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, 2013, 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 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 (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(atomic_read(&pool->qpi_ref) > 1);
82 atomic_dec(&pool->qpi_ref);
86 * Hash functions for qmt_pool_info management
90 qpi_hash_hash(struct cfs_hash *hs, const void *key, unsigned mask)
92 return cfs_hash_u32_hash(*((__u32 *)key), mask);
95 static void *qpi_hash_key(struct hlist_node *hnode)
97 struct qmt_pool_info *pool;
98 pool = hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
99 return &pool->qpi_key;
102 static int qpi_hash_keycmp(const void *key, struct hlist_node *hnode)
104 struct qmt_pool_info *pool;
105 pool = hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
106 return pool->qpi_key == *((__u32 *)key);
109 static void *qpi_hash_object(struct hlist_node *hnode)
111 return hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
114 static void qpi_hash_get(struct cfs_hash *hs, struct hlist_node *hnode)
116 struct qmt_pool_info *pool;
117 pool = hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
121 static void qpi_hash_put_locked(struct cfs_hash *hs, struct hlist_node *hnode)
123 struct qmt_pool_info *pool;
124 pool = hlist_entry(hnode, struct qmt_pool_info, qpi_hash);
125 qpi_putref_locked(pool);
128 static void qpi_hash_exit(struct cfs_hash *hs, struct hlist_node *hnode)
130 CERROR("Should not have any item left!\n");
133 /* vector of hash operations */
134 static struct cfs_hash_ops qpi_hash_ops = {
135 .hs_hash = qpi_hash_hash,
136 .hs_key = qpi_hash_key,
137 .hs_keycmp = qpi_hash_keycmp,
138 .hs_object = qpi_hash_object,
139 .hs_get = qpi_hash_get,
140 .hs_put_locked = qpi_hash_put_locked,
141 .hs_exit = qpi_hash_exit
144 /* some procfs helpers */
145 static int qpi_state_seq_show(struct seq_file *m, void *data)
147 struct qmt_pool_info *pool = m->private;
150 LASSERT(pool != NULL);
152 seq_printf(m, "pool:\n"
156 " least qunit: %lu\n",
157 pool->qpi_key & 0x0000ffff,
158 RES_NAME(pool->qpi_key >> 16),
159 atomic_read(&pool->qpi_ref),
160 pool->qpi_least_qunit);
162 for (type = 0; type < MAXQUOTAS; type++)
163 seq_printf(m, " %s:\n"
167 pool->qpi_slv_nr[type],
168 atomic_read(&pool->qpi_site[type]->lqs_hash->hs_count));
172 LPROC_SEQ_FOPS_RO(qpi_state);
174 static struct lprocfs_vars lprocfs_quota_qpi_vars[] = {
176 .fops = &qpi_state_fops },
181 * Allocate a new qmt_pool_info structure and add it to the pool hash table
184 * \param env - is the environment passed by the caller
185 * \param qmt - is the quota master target
186 * \param pool_id - is the 16-bit pool identifier of the new pool to add
187 * \param pool_type - is the resource type of this pool instance, either
188 * LQUOTA_RES_MD or LQUOTA_RES_DT.
190 * \retval - 0 on success, appropriate error on failure
192 static int qmt_pool_alloc(const struct lu_env *env, struct qmt_device *qmt,
193 int pool_id, int pool_type)
195 struct qmt_thread_info *qti = qmt_info(env);
196 struct qmt_pool_info *pool;
203 INIT_LIST_HEAD(&pool->qpi_linkage);
205 /* assign key used by hash functions */
206 pool->qpi_key = pool_id + (pool_type << 16);
208 /* initialize refcount to 1, hash table will then grab an additional
210 atomic_set(&pool->qpi_ref, 1);
212 /* set up least qunit size to use for this pool */
213 pool->qpi_least_qunit = LQUOTA_LEAST_QUNIT(pool_type);
215 /* create pool proc directory */
216 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
217 pool->qpi_proc = lprocfs_register(qti->qti_buf, qmt->qmt_proc,
218 lprocfs_quota_qpi_vars, pool);
219 if (IS_ERR(pool->qpi_proc)) {
220 rc = PTR_ERR(pool->qpi_proc);
221 CERROR("%s: failed to create proc entry for pool %s (%d)\n",
222 qmt->qmt_svname, qti->qti_buf, rc);
223 pool->qpi_proc = NULL;
227 /* grab reference on master target that this pool belongs to */
228 lu_device_get(qmt2lu_dev(qmt));
229 lu_ref_add(&qmt2lu_dev(qmt)->ld_reference, "pool", pool);
232 /* add to qmt hash table */
233 rc = cfs_hash_add_unique(qmt->qmt_pool_hash, &pool->qpi_key,
236 CERROR("%s: failed to add pool %s to qmt hash (%d)\n",
237 qmt->qmt_svname, qti->qti_buf, rc);
241 /* add to qmt pool list */
242 list_add_tail(&pool->qpi_linkage, &qmt->qmt_pool_list);
246 /* this frees the pool structure since refcount is equal to 1 */
247 qpi_putref(env, pool);
252 * Delete a qmt_pool_info instance and all structures associated.
254 * \param env - is the environment passed by the caller
255 * \param pool - is the qmt_pool_info structure to free
257 static void qmt_pool_free(const struct lu_env *env, struct qmt_pool_info *pool)
262 /* release proc entry */
263 if (pool->qpi_proc) {
264 lprocfs_remove(&pool->qpi_proc);
265 pool->qpi_proc = NULL;
268 /* release per-quota type site used to manage quota entries as well as
269 * references to global index files */
270 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
271 /* release lqe storing grace time */
272 if (pool->qpi_grace_lqe[qtype] != NULL)
273 lqe_putref(pool->qpi_grace_lqe[qtype]);
276 if (pool->qpi_site[qtype] != NULL &&
277 !IS_ERR(pool->qpi_site[qtype]))
278 lquota_site_free(env, pool->qpi_site[qtype]);
279 /* release reference to global index */
280 if (pool->qpi_glb_obj[qtype] != NULL &&
281 !IS_ERR(pool->qpi_glb_obj[qtype]))
282 lu_object_put(env, &pool->qpi_glb_obj[qtype]->do_lu);
285 /* release reference on pool directory */
286 if (pool->qpi_root != NULL && !IS_ERR(pool->qpi_root))
287 lu_object_put(env, &pool->qpi_root->do_lu);
289 /* release reference on the master target */
290 if (pool->qpi_qmt != NULL) {
291 struct lu_device *ld = qmt2lu_dev(pool->qpi_qmt);
294 lu_ref_del(&ld->ld_reference, "pool", pool);
295 pool->qpi_qmt = NULL;
298 LASSERT(list_empty(&pool->qpi_linkage));
303 * Look-up a pool in the hash table based on the pool ID and type.
305 * \param env - is the environment passed by the caller
306 * \param qmt - is the quota master target
307 * \param pool_id - is the 16-bit identifier of the pool to look up
308 * \param pool_type - is the type of this pool, either LQUOTA_RES_MD or
311 static struct qmt_pool_info *qmt_pool_lookup(const struct lu_env *env,
312 struct qmt_device *qmt,
313 int pool_id, int pool_type)
315 struct qmt_pool_info *pool;
319 LASSERT(qmt->qmt_pool_hash != NULL);
321 /* look-up pool in hash table */
322 key = pool_id + (pool_type << 16);
323 pool = cfs_hash_lookup(qmt->qmt_pool_hash, (void *)&key);
325 /* this qmt isn't managing this pool! */
326 CERROR("%s: looking up quota entry for a pool (0x%x/%d) which "
327 "isn't managed by this quota master target\n",
328 qmt->qmt_svname, pool_id, pool_type);
329 RETURN(ERR_PTR(-ENOENT));
335 * Functions implementing the pool API, used by the qmt handlers
339 * Destroy all pools which are still in the hash table and free the pool
342 * \param env - is the environment passed by the caller
343 * \param qmt - is the quota master target
346 void qmt_pool_fini(const struct lu_env *env, struct qmt_device *qmt)
348 struct qmt_pool_info *pool;
349 struct list_head *pos, *n;
352 if (qmt->qmt_pool_hash == NULL)
355 /* parse list of pool and destroy each element */
356 list_for_each_safe(pos, n, &qmt->qmt_pool_list) {
357 pool = list_entry(pos, struct qmt_pool_info,
359 /* remove from hash */
360 cfs_hash_del(qmt->qmt_pool_hash, &pool->qpi_key,
363 /* remove from list */
364 list_del_init(&pool->qpi_linkage);
366 /* release extra reference taken in qmt_pool_alloc */
367 qpi_putref(env, pool);
369 LASSERT(list_empty(&qmt->qmt_pool_list));
371 cfs_hash_putref(qmt->qmt_pool_hash);
372 qmt->qmt_pool_hash = NULL;
377 * Initialize pool configure for the quota master target. For now, we only
378 * support the default data (i.e. all OSTs) and metadata (i.e. all the MDTs)
379 * pool which are instantiated in this function.
381 * \param env - is the environment passed by the caller
382 * \param qmt - is the quota master target for which we have to initializa the
385 * \retval - 0 on success, appropriate error on failure
387 int qmt_pool_init(const struct lu_env *env, struct qmt_device *qmt)
392 /* initialize pool hash table */
393 qmt->qmt_pool_hash = cfs_hash_create("POOL_HASH",
396 HASH_POOLS_BKT_BITS, 0,
401 if (qmt->qmt_pool_hash == NULL) {
402 CERROR("%s: failed to create pool hash table\n",
407 /* initialize pool list */
408 INIT_LIST_HEAD(&qmt->qmt_pool_list);
410 /* Instantiate pool master for the default data and metadata pool (both
411 * have pool ID equals to 0).
412 * This code will have to be revisited once we support quota on
413 * non-default pools */
414 for (res = LQUOTA_FIRST_RES; res < LQUOTA_LAST_RES; res++) {
415 rc = qmt_pool_alloc(env, qmt, 0, res);
421 qmt_pool_fini(env, qmt);
426 static int qmt_slv_cnt(const struct lu_env *env, struct lu_fid *glb_fid,
427 char *slv_name, struct lu_fid *slv_fid, void *arg)
438 * Set up on-disk index files associated with each pool.
440 * \param env - is the environment passed by the caller
441 * \param qmt - is the quota master target for which we have to initializa the
443 * \param qmt_root - is the on-disk directory created for the QMT.
445 * \retval - 0 on success, appropriate error on failure
447 int qmt_pool_prepare(const struct lu_env *env, struct qmt_device *qmt,
448 struct dt_object *qmt_root)
450 struct qmt_thread_info *qti = qmt_info(env);
451 struct lquota_glb_rec *rec = &qti->qti_glb_rec;
452 struct qmt_pool_info *pool;
453 struct dt_device *dev = NULL;
454 dt_obj_version_t version;
455 struct list_head *pos;
459 LASSERT(qmt->qmt_pool_hash != NULL);
461 /* iterate over each pool in the hash and allocate a quota site for each
462 * one. This involves creating a global index file on disk */
463 list_for_each(pos, &qmt->qmt_pool_list) {
464 struct dt_object *obj;
465 int pool_type, pool_id;
466 struct lquota_entry *lqe;
468 pool = list_entry(pos, struct qmt_pool_info,
471 pool_id = pool->qpi_key & 0x0000ffff;
472 pool_type = pool->qpi_key >> 16;
474 dev = pool->qpi_qmt->qmt_child;
476 /* allocate directory for this pool */
477 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
478 obj = lquota_disk_dir_find_create(env, qmt->qmt_child, qmt_root,
481 RETURN(PTR_ERR(obj));
482 pool->qpi_root = obj;
484 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
485 /* Generating FID of global index in charge of storing
486 * settings for this quota type */
487 lquota_generate_fid(&qti->qti_fid, pool_id, pool_type,
490 /* open/create the global index file for this quota
492 obj = lquota_disk_glb_find_create(env, dev,
494 &qti->qti_fid, false);
497 CERROR("%s: failed to create glb index copy for"
498 " %s type (%d)\n", qmt->qmt_svname,
499 QTYPE_NAME(qtype), rc);
503 pool->qpi_glb_obj[qtype] = obj;
505 version = dt_version_get(env, obj);
506 /* set default grace time for newly created index */
508 rec->qbr_hardlimit = 0;
509 rec->qbr_softlimit = 0;
510 rec->qbr_granted = 0;
511 rec->qbr_time = pool_type == LQUOTA_RES_MD ?
512 MAX_IQ_TIME : MAX_DQ_TIME;
514 rc = lquota_disk_write_glb(env, obj, 0, rec);
516 CERROR("%s: failed to set default "
517 "grace time for %s type (%d)\n",
519 QTYPE_NAME(qtype), rc);
523 rc = lquota_disk_update_ver(env, dev, obj, 1);
525 CERROR("%s: failed to set initial "
526 "version for %s type (%d)\n",
528 QTYPE_NAME(qtype), rc);
533 /* create quota entry site for this quota type */
534 pool->qpi_site[qtype] = lquota_site_alloc(env, pool,
537 if (IS_ERR(pool->qpi_site[qtype])) {
538 rc = PTR_ERR(pool->qpi_site[qtype]);
539 CERROR("%s: failed to create site for %s type "
540 "(%d)\n", qmt->qmt_svname,
541 QTYPE_NAME(qtype), rc);
545 /* count number of slaves which already connected to
546 * the master in the past */
547 pool->qpi_slv_nr[qtype] = 0;
548 rc = lquota_disk_for_each_slv(env, pool->qpi_root,
551 &pool->qpi_slv_nr[qtype]);
553 CERROR("%s: failed to scan & count slave "
554 "indexes for %s type (%d)\n",
555 qmt->qmt_svname, QTYPE_NAME(qtype), rc);
559 /* Global grace time is stored in quota settings of
561 qti->qti_id.qid_uid = 0;
563 /* look-up quota entry storing grace time */
564 lqe = lqe_locate(env, pool->qpi_site[qtype],
567 RETURN(PTR_ERR(lqe));
568 pool->qpi_grace_lqe[qtype] = lqe;
569 #ifdef CONFIG_PROC_FS
570 /* add procfs file to dump the global index, mostly for
571 * debugging purpose */
572 sprintf(qti->qti_buf, "glb-%s", QTYPE_NAME(qtype));
573 rc = lprocfs_seq_create(pool->qpi_proc, qti->qti_buf,
574 0444, &lprocfs_quota_seq_fops,
577 CWARN("%s: Error adding procfs file for global"
578 "quota index "DFID", rc:%d\n",
579 qmt->qmt_svname, PFID(&qti->qti_fid), rc);
588 * Handle new slave connection. Called when a slave enqueues the global quota
589 * lock at the beginning of the reintegration procedure.
591 * \param env - is the environment passed by the caller
592 * \parap qmt - is the quota master target handling this request
593 * \param glb_fid - is the fid of the global index file
594 * \param slv_fid - is the fid of the newly created slave index file
595 * \param slv_ver - is the current version of the slave index file
596 * \param uuid - is the uuid of slave which is (re)connecting to the master
599 * \retval - 0 on success, appropriate error on failure
601 int qmt_pool_new_conn(const struct lu_env *env, struct qmt_device *qmt,
602 struct lu_fid *glb_fid, struct lu_fid *slv_fid,
603 __u64 *slv_ver, struct obd_uuid *uuid)
605 struct qmt_pool_info *pool;
606 struct dt_object *slv_obj;
607 int pool_id, pool_type, qtype;
608 bool created = false;
611 /* extract pool info from global index FID */
612 rc = lquota_extract_fid(glb_fid, &pool_id, &pool_type, &qtype);
616 /* look-up pool in charge of this global index FID */
617 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
619 RETURN(PTR_ERR(pool));
621 /* look-up slave index file */
622 slv_obj = lquota_disk_slv_find(env, qmt->qmt_child, pool->qpi_root,
624 if (IS_ERR(slv_obj) && PTR_ERR(slv_obj) == -ENOENT) {
625 /* create slave index file */
626 slv_obj = lquota_disk_slv_find_create(env, qmt->qmt_child,
627 pool->qpi_root, glb_fid,
631 if (IS_ERR(slv_obj)) {
632 rc = PTR_ERR(slv_obj);
633 CERROR("%s: failed to create quota slave index file for %s (%d)"
634 "\n", qmt->qmt_svname, obd_uuid2str(uuid), rc);
638 /* retrieve slave fid & current object version */
639 memcpy(slv_fid, lu_object_fid(&slv_obj->do_lu), sizeof(*slv_fid));
640 *slv_ver = dt_version_get(env, slv_obj);
641 lu_object_put(env, &slv_obj->do_lu);
643 pool->qpi_slv_nr[qtype]++;
645 qpi_putref(env, pool);
650 * Look-up a lquota_entry in the pool hash and allocate it if not found.
652 * \param env - is the environment passed by the caller
653 * \param qmt - is the quota master target for which we have to initializa the
655 * \param pool_id - is the 16-bit identifier of the pool
656 * \param pool_type - is the pool type, either LQUOTA_RES_MD or LQUOTA_RES_DT.
657 * \param qtype - is the quota type, either user or group.
658 * \param qid - is the quota ID to look-up
660 * \retval - valid pointer to lquota entry on success, appropriate error on
663 struct lquota_entry *qmt_pool_lqe_lookup(const struct lu_env *env,
664 struct qmt_device *qmt,
665 int pool_id, int pool_type,
666 int qtype, union lquota_id *qid)
668 struct qmt_pool_info *pool;
669 struct lquota_entry *lqe;
672 /* look-up pool responsible for this global index FID */
673 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
675 RETURN((void *)pool);
677 if (qid->qid_uid == 0) {
678 /* caller wants to access grace time, no need to look up the
679 * entry since we keep a reference on ID 0 all the time */
680 lqe = pool->qpi_grace_lqe[qtype];
685 /* now that we have the pool, let's look-up the quota entry in the
686 * right quota site */
687 lqe = lqe_locate(env, pool->qpi_site[qtype], qid);
689 qpi_putref(env, pool);