure for the quota master target. For now, we only
376 * support the default data (i.e. all OSTs) and metadata (i.e. all the MDTs)
377 * pool which are instantiated in this function.
379 * \param env - is the environment passed by the caller
380 * \param qmt - is the quota master target for which we have to initializa the
383 * \retval - 0 on success, appropriate error on failure
385 int qmt_pool_init(const struct lu_env *env, struct qmt_device *qmt)
390 /* initialize pool hash table */
391 qmt->qmt_pool_hash = cfs_hash_create("POOL_HASH",
394 HASH_POOLS_BKT_BITS, 0,
399 if (qmt->qmt_pool_hash == NULL) {
400 CERROR("%s: failed to create pool hash table\n",
405 /* initialize pool list */
406 CFS_INIT_LIST_HEAD(&qmt->qmt_pool_list);
408 /* Instantiate pool master for the default data and metadata pool (both
409 * have pool ID equals to 0).
410 * This code will have to be revisited once we support quota on
411 * non-default pools */
412 for (res = LQUOTA_FIRST_RES; res < LQUOTA_LAST_RES; res++) {
413 rc = qmt_pool_alloc(env, qmt, 0, res);
419 qmt_pool_fini(env, qmt);
424 static int qmt_slv_cnt(const struct lu_env *env, struct lu_fid *glb_fid,
425 char *slv_name, struct lu_fid *slv_fid, void *arg)
436 * Set up on-disk index files associated with each pool.
438 * \param env - is the environment passed by the caller
439 * \param qmt - is the quota master target for which we have to initializa the
441 * \param qmt_root - is the on-disk directory created for the QMT.
443 * \retval - 0 on success, appropriate error on failure
445 int qmt_pool_prepare(const struct lu_env *env, struct qmt_device *qmt,
446 struct dt_object *qmt_root)
448 struct qmt_thread_info *qti = qmt_info(env);
449 struct lquota_glb_rec *rec = &qti->qti_glb_rec;
450 struct qmt_pool_info *pool;
451 struct dt_device *dev = NULL;
452 dt_obj_version_t version;
457 LASSERT(qmt->qmt_pool_hash != NULL);
459 /* iterate over each pool in the hash and allocate a quota site for each
460 * one. This involves creating a global index file on disk */
461 cfs_list_for_each(pos, &qmt->qmt_pool_list) {
462 struct dt_object *obj;
463 int pool_type, pool_id;
464 struct lquota_entry *lqe;
466 pool = cfs_list_entry(pos, struct qmt_pool_info,
469 pool_id = pool->qpi_key & 0x0000ffff;
470 pool_type = pool->qpi_key >> 16;
472 dev = pool->qpi_qmt->qmt_child;
474 /* allocate directory for this pool */
475 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
476 obj = lquota_disk_dir_find_create(env, qmt->qmt_child, qmt_root,
479 RETURN(PTR_ERR(obj));
480 pool->qpi_root = obj;
482 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
483 /* Generating FID of global index in charge of storing
484 * settings for this quota type */
485 lquota_generate_fid(&qti->qti_fid, pool_id, pool_type,
488 /* open/create the global index file for this quota
490 obj = lquota_disk_glb_find_create(env, dev,
492 &qti->qti_fid, false);
495 CERROR("%s: failed to create glb index copy for"
496 " %s type (%d)\n", qmt->qmt_svname,
497 QTYPE_NAME(qtype), rc);
501 pool->qpi_glb_obj[qtype] = obj;
503 version = dt_version_get(env, obj);
504 /* set default grace time for newly created index */
506 rec->qbr_hardlimit = 0;
507 rec->qbr_softlimit = 0;
508 rec->qbr_granted = 0;
509 rec->qbr_time = pool_type == LQUOTA_RES_MD ?
510 MAX_IQ_TIME : MAX_DQ_TIME;
512 rc = lquota_disk_write_glb(env, obj, 0, rec);
514 CERROR("%s: failed to set default "
515 "grace time for %s type (%d)\n",
517 QTYPE_NAME(qtype), rc);
521 rc = lquota_disk_update_ver(env, dev, obj, 1);
523 CERROR("%s: failed to set initial "
524 "version for %s type (%d)\n",
526 QTYPE_NAME(qtype), rc);
531 /* create quota entry site for this quota type */
532 pool->qpi_site[qtype] = lquota_site_alloc(env, pool,
535 if (IS_ERR(pool->qpi_site[qtype])) {
536 rc = PTR_ERR(pool->qpi_site[qtype]);
537 CERROR("%s: failed to create site for %s type "
538 "(%d)\n", qmt->qmt_svname,
539 QTYPE_NAME(qtype), rc);
543 /* count number of slaves which already connected to
544 * the master in the past */
545 pool->qpi_slv_nr[qtype] = 0;
546 rc = lquota_disk_for_each_slv(env, pool->qpi_root,
549 &pool->qpi_slv_nr[qtype]);
551 CERROR("%s: failed to scan & count slave "
552 "indexes for %s type (%d)\n",
553 qmt->qmt_svname, QTYPE_NAME(qtype), rc);
557 /* Global grace time is stored in quota settings of
559 qti->qti_id.qid_uid = 0;
561 /* look-up quota entry storing grace time */
562 lqe = lqe_locate(env, pool->qpi_site[qtype],
565 RETURN(PTR_ERR(lqe));
566 pool->qpi_grace_lqe[qtype] = lqe;
568 /* add procfs file to dump the global index, mostly for
569 * debugging purpose */
570 sprintf(qti->qti_buf, "glb-%s", QTYPE_NAME(qtype));
571 rc = lprocfs_seq_create(pool->qpi_proc, qti->qti_buf,
572 0444, &lprocfs_quota_seq_fops,
575 CWARN("%s: Error adding procfs file for global"
576 "quota index "DFID", rc:%d\n",
577 qmt->qmt_svname, PFID(&qti->qti_fid), rc);
586 * Handle new slave connection. Called when a slave enqueues the global quota
587 * lock at the beginning of the reintegration procedure.
589 * \param env - is the environment passed by the caller
590 * \parap qmt - is the quota master target handling this request
591 * \param glb_fid - is the fid of the global index file
592 * \param slv_fid - is the fid of the newly created slave index file
593 * \param slv_ver - is the current version of the slave index file
594 * \param uuid - is the uuid of slave which is (re)connecting to the master
597 * \retval - 0 on success, appropriate error on failure
599 int qmt_pool_new_conn(const struct lu_env *env, struct qmt_device *qmt,
600 struct lu_fid *glb_fid, struct lu_fid *slv_fid,
601 __u64 *slv_ver, struct obd_uuid *uuid)
603 struct qmt_pool_info *pool;
604 struct dt_object *slv_obj;
605 int pool_id, pool_type, qtype;
606 bool created = false;
609 /* extract pool info from global index FID */
610 rc = lquota_extract_fid(glb_fid, &pool_id, &pool_type, &qtype);
614 /* look-up pool in charge of this global index FID */
615 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
617 RETURN(PTR_ERR(pool));
619 /* look-up slave index file */
620 slv_obj = lquota_disk_slv_find(env, qmt->qmt_child, pool->qpi_root,
622 if (IS_ERR(slv_obj) && PTR_ERR(slv_obj) == -ENOENT) {
623 /* create slave index file */
624 slv_obj = lquota_disk_slv_find_create(env, qmt->qmt_child,
625 pool->qpi_root, glb_fid,
629 if (IS_ERR(slv_obj)) {
630 rc = PTR_ERR(slv_obj);
631 CERROR("%s: failed to create quota slave index file for %s (%d)"
632 "\n", qmt->qmt_svname, obd_uuid2str(uuid), rc);
636 /* retrieve slave fid & current object version */
637 memcpy(slv_fid, lu_object_fid(&slv_obj->do_lu), sizeof(*slv_fid));
638 *slv_ver = dt_version_get(env, slv_obj);
639 lu_object_put(env, &slv_obj->do_lu);
641 pool->qpi_slv_nr[qtype]++;
643 qpi_putref(env, pool);
648 * Look-up a lquota_entry in the pool hash and allocate it if not found.
650 * \param env - is the environment passed by the caller
651 * \param qmt - is the quota master target for which we have to initializa the
653 * \param pool_id - is the 16-bit identifier of the pool
654 * \param pool_type - is the pool type, either LQUOTA_RES_MD or LQUOTA_RES_DT.
655 * \param qtype - is the quota type, either user or group.
656 * \param qid - is the quota ID to look-up
658 * \retval - valid pointer to lquota entry on success, appropriate error on
661 struct lquota_entry *qmt_pool_lqe_lookup(const struct lu_env *env,
662 struct qmt_device *qmt,
663 int pool_id, int pool_type,
664 int qtype, union lquota_id *qid)
666 struct qmt_pool_info *pool;
667 struct lquota_entry *lqe;
670 /* look-up pool responsible for this global index FID */
671 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
673 RETURN((void *)pool);
675 if (qid->qid_uid == 0) {
676 /* caller wants to access grace time, no need to look up the
677 * entry since we keep a reference on ID 0 all the time */
678 lqe = pool->qpi_grace_lqe[qtype];
683 /* now that we have the pool, let's look-up the quota entry in the
684 * right quota site */
685 lqe = lqe_locate(env, pool->qpi_site[qtype], qid);
687 qpi_putref(env, pool);