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.
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 < LL_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 int qpi_soft_least_qunit_seq_show(struct seq_file *m, void *data)
176 struct qmt_pool_info *pool = m->private;
177 LASSERT(pool != NULL);
179 seq_printf(m, "%lu\n", pool->qpi_soft_least_qunit);
184 qpi_soft_least_qunit_seq_write(struct file *file, const char __user *buffer,
185 size_t count, loff_t *off)
187 struct qmt_pool_info *pool;
191 pool = ((struct seq_file *)file->private_data)->private;
192 LASSERT(pool != NULL);
194 /* Not tuneable for inode limit */
195 if (pool->qpi_key >> 16 != LQUOTA_RES_DT)
198 rc = lprocfs_str_to_s64(buffer, count, &least_qunit);
202 /* Miminal qpi_soft_least_qunit */
203 qunit = pool->qpi_least_qunit << 2;
204 /* The value must be power of miminal qpi_soft_least_qunit, see
205 * how the qunit is adjusted in qmt_adjust_qunit(). */
206 while (qunit > 0 && qunit < least_qunit)
209 qunit = INT_MAX & ~3;
211 pool->qpi_soft_least_qunit = qunit;
214 LPROC_SEQ_FOPS(qpi_soft_least_qunit);
216 static struct lprocfs_vars lprocfs_quota_qpi_vars[] = {
218 .fops = &qpi_state_fops },
219 { .name = "soft_least_qunit",
220 .fops = &qpi_soft_least_qunit_fops },
225 * Allocate a new qmt_pool_info structure and add it to the pool hash table
228 * \param env - is the environment passed by the caller
229 * \param qmt - is the quota master target
230 * \param pool_id - is the 16-bit pool identifier of the new pool to add
231 * \param pool_type - is the resource type of this pool instance, either
232 * LQUOTA_RES_MD or LQUOTA_RES_DT.
234 * \retval - 0 on success, appropriate error on failure
236 static int qmt_pool_alloc(const struct lu_env *env, struct qmt_device *qmt,
237 int pool_id, int pool_type)
239 struct qmt_thread_info *qti = qmt_info(env);
240 struct qmt_pool_info *pool;
247 INIT_LIST_HEAD(&pool->qpi_linkage);
249 /* assign key used by hash functions */
250 pool->qpi_key = pool_id + (pool_type << 16);
252 /* initialize refcount to 1, hash table will then grab an additional
254 atomic_set(&pool->qpi_ref, 1);
256 /* set up least qunit size to use for this pool */
257 pool->qpi_least_qunit = LQUOTA_LEAST_QUNIT(pool_type);
258 if (pool_type == LQUOTA_RES_DT)
259 pool->qpi_soft_least_qunit = pool->qpi_least_qunit << 2;
261 pool->qpi_soft_least_qunit = pool->qpi_least_qunit;
263 /* create pool proc directory */
264 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
265 pool->qpi_proc = lprocfs_register(qti->qti_buf, qmt->qmt_proc,
266 lprocfs_quota_qpi_vars, pool);
267 if (IS_ERR(pool->qpi_proc)) {
268 rc = PTR_ERR(pool->qpi_proc);
269 CERROR("%s: failed to create proc entry for pool %s (%d)\n",
270 qmt->qmt_svname, qti->qti_buf, rc);
271 pool->qpi_proc = NULL;
275 /* grab reference on master target that this pool belongs to */
276 lu_device_get(qmt2lu_dev(qmt));
277 lu_ref_add(&qmt2lu_dev(qmt)->ld_reference, "pool", pool);
280 /* add to qmt hash table */
281 rc = cfs_hash_add_unique(qmt->qmt_pool_hash, &pool->qpi_key,
284 CERROR("%s: failed to add pool %s to qmt hash (%d)\n",
285 qmt->qmt_svname, qti->qti_buf, rc);
289 /* add to qmt pool list */
290 list_add_tail(&pool->qpi_linkage, &qmt->qmt_pool_list);
294 /* this frees the pool structure since refcount is equal to 1 */
295 qpi_putref(env, pool);
300 * Delete a qmt_pool_info instance and all structures associated.
302 * \param env - is the environment passed by the caller
303 * \param pool - is the qmt_pool_info structure to free
305 static void qmt_pool_free(const struct lu_env *env, struct qmt_pool_info *pool)
310 /* release proc entry */
311 if (pool->qpi_proc) {
312 lprocfs_remove(&pool->qpi_proc);
313 pool->qpi_proc = NULL;
316 /* release per-quota type site used to manage quota entries as well as
317 * references to global index files */
318 for (qtype = 0; qtype < LL_MAXQUOTAS; qtype++) {
319 /* release lqe storing grace time */
320 if (pool->qpi_grace_lqe[qtype] != NULL)
321 lqe_putref(pool->qpi_grace_lqe[qtype]);
324 if (pool->qpi_site[qtype] != NULL &&
325 !IS_ERR(pool->qpi_site[qtype]))
326 lquota_site_free(env, pool->qpi_site[qtype]);
327 /* release reference to global index */
328 if (pool->qpi_glb_obj[qtype] != NULL &&
329 !IS_ERR(pool->qpi_glb_obj[qtype]))
330 dt_object_put(env, pool->qpi_glb_obj[qtype]);
333 /* release reference on pool directory */
334 if (pool->qpi_root != NULL && !IS_ERR(pool->qpi_root))
335 dt_object_put(env, pool->qpi_root);
337 /* release reference on the master target */
338 if (pool->qpi_qmt != NULL) {
339 struct lu_device *ld = qmt2lu_dev(pool->qpi_qmt);
342 lu_ref_del(&ld->ld_reference, "pool", pool);
343 pool->qpi_qmt = NULL;
346 LASSERT(list_empty(&pool->qpi_linkage));
351 * Look-up a pool in the hash table based on the pool ID and type.
353 * \param env - is the environment passed by the caller
354 * \param qmt - is the quota master target
355 * \param pool_id - is the 16-bit identifier of the pool to look up
356 * \param pool_type - is the type of this pool, either LQUOTA_RES_MD or
359 static struct qmt_pool_info *qmt_pool_lookup(const struct lu_env *env,
360 struct qmt_device *qmt,
361 int pool_id, int pool_type)
363 struct qmt_pool_info *pool;
367 LASSERT(qmt->qmt_pool_hash != NULL);
369 /* look-up pool in hash table */
370 key = pool_id + (pool_type << 16);
371 pool = cfs_hash_lookup(qmt->qmt_pool_hash, (void *)&key);
373 /* this qmt isn't managing this pool! */
374 CERROR("%s: looking up quota entry for a pool (0x%x/%d) which "
375 "isn't managed by this quota master target\n",
376 qmt->qmt_svname, pool_id, pool_type);
377 RETURN(ERR_PTR(-ENOENT));
383 * Functions implementing the pool API, used by the qmt handlers
387 * Destroy all pools which are still in the hash table and free the pool
390 * \param env - is the environment passed by the caller
391 * \param qmt - is the quota master target
394 void qmt_pool_fini(const struct lu_env *env, struct qmt_device *qmt)
396 struct qmt_pool_info *pool;
397 struct list_head *pos, *n;
400 if (qmt->qmt_pool_hash == NULL)
403 /* parse list of pool and destroy each element */
404 list_for_each_safe(pos, n, &qmt->qmt_pool_list) {
405 pool = list_entry(pos, struct qmt_pool_info,
407 /* remove from hash */
408 cfs_hash_del(qmt->qmt_pool_hash, &pool->qpi_key,
411 /* remove from list */
412 list_del_init(&pool->qpi_linkage);
414 /* release extra reference taken in qmt_pool_alloc */
415 qpi_putref(env, pool);
417 LASSERT(list_empty(&qmt->qmt_pool_list));
419 cfs_hash_putref(qmt->qmt_pool_hash);
420 qmt->qmt_pool_hash = NULL;
425 * Initialize pool configure for the quota master target. For now, we only
426 * support the default data (i.e. all OSTs) and metadata (i.e. all the MDTs)
427 * pool which are instantiated in this function.
429 * \param env - is the environment passed by the caller
430 * \param qmt - is the quota master target for which we have to initialize the
433 * \retval - 0 on success, appropriate error on failure
435 int qmt_pool_init(const struct lu_env *env, struct qmt_device *qmt)
440 /* initialize pool hash table */
441 qmt->qmt_pool_hash = cfs_hash_create("POOL_HASH",
444 HASH_POOLS_BKT_BITS, 0,
449 if (qmt->qmt_pool_hash == NULL) {
450 CERROR("%s: failed to create pool hash table\n",
455 /* initialize pool list */
456 INIT_LIST_HEAD(&qmt->qmt_pool_list);
458 /* Instantiate pool master for the default data and metadata pool (both
459 * have pool ID equals to 0).
460 * This code will have to be revisited once we support quota on
461 * non-default pools */
462 for (res = LQUOTA_FIRST_RES; res < LQUOTA_LAST_RES; res++) {
463 rc = qmt_pool_alloc(env, qmt, 0, res);
469 qmt_pool_fini(env, qmt);
474 static int qmt_slv_cnt(const struct lu_env *env, struct lu_fid *glb_fid,
475 char *slv_name, struct lu_fid *slv_fid, void *arg)
486 * Set up on-disk index files associated with each pool.
488 * \param env - is the environment passed by the caller
489 * \param qmt - is the quota master target for which we have to initialize the
491 * \param qmt_root - is the on-disk directory created for the QMT.
493 * \retval - 0 on success, appropriate error on failure
495 int qmt_pool_prepare(const struct lu_env *env, struct qmt_device *qmt,
496 struct dt_object *qmt_root)
498 struct qmt_thread_info *qti = qmt_info(env);
499 struct lquota_glb_rec *rec = &qti->qti_glb_rec;
500 struct qmt_pool_info *pool;
501 struct dt_device *dev = NULL;
502 dt_obj_version_t version;
503 struct list_head *pos;
507 LASSERT(qmt->qmt_pool_hash != NULL);
509 /* iterate over each pool in the hash and allocate a quota site for each
510 * one. This involves creating a global index file on disk */
511 list_for_each(pos, &qmt->qmt_pool_list) {
512 struct dt_object *obj;
513 int pool_type, pool_id;
514 struct lquota_entry *lqe;
516 pool = list_entry(pos, struct qmt_pool_info,
519 pool_id = pool->qpi_key & 0x0000ffff;
520 pool_type = pool->qpi_key >> 16;
522 dev = pool->qpi_qmt->qmt_child;
524 /* allocate directory for this pool */
525 sprintf(qti->qti_buf, "%s-0x%x", RES_NAME(pool_type), pool_id);
526 obj = lquota_disk_dir_find_create(env, qmt->qmt_child, qmt_root,
529 RETURN(PTR_ERR(obj));
530 pool->qpi_root = obj;
532 for (qtype = 0; qtype < LL_MAXQUOTAS; qtype++) {
533 /* Generating FID of global index in charge of storing
534 * settings for this quota type */
535 lquota_generate_fid(&qti->qti_fid, pool_id, pool_type,
538 /* open/create the global index file for this quota
540 obj = lquota_disk_glb_find_create(env, dev,
542 &qti->qti_fid, false);
545 CERROR("%s: failed to create glb index copy for %s type: rc = %d\n",
546 qmt->qmt_svname, qtype_name(qtype), rc);
550 pool->qpi_glb_obj[qtype] = obj;
552 version = dt_version_get(env, obj);
553 /* set default grace time for newly created index */
555 rec->qbr_hardlimit = 0;
556 rec->qbr_softlimit = 0;
557 rec->qbr_granted = 0;
558 rec->qbr_time = pool_type == LQUOTA_RES_MD ?
559 MAX_IQ_TIME : MAX_DQ_TIME;
561 rc = lquota_disk_write_glb(env, obj, 0, rec);
563 CERROR("%s: failed to set default grace time for %s type: rc = %d\n",
564 qmt->qmt_svname, qtype_name(qtype), rc);
568 rc = lquota_disk_update_ver(env, dev, obj, 1);
570 CERROR("%s: failed to set initial version for %s type: rc = %d\n",
571 qmt->qmt_svname, qtype_name(qtype), rc);
576 /* create quota entry site for this quota type */
577 pool->qpi_site[qtype] = lquota_site_alloc(env, pool,
580 if (IS_ERR(pool->qpi_site[qtype])) {
581 rc = PTR_ERR(pool->qpi_site[qtype]);
582 CERROR("%s: failed to create site for %s type: rc = %d\n",
583 qmt->qmt_svname, qtype_name(qtype), rc);
587 /* count number of slaves which already connected to
588 * the master in the past */
589 pool->qpi_slv_nr[qtype] = 0;
590 rc = lquota_disk_for_each_slv(env, pool->qpi_root,
593 &pool->qpi_slv_nr[qtype]);
595 CERROR("%s: failed to scan & count slave indexes for %s type: rc = %d\n",
596 qmt->qmt_svname, qtype_name(qtype), rc);
600 /* Global grace time is stored in quota settings of
602 qti->qti_id.qid_uid = 0;
604 /* look-up quota entry storing grace time */
605 lqe = lqe_locate(env, pool->qpi_site[qtype],
608 RETURN(PTR_ERR(lqe));
609 pool->qpi_grace_lqe[qtype] = lqe;
610 #ifdef CONFIG_PROC_FS
611 /* add procfs file to dump the global index, mostly for
612 * debugging purpose */
613 snprintf(qti->qti_buf, MTI_NAME_MAXLEN,
614 "glb-%s", qtype_name(qtype));
615 rc = lprocfs_seq_create(pool->qpi_proc, qti->qti_buf,
616 0444, &lprocfs_quota_seq_fops,
619 CWARN("%s: Error adding procfs file for global"
620 "quota index "DFID", rc:%d\n",
621 qmt->qmt_svname, PFID(&qti->qti_fid), rc);
630 * Handle new slave connection. Called when a slave enqueues the global quota
631 * lock at the beginning of the reintegration procedure.
633 * \param env - is the environment passed by the caller
634 * \parap qmt - is the quota master target handling this request
635 * \param glb_fid - is the fid of the global index file
636 * \param slv_fid - is the fid of the newly created slave index file
637 * \param slv_ver - is the current version of the slave index file
638 * \param uuid - is the uuid of slave which is (re)connecting to the master
641 * \retval - 0 on success, appropriate error on failure
643 int qmt_pool_new_conn(const struct lu_env *env, struct qmt_device *qmt,
644 struct lu_fid *glb_fid, struct lu_fid *slv_fid,
645 __u64 *slv_ver, struct obd_uuid *uuid)
647 struct qmt_pool_info *pool;
648 struct dt_object *slv_obj;
649 int pool_id, pool_type, qtype;
650 bool created = false;
653 /* extract pool info from global index FID */
654 rc = lquota_extract_fid(glb_fid, &pool_id, &pool_type, &qtype);
658 /* look-up pool in charge of this global index FID */
659 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
661 RETURN(PTR_ERR(pool));
663 /* look-up slave index file */
664 slv_obj = lquota_disk_slv_find(env, qmt->qmt_child, pool->qpi_root,
666 if (IS_ERR(slv_obj) && PTR_ERR(slv_obj) == -ENOENT) {
667 /* create slave index file */
668 slv_obj = lquota_disk_slv_find_create(env, qmt->qmt_child,
669 pool->qpi_root, glb_fid,
673 if (IS_ERR(slv_obj)) {
674 rc = PTR_ERR(slv_obj);
675 CERROR("%s: failed to create quota slave index file for %s (%d)"
676 "\n", qmt->qmt_svname, obd_uuid2str(uuid), rc);
680 /* retrieve slave fid & current object version */
681 memcpy(slv_fid, lu_object_fid(&slv_obj->do_lu), sizeof(*slv_fid));
682 *slv_ver = dt_version_get(env, slv_obj);
683 dt_object_put(env, slv_obj);
685 pool->qpi_slv_nr[qtype]++;
687 qpi_putref(env, pool);
692 * Look-up a lquota_entry in the pool hash and allocate it if not found.
694 * \param env - is the environment passed by the caller
695 * \param qmt - is the quota master target for which we have to initialize the
697 * \param pool_id - is the 16-bit identifier of the pool
698 * \param pool_type - is the pool type, either LQUOTA_RES_MD or LQUOTA_RES_DT.
699 * \param qtype - is the quota type, either user or group.
700 * \param qid - is the quota ID to look-up
702 * \retval - valid pointer to lquota entry on success, appropriate error on
705 struct lquota_entry *qmt_pool_lqe_lookup(const struct lu_env *env,
706 struct qmt_device *qmt,
707 int pool_id, int pool_type,
708 int qtype, union lquota_id *qid)
710 struct qmt_pool_info *pool;
711 struct lquota_entry *lqe;
714 /* look-up pool responsible for this global index FID */
715 pool = qmt_pool_lookup(env, qmt, pool_id, pool_type);
717 RETURN((void *)pool);
719 if (qid->qid_uid == 0) {
720 /* caller wants to access grace time, no need to look up the
721 * entry since we keep a reference on ID 0 all the time */
722 lqe = pool->qpi_grace_lqe[qtype];
727 /* now that we have the pool, let's look-up the quota entry in the
728 * right quota site */
729 lqe = lqe_locate(env, pool->qpi_site[qtype], qid);
731 qpi_putref(env, pool);