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>
31 #define DEBUG_SUBSYSTEM S_LQUOTA
33 #include <lustre_dlm.h>
34 #include <obd_class.h>
36 #include "qmt_internal.h"
38 /* intent policy function called from mdt_intent_opc() when the intent is of
40 int qmt_intent_policy(const struct lu_env *env, struct lu_device *ld,
41 struct ptlrpc_request *req, struct ldlm_lock **lockp,
44 struct qmt_device *qmt = lu2qmt_dev(ld);
45 struct ldlm_intent *it;
46 struct quota_body *reqbody;
47 struct quota_body *repbody;
48 struct obd_uuid *uuid;
49 struct lquota_lvb *lvb;
50 struct ldlm_resource *res = (*lockp)->l_resource;
54 req_capsule_extend(&req->rq_pill, &RQF_LDLM_INTENT_QUOTA);
55 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
56 ldlm_lvbo_size(*lockp));
58 /* extract quota body and intent opc */
59 it = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
61 RETURN(err_serious(-EFAULT));
63 reqbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
65 RETURN(err_serious(-EFAULT));
68 rc = req_capsule_server_pack(&req->rq_pill);
70 CERROR("Can't pack response, rc %d\n", rc);
71 RETURN(err_serious(rc));
74 repbody = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_BODY);
76 RETURN(err_serious(-EFAULT));
78 uuid = &(*lockp)->l_export->exp_client_uuid;
81 case IT_QUOTA_DQACQ: {
82 struct lquota_entry *lqe;
83 struct ldlm_lock *lock;
85 if (res->lr_name.name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] == 0)
86 /* acquire on global lock? something is wrong ... */
87 GOTO(out, rc = -EPROTO);
89 /* verify global lock isn't stale */
90 if (!lustre_handle_is_used(&reqbody->qb_glb_lockh))
91 GOTO(out, rc = -ENOLCK);
93 lock = ldlm_handle2lock(&reqbody->qb_glb_lockh);
95 GOTO(out, rc = -ENOLCK);
98 lqe = res->lr_lvb_data;
102 /* acquire quota space */
103 rc = qmt_dqacq0(env, lqe, qmt, uuid, reqbody->qb_flags,
104 reqbody->qb_count, reqbody->qb_usage,
113 /* new connection from slave */
115 if (res->lr_name.name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] != 0)
116 /* connection on per-ID lock? something is wrong ... */
117 GOTO(out, rc = -EPROTO);
119 rc = qmt_pool_new_conn(env, qmt, &reqbody->qb_fid,
120 &repbody->qb_slv_fid,
121 &repbody->qb_slv_ver, uuid);
127 CERROR("%s: invalid intent opcode: "LPU64"\n", qmt->qmt_svname,
129 GOTO(out, rc = err_serious(-EINVAL));
132 /* on success, pack lvb in reply */
133 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
134 lvb_len = ldlm_lvbo_size(*lockp);
135 lvb_len = ldlm_lvbo_fill(*lockp, lvb, lvb_len);
137 GOTO(out, rc = lvb_len);
139 req_capsule_shrink(&req->rq_pill, &RMF_DLM_LVB, lvb_len, RCL_SERVER);
146 * Initialize quota LVB associated with quota indexes.
147 * Called with res->lr_lvb_sem held
149 int qmt_lvbo_init(struct lu_device *ld, struct ldlm_resource *res)
152 struct qmt_thread_info *qti;
153 struct qmt_device *qmt = lu2qmt_dev(ld);
154 int pool_id, pool_type, qtype;
158 LASSERT(res != NULL);
160 if (res->lr_type != LDLM_PLAIN)
163 if (res->lr_lvb_data ||
164 res->lr_name.name[LUSTRE_RES_ID_SEQ_OFF] != FID_SEQ_QUOTA_GLB)
171 /* initialize environment */
172 rc = lu_env_init(env, LCT_MD_THREAD);
177 /* extract global index FID and quota identifier */
178 fid_extract_from_quota_res(&qti->qti_fid, &qti->qti_id, &res->lr_name);
180 /* sanity check the global index FID */
181 rc = lquota_extract_fid(&qti->qti_fid, &pool_id, &pool_type, &qtype);
183 CERROR("can't extract pool information from FID "DFID"\n",
184 PFID(&qti->qti_fid));
188 if (res->lr_name.name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] != 0) {
189 /* no ID quota lock associated with UID/GID 0 or with a seq 0,
190 * we are thus dealing with an ID lock. */
191 struct lquota_entry *lqe;
193 /* Find the quota entry associated with the quota id */
194 lqe = qmt_pool_lqe_lookup(env, qmt, pool_id, pool_type, qtype,
197 GOTO(out, rc = PTR_ERR(lqe));
199 /* store reference to lqe in lr_lvb_data */
200 res->lr_lvb_data = lqe;
201 LQUOTA_DEBUG(lqe, "initialized res lvb");
203 struct dt_object *obj;
205 /* lookup global index */
206 obj = dt_locate(env, qmt->qmt_child, &qti->qti_fid);
208 GOTO(out, rc = PTR_ERR(obj));
209 if (!dt_object_exists(obj)) {
210 lu_object_put(env, &obj->do_lu);
211 GOTO(out, rc = -ENOENT);
214 /* store reference to global index object in lr_lvb_data */
215 res->lr_lvb_data = obj;
216 CDEBUG(D_QUOTA, DFID" initialized lvb\n", PFID(&qti->qti_fid));
219 res->lr_lvb_len = sizeof(struct lquota_lvb);
229 * Update LVB associated with the global quota index.
230 * This function is called from the DLM itself after a glimpse callback, in this
231 * case valid ptlrpc request is passed.
233 int qmt_lvbo_update(struct lu_device *ld, struct ldlm_resource *res,
234 struct ptlrpc_request *req, int increase_only)
237 struct qmt_thread_info *qti;
238 struct qmt_device *qmt = lu2qmt_dev(ld);
239 struct lquota_entry *lqe;
240 struct lquota_lvb *lvb;
241 struct ldlm_lock *lock;
242 struct obd_export *exp;
246 LASSERT(res != NULL);
251 if (res->lr_name.name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] == 0)
252 /* no need to update lvb for global quota locks */
255 lvb = req_capsule_server_swab_get(&req->rq_pill, &RMF_DLM_LVB,
256 lustre_swab_lquota_lvb);
258 CERROR("%s: failed to extract lvb from request\n",
263 lqe = res->lr_lvb_data;
264 LASSERT(lqe != NULL);
267 LQUOTA_DEBUG(lqe, "releasing:"LPU64" may release:"LPU64,
268 lvb->lvb_id_rel, lvb->lvb_id_may_rel);
270 if (lvb->lvb_id_rel == 0) {
271 /* nothing to release */
272 if (lvb->lvb_id_may_rel != 0)
273 /* but might still release later ... */
274 lqe->lqe_may_rel += lvb->lvb_id_may_rel;
275 GOTO(out_lqe, rc = 0);
278 /* allocate environement */
281 GOTO(out_lqe, rc = -ENOMEM);
283 /* initialize environment */
284 rc = lu_env_init(env, LCT_MD_THREAD);
289 /* The request is a glimpse callback which was sent via the
290 * reverse import to the slave. What we care about here is the
291 * export associated with the slave and req->rq_export is
292 * definitely not what we are looking for (it is actually set to
294 * Therefore we extract the lock from the request argument
295 * and use lock->l_export. */
296 lock = ldlm_request_lock(req);
298 CERROR("%s: failed to get lock from request!\n",
300 GOTO(out_env_init, rc = PTR_ERR(lock));
303 exp = class_export_get(lock->l_export);
305 CERROR("%s: failed to get export from lock!\n",
307 GOTO(out_env_init, rc = -EFAULT);
310 /* release quota space */
311 rc = qmt_dqacq0(env, lqe, qmt, &exp->exp_client_uuid,
312 QUOTA_DQACQ_FL_REL, lvb->lvb_id_rel, 0, &qti->qti_body);
313 if (rc || qti->qti_body.qb_count != lvb->lvb_id_rel)
314 LQUOTA_ERROR(lqe, "failed to release quota space on glimpse "
315 LPU64"!="LPU64" rc:%d\n", qti->qti_body.qb_count,
316 lvb->lvb_id_rel, rc);
317 class_export_put(exp);
319 GOTO(out_env_init, rc);
331 * Report size of lvb to ldlm layer in order to allocate lvb buffer
332 * As far as quota locks are concerned, the size is static and is the same
333 * for both global and per-ID locks which shares the same lvb format.
335 int qmt_lvbo_size(struct lu_device *ld, struct ldlm_lock *lock)
337 return sizeof(struct lquota_lvb);
341 * Fill request buffer with quota lvb
343 int qmt_lvbo_fill(struct lu_device *ld, struct ldlm_lock *lock, void *lvb,
346 struct ldlm_resource *res = lock->l_resource;
347 struct lquota_lvb *qlvb = lvb;
350 LASSERT(res != NULL);
352 if (res->lr_type != LDLM_PLAIN || res->lr_lvb_data == NULL ||
353 res->lr_name.name[LUSTRE_RES_ID_SEQ_OFF] != FID_SEQ_QUOTA_GLB)
356 if (res->lr_name.name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] != 0) {
357 /* no ID quota lock associated with UID/GID 0 or with a seq 0,
358 * we are thus dealing with an ID lock. */
359 struct lquota_entry *lqe = res->lr_lvb_data;
361 /* return current qunit value & edquot flags in lvb */
363 qlvb->lvb_id_qunit = lqe->lqe_qunit;
366 qlvb->lvb_flags = LQUOTA_FL_EDQUOT;
369 /* global quota lock */
372 struct dt_object *obj = res->lr_lvb_data;
378 /* initialize environment */
379 rc = lu_env_init(env, LCT_LOCAL);
385 /* return current version of global index */
386 qlvb->lvb_glb_ver = dt_version_get(env, obj);
392 RETURN(sizeof(struct lquota_lvb));
396 * Free lvb associated with a given ldlm resource
397 * we don't really allocate a lvb, lr_lvb_data just points to
398 * the appropriate backend structures.
400 int qmt_lvbo_free(struct lu_device *ld, struct ldlm_resource *res)
404 if (res->lr_lvb_data == NULL)
407 if (res->lr_name.name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] != 0) {
408 struct lquota_entry *lqe = res->lr_lvb_data;
410 /* release lqe reference */
413 struct dt_object *obj = res->lr_lvb_data;
421 /* initialize environment */
422 rc = lu_env_init(env, LCT_LOCAL);
428 /* release object reference */
429 lu_object_put(env, &obj->do_lu);
434 res->lr_lvb_data = NULL;
440 typedef int (*qmt_glimpse_cb_t)(const struct lu_env *, struct qmt_device *,
441 struct obd_uuid *, union ldlm_gl_desc *,
444 * Send glimpse callback to slaves holding a lock on resource \res.
445 * This is used to notify slaves of new quota settings or to claim quota space
448 * \param env - is the environment passed by the caller
449 * \param qmt - is the quota master target
450 * \param res - is the dlm resource associated with the quota object
451 * \param desc - is the glimpse descriptor to pack in glimpse callback
452 * \param cb - is the callback function called on every lock and determine
453 * whether a glimpse should be issued
454 * \param arg - is an opaq parameter passed to the callback function
456 static int qmt_glimpse_lock(const struct lu_env *env, struct qmt_device *qmt,
457 struct ldlm_resource *res, union ldlm_gl_desc *desc,
458 qmt_glimpse_cb_t cb, void *arg)
460 cfs_list_t *tmp, *pos;
461 CFS_LIST_HEAD(gl_list);
466 /* scan list of granted locks */
467 cfs_list_for_each(pos, &res->lr_granted) {
468 struct ldlm_glimpse_work *work;
469 struct ldlm_lock *lock;
470 struct obd_uuid *uuid;
472 lock = cfs_list_entry(pos, struct ldlm_lock, l_res_link);
473 LASSERT(lock->l_export);
474 uuid = &lock->l_export->exp_client_uuid;
477 rc = cb(env, qmt, uuid, desc, arg);
479 /* slave should not be notified */
482 /* something wrong happened, we still notify */
483 CERROR("%s: callback function failed to "
484 "determine whether slave %s should be "
485 "notified (%d)\n", qmt->qmt_svname,
486 obd_uuid2str(uuid), rc);
491 CERROR("%s: failed to notify %s\n", qmt->qmt_svname,
496 cfs_list_add_tail(&work->gl_list, &gl_list);
497 work->gl_lock = LDLM_LOCK_GET(lock);
499 work->gl_desc = desc;
504 if (cfs_list_empty(&gl_list)) {
505 CDEBUG(D_QUOTA, "%s: nobody to notify\n", qmt->qmt_svname);
509 /* issue glimpse callbacks to all connected slaves */
510 rc = ldlm_glimpse_locks(res, &gl_list);
512 cfs_list_for_each_safe(pos, tmp, &gl_list) {
513 struct ldlm_glimpse_work *work;
515 work = cfs_list_entry(pos, struct ldlm_glimpse_work, gl_list);
517 cfs_list_del(&work->gl_list);
518 CERROR("%s: failed to notify %s of new quota settings\n",
520 obd_uuid2str(&work->gl_lock->l_export->exp_client_uuid));
521 LDLM_LOCK_RELEASE(work->gl_lock);
529 * Send glimpse request to all global quota locks to push new quota setting to
532 * \param env - is the environment passed by the caller
533 * \param lqe - is the lquota entry which has new settings
534 * \param ver - is the version associated with the setting change
536 void qmt_glb_lock_notify(const struct lu_env *env, struct lquota_entry *lqe,
539 struct qmt_thread_info *qti = qmt_info(env);
540 struct qmt_pool_info *pool = lqe2qpi(lqe);
541 struct ldlm_resource *res = NULL;
545 lquota_generate_fid(&qti->qti_fid, pool->qpi_key & 0x0000ffff,
546 pool->qpi_key >> 16, lqe->lqe_site->lqs_qtype);
548 /* send glimpse callback to notify slaves of new quota settings */
549 qti->qti_gl_desc.lquota_desc.gl_id = lqe->lqe_id;
550 qti->qti_gl_desc.lquota_desc.gl_flags = 0;
551 qti->qti_gl_desc.lquota_desc.gl_hardlimit = lqe->lqe_hardlimit;
552 qti->qti_gl_desc.lquota_desc.gl_softlimit = lqe->lqe_softlimit;
553 qti->qti_gl_desc.lquota_desc.gl_time = lqe->lqe_gracetime;
554 qti->qti_gl_desc.lquota_desc.gl_ver = ver;
556 /* look up ldlm resource associated with global index */
557 fid_build_reg_res_name(&qti->qti_fid, &qti->qti_resid);
558 res = ldlm_resource_get(pool->qpi_qmt->qmt_ns, NULL, &qti->qti_resid,
561 /* this might happen if no slaves have enqueued global quota
563 LQUOTA_DEBUG(lqe, "failed to lookup ldlm resource associated "
564 "with "DFID, PFID(&qti->qti_fid));
568 rc = qmt_glimpse_lock(env, pool->qpi_qmt, res, &qti->qti_gl_desc,
570 ldlm_resource_putref(res);
574 /* Callback function used to select locks that should be glimpsed when
575 * broadcasting the new qunit value */
576 static int qmt_id_lock_cb(const struct lu_env *env, struct qmt_device *qmt,
577 struct obd_uuid *uuid, union ldlm_gl_desc *desc,
580 struct obd_uuid *slv_uuid = arg;
582 if (slv_uuid != NULL && obd_uuid_equals(uuid, slv_uuid))
588 * Send glimpse request on per-ID lock to push new qunit value to slave.
590 * \param env - is the environment passed by the caller
591 * \param qmt - is the quota master target device
592 * \param lqe - is the lquota entry with the new qunit value
593 * \param uuid - is the uuid of the slave acquiring space, if any
595 static void qmt_id_lock_glimpse(const struct lu_env *env,
596 struct qmt_device *qmt,
597 struct lquota_entry *lqe, struct obd_uuid *uuid)
599 struct qmt_thread_info *qti = qmt_info(env);
600 struct qmt_pool_info *pool = lqe2qpi(lqe);
601 struct ldlm_resource *res = NULL;
605 if (!lqe->lqe_enforced)
608 lquota_generate_fid(&qti->qti_fid, pool->qpi_key & 0x0000ffff,
609 pool->qpi_key >> 16, lqe->lqe_site->lqs_qtype);
610 fid_build_quota_res_name(&qti->qti_fid, &lqe->lqe_id, &qti->qti_resid);
611 res = ldlm_resource_get(qmt->qmt_ns, NULL, &qti->qti_resid, LDLM_PLAIN,
614 /* this might legitimately happens if slaves haven't had the
615 * opportunity to enqueue quota lock yet. */
616 LQUOTA_DEBUG(lqe, "failed to lookup ldlm resource for per-ID "
617 "lock "DFID, PFID(&qti->qti_fid));
619 if (lqe->lqe_revoke_time == 0 &&
620 lqe->lqe_qunit == pool->qpi_least_qunit)
621 lqe->lqe_revoke_time = cfs_time_current_64();
622 lqe_write_unlock(lqe);
627 /* The purpose of glimpse callback on per-ID lock is twofold:
628 * - notify slaves of new qunit value and hope they will release some
629 * spare quota space in return
630 * - notify slaves that master ran out of quota space and there is no
631 * need to send acquire request any more until further notice */
633 /* fill glimpse descriptor with lqe settings */
635 qti->qti_gl_desc.lquota_desc.gl_flags = LQUOTA_FL_EDQUOT;
637 qti->qti_gl_desc.lquota_desc.gl_flags = 0;
638 qti->qti_gl_desc.lquota_desc.gl_qunit = lqe->lqe_qunit;
640 if (lqe->lqe_revoke_time == 0 &&
641 qti->qti_gl_desc.lquota_desc.gl_qunit == pool->qpi_least_qunit)
642 /* reset lqe_may_rel, it will be updated on glimpse callback
643 * replies if needed */
644 lqe->lqe_may_rel = 0;
646 /* The rebalance thread is the only thread which can issue glimpses */
647 LASSERT(!lqe->lqe_gl);
649 lqe_write_unlock(lqe);
651 /* issue glimpse callback to slaves */
652 rc = qmt_glimpse_lock(env, qmt, res, &qti->qti_gl_desc,
653 uuid ? qmt_id_lock_cb : NULL, (void *)uuid);
656 if (lqe->lqe_revoke_time == 0 &&
657 qti->qti_gl_desc.lquota_desc.gl_qunit == pool->qpi_least_qunit &&
658 lqe->lqe_qunit == pool->qpi_least_qunit) {
659 lqe->lqe_revoke_time = cfs_time_current_64();
660 qmt_adjust_edquot(lqe, cfs_time_current_sec());
662 LASSERT(lqe->lqe_gl);
664 lqe_write_unlock(lqe);
666 ldlm_resource_putref(res);
671 * Schedule a glimpse request on per-ID locks to push new qunit value or
672 * edquot flag to quota slaves.
674 * \param qmt - is the quota master target device
675 * \param lqe - is the lquota entry with the new qunit value
677 void qmt_id_lock_notify(struct qmt_device *qmt, struct lquota_entry *lqe)
683 spin_lock(&qmt->qmt_reba_lock);
684 if (!qmt->qmt_stopping && cfs_list_empty(&lqe->lqe_link)) {
685 cfs_list_add_tail(&lqe->lqe_link, &qmt->qmt_reba_list);
688 spin_unlock(&qmt->qmt_reba_lock);
691 wake_up(&qmt->qmt_reba_thread.t_ctl_waitq);
698 * The rebalance thread is in charge of sending glimpse callbacks on per-ID
699 * quota locks owned by slaves in order to notify them of:
700 * - a qunit shrink in which case slaves might release quota space back in
702 * - set/clear edquot flag used to cache the "quota exhausted" state of the
703 * master. When the flag is set, slaves know that there is no need to
704 * try to acquire quota from the master since this latter has already
705 * distributed all the space.
707 static int qmt_reba_thread(void *arg)
709 struct qmt_device *qmt = (struct qmt_device *)arg;
710 struct ptlrpc_thread *thread = &qmt->qmt_reba_thread;
711 struct l_wait_info lwi = { 0 };
713 struct lquota_entry *lqe, *tmp;
721 rc = lu_env_init(env, LCT_MD_THREAD);
723 CERROR("%s: failed to init env.", qmt->qmt_svname);
728 thread_set_flags(thread, SVC_RUNNING);
729 wake_up(&thread->t_ctl_waitq);
732 l_wait_event(thread->t_ctl_waitq,
733 !cfs_list_empty(&qmt->qmt_reba_list) ||
734 !thread_is_running(thread), &lwi);
736 spin_lock(&qmt->qmt_reba_lock);
737 cfs_list_for_each_entry_safe(lqe, tmp, &qmt->qmt_reba_list,
739 cfs_list_del_init(&lqe->lqe_link);
740 spin_unlock(&qmt->qmt_reba_lock);
742 if (thread_is_running(thread))
743 qmt_id_lock_glimpse(env, qmt, lqe, NULL);
746 spin_lock(&qmt->qmt_reba_lock);
748 spin_unlock(&qmt->qmt_reba_lock);
750 if (!thread_is_running(thread))
755 thread_set_flags(thread, SVC_STOPPED);
756 wake_up(&thread->t_ctl_waitq);
761 * Start rebalance thread. Called when the QMT is being setup
763 int qmt_start_reba_thread(struct qmt_device *qmt)
765 struct ptlrpc_thread *thread = &qmt->qmt_reba_thread;
766 struct l_wait_info lwi = { 0 };
767 struct task_struct *task;
770 task = kthread_run(qmt_reba_thread, (void *)qmt,
771 "qmt_reba_%s", qmt->qmt_svname);
773 CERROR("%s: failed to start rebalance thread (%ld)\n",
774 qmt->qmt_svname, PTR_ERR(task));
775 thread_set_flags(thread, SVC_STOPPED);
776 RETURN(PTR_ERR(task));
779 l_wait_event(thread->t_ctl_waitq,
780 thread_is_running(thread) || thread_is_stopped(thread),
787 * Stop rebalance thread. Called when the QMT is about to shutdown.
789 void qmt_stop_reba_thread(struct qmt_device *qmt)
791 struct ptlrpc_thread *thread = &qmt->qmt_reba_thread;
793 if (!thread_is_stopped(thread)) {
794 struct l_wait_info lwi = { 0 };
796 thread_set_flags(thread, SVC_STOPPING);
797 wake_up(&thread->t_ctl_waitq);
799 l_wait_event(thread->t_ctl_waitq, thread_is_stopped(thread),
802 LASSERT(cfs_list_empty(&qmt->qmt_reba_list));