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, 2016, Intel Corporation.
25 * Use is subject to license terms.
28 #ifndef _QMT_INTERNAL_H
29 #define _QMT_INTERNAL_H
31 #include "lquota_internal.h"
34 * The Quota Master Target Device.
35 * The qmt is responsible for:
36 * - all interactions with MDT0 (provide request handlers, share ldlm namespace,
37 * manage ldlm lvbo, ...)
38 * - all quota lock management (i.e. global quota locks as well as per-ID locks)
39 * - manage the quota pool configuration
41 * That's the structure MDT0 connects to in mdt_quota_init().
44 /* Super-class. dt_device/lu_device for this master target */
45 struct dt_device qmt_dt_dev;
47 /* service name of this qmt */
48 char qmt_svname[MAX_OBD_NAME];
50 /* Reference to the next device in the side stack
51 * The child device is actually the OSD device where we store the quota
53 struct obd_export *qmt_child_exp;
54 struct dt_device *qmt_child;
56 /* pointer to ldlm namespace to be used for quota locks */
57 struct ldlm_namespace *qmt_ns;
59 /* Hash table containing a qmt_pool_info structure for each pool
60 * this quota master is in charge of. We only have 2 pools in this
61 * hash for the time being:
62 * - one for quota management on the default metadata pool
63 * - one for quota managment on the default data pool
65 * Once we support quota on non-default pools, then more pools will
66 * be added to this hash table and pool master setup would have to be
67 * handled via configuration logs */
68 struct cfs_hash *qmt_pool_hash;
70 /* List of pools managed by this master target */
71 struct list_head qmt_pool_list;
73 /* procfs root directory for this qmt */
74 struct proc_dir_entry *qmt_proc;
76 /* dedicated thread in charge of space rebalancing */
77 struct ptlrpc_thread qmt_reba_thread;
79 /* list of lqe entry which need space rebalancing */
80 struct list_head qmt_reba_list;
82 /* lock protecting rebalancing list */
83 spinlock_t qmt_reba_lock;
85 unsigned long qmt_stopping:1; /* qmt is stopping */
90 * Per-pool quota information.
91 * The qmt creates one such structure for each pool
92 * with quota enforced. All the structures are kept in a hash which is used to
93 * determine whether or not quota is enforced for a given pool.
94 * We currently only support the default data pool and default metadata pool
97 struct qmt_pool_info {
98 /* link to qmt's pool hash */
99 struct hlist_node qpi_hash;
101 /* chained list of all pools managed by the same qmt */
102 struct list_head qpi_linkage;
104 /* Pool key composed of pool_id | (pool_type << 16)
105 * Only pool ID 0 is supported for now and the pool type is either
106 * QUOTA_RES_MD or QUOTA_RES_DT.
107 * immutable after creation. */
110 /* track users of this pool instance */
113 /* back pointer to master target
114 * immutable after creation. */
115 struct qmt_device *qpi_qmt;
117 /* pointer to dt object associated with global indexes for both user
119 struct dt_object *qpi_glb_obj[LL_MAXQUOTAS];
121 /* A pool supports two different quota types: user and group quota.
122 * Each quota type has its own global index and lquota_entry hash table.
124 struct lquota_site *qpi_site[LL_MAXQUOTAS];
126 /* number of slaves registered for each quota types */
127 int qpi_slv_nr[LL_MAXQUOTAS];
129 /* reference on lqe (ID 0) storing grace time. */
130 struct lquota_entry *qpi_grace_lqe[LL_MAXQUOTAS];
132 /* procfs root directory for this pool */
133 struct proc_dir_entry *qpi_proc;
135 /* pool directory where all indexes related to this pool instance are
137 struct dt_object *qpi_root;
139 /* Global quota parameters which apply to all quota type */
140 /* the least value of qunit */
141 unsigned long qpi_least_qunit;
143 /* Least value of qunit when soft limit is exceeded.
145 * When soft limit is exceeded, qunit will be shrinked to least_qunit
146 * (1M for block limit), that results in significant write performance
147 * drop since the client will turn to sync write from now on.
149 * To retain the write performance in an acceptable level, we choose
150 * to sacrifice grace time accuracy a bit and use a larger least_qunit
151 * when soft limit is exceeded. It's (qpi_least_qunit * 4) by default,
152 * and user may enlarge it via procfs to get even better performance
153 * (with the cost of losing more grace time accuracy).
155 * See qmt_calc_softlimit().
157 unsigned long qpi_soft_least_qunit;
161 * Helper routines and prototypes
164 /* helper routine to find qmt_pool_info associated a lquota_entry */
165 static inline struct qmt_pool_info *lqe2qpi(struct lquota_entry *lqe)
167 LASSERT(lqe_is_master(lqe));
168 return (struct qmt_pool_info *)lqe->lqe_site->lqs_parent;
171 /* return true if someone holds either a read or write lock on the lqe */
172 static inline bool lqe_is_locked(struct lquota_entry *lqe)
174 LASSERT(lqe_is_master(lqe));
175 if (down_write_trylock(&lqe->lqe_sem) == 0)
177 lqe_write_unlock(lqe);
181 /* value to be restored if someone wrong happens during lqe writeback */
182 struct qmt_lqe_restore {
190 /* Common data shared by qmt handlers */
191 struct qmt_thread_info {
192 union lquota_rec qti_rec;
193 union lquota_id qti_id;
194 char qti_buf[MTI_NAME_MAXLEN];
195 struct lu_fid qti_fid;
196 struct ldlm_res_id qti_resid;
197 union ldlm_gl_desc qti_gl_desc;
198 struct quota_body qti_body;
199 struct qmt_lqe_restore qti_restore;
202 extern struct lu_context_key qmt_thread_key;
204 /* helper function to extract qmt_thread_info from current environment */
206 struct qmt_thread_info *qmt_info(const struct lu_env *env)
208 struct qmt_thread_info *info;
210 info = lu_context_key_get(&env->le_ctx, &qmt_thread_key);
212 lu_env_refill((struct lu_env *)env);
213 info = lu_context_key_get(&env->le_ctx, &qmt_thread_key);
219 /* helper routine to convert a lu_device into a qmt_device */
220 static inline struct qmt_device *lu2qmt_dev(struct lu_device *ld)
222 return container_of0(lu2dt_dev(ld), struct qmt_device, qmt_dt_dev);
225 /* helper routine to convert a qmt_device into lu_device */
226 static inline struct lu_device *qmt2lu_dev(struct qmt_device *qmt)
228 return &qmt->qmt_dt_dev.dd_lu_dev;
231 #define LQE_ROOT(lqe) (lqe2qpi(lqe)->qpi_root)
232 #define LQE_GLB_OBJ(lqe) (lqe2qpi(lqe)->qpi_glb_obj[lqe->lqe_site->lqs_qtype])
234 /* helper function returning grace time to use for a given lquota entry */
235 static inline __u64 qmt_lqe_grace(struct lquota_entry *lqe)
237 struct qmt_pool_info *pool = lqe2qpi(lqe);
238 struct lquota_entry *grace_lqe;
240 grace_lqe = pool->qpi_grace_lqe[lqe->lqe_site->lqs_qtype];
241 LASSERT(grace_lqe != NULL);
243 return grace_lqe->lqe_gracetime;
246 static inline void qmt_restore(struct lquota_entry *lqe,
247 struct qmt_lqe_restore *restore)
249 lqe->lqe_hardlimit = restore->qlr_hardlimit;
250 lqe->lqe_softlimit = restore->qlr_softlimit;
251 lqe->lqe_gracetime = restore->qlr_gracetime;
252 lqe->lqe_granted = restore->qlr_granted;
253 lqe->lqe_qunit = restore->qlr_qunit;
256 #define QMT_GRANT(lqe, slv, cnt) \
258 (lqe)->lqe_granted += (cnt); \
261 #define QMT_REL(lqe, slv, cnt) \
263 (lqe)->lqe_granted -= (cnt); \
267 /* helper routine returning true when reached hardlimit */
268 static inline bool qmt_hard_exhausted(struct lquota_entry *lqe)
270 if (lqe->lqe_hardlimit != 0 && lqe->lqe_granted >= lqe->lqe_hardlimit)
275 /* helper routine returning true when reached softlimit */
276 static inline bool qmt_soft_exhausted(struct lquota_entry *lqe, __u64 now)
278 if (lqe->lqe_softlimit != 0 && lqe->lqe_granted > lqe->lqe_softlimit &&
279 lqe->lqe_gracetime != 0 && now >= lqe->lqe_gracetime)
284 /* helper routine returning true when the id has run out of quota space:
285 * - reached hardlimit
287 * - reached softlimit and grace time expired already */
288 static inline bool qmt_space_exhausted(struct lquota_entry *lqe, __u64 now)
290 return (qmt_hard_exhausted(lqe) || qmt_soft_exhausted(lqe, now));
293 /* number of seconds to wait for slaves to release quota space after
295 #define QMT_REBA_TIMEOUT 2
298 void qmt_pool_fini(const struct lu_env *, struct qmt_device *);
299 int qmt_pool_init(const struct lu_env *, struct qmt_device *);
300 int qmt_pool_prepare(const struct lu_env *, struct qmt_device *,
302 int qmt_pool_new_conn(const struct lu_env *, struct qmt_device *,
303 struct lu_fid *, struct lu_fid *, __u64 *,
305 struct lquota_entry *qmt_pool_lqe_lookup(const struct lu_env *,
306 struct qmt_device *, int, int, int,
309 extern struct lquota_entry_operations qmt_lqe_ops;
310 struct thandle *qmt_trans_start_with_slv(const struct lu_env *,
311 struct lquota_entry *,
313 struct qmt_lqe_restore *);
314 struct thandle *qmt_trans_start(const struct lu_env *, struct lquota_entry *,
315 struct qmt_lqe_restore *);
316 int qmt_glb_write(const struct lu_env *, struct thandle *,
317 struct lquota_entry *, __u32, __u64 *);
318 int qmt_slv_write(const struct lu_env *, struct thandle *,
319 struct lquota_entry *, struct dt_object *, __u32, __u64 *,
321 int qmt_slv_read(const struct lu_env *, struct lquota_entry *,
322 struct dt_object *, __u64 *);
323 int qmt_validate_limits(struct lquota_entry *, __u64, __u64);
324 void qmt_adjust_qunit(const struct lu_env *, struct lquota_entry *);
325 void qmt_adjust_edquot(struct lquota_entry *, __u64);
326 void qmt_revalidate(const struct lu_env *, struct lquota_entry *);
327 __u64 qmt_alloc_expand(struct lquota_entry *, __u64, __u64);
330 int qmt_dqacq0(const struct lu_env *, struct lquota_entry *,
331 struct qmt_device *, struct obd_uuid *, __u32, __u64, __u64,
332 struct quota_body *);
335 int qmt_intent_policy(const struct lu_env *, struct lu_device *,
336 struct ptlrpc_request *, struct ldlm_lock **, int);
337 int qmt_lvbo_init(struct lu_device *, struct ldlm_resource *);
338 int qmt_lvbo_update(struct lu_device *, struct ldlm_resource *,
339 struct ptlrpc_request *, int);
340 int qmt_lvbo_size(struct lu_device *, struct ldlm_lock *);
341 int qmt_lvbo_fill(struct lu_device *, struct ldlm_lock *, void *, int);
342 int qmt_lvbo_free(struct lu_device *, struct ldlm_resource *);
343 int qmt_start_reba_thread(struct qmt_device *);
344 void qmt_stop_reba_thread(struct qmt_device *);
345 void qmt_glb_lock_notify(const struct lu_env *, struct lquota_entry *, __u64);
346 void qmt_id_lock_notify(struct qmt_device *, struct lquota_entry *);
347 #endif /* _QMT_INTERNAL_H */