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 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
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 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 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[] = {
175 .fops = &qpi_state_fops },
180 * Allocate a new qmt_pool_info structure and add it to the pool hash table
183 * \param env - is the environment passed by the caller
184 * \param qmt - is the quota master target
185 * \param pool_id - is the 16-bit pool identifier of the new pool to add
186 * \param pool_type - is the resource type of this pool instance, either
187 * LQUOTA_RES_MD or LQUOTA_RES_DT.
189 * \retval - 0 on success, appropriate error on failure
191 static int qmt_pool_alloc(const struct lu_env *env, struct qmt_device *qmt,
192 int pool_id, int pool_type)
194 struct qmt_thread_info *qti = qmt_info(env);
195 struct qmt_pool_info *pool;
202 CFS_INIT_LIST_HEAD(&pool->qpi_linkage);
204 /* assign key used by hash functions */
205 pool->qpi_key = pool_id + (pool_type << 16);
207 /* initialize refcount to 1, hash table will then grab an additional
209 atomic_set(&pool->qpi_ref, 1);
211 /* set up least qunit size to use for this pool */
212 pool->qpi_least_qunit = LQUOTA_LEAST_QUNIT(pool_type);
214 /* create pool proc directory */
215 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
216 pool->qpi_proc = lprocfs_seq_register(qti->qti_buf, qmt->qmt_proc,
217 lprocfs_quota_qpi_vars, pool);
218 if (IS_ERR(pool->qpi_proc)) {
219 rc = PTR_ERR(pool->qpi_proc);
220 CERROR("%s: failed to create proc entry for pool %s (%d)\n",
221 qmt->qmt_svname, qti->qti_buf, rc);
222 pool->qpi_proc = NULL;
226 /* grab reference on master target that this pool belongs to */
227 lu_device_get(qmt2lu_dev(qmt));
228 lu_ref_add(&qmt2lu_dev(qmt)->ld_reference, "pool", pool);
231 /* add to qmt hash table */
232 rc = cfs_hash_add_unique(qmt->qmt_pool_hash, &pool->qpi_key,
235 CERROR("%s: failed to add pool %s to qmt hash (%d)\n",
236 qmt->qmt_svname, qti->qti_buf, rc);
240 /* add to qmt pool list */
241 cfs_list_add_tail(&pool->qpi_linkage, &qmt->qmt_pool_list);
245 /* this frees the pool structure since refcount is equal to 1 */
246 qpi_putref(env, pool);
251 * Delete a qmt_pool_info instance and all structures associated.
253 * \param env - is the environment passed by the caller
254 * \param pool - is the qmt_pool_info structure to free
256 static void qmt_pool_free(const struct lu_env *env, struct qmt_pool_info *pool)
261 /* release proc entry */
262 if (pool->qpi_proc) {
263 lprocfs_remove(&pool->qpi_proc);
264 pool->qpi_proc = NULL;
267 /* release per-quota type site used to manage quota entries as well as
268 * references to global index files */
269 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
270 /* release lqe storing grace time */
271 if (pool->qpi_grace_lqe[qtype] != NULL)
272 lqe_putref(pool->qpi_grace_lqe[qtype]);
275 if (pool->qpi_site[qtype] != NULL &&
276 !IS_ERR(pool->qpi_site[qtype]))
277 lquota_site_free(env, pool->qpi_site[qtype]);
278 /* release reference to global index */
279 if (pool->qpi_glb_obj[qtype] != NULL &&
280 !IS_ERR(pool->qpi_glb_obj[qtype]))
281 lu_object_put(env, &pool->qpi_glb_obj[qtype]->do_lu);
284 /* release reference on pool directory */
285 if (pool->qpi_root != NULL && !IS_ERR(pool->qpi_root))
286 lu_object_put(env, &pool->qpi_root->do_lu);
288 /* release reference on the master target */
289 if (pool->qpi_qmt != NULL) {
290 struct lu_device *ld = qmt2lu_dev(pool->qpi_qmt);
293 lu_ref_del(&ld->ld_reference, "pool", pool);
294 pool->qpi_qmt = NULL;
297 LASSERT(cfs_list_empty(&pool->qpi_linkage));
302 * Look-up a pool in the hash table based on the pool ID and type.
304 * \param env - is the environment passed by the caller
305 * \param qmt - is the quota master target
306 * \param pool_id - is the 16-bit identifier of the pool to look up
307 * \param pool_type - is the type of this pool, either LQUOTA_RES_MD or
310 static struct qmt_pool_info *qmt_pool_lookup(const struct lu_env *env,
311 struct qmt_device *qmt,
312 int pool_id, int pool_type)
314 struct qmt_pool_info *pool;
318 LASSERT(qmt->qmt_pool_hash != NULL);
320 /* look-up pool in hash table */
321 key = pool_id + (pool_type << 16);
322 pool = cfs_hash_lookup(qmt->qmt_pool_hash, (void *)&key);
324 /* this qmt isn't managing this pool! */
325 CERROR("%s: looking up quota entry for a pool (0x%x/%d) which "
326 "isn't managed by this quota master target\n",
327 qmt->qmt_svname, pool_id, pool_type);
328 RETURN(ERR_PTR(-ENOENT));
334 * Functions implementing the pool API, used by the qmt handlers
338 * Destroy all pools which are still in the hash table and free the pool
341 * \param env - is the environment passed by the caller
342 * \param qmt - is the quota master target
345 void qmt_pool_fini(const struct lu_env *env, struct qmt_device *qmt)
347 struct qmt_pool_info *pool;
351 if (qmt->qmt_pool_hash == NULL)
354 /* parse list of pool and destroy each element */
355 cfs_list_for_each_safe(pos, n, &qmt->qmt_pool_list) {
356 pool = cfs_list_entry(pos, struct qmt_pool_info,
358 /* remove from hash */
359 cfs_hash_del(qmt->qmt_pool_hash, &pool->qpi_key,
362 /* remove from list */
363 cfs_list_del_init(&pool->qpi_linkage);
365 /* release extra reference taken in qmt_pool_alloc */
366 qpi_putref(env, pool);
368 LASSERT(cfs_list_empty(&qmt->qmt_pool_list));
370 cfs_hash_putref(qmt->qmt_pool_hash);
371 qmt->qmt_pool_hash = NULL;
376 * Initialize pool configure for the quota master target. For now, we only
377 * support the default data (i.e. all OSTs) and metadata (i.e. all the MDTs)
378 * pool which are instantiated in this function.
380 * \param env - is the environment passed by the caller
381 * \param qmt - is the quota master target for which we have to initializa the
384 * \retval - 0 on success, appropriate error on failure
386 int qmt_pool_init(const struct lu_env *env, struct qmt_device *qmt)
391 /* initialize pool hash table */
392 qmt->qmt_pool_hash = cfs_hash_create("POOL_HASH",
395 HASH_POOLS_BKT_BITS, 0,
400 if (qmt->qmt_pool_hash == NULL) {
401 CERROR("%s: failed to create pool hash table\n",
406 /* initialize pool list */
407 CFS_INIT_LIST_HEAD(&qmt->qmt_pool_list);
409 /* Instantiate pool master for the default data and metadata pool (both
410 * have pool ID equals to 0).
411 * This code will have to be revisited once we support quota on
412 * non-default pools */
413 for (res = LQUOTA_FIRST_RES; res < LQUOTA_LAST_RES; res++) {
414 rc = qmt_pool_alloc(env, qmt, 0, res);
420 qmt_pool_fini(env, qmt);
425 static int qmt_slv_cnt(const struct lu_env *env, struct lu_fid *glb_fid,
426 char *slv_name, struct lu_fid *slv_fid, void *arg)
437 * Set up on-disk index files associated with each pool.
439 * \param env - is the environment passed by the caller
440 * \param qmt - is the quota master target for which we have to initializa the
442 * \param qmt_root - is the on-disk directory created for the QMT.
444 * \retval - 0 on success, appropriate error on failure
446 int qmt_pool_prepare(const struct lu_env *env, struct qmt_device *qmt,
447 struct dt_object *qmt_root)
449 struct qmt_thread_info *qti = qmt_info(env);
450 struct lquota_glb_rec *rec = &qti->qti_glb_rec;
451 struct qmt_pool_info *pool;
452 struct dt_device *dev = NULL;
453 dt_obj_version_t version;
458 LASSERT(qmt->qmt_pool_hash != NULL);
460 /* iterate over each pool in the hash and allocate a quota site for each
461 * one. This involves creating a global index file on disk */
462 cfs_list_for_each(pos, &qmt->qmt_pool_list) {
463 struct dt_object *obj;
464 int pool_type, pool_id;
465 struct lquota_entry *lqe;
467 pool = cfs_list_entry(pos, struct qmt_pool_info,
470 pool_id = pool->qpi_key & 0x0000ffff;
471 pool_type = pool->qpi_key >> 16;
473 dev = pool->qpi_qmt->qmt_child;
475 /* allocate directory for this pool */
476 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
477 obj = lquota_disk_dir_find_create(env, qmt->qmt_child, qmt_root,
480 RETURN(PTR_ERR(obj));
481 pool->qpi_root = obj;
483 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
484 /* Generating FID of global index in charge of storing
485 * settings for this quota type */
486 lquota_generate_fid(&qti->qti_fid, pool_id, pool_type,
489 /* open/create the global index file for this quota
491 obj = lquota_disk_glb_find_create(env, dev,
493 &qti->qti_fid, false);
496 CERROR("%s: failed to create glb index copy for"
497 " %s type (%d)\n", qmt->qmt_svname,
498 QTYPE_NAME(qtype), rc);
502 pool->qpi_glb_obj[qtype] = obj;
504 version = dt_version_get(env, obj);
505 /* set default grace time for newly created index */
507 rec->qbr_hardlimit = 0;
508 rec->qbr_softlimit = 0;
509 rec->qbr_granted = 0;
510 rec->qbr_time = pool_type == LQUOTA_RES_MD ?
511 MAX_IQ_TIME : MAX_DQ_TIME;
513 rc = lquota_disk_write_glb(env, obj, 0, rec);
515 CERROR("%s: failed to set default "
516 "grace time for %s type (%d)\n",
518 QTYPE_NAME(qtype), rc);
522 rc = lquota_disk_update_ver(env, dev, obj, 1);
524 CERROR("%s: failed to set initial "
525 "version for %s type (%d)\n",
527 QTYPE_NAME(qtype), rc);
532 /* create quota entry site for this quota type */
533 pool->qpi_site[qtype] = lquota_site_alloc(env, pool,
536 if (IS_ERR(pool->qpi_site[qtype])) {
537 rc = PTR_ERR(pool->qpi_site[qtype]);
538 CERROR("%s: failed to create site for %s type "
539 "(%d)\n", qmt->qmt_svname,
540 QTYPE_NAME(qtype), rc);
544 /* count number of slaves which already connected to
545 * the master in the past */
546 pool->qpi_slv_nr[qtype] = 0;
547 rc = lquota_disk_for_each_slv(env, pool->qpi_root,
550 &pool->qpi_slv_nr[qtype]);
552 CERROR("%s: failed to scan & count slave "
553 "indexes for %s type (%d)\n",
554 qmt->qmt_svname, QTYPE_NAME(qtype), rc);
558 /* Global grace time is stored in quota settings of
560 qti->qti_id.qid_uid = 0;
562 /* look-up quota entry storing grace time */
563 lqe = lqe_locate(env, pool->qpi_site[qtype],
566 RETURN(PTR_ERR(lqe));
567 pool->qpi_grace_lqe[qtype] = lqe;
569 /* add procfs file to dump the global index, mostly for
570 * debugging purpose */
571 sprintf(qti->qti_buf, "glb-%s", QTYPE_NAME(qtype));
572 rc = lprocfs_seq_create(pool->qpi_proc, qti->qti_buf,
573 0444, &lprocfs_quota_seq_fops,
576 CWARN("%s: Error adding procfs file for global"
577 "quota index "DFID", rc:%d\n",
578 qmt->qmt_svname, PFID(&qti->qti_fid), rc);
587 * Handle new slave connection. Called when a slave enqueues the global quota
588 * lock at the beginning of the reintegration procedure.
590 * \param env - is the environment passed by the caller
591 * \parap qmt - is the quota master target handling this request
592 * \param glb_fid - is the fid of the global index file
593 * \param slv_fid - is the fid of the newly created slave index file
594 * \param slv_ver - is the current version of the slave index file
595 * \param uuid - is the uuid of slave which is (re)connecting to the master
598 * \retval - 0 on success, appropriate error on failure
600 int qmt_pool_new_conn(const struct lu_env *env, struct qmt_device *qmt,
601 struct lu_fid *glb_fid, struct lu_fid *slv_fid,
602 __u64 *slv_ver, struct obd_uuid *uuid)
604 struct qmt_pool_info *pool;
605 struct dt_object *slv_obj;
606 int pool_id, pool_type, qtype;
607 bool created = false;
610 /* extract pool info from global index FID */
611 rc = lquota_extract_fid(glb_fid, &pool_id, &pool_type, &qtype);
615 /* look-up pool in charge of this global index FID */
616 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
618 RETURN(PTR_ERR(pool));
620 /* look-up slave index file */
621 slv_obj = lquota_disk_slv_find(env, qmt->qmt_child, pool->qpi_root,
623 if (IS_ERR(slv_obj) && PTR_ERR(slv_obj) == -ENOENT) {
624 /* create slave index file */
625 slv_obj = lquota_disk_slv_find_create(env, qmt->qmt_child,
626 pool->qpi_root, glb_fid,
630 if (IS_ERR(slv_obj)) {
631 rc = PTR_ERR(slv_obj);
632 CERROR("%s: failed to create quota slave index file for %s (%d)"
633 "\n", qmt->qmt_svname, obd_uuid2str(uuid), rc);
637 /* retrieve slave fid & current object version */
638 memcpy(slv_fid, lu_object_fid(&slv_obj->do_lu), sizeof(*slv_fid));
639 *slv_ver = dt_version_get(env, slv_obj);
640 lu_object_put(env, &slv_obj->do_lu);
642 pool->qpi_slv_nr[qtype]++;
644 qpi_putref(env, pool);
649 * Look-up a lquota_entry in the pool hash and allocate it if not found.
651 * \param env - is the environment passed by the caller
652 * \param qmt - is the quota master target for which we have to initializa the
654 * \param pool_id - is the 16-bit identifier of the pool
655 * \param pool_type - is the pool type, either LQUOTA_RES_MD or LQUOTA_RES_DT.
656 * \param qtype - is the quota type, either user or group.
657 * \param qid - is the quota ID to look-up
659 * \retval - valid pointer to lquota entry on success, appropriate error on
662 struct lquota_entry *qmt_pool_lqe_lookup(const struct lu_env *env,
663 struct qmt_device *qmt,
664 int pool_id, int pool_type,
665 int qtype, union lquota_id *qid)
667 struct qmt_pool_info *pool;
668 struct lquota_entry *lqe;
671 /* look-up pool responsible for this global index FID */
672 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
674 RETURN((void *)pool);
676 if (qid->qid_uid == 0) {
677 /* caller wants to access grace time, no need to look up the
678 * entry since we keep a reference on ID 0 all the time */
679 lqe = pool->qpi_grace_lqe[qtype];
684 /* now that we have the pool, let's look-up the quota entry in the
685 * right quota site */
686 lqe = lqe_locate(env, pool->qpi_site[qtype], qid);
688 qpi_putref(env, pool);