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, 2017, 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 "qsd_internal.h"
36 * helper function bumping lqe_pending_req if there is no quota request in
37 * flight for the lquota entry \a lqe. Otherwise, EBUSY is returned.
39 static inline int qsd_request_enter(struct lquota_entry *lqe)
41 /* is there already a quota request in flight? */
42 if (lqe->lqe_pending_req != 0) {
43 LQUOTA_DEBUG(lqe, "already a request in flight");
47 if (lqe->lqe_pending_rel != 0) {
48 LQUOTA_ERROR(lqe, "no request in flight with pending_rel=%llu",
49 lqe->lqe_pending_rel);
53 lqe->lqe_pending_req++;
58 * Companion of qsd_request_enter() dropping lqe_pending_req to 0.
60 static inline void qsd_request_exit(struct lquota_entry *lqe)
62 if (lqe->lqe_pending_req != 1) {
63 LQUOTA_ERROR(lqe, "lqe_pending_req != 1!!!");
66 lqe->lqe_pending_req--;
67 lqe->lqe_pending_rel = 0;
68 wake_up_all(&lqe->lqe_waiters);
72 * Check whether a qsd instance is all set to send quota request to master.
73 * This includes checking whether:
74 * - the connection to master is set up and usable,
75 * - the qsd isn't stopping
76 * - reintegration has been successfully completed and all indexes are
79 * \param lqe - is the lquota entry for which we would like to send an quota
81 * \param lockh - is the remote handle of the global lock returned on success
83 * \retval 0 on success, appropriate error on failure
85 static int qsd_ready(struct lquota_entry *lqe, struct lustre_handle *lockh)
87 struct qsd_qtype_info *qqi = lqe2qqi(lqe);
88 struct qsd_instance *qsd = qqi->qqi_qsd;
89 struct obd_import *imp = NULL;
90 struct ldlm_lock *lock;
93 read_lock(&qsd->qsd_lock);
94 /* is the qsd about to shut down? */
95 if (qsd->qsd_stopping) {
96 read_unlock(&qsd->qsd_lock);
97 LQUOTA_DEBUG(lqe, "dropping quota req since qsd is stopping");
98 /* Target is about to shut down, client will retry */
102 /* is the connection to the quota master ready? */
103 if (qsd->qsd_exp_valid)
104 imp = class_exp2cliimp(qsd->qsd_exp);
105 if (imp == NULL || imp->imp_invalid) {
106 read_unlock(&qsd->qsd_lock);
107 LQUOTA_DEBUG(lqe, "connection to master not ready");
111 /* In most case, reintegration must have been triggered (when enable
112 * quota or on OST start), however, in rare race condition (enabling
113 * quota when starting OSTs), we might miss triggering reintegration
116 * If the previous reintegration failed for some reason, we'll
117 * re-trigger it here as well. */
118 if (!qqi->qqi_glb_uptodate || !qqi->qqi_slv_uptodate) {
119 read_unlock(&qsd->qsd_lock);
120 LQUOTA_DEBUG(lqe, "not up-to-date, dropping request and "
121 "kicking off reintegration");
122 qsd_start_reint_thread(qqi);
123 RETURN(-EINPROGRESS);
126 /* Fill the remote global lock handle, master will check this handle
127 * to see if the slave is sending request with stale lock */
128 lustre_handle_copy(lockh, &qqi->qqi_lockh);
129 read_unlock(&qsd->qsd_lock);
131 if (!lustre_handle_is_used(lockh))
134 lock = ldlm_handle2lock(lockh);
138 /* return remote lock handle to be packed in quota request */
139 lustre_handle_copy(lockh, &lock->l_remote_handle);
146 * Check whether any quota space adjustment (pre-acquire/release/report) is
147 * needed for a given quota ID. If a non-null \a qbody is passed, then the
148 * \a qbody structure (qb_count/flags/usage) is filled with appropriate data
149 * to be packed in the quota request.
151 * \param lqe - is the lquota entry for which we would like to adjust quota
153 * \param qbody - is the quota body to fill, if not NULL.
155 * \retval true - space adjustment is required and \a qbody is filled, if not
157 * \retval false - no space adjustment required
159 static bool qsd_calc_adjust(struct lquota_entry *lqe, struct quota_body *qbody)
161 __u64 usage, granted;
164 usage = lqe->lqe_usage;
165 usage += lqe->lqe_pending_write + lqe->lqe_waiting_write;
166 granted = lqe->lqe_granted;
171 if (!lqe->lqe_enforced) {
172 /* quota not enforced any more for this ID */
174 /* release all quota space unconditionally */
175 LQUOTA_DEBUG(lqe, "not enforced, releasing all space");
177 qbody->qb_count = granted;
178 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
185 if (!lustre_handle_is_used(&lqe->lqe_lockh)) {
186 /* No valid per-ID lock
187 * When reporting quota (during reintegration or on setquota
188 * glimpse), we should release granted space if usage is 0.
189 * Otherwise, if the usage is less than granted, we need to
190 * acquire the per-ID lock to make sure the unused grant can be
191 * reclaimed by per-ID lock glimpse. */
193 /* no on-disk usage and no outstanding activity, release
196 LQUOTA_DEBUG(lqe, "no usage, releasing all "
199 qbody->qb_count = granted;
200 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
204 LQUOTA_DEBUG(lqe, "no usage + no granted, nothing to "
209 if (lqe->lqe_usage < lqe->lqe_granted) {
210 /* holding quota space w/o any lock, enqueue per-ID lock
212 LQUOTA_DEBUG(lqe, "(re)acquiring per-ID lock");
215 qbody->qb_flags = QUOTA_DQACQ_FL_ACQ;
220 if (lqe->lqe_usage > lqe->lqe_granted) {
221 /* quota overrun, report usage */
222 LQUOTA_DEBUG(lqe, "overrun, reporting usage");
224 qbody->qb_usage = lqe->lqe_usage;
225 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
229 LQUOTA_DEBUG(lqe, "granted matches usage, nothing to do");
234 * Apply good old quota qunit adjustment logic which has been around
236 * 1. release spare quota space? */
237 if (granted > usage + lqe->lqe_qunit) {
238 /* pre-release quota space */
241 qbody->qb_count = granted - usage;
242 /* if usage == 0, release all granted space */
244 /* try to keep one qunit of quota space */
245 qbody->qb_count -= lqe->lqe_qunit;
246 /* but don't release less than qtune to avoid releasing
248 if (qbody->qb_count < lqe->lqe_qtune)
249 qbody->qb_count = lqe->lqe_qtune;
251 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
255 /* 2. Any quota overrun? */
256 if (lqe->lqe_usage > lqe->lqe_granted) {
257 /* we overconsumed quota space, we report usage in request so
258 * that master can adjust it unconditionally */
261 qbody->qb_usage = lqe->lqe_usage;
262 granted = lqe->lqe_usage;
263 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
266 /* 3. Time to pre-acquire? */
267 if (!lqe->lqe_edquot && !lqe->lqe_nopreacq && usage > 0 &&
268 lqe->lqe_qunit != 0 && granted < usage + lqe->lqe_qtune) {
269 /* To pre-acquire quota space, we report how much spare quota
270 * space the slave currently owns, then the master will grant us
271 * back how much we can pretend given the current state of
275 if (granted <= usage)
278 qbody->qb_count = granted - usage;
279 qbody->qb_flags |= QUOTA_DQACQ_FL_PREACQ;
284 RETURN(qbody->qb_flags != 0);
290 * Helper function returning true when quota space need to be adjusted (some
291 * unused space should be free or pre-acquire) and false otherwise.
293 static inline bool qsd_adjust_needed(struct lquota_entry *lqe)
295 return qsd_calc_adjust(lqe, NULL);
299 * Callback function called when an acquire/release request sent to the master
302 static void qsd_req_completion(const struct lu_env *env,
303 struct qsd_qtype_info *qqi,
304 struct quota_body *reqbody,
305 struct quota_body *repbody,
306 struct lustre_handle *lockh,
307 struct lquota_lvb *lvb,
310 struct lquota_entry *lqe = (struct lquota_entry *)arg;
311 struct qsd_thread_info *qti;
313 bool adjust = false, cancel = false;
316 LASSERT(qqi != NULL && lqe != NULL);
318 /* environment passed by ptlrpcd is mostly used by CLIO and hasn't the
320 rc = lu_env_refill_by_tags((struct lu_env *)env, LCT_DT_THREAD, 0);
322 LQUOTA_ERROR(lqe, "failed to refill environmnent %d", rc);
324 /* can't afford to adjust quota space with no suitable lu_env */
325 GOTO(out_noadjust, rc);
330 LQUOTA_DEBUG(lqe, "DQACQ returned %d, flags:0x%x", ret,
333 /* despite -EDQUOT & -EINPROGRESS errors, the master might still
334 * grant us back quota space to adjust quota overrun */
335 if (ret != 0 && ret != -EDQUOT && ret != -EINPROGRESS) {
336 if (ret != -ETIMEDOUT && ret != -ENOTCONN &&
337 ret != -ESHUTDOWN && ret != -EAGAIN)
338 /* print errors only if return code is unexpected */
339 LQUOTA_ERROR(lqe, "DQACQ failed with %d, flags:0x%x",
340 ret, reqbody->qb_flags);
344 /* Set the lqe_lockh */
345 if (lustre_handle_is_used(lockh) &&
346 !lustre_handle_equal(lockh, &lqe->lqe_lockh))
347 lustre_handle_copy(&lqe->lqe_lockh, lockh);
349 /* If the replied qb_count is zero, it means master didn't process
350 * the DQACQ since the limit for this ID has been removed, so we
351 * should not update quota entry & slave index copy neither. */
352 if (repbody != NULL && repbody->qb_count != 0) {
353 LQUOTA_DEBUG(lqe, "DQACQ qb_count:%llu", repbody->qb_count);
355 if (req_is_rel(reqbody->qb_flags)) {
356 if (lqe->lqe_granted < repbody->qb_count) {
357 LQUOTA_ERROR(lqe, "can't release more space "
358 "than owned %llu<%llu",
361 lqe->lqe_granted = 0;
363 lqe->lqe_granted -= repbody->qb_count;
365 /* Cancel the per-ID lock initiatively when there
366 * isn't any usage & grant, which can avoid master
367 * sending glimpse unnecessarily to this slave on
369 if (!lqe->lqe_pending_write && !lqe->lqe_granted &&
370 !lqe->lqe_waiting_write && !lqe->lqe_usage)
373 lqe->lqe_granted += repbody->qb_count;
375 qti->qti_rec.lqr_slv_rec.qsr_granted = lqe->lqe_granted;
376 lqe_write_unlock(lqe);
378 /* Update the slave index file in the dedicated thread. So far,
379 * We don't update the version of slave index copy on DQACQ.
380 * No locking is necessary since nobody can change
381 * lqe->lqe_granted while lqe->lqe_pending_req > 0 */
382 qsd_upd_schedule(qqi, lqe, &lqe->lqe_id, &qti->qti_rec, 0,
387 /* extract information from lvb */
388 if (ret == 0 && lvb != NULL) {
389 if (lvb->lvb_id_qunit != 0)
390 qsd_set_qunit(lqe, lvb->lvb_id_qunit);
391 qsd_set_edquot(lqe, !!(lvb->lvb_flags & LQUOTA_FL_EDQUOT));
392 } else if (repbody != NULL && repbody->qb_qunit != 0) {
393 qsd_set_qunit(lqe, repbody->qb_qunit);
396 /* turn off pre-acquire if it failed with -EDQUOT. This is done to avoid
397 * flooding the master with acquire request. Pre-acquire will be turned
398 * on again as soon as qunit is modified */
399 if (req_is_preacq(reqbody->qb_flags) && ret == -EDQUOT)
400 lqe->lqe_nopreacq = true;
402 adjust = qsd_adjust_needed(lqe);
403 if (reqbody && req_is_acq(reqbody->qb_flags) && ret != -EDQUOT) {
404 lqe->lqe_acq_rc = ret;
405 lqe->lqe_acq_time = ktime_get_seconds();
408 qsd_request_exit(lqe);
409 lqe_write_unlock(lqe);
411 /* release reference on per-ID lock */
412 if (lustre_handle_is_used(lockh))
413 ldlm_lock_decref(lockh, qsd_id_einfo.ei_mode);
416 qsd_adjust_schedule(lqe, false, true);
418 if (!ret || ret == -EDQUOT)
419 qsd_adjust_schedule(lqe, false, false);
421 qsd_adjust_schedule(lqe, true, false);
431 * Try to consume local quota space.
433 * \param lqe - is the qid entry to be processed
434 * \param space - is the amount of quota space needed to complete the operation
436 * \retval 0 - success
437 * \retval -EDQUOT - out of quota
438 * \retval -EAGAIN - need to acquire space from master
440 static int qsd_acquire_local(struct lquota_entry *lqe, __u64 space)
446 if (!lqe->lqe_enforced)
447 /* not enforced any more, we are good */
451 /* use latest usage */
452 usage = lqe->lqe_usage;
453 /* take pending write into account */
454 usage += lqe->lqe_pending_write;
456 if (space + usage <= lqe->lqe_granted - lqe->lqe_pending_rel) {
457 /* Yay! we got enough space */
458 lqe->lqe_pending_write += space;
459 lqe->lqe_waiting_write -= space;
461 /* lqe_edquot flag is used to avoid flooding dqacq requests when
462 * the user is over quota, however, the lqe_edquot could be stale
463 * sometimes due to the race reply of dqacq vs. id lock glimpse
464 * (see LU-4505), so we revalidate it every 5 seconds. */
465 } else if (lqe->lqe_edquot &&
466 (lqe->lqe_edquot_time > ktime_get_seconds() - 5)) {
471 lqe_write_unlock(lqe);
477 * Compute how much quota space should be acquire from the master based
478 * on how much is currently granted to this slave and pending/waiting
481 * \param lqe - is the lquota entry for which we would like to adjust quota
483 * \param qbody - is the quota body of the acquire request to fill
485 * \retval true - space acquisition is needed and qbody is filled
486 * \retval false - no space acquisition required
488 static inline bool qsd_calc_acquire(struct lquota_entry *lqe,
489 struct quota_body *qbody)
491 __u64 usage, granted;
493 usage = lqe->lqe_usage;
494 usage += lqe->lqe_pending_write + lqe->lqe_waiting_write;
495 granted = lqe->lqe_granted;
499 /* if we overconsumed quota space, we report usage in request so that
500 * master can adjust it unconditionally */
501 if (lqe->lqe_usage > lqe->lqe_granted) {
502 qbody->qb_usage = lqe->lqe_usage;
503 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
504 granted = lqe->lqe_usage;
507 /* acquire as much as needed, but not more */
508 if (usage > granted) {
509 qbody->qb_count = usage - granted;
510 qbody->qb_flags |= QUOTA_DQACQ_FL_ACQ;
513 return qbody->qb_flags != 0;
517 * Acquire quota space from master.
518 * There are at most 1 in-flight dqacq/dqrel.
520 * \param env - the environment passed by the caller
521 * \param lqe - is the qid entry to be processed
523 * \retval 0 - success
524 * \retval -EDQUOT - out of quota
525 * \retval -EINPROGRESS - inform client to retry write/create
526 * \retval -EBUSY - already a quota request in flight
527 * \retval -ve - other appropriate errors
529 static int qsd_acquire_remote(const struct lu_env *env,
530 struct lquota_entry *lqe)
532 struct qsd_thread_info *qti = qsd_info(env);
533 struct quota_body *qbody = &qti->qti_body;
534 struct qsd_instance *qsd;
535 struct qsd_qtype_info *qqi;
539 memset(qbody, 0, sizeof(*qbody));
540 rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
549 /* is quota really enforced for this id? */
550 if (!lqe->lqe_enforced) {
551 lqe_write_unlock(lqe);
552 LQUOTA_DEBUG(lqe, "quota not enforced any more");
556 /* fill qb_count & qb_flags */
557 if (!qsd_calc_acquire(lqe, qbody)) {
558 lqe_write_unlock(lqe);
559 LQUOTA_DEBUG(lqe, "No acquire required");
563 /* check whether an acquire request completed recently */
564 if (lqe->lqe_acq_rc != 0 &&
565 lqe->lqe_acq_time > ktime_get_seconds() - 1) {
566 lqe_write_unlock(lqe);
567 LQUOTA_DEBUG(lqe, "using cached return code %d", lqe->lqe_acq_rc);
568 RETURN(lqe->lqe_acq_rc);
571 /* only 1 quota request in flight for a given ID is allowed */
572 rc = qsd_request_enter(lqe);
574 lqe_write_unlock(lqe);
578 lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
579 lqe_write_unlock(lqe);
581 /* hold a refcount until completion */
584 /* fill other quota body fields */
585 qbody->qb_fid = qqi->qqi_fid;
586 qbody->qb_id = lqe->lqe_id;
588 /* check whether we already own a valid lock for this ID */
589 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
591 struct lquota_lvb *lvb;
596 qsd_req_completion(env, qqi, qbody, NULL,
597 &qti->qti_lockh, NULL, lqe, rc);
600 /* no lock found, should use intent */
601 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, true,
602 IT_QUOTA_DQACQ, qsd_req_completion,
603 qqi, lvb, (void *)lqe);
605 /* lock found, should use regular dqacq */
606 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, true,
607 qsd_req_completion, qqi, &qti->qti_lockh,
611 /* the completion function will be called by qsd_send_dqacq or
617 * Acquire \a space of quota space in order to complete an operation.
618 * Try to consume local quota space first and send acquire request to quota
619 * master if required.
621 * \param env - the environment passed by the caller
622 * \param lqe - is the qid entry to be processed
623 * \param space - is the amount of quota required for the operation
624 * \param ret - is the return code (-EDQUOT, -EINPROGRESS, ...)
626 * \retval true - exit from l_wait_event and real return value in \a ret
627 * \retval false - continue waiting
629 static bool qsd_acquire(const struct lu_env *env, struct lquota_entry *lqe,
630 long long space, int *ret)
635 for (count = 0; rc == 0; count++) {
636 LQUOTA_DEBUG(lqe, "acquiring:%lld count=%d", space, count);
638 if (lqe2qqi(lqe)->qqi_qsd->qsd_stopping) {
643 /* refresh disk usage */
644 rc = qsd_refresh_usage(env, lqe);
648 /* try to consume local quota space first */
649 rc = qsd_acquire_local(lqe, space);
651 /* rc == 0, Wouhou! enough local quota space
652 * rc < 0, something bad happened */
655 /* if we have gotten some quota and stil wait more quota,
656 * it's better to give QMT some time to reclaim from clients */
658 set_current_state(TASK_INTERRUPTIBLE);
659 schedule_timeout(cfs_time_seconds(1));
662 /* need to acquire more quota space from master */
663 rc = qsd_acquire_remote(env, lqe);
667 /* already a request in flight, continue waiting */
670 RETURN(true); /* exit from l_wait_event */
674 * Quota enforcement handler. If local quota can satisfy this operation,
675 * return success, otherwise, acquire more quota from master.
676 * (for write operation, if master isn't available at this moment, return
677 * -EINPROGRESS to inform client to retry the write)
679 * \param env - the environment passed by the caller
680 * \param qsd - is the qsd instance associated with the device in charge
682 * \param qid - is the qid information attached in the transaction handle
683 * \param space - is the space required by the operation
684 * \param flags - if the operation is write, return caller no user/group
685 * and sync commit flags
687 * \retval 0 - success
688 * \retval -EDQUOT - out of quota
689 * \retval -EINPROGRESS - inform client to retry write
690 * \retval -ve - other appropriate errors
692 static int qsd_op_begin0(const struct lu_env *env, struct qsd_qtype_info *qqi,
693 struct lquota_id_info *qid, long long space,
694 enum osd_quota_local_flags *local_flags)
696 struct lquota_entry *lqe;
697 enum osd_quota_local_flags qtype_flag = 0;
698 int rc, ret = -EINPROGRESS;
701 if (qid->lqi_qentry != NULL) {
702 /* we already had to deal with this id for this transaction */
703 lqe = qid->lqi_qentry;
704 if (!lqe->lqe_enforced)
707 /* look up lquota entry associated with qid */
708 lqe = lqe_locate(env, qqi->qqi_site, &qid->lqi_id);
710 RETURN(PTR_ERR(lqe));
711 if (!lqe->lqe_enforced) {
715 qid->lqi_qentry = lqe;
716 /* lqe will be released in qsd_op_end() */
720 /* when space is negative or null, we don't need to consume
721 * quota space. That said, we still want to perform space
722 * adjustments in qsd_op_end, so we return here, but with
723 * a reference on the lqe */
724 if (local_flags != NULL) {
725 rc = qsd_refresh_usage(env, lqe);
731 LQUOTA_DEBUG(lqe, "op_begin space:%lld", space);
734 lqe->lqe_waiting_write += space;
735 lqe_write_unlock(lqe);
737 /* acquire quota space for the operation, cap overall wait time to
738 * prevent a service thread from being stuck for too long */
739 rc = wait_event_idle_timeout(
740 lqe->lqe_waiters, qsd_acquire(env, lqe, space, &ret),
741 cfs_time_seconds(qsd_wait_timeout(qqi->qqi_qsd)));
743 if (rc > 0 && ret == 0) {
744 qid->lqi_space += space;
752 LQUOTA_DEBUG(lqe, "acquire quota failed:%d", rc);
755 lqe->lqe_waiting_write -= space;
757 if (local_flags && lqe->lqe_pending_write != 0)
758 /* Inform OSD layer that there are pending writes.
759 * It might want to retry after a sync if appropriate */
760 *local_flags |= QUOTA_FL_SYNC;
761 lqe_write_unlock(lqe);
763 /* convert recoverable error into -EINPROGRESS, client will
765 if (rc == -ETIMEDOUT || rc == -ENOTCONN || rc == -ENOLCK ||
766 rc == -EAGAIN || rc == -EINTR) {
768 } else if (rc == -ESRCH) {
770 LQUOTA_ERROR(lqe, "ID isn't enforced on master, it "
771 "probably due to a legeal race, if this "
772 "message is showing up constantly, there "
773 "could be some inconsistence between "
774 "master & slave, and quota reintegration "
775 "needs be re-triggered.");
779 if (local_flags != NULL) {
781 LASSERT(qid->lqi_is_blk);
783 *local_flags |= lquota_over_fl(qqi->qqi_qtype);
788 usage = lqe->lqe_pending_write;
789 usage += lqe->lqe_waiting_write;
790 if (lqe->lqe_qunit != 0 && (usage % lqe->lqe_qunit >
791 qqi->qqi_qsd->qsd_sync_threshold))
792 usage += qqi->qqi_qsd->qsd_sync_threshold;
794 usage += lqe->lqe_usage;
796 qtype_flag = lquota_over_fl(qqi->qqi_qtype);
797 /* if we should notify client to start sync write */
798 if (usage >= lqe->lqe_granted - lqe->lqe_pending_rel)
799 *local_flags |= qtype_flag;
801 *local_flags &= ~qtype_flag;
802 lqe_read_unlock(lqe);
809 * helper function comparing two lquota_id_info structures
811 static inline bool qid_equal(struct lquota_id_info *q1,
812 struct lquota_id_info *q2)
814 if (q1->lqi_is_blk != q2->lqi_is_blk || q1->lqi_type != q2->lqi_type)
816 return (q1->lqi_id.qid_uid == q2->lqi_id.qid_uid) ? true : false;
820 * Enforce quota, it's called in the declaration of each operation.
821 * qsd_op_end() will then be called later once all the operations have been
822 * completed in order to release/adjust the quota space.
824 * \param env - the environment passed by the caller
825 * \param qsd - is the qsd instance associated with the device in charge of
827 * \param trans - is the quota transaction information
828 * \param qi - qid & space required by current operation
829 * \param flags - if the operation is write, return caller no user/group and
832 * \retval 0 - success
833 * \retval -EDQUOT - out of quota
834 * \retval -EINPROGRESS - inform client to retry write
835 * \retval -ve - other appropriate errors
837 int qsd_op_begin(const struct lu_env *env, struct qsd_instance *qsd,
838 struct lquota_trans *trans, struct lquota_id_info *qi,
839 enum osd_quota_local_flags *local_flags)
845 if (unlikely(qsd == NULL))
848 if (qsd->qsd_dev->dd_rdonly)
851 /* We don't enforce quota until the qsd_instance is started */
852 read_lock(&qsd->qsd_lock);
853 if (!qsd->qsd_started) {
854 read_unlock(&qsd->qsd_lock);
857 read_unlock(&qsd->qsd_lock);
859 /* ignore block quota on MDTs, ignore inode quota on OSTs */
860 if ((!qsd->qsd_is_md && !qi->lqi_is_blk) ||
861 (qsd->qsd_is_md && qi->lqi_is_blk))
864 /* ignore quota enforcement request when:
865 * - quota isn't enforced for this quota type
866 * or - the user/group is root
867 * or - quota accounting isn't enabled */
868 if (!qsd_type_enabled(qsd, qi->lqi_type) || qi->lqi_id.qid_uid == 0 ||
869 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed)
872 LASSERTF(trans->lqt_id_cnt <= QUOTA_MAX_TRANSIDS, "id_cnt=%d\n",
874 /* check whether we already allocated a slot for this id */
875 for (i = 0; i < trans->lqt_id_cnt; i++) {
876 if (qid_equal(qi, &trans->lqt_ids[i])) {
883 if (unlikely(i >= QUOTA_MAX_TRANSIDS)) {
884 CERROR("%s: more than %d qids enforced for a "
885 "transaction?\n", qsd->qsd_svname, i);
890 trans->lqt_ids[i].lqi_id = qi->lqi_id;
891 trans->lqt_ids[i].lqi_type = qi->lqi_type;
892 trans->lqt_ids[i].lqi_is_blk = qi->lqi_is_blk;
896 /* manage quota enforcement for this ID */
897 rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type],
898 &trans->lqt_ids[i], qi->lqi_space, local_flags);
901 EXPORT_SYMBOL(qsd_op_begin);
904 * Adjust quota space (by acquiring or releasing) hold by the quota slave.
905 * This function is called after each quota request completion and during
906 * reintegration in order to report usage or re-acquire quota locks.
907 * Space adjustment is aborted if there is already a quota request in flight
910 * \param env - the environment passed by the caller
911 * \param lqe - is the qid entry to be processed
913 * \retval 0 on success, appropriate errors on failure
915 int qsd_adjust(const struct lu_env *env, struct lquota_entry *lqe)
917 struct qsd_thread_info *qti = qsd_info(env);
918 struct quota_body *qbody = &qti->qti_body;
919 struct qsd_instance *qsd;
920 struct qsd_qtype_info *qqi;
925 memset(qbody, 0, sizeof(*qbody));
926 rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
928 /* add to adjust list again to trigger adjustment later when
930 LQUOTA_DEBUG(lqe, "delaying adjustment since qsd isn't ready");
931 qsd_adjust_schedule(lqe, true, false);
938 if (qsd->qsd_dev->dd_rdonly)
943 /* fill qb_count & qb_flags */
944 if (!qsd_calc_adjust(lqe, qbody)) {
945 lqe_write_unlock(lqe);
946 LQUOTA_DEBUG(lqe, "no adjustment required");
950 /* only 1 quota request in flight for a given ID is allowed */
951 rc = qsd_request_enter(lqe);
953 /* already a request in flight, space adjustment will be run
954 * again on request completion */
955 lqe_write_unlock(lqe);
959 if (req_is_rel(qbody->qb_flags))
960 lqe->lqe_pending_rel = qbody->qb_count;
961 lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
962 lqe_write_unlock(lqe);
964 /* hold a refcount until completion */
967 /* fill other quota body fields */
968 qbody->qb_fid = qqi->qqi_fid;
969 qbody->qb_id = lqe->lqe_id;
971 if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
972 /* check whether we own a valid lock for this ID */
973 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
975 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
976 if (req_is_preacq(qbody->qb_flags)) {
977 if (req_has_rep(qbody->qb_flags))
978 /* still want to report usage */
979 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
981 /* no pre-acquire if no per-ID lock */
982 GOTO(out, rc = -ENOLCK);
984 /* no lock found, should use intent */
987 } else if (req_is_acq(qbody->qb_flags) &&
988 qbody->qb_count == 0) {
989 /* found cached lock, no need to acquire */
993 /* release and report don't need a per-ID lock */
994 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
998 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, false,
999 qsd_req_completion, qqi, &qti->qti_lockh,
1002 struct lquota_lvb *lvb;
1006 GOTO(out, rc = -ENOMEM);
1008 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, false,
1009 IT_QUOTA_DQACQ, qsd_req_completion,
1010 qqi, lvb, (void *)lqe);
1012 /* the completion function will be called by qsd_send_dqacq or
1013 * qsd_intent_lock */
1016 qsd_req_completion(env, qqi, qbody, NULL, &qti->qti_lockh, NULL, lqe,
1022 * Post quota operation, pre-acquire/release quota from master.
1024 * \param env - the environment passed by the caller
1025 * \param qsd - is the qsd instance attached to the OSD device which
1026 * is handling the operation.
1027 * \param qqi - is the qsd_qtype_info structure associated with the quota ID
1028 * subject to the operation
1029 * \param qid - stores information related to his ID for the operation
1030 * which has just completed
1032 * \retval 0 - success
1033 * \retval -ve - failure
1035 static void qsd_op_end0(const struct lu_env *env, struct qsd_qtype_info *qqi,
1036 struct lquota_id_info *qid)
1038 struct lquota_entry *lqe;
1042 lqe = qid->lqi_qentry;
1045 qid->lqi_qentry = NULL;
1047 /* refresh cached usage if a suitable environment is passed */
1049 qsd_refresh_usage(env, lqe);
1051 lqe_write_lock(lqe);
1052 if (qid->lqi_space > 0)
1053 lqe->lqe_pending_write -= qid->lqi_space;
1055 adjust = qsd_adjust_needed(lqe);
1058 lqe_write_unlock(lqe);
1061 /* pre-acquire/release quota space is needed */
1063 qsd_adjust(env, lqe);
1065 /* no suitable environment, handle adjustment in
1066 * separate thread context */
1067 qsd_adjust_schedule(lqe, false, false);
1074 * Post quota operation. It's called after each operation transaction stopped.
1076 * \param env - the environment passed by the caller
1077 * \param qsd - is the qsd instance associated with device which is handling
1079 * \param qids - all qids information attached in the transaction handle
1080 * \param count - is the number of qid entries in the qids array.
1082 * \retval 0 - success
1083 * \retval -ve - failure
1085 void qsd_op_end(const struct lu_env *env, struct qsd_instance *qsd,
1086 struct lquota_trans *trans)
1091 if (unlikely(qsd == NULL))
1094 if (qsd->qsd_dev->dd_rdonly)
1097 /* We don't enforce quota until the qsd_instance is started */
1098 read_lock(&qsd->qsd_lock);
1099 if (!qsd->qsd_started) {
1100 read_unlock(&qsd->qsd_lock);
1103 read_unlock(&qsd->qsd_lock);
1105 LASSERT(trans != NULL);
1107 for (i = 0; i < trans->lqt_id_cnt; i++) {
1108 struct qsd_qtype_info *qqi;
1110 if (trans->lqt_ids[i].lqi_qentry == NULL)
1113 qqi = qsd->qsd_type_array[trans->lqt_ids[i].lqi_type];
1114 qsd_op_end0(env, qqi, &trans->lqt_ids[i]);
1117 /* reset id_count to 0 so that a second accidental call to qsd_op_end()
1118 * does not result in failure */
1119 trans->lqt_id_cnt = 0;
1122 EXPORT_SYMBOL(qsd_op_end);
1125 * Trigger pre-acquire/release if necessary.
1126 * It's only used by ldiskfs osd so far. When unlink a file in ldiskfs, the
1127 * quota accounting isn't updated when the transaction stopped. Instead, it'll
1128 * be updated on the final iput, so qsd_op_adjust() will be called then (in
1129 * osd_object_delete()) to trigger quota release if necessary.
1131 * \param env - the environment passed by the caller
1132 * \param qsd - is the qsd instance associated with the device in charge
1134 * \param qid - is the lquota ID of the user/group for which to trigger
1135 * quota space adjustment
1136 * \param qtype - is the quota type (USRQUOTA or GRPQUOTA)
1138 void qsd_op_adjust(const struct lu_env *env, struct qsd_instance *qsd,
1139 union lquota_id *qid, int qtype)
1141 struct lquota_entry *lqe;
1142 struct qsd_qtype_info *qqi;
1146 if (unlikely(qsd == NULL))
1149 /* We don't enforce quota until the qsd_instance is started */
1150 read_lock(&qsd->qsd_lock);
1151 if (!qsd->qsd_started) {
1152 read_unlock(&qsd->qsd_lock);
1155 read_unlock(&qsd->qsd_lock);
1157 qqi = qsd->qsd_type_array[qtype];
1160 if (!qsd_type_enabled(qsd, qtype) || qqi->qqi_acct_obj == NULL ||
1164 read_lock(&qsd->qsd_lock);
1165 if (!qsd->qsd_started) {
1166 read_unlock(&qsd->qsd_lock);
1169 read_unlock(&qsd->qsd_lock);
1171 lqe = lqe_locate(env, qqi->qqi_site, qid);
1173 CERROR("%s: fail to locate lqe for id:%llu, type:%d\n",
1174 qsd->qsd_svname, qid->qid_uid, qtype);
1178 qsd_refresh_usage(env, lqe);
1181 adjust = qsd_adjust_needed(lqe);
1182 lqe_read_unlock(lqe);
1185 qsd_adjust(env, lqe);
1190 EXPORT_SYMBOL(qsd_op_adjust);