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(&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 (lqe->lqe_is_reset) {
356 lqe->lqe_granted = 0;
357 } else if (req_is_rel(reqbody->qb_flags)) {
358 if (lqe->lqe_granted < repbody->qb_count) {
359 LQUOTA_ERROR(lqe, "can't release more space "
360 "than owned %llu<%llu",
363 lqe->lqe_granted = 0;
365 lqe->lqe_granted -= repbody->qb_count;
367 /* Cancel the per-ID lock initiatively when there
368 * isn't any usage & grant, which can avoid master
369 * sending glimpse unnecessarily to this slave on
371 if (!lqe->lqe_pending_write && !lqe->lqe_granted &&
372 !lqe->lqe_waiting_write && !lqe->lqe_usage)
375 lqe->lqe_granted += repbody->qb_count;
377 qti->qti_rec.lqr_slv_rec.qsr_granted = lqe->lqe_granted;
378 lqe_write_unlock(lqe);
380 /* Update the slave index file in the dedicated thread. So far,
381 * We don't update the version of slave index copy on DQACQ.
382 * No locking is necessary since nobody can change
383 * lqe->lqe_granted while lqe->lqe_pending_req > 0 */
384 if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_GRANT))
385 qti->qti_rec.lqr_slv_rec.qsr_granted =
386 0xFFFFFFFFFFDEC80CULL;
387 qsd_upd_schedule(qqi, lqe, &lqe->lqe_id, &qti->qti_rec, 0,
392 /* extract information from lvb */
393 if (ret == 0 && lvb != NULL) {
394 if (lvb->lvb_id_qunit != 0)
395 qsd_set_qunit(lqe, lvb->lvb_id_qunit);
396 qsd_set_edquot(lqe, !!(lvb->lvb_flags & LQUOTA_FL_EDQUOT));
397 } else if (repbody != NULL && repbody->qb_qunit != 0) {
398 qsd_set_qunit(lqe, repbody->qb_qunit);
401 /* turn off pre-acquire if it failed with -EDQUOT. This is done to avoid
402 * flooding the master with acquire request. Pre-acquire will be turned
403 * on again as soon as qunit is modified */
404 if (req_is_preacq(reqbody->qb_flags) && ret == -EDQUOT)
405 lqe->lqe_nopreacq = true;
407 adjust = qsd_adjust_needed(lqe);
408 if (reqbody && req_is_acq(reqbody->qb_flags) && ret != -EDQUOT) {
409 lqe->lqe_acq_rc = ret;
410 lqe->lqe_acq_time = ktime_get_seconds();
413 qsd_request_exit(lqe);
414 lqe_write_unlock(lqe);
416 /* release reference on per-ID lock */
417 if (lustre_handle_is_used(lockh))
418 ldlm_lock_decref(lockh, qsd_id_einfo.ei_mode);
421 qsd_adjust_schedule(lqe, false, true);
423 if (!ret || ret == -EDQUOT)
424 qsd_adjust_schedule(lqe, false, false);
426 qsd_adjust_schedule(lqe, true, false);
436 * Try to consume local quota space.
438 * \param lqe - is the qid entry to be processed
439 * \param space - is the amount of quota space needed to complete the operation
441 * \retval 0 - success
442 * \retval -EDQUOT - out of quota
443 * \retval -EAGAIN - need to acquire space from master
445 static int qsd_acquire_local(struct lquota_entry *lqe, __u64 space)
451 if (!lqe->lqe_enforced)
452 /* not enforced any more, we are good */
456 /* use latest usage */
457 usage = lqe->lqe_usage;
458 /* take pending write into account */
459 usage += lqe->lqe_pending_write;
461 if (space + usage <= lqe->lqe_granted - lqe->lqe_pending_rel) {
462 /* Yay! we got enough space */
463 lqe->lqe_pending_write += space;
464 lqe->lqe_waiting_write -= space;
466 /* lqe_edquot flag is used to avoid flooding dqacq requests when
467 * the user is over quota, however, the lqe_edquot could be stale
468 * sometimes due to the race reply of dqacq vs. id lock glimpse
469 * (see LU-4505), so we revalidate it every 5 seconds. */
470 } else if (lqe->lqe_edquot &&
471 (lqe->lqe_edquot_time > ktime_get_seconds() - 5)) {
476 lqe_write_unlock(lqe);
482 * Compute how much quota space should be acquire from the master based
483 * on how much is currently granted to this slave and pending/waiting
486 * \param lqe - is the lquota entry for which we would like to adjust quota
488 * \param qbody - is the quota body of the acquire request to fill
490 * \retval true - space acquisition is needed and qbody is filled
491 * \retval false - no space acquisition required
493 static inline bool qsd_calc_acquire(struct lquota_entry *lqe,
494 struct quota_body *qbody)
496 __u64 usage, granted;
498 usage = lqe->lqe_usage;
499 usage += lqe->lqe_pending_write + lqe->lqe_waiting_write;
500 granted = lqe->lqe_granted;
504 /* if we overconsumed quota space, we report usage in request so that
505 * master can adjust it unconditionally */
506 if (lqe->lqe_usage > lqe->lqe_granted) {
507 qbody->qb_usage = lqe->lqe_usage;
508 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
509 granted = lqe->lqe_usage;
512 /* acquire as much as needed, but not more */
513 if (usage > granted) {
514 qbody->qb_count = usage - granted;
515 qbody->qb_flags |= QUOTA_DQACQ_FL_ACQ;
518 return qbody->qb_flags != 0;
522 * Acquire quota space from master.
523 * There are at most 1 in-flight dqacq/dqrel.
525 * \param env - the environment passed by the caller
526 * \param lqe - is the qid entry to be processed
528 * \retval 0 - success
529 * \retval -EDQUOT - out of quota
530 * \retval -EINPROGRESS - inform client to retry write/create
531 * \retval -EBUSY - already a quota request in flight
532 * \retval -ve - other appropriate errors
534 static int qsd_acquire_remote(const struct lu_env *env,
535 struct lquota_entry *lqe)
537 struct qsd_thread_info *qti = qsd_info(env);
538 struct quota_body *qbody = &qti->qti_body;
539 struct qsd_instance *qsd;
540 struct qsd_qtype_info *qqi;
544 memset(qbody, 0, sizeof(*qbody));
545 rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
554 /* is quota really enforced for this id? */
555 if (!lqe->lqe_enforced) {
556 lqe_write_unlock(lqe);
557 LQUOTA_DEBUG(lqe, "quota not enforced any more");
561 /* fill qb_count & qb_flags */
562 if (!qsd_calc_acquire(lqe, qbody)) {
563 lqe_write_unlock(lqe);
564 LQUOTA_DEBUG(lqe, "No acquire required");
568 /* check whether an acquire request completed recently */
569 if (lqe->lqe_acq_rc != 0 &&
570 lqe->lqe_acq_time > ktime_get_seconds() - 1) {
571 lqe_write_unlock(lqe);
572 LQUOTA_DEBUG(lqe, "using cached return code %d", lqe->lqe_acq_rc);
573 RETURN(lqe->lqe_acq_rc);
576 /* only 1 quota request in flight for a given ID is allowed */
577 rc = qsd_request_enter(lqe);
579 lqe_write_unlock(lqe);
583 lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
584 lqe_write_unlock(lqe);
586 /* hold a refcount until completion */
589 /* fill other quota body fields */
590 qbody->qb_fid = qqi->qqi_fid;
591 qbody->qb_id = lqe->lqe_id;
593 /* check whether we already own a valid lock for this ID */
594 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
596 struct lquota_lvb *lvb;
601 qsd_req_completion(env, qqi, qbody, NULL,
602 &qti->qti_lockh, NULL, lqe, rc);
605 /* no lock found, should use intent */
606 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, true,
607 IT_QUOTA_DQACQ, qsd_req_completion,
608 qqi, lvb, (void *)lqe);
610 /* lock found, should use regular dqacq */
611 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, true,
612 qsd_req_completion, qqi, &qti->qti_lockh,
616 /* the completion function will be called by qsd_send_dqacq or
622 * Acquire \a space of quota space in order to complete an operation.
623 * Try to consume local quota space first and send acquire request to quota
624 * master if required.
626 * \param env - the environment passed by the caller
627 * \param lqe - is the qid entry to be processed
628 * \param space - is the amount of quota required for the operation
629 * \param ret - is the return code (-EDQUOT, -EINPROGRESS, ...)
631 * \retval true - stop waiting in wait_event_idle_timeout,
632 * and real return value in \a ret
633 * \retval false - continue waiting
635 static bool qsd_acquire(const struct lu_env *env, struct lquota_entry *lqe,
636 long long space, int *ret)
639 int wait_pending = 0;
640 struct qsd_qtype_info *qqi = lqe2qqi(lqe);
644 for (count = 0; rc == 0; count++) {
645 LQUOTA_DEBUG(lqe, "acquiring:%lld count=%d", space, count);
647 if (lqe2qqi(lqe)->qqi_qsd->qsd_stopping) {
652 /* refresh disk usage */
653 rc = qsd_refresh_usage(env, lqe);
657 /* try to consume local quota space first */
658 rc = qsd_acquire_local(lqe, space);
660 /* rc == 0, Wouhou! enough local quota space
661 * rc < 0, something bad happened */
664 * There might be a window that commit transaction
665 * have updated usage but pending write doesn't change
666 * wait for it before acquiring remotely.
668 if (lqe->lqe_pending_write >= space && !wait_pending) {
670 dt_sync(env, qqi->qqi_qsd->qsd_dev);
674 /* if we have gotten some quota and stil wait more quota,
675 * it's better to give QMT some time to reclaim from clients */
677 schedule_timeout_interruptible(cfs_time_seconds(1));
679 /* need to acquire more quota space from master */
680 rc = qsd_acquire_remote(env, lqe);
684 /* already a request in flight, continue waiting */
691 * Quota enforcement handler. If local quota can satisfy this operation,
692 * return success, otherwise, acquire more quota from master.
693 * (for write operation, if master isn't available at this moment, return
694 * -EINPROGRESS to inform client to retry the write)
696 * \param env - the environment passed by the caller
697 * \param qsd - is the qsd instance associated with the device in charge
699 * \param qid - is the qid information attached in the transaction handle
700 * \param space - is the space required by the operation
701 * \param flags - if the operation is write, return caller no user/group
702 * and sync commit flags
704 * \retval 0 - success
705 * \retval -EDQUOT - out of quota
706 * \retval -EINPROGRESS - inform client to retry write
707 * \retval -ve - other appropriate errors
709 static int qsd_op_begin0(const struct lu_env *env, struct qsd_qtype_info *qqi,
710 struct lquota_id_info *qid, long long space,
711 enum osd_quota_local_flags *local_flags)
713 struct lquota_entry *lqe;
714 enum osd_quota_local_flags qtype_flag = 0;
715 int rc, ret = -EINPROGRESS;
718 if (qid->lqi_qentry != NULL) {
719 /* we already had to deal with this id for this transaction */
720 lqe = qid->lqi_qentry;
721 if (!lqe->lqe_enforced)
724 /* look up lquota entry associated with qid */
725 lqe = lqe_locate(env, qqi->qqi_site, &qid->lqi_id);
727 RETURN(PTR_ERR(lqe));
728 if (!lqe->lqe_enforced) {
732 qid->lqi_qentry = lqe;
733 /* lqe will be released in qsd_op_end() */
737 /* when space is negative or null, we don't need to consume
738 * quota space. That said, we still want to perform space
739 * adjustments in qsd_op_end, so we return here, but with
740 * a reference on the lqe */
741 if (local_flags != NULL) {
742 rc = qsd_refresh_usage(env, lqe);
748 LQUOTA_DEBUG(lqe, "op_begin space:%lld", space);
751 lqe->lqe_waiting_write += space;
752 lqe_write_unlock(lqe);
754 /* acquire quota space for the operation, cap overall wait time to
755 * prevent a service thread from being stuck for too long */
756 rc = wait_event_idle_timeout(
757 lqe->lqe_waiters, qsd_acquire(env, lqe, space, &ret),
758 cfs_time_seconds(qsd_wait_timeout(qqi->qqi_qsd)));
760 if (rc > 0 && ret == 0) {
761 qid->lqi_space += space;
769 LQUOTA_DEBUG(lqe, "acquire quota failed:%d", rc);
772 lqe->lqe_waiting_write -= space;
774 if (local_flags && lqe->lqe_pending_write != 0)
775 /* Inform OSD layer that there are pending writes.
776 * It might want to retry after a sync if appropriate */
777 *local_flags |= QUOTA_FL_SYNC;
778 lqe_write_unlock(lqe);
780 /* convert recoverable error into -EINPROGRESS, client will
782 if (rc == -ETIMEDOUT || rc == -ENOTCONN || rc == -ENOLCK ||
783 rc == -EAGAIN || rc == -EINTR) {
785 } else if (rc == -ESRCH) {
787 LQUOTA_ERROR(lqe, "ID isn't enforced on master, it "
788 "probably due to a legeal race, if this "
789 "message is showing up constantly, there "
790 "could be some inconsistence between "
791 "master & slave, and quota reintegration "
792 "needs be re-triggered.");
796 if (local_flags != NULL) {
798 LASSERT(qid->lqi_is_blk);
800 *local_flags |= lquota_over_fl(qqi->qqi_qtype);
805 usage = lqe->lqe_pending_write;
806 usage += lqe->lqe_waiting_write;
807 /* There is a chance to successfully grant more quota
808 * but get edquot flag through glimpse. */
809 if (lqe->lqe_edquot || (lqe->lqe_qunit != 0 &&
810 (usage % lqe->lqe_qunit >
811 qqi->qqi_qsd->qsd_sync_threshold)))
812 usage += qqi->qqi_qsd->qsd_sync_threshold;
814 usage += lqe->lqe_usage;
816 qtype_flag = lquota_over_fl(qqi->qqi_qtype);
817 /* if we should notify client to start sync write */
818 if (usage >= lqe->lqe_granted - lqe->lqe_pending_rel)
819 *local_flags |= qtype_flag;
821 *local_flags &= ~qtype_flag;
822 lqe_read_unlock(lqe);
829 * helper function comparing two lquota_id_info structures
831 static inline bool qid_equal(struct lquota_id_info *q1,
832 struct lquota_id_info *q2)
834 if (q1->lqi_is_blk != q2->lqi_is_blk || q1->lqi_type != q2->lqi_type)
836 return (q1->lqi_id.qid_uid == q2->lqi_id.qid_uid) ? true : false;
840 * Enforce quota, it's called in the declaration of each operation.
841 * qsd_op_end() will then be called later once all the operations have been
842 * completed in order to release/adjust the quota space.
844 * \param env - the environment passed by the caller
845 * \param qsd - is the qsd instance associated with the device in charge of
847 * \param trans - is the quota transaction information
848 * \param qi - qid & space required by current operation
849 * \param flags - if the operation is write, return caller no user/group and
852 * \retval 0 - success
853 * \retval -EDQUOT - out of quota
854 * \retval -EINPROGRESS - inform client to retry write
855 * \retval -ve - other appropriate errors
857 int qsd_op_begin(const struct lu_env *env, struct qsd_instance *qsd,
858 struct lquota_trans *trans, struct lquota_id_info *qi,
859 enum osd_quota_local_flags *local_flags)
865 /* fast path, ignore quota enforcement request for root owned files */
866 if (qi->lqi_id.qid_uid == 0)
869 if (unlikely(qsd == NULL))
872 if (qsd->qsd_dev->dd_rdonly)
875 /* We don't enforce quota until the qsd_instance is started */
876 read_lock(&qsd->qsd_lock);
877 if (!qsd->qsd_started) {
878 read_unlock(&qsd->qsd_lock);
881 read_unlock(&qsd->qsd_lock);
883 /* ignore block quota on MDTs, ignore inode quota on OSTs */
884 if ((!qsd->qsd_is_md && !qi->lqi_is_blk) ||
885 (qsd->qsd_is_md && qi->lqi_is_blk))
888 /* ignore quota enforcement request when:
889 * - quota isn't enforced for this quota type
890 * or - the user/group is root
891 * or - quota accounting isn't enabled */
892 if (!qsd_type_enabled(qsd, qi->lqi_type) ||
893 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed)
896 if (local_flags && qi->lqi_id.qid_projid && qsd->qsd_root_prj_enable)
897 *local_flags |= QUOTA_FL_ROOT_PRJQUOTA;
899 LASSERTF(trans->lqt_id_cnt <= QUOTA_MAX_TRANSIDS, "id_cnt=%d\n",
901 /* check whether we already allocated a slot for this id */
902 for (i = 0; i < trans->lqt_id_cnt; i++) {
903 if (qid_equal(qi, &trans->lqt_ids[i])) {
910 if (unlikely(i >= QUOTA_MAX_TRANSIDS)) {
911 CERROR("%s: more than %d qids enforced for a "
912 "transaction?\n", qsd->qsd_svname, i);
917 trans->lqt_ids[i].lqi_id = qi->lqi_id;
918 trans->lqt_ids[i].lqi_type = qi->lqi_type;
919 trans->lqt_ids[i].lqi_is_blk = qi->lqi_is_blk;
923 /* manage quota enforcement for this ID */
924 rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type],
925 &trans->lqt_ids[i], qi->lqi_space, local_flags);
928 EXPORT_SYMBOL(qsd_op_begin);
931 * Adjust quota space (by acquiring or releasing) hold by the quota slave.
932 * This function is called after each quota request completion and during
933 * reintegration in order to report usage or re-acquire quota locks.
934 * Space adjustment is aborted if there is already a quota request in flight
937 * \param env - the environment passed by the caller
938 * \param lqe - is the qid entry to be processed
940 * \retval 0 on success, appropriate errors on failure
942 int qsd_adjust(const struct lu_env *env, struct lquota_entry *lqe)
944 struct qsd_thread_info *qti = qsd_info(env);
945 struct quota_body *qbody = &qti->qti_body;
946 struct qsd_instance *qsd;
947 struct qsd_qtype_info *qqi;
952 memset(qbody, 0, sizeof(*qbody));
953 rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
955 /* add to adjust list again to trigger adjustment later when
957 LQUOTA_DEBUG(lqe, "delaying adjustment since qsd isn't ready");
958 qsd_adjust_schedule(lqe, true, false);
965 if (qsd->qsd_dev->dd_rdonly)
970 /* fill qb_count & qb_flags */
971 if (!qsd_calc_adjust(lqe, qbody)) {
972 lqe_write_unlock(lqe);
973 LQUOTA_DEBUG(lqe, "no adjustment required");
977 /* only 1 quota request in flight for a given ID is allowed */
978 rc = qsd_request_enter(lqe);
980 /* already a request in flight, space adjustment will be run
981 * again on request completion */
982 lqe_write_unlock(lqe);
986 if (req_is_rel(qbody->qb_flags))
987 lqe->lqe_pending_rel = qbody->qb_count;
988 lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
989 lqe_write_unlock(lqe);
991 /* hold a refcount until completion */
994 /* fill other quota body fields */
995 qbody->qb_fid = qqi->qqi_fid;
996 qbody->qb_id = lqe->lqe_id;
998 if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
999 /* check whether we own a valid lock for this ID */
1000 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
1002 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
1003 if (req_is_preacq(qbody->qb_flags)) {
1004 if (req_has_rep(qbody->qb_flags))
1005 /* still want to report usage */
1006 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
1008 /* no pre-acquire if no per-ID lock */
1009 GOTO(out, rc = -ENOLCK);
1011 /* no lock found, should use intent */
1014 } else if (req_is_acq(qbody->qb_flags) &&
1015 qbody->qb_count == 0) {
1016 /* found cached lock, no need to acquire */
1020 /* release and report don't need a per-ID lock */
1021 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
1025 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, false,
1026 qsd_req_completion, qqi, &qti->qti_lockh,
1029 struct lquota_lvb *lvb;
1033 GOTO(out, rc = -ENOMEM);
1035 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, false,
1036 IT_QUOTA_DQACQ, qsd_req_completion,
1037 qqi, lvb, (void *)lqe);
1039 /* the completion function will be called by qsd_send_dqacq or
1040 * qsd_intent_lock */
1043 qsd_req_completion(env, qqi, qbody, NULL, &qti->qti_lockh, NULL, lqe,
1049 * Post quota operation, pre-acquire/release quota from master.
1051 * \param env - the environment passed by the caller
1052 * \param qsd - is the qsd instance attached to the OSD device which
1053 * is handling the operation.
1054 * \param qqi - is the qsd_qtype_info structure associated with the quota ID
1055 * subject to the operation
1056 * \param qid - stores information related to his ID for the operation
1057 * which has just completed
1059 * \retval 0 - success
1060 * \retval -ve - failure
1062 static void qsd_op_end0(const struct lu_env *env, struct qsd_qtype_info *qqi,
1063 struct lquota_id_info *qid)
1065 struct lquota_entry *lqe;
1069 lqe = qid->lqi_qentry;
1072 qid->lqi_qentry = NULL;
1074 /* refresh cached usage if a suitable environment is passed */
1076 qsd_refresh_usage(env, lqe);
1078 lqe_write_lock(lqe);
1079 if (qid->lqi_space > 0) {
1080 if (lqe->lqe_pending_write < qid->lqi_space) {
1082 "More pending write quota to reduce (pending %llu, space %lld)\n",
1083 lqe->lqe_pending_write, qid->lqi_space);
1084 lqe->lqe_pending_write = 0;
1086 lqe->lqe_pending_write -= qid->lqi_space;
1090 adjust = qsd_adjust_needed(lqe);
1093 lqe_write_unlock(lqe);
1096 /* pre-acquire/release quota space is needed */
1098 qsd_adjust(env, lqe);
1100 /* no suitable environment, handle adjustment in
1101 * separate thread context */
1102 qsd_adjust_schedule(lqe, false, false);
1109 * Post quota operation. It's called after each operation transaction stopped.
1111 * \param env - the environment passed by the caller
1112 * \param qsd - is the qsd instance associated with device which is handling
1114 * \param qids - all qids information attached in the transaction handle
1115 * \param count - is the number of qid entries in the qids array.
1117 * \retval 0 - success
1118 * \retval -ve - failure
1120 void qsd_op_end(const struct lu_env *env, struct qsd_instance *qsd,
1121 struct lquota_trans *trans)
1126 if (unlikely(qsd == NULL))
1129 if (qsd->qsd_dev->dd_rdonly)
1132 /* We don't enforce quota until the qsd_instance is started */
1133 read_lock(&qsd->qsd_lock);
1134 if (!qsd->qsd_started) {
1135 read_unlock(&qsd->qsd_lock);
1138 read_unlock(&qsd->qsd_lock);
1140 LASSERT(trans != NULL);
1142 for (i = 0; i < trans->lqt_id_cnt; i++) {
1143 struct qsd_qtype_info *qqi;
1145 if (trans->lqt_ids[i].lqi_qentry == NULL)
1148 qqi = qsd->qsd_type_array[trans->lqt_ids[i].lqi_type];
1149 qsd_op_end0(env, qqi, &trans->lqt_ids[i]);
1152 /* reset id_count to 0 so that a second accidental call to qsd_op_end()
1153 * does not result in failure */
1154 trans->lqt_id_cnt = 0;
1157 EXPORT_SYMBOL(qsd_op_end);
1159 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1160 * setattr needs. As a reminder, only the root user can change ownership of
1161 * a file, that's why EDQUOT & EINPROGRESS errors are discarded
1163 int qsd_transfer(const struct lu_env *env, struct qsd_instance *qsd,
1164 struct lquota_trans *trans, unsigned int qtype,
1165 u64 orig_id, u64 new_id, u64 bspace,
1166 struct lquota_id_info *qi)
1173 LASSERT(qtype < LL_MAXQUOTAS);
1174 if (qtype == PRJQUOTA)
1175 if (!projid_valid(make_kprojid(&init_user_ns, new_id)))
1178 qi->lqi_type = qtype;
1180 /* inode accounting */
1181 qi->lqi_is_blk = false;
1183 /* one more inode for the new owner ... */
1184 qi->lqi_id.qid_uid = new_id;
1186 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1187 if (rc == -EDQUOT || rc == -EINPROGRESS)
1192 /* and one less inode for the current id */
1193 qi->lqi_id.qid_uid = orig_id;
1195 /* can't get EDQUOT when reducing usage */
1196 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1197 if (rc == -EINPROGRESS)
1202 /* block accounting */
1203 qi->lqi_is_blk = true;
1205 /* more blocks for the new owner ... */
1206 qi->lqi_id.qid_uid = new_id;
1207 qi->lqi_space = bspace;
1208 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1209 if (rc == -EDQUOT || rc == -EINPROGRESS)
1214 /* and finally less blocks for the current owner */
1215 qi->lqi_id.qid_uid = orig_id;
1216 qi->lqi_space = -bspace;
1217 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1218 /* can't get EDQUOT when reducing usage */
1219 if (rc == -EINPROGRESS)
1223 EXPORT_SYMBOL(qsd_transfer);
1226 * Trigger pre-acquire/release if necessary.
1227 * It's only used by ldiskfs osd so far. When unlink a file in ldiskfs, the
1228 * quota accounting isn't updated when the transaction stopped. Instead, it'll
1229 * be updated on the final iput, so qsd_op_adjust() will be called then (in
1230 * osd_object_delete()) to trigger quota release if necessary.
1232 * \param env - the environment passed by the caller
1233 * \param qsd - is the qsd instance associated with the device in charge
1235 * \param qid - is the lquota ID of the user/group for which to trigger
1236 * quota space adjustment
1237 * \param qtype - is the quota type (USRQUOTA or GRPQUOTA)
1239 void qsd_op_adjust(const struct lu_env *env, struct qsd_instance *qsd,
1240 union lquota_id *qid, int qtype)
1242 struct lquota_entry *lqe;
1243 struct qsd_qtype_info *qqi;
1247 if (unlikely(qsd == NULL))
1250 /* We don't enforce quota until the qsd_instance is started */
1251 read_lock(&qsd->qsd_lock);
1252 if (!qsd->qsd_started) {
1253 read_unlock(&qsd->qsd_lock);
1256 read_unlock(&qsd->qsd_lock);
1258 qqi = qsd->qsd_type_array[qtype];
1261 if (!qsd_type_enabled(qsd, qtype) || qqi->qqi_acct_obj == NULL ||
1262 qid->qid_uid == 0 || qsd->qsd_dev->dd_rdonly)
1265 read_lock(&qsd->qsd_lock);
1266 if (!qsd->qsd_started) {
1267 read_unlock(&qsd->qsd_lock);
1270 read_unlock(&qsd->qsd_lock);
1272 lqe = lqe_locate(env, qqi->qqi_site, qid);
1274 CERROR("%s: fail to locate lqe for id:%llu, type:%d\n",
1275 qsd->qsd_svname, qid->qid_uid, qtype);
1279 qsd_refresh_usage(env, lqe);
1282 adjust = qsd_adjust_needed(lqe);
1283 lqe_read_unlock(lqe);
1286 qsd_adjust(env, lqe);
1291 EXPORT_SYMBOL(qsd_op_adjust);
1294 * Reserve or free quota.
1296 * Currently, It's used to reserve quota space before changing the file's group
1297 * for normal user and free the reserved quota after the group change.
1299 * \param env - the environment passed by the caller
1300 * \param qsd - is the qsd instance associated with the device in charge of
1302 * \param qi - qid & space required by current operation
1304 * \retval 0 - success
1305 * \retval -EDQUOT - out of quota
1306 * \retval -EINPROGRESS - inform client to retry write
1307 * \retval -ve - other appropriate errors
1309 int qsd_reserve_or_free_quota(const struct lu_env *env,
1310 struct qsd_instance *qsd,
1311 struct lquota_id_info *qi)
1313 struct qsd_qtype_info *qqi;
1314 bool is_free = qi->lqi_space < 0;
1319 if (unlikely(qsd == NULL))
1322 if (qsd->qsd_dev->dd_rdonly)
1326 qi->lqi_space *= -1;
1328 /* We don't enforce quota until the qsd_instance is started */
1329 read_lock(&qsd->qsd_lock);
1330 if (!qsd->qsd_started) {
1331 read_unlock(&qsd->qsd_lock);
1334 read_unlock(&qsd->qsd_lock);
1336 qqi = qsd->qsd_type_array[qi->lqi_type];
1339 CDEBUG(D_QUOTA, "type %s, acct %s, free %d, count %llu\n",
1340 qsd_type_enabled(qsd, qi->lqi_type) ? "enabled" : "disabled",
1341 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed ? "failed" :
1342 "succeed", is_free, qi->lqi_space);
1344 /* ignore quota enforcement request when:
1345 * - quota isn't enforced for this quota type
1346 * or - the user/group is root
1347 * or - quota accounting isn't enabled
1350 (!qsd_type_enabled(qsd, qi->lqi_type) || qi->lqi_id.qid_uid == 0 ||
1351 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed))
1355 qsd_op_end0(env, qsd->qsd_type_array[qi->lqi_type], qi);
1357 long long qspace = qi->lqi_space;
1359 /* the acquired quota will add to lqi_space in qsd_op_begin0 */
1361 rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type], qi,
1363 if (rc && qi->lqi_qentry) {
1364 lqe_putref(qi->lqi_qentry);
1365 qi->lqi_qentry = NULL;
1369 CDEBUG(D_QUOTA, "%s quota: type %i, uid %llu, gid %llu, space %llu\n",
1370 is_free ? "Free" : "Reserve", qi->lqi_type, qi->lqi_id.qid_uid,
1371 qi->lqi_id.qid_gid, qi->lqi_space);
1375 EXPORT_SYMBOL(qsd_reserve_or_free_quota);