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. revoke all extra grant
238 if (lqe->lqe_revoke) {
241 LQUOTA_DEBUG(lqe, "revoke pre-acquired quota: %llu - %llu\n",
243 qbody->qb_count = granted - usage;
244 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
248 /* 2. release spare quota space? */
249 if (granted > usage + lqe->lqe_qunit) {
250 /* pre-release quota space */
253 qbody->qb_count = granted - usage;
254 /* if usage == 0, release all granted space */
256 /* try to keep one qunit of quota space */
257 qbody->qb_count -= lqe->lqe_qunit;
258 /* but don't release less than qtune to avoid releasing
260 if (qbody->qb_count < lqe->lqe_qtune)
261 qbody->qb_count = lqe->lqe_qtune;
263 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
267 /* 3. Any quota overrun? */
268 if (lqe->lqe_usage > lqe->lqe_granted) {
269 /* we overconsumed quota space, we report usage in request so
270 * that master can adjust it unconditionally */
273 qbody->qb_usage = lqe->lqe_usage;
274 granted = lqe->lqe_usage;
275 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
278 /* 4. Time to pre-acquire? */
279 if (!lqe->lqe_edquot && !lqe->lqe_nopreacq && usage > 0 &&
280 lqe->lqe_qunit != 0 && granted < usage + lqe->lqe_qtune) {
281 /* To pre-acquire quota space, we report how much spare quota
282 * space the slave currently owns, then the master will grant us
283 * back how much we can pretend given the current state of
287 if (granted <= usage)
290 qbody->qb_count = granted - usage;
291 qbody->qb_flags |= QUOTA_DQACQ_FL_PREACQ;
296 RETURN(qbody->qb_flags != 0);
302 * Helper function returning true when quota space need to be adjusted (some
303 * unused space should be free or pre-acquire) and false otherwise.
305 static inline bool qsd_adjust_needed(struct lquota_entry *lqe)
307 return qsd_calc_adjust(lqe, NULL);
311 * Callback function called when an acquire/release request sent to the master
314 static void qsd_req_completion(const struct lu_env *env,
315 struct qsd_qtype_info *qqi,
316 struct quota_body *reqbody,
317 struct quota_body *repbody,
318 struct lustre_handle *lockh,
319 struct lquota_lvb *lvb,
322 struct lquota_entry *lqe = (struct lquota_entry *)arg;
323 struct qsd_thread_info *qti;
325 bool adjust = false, cancel = false;
328 LASSERT(qqi != NULL && lqe != NULL);
330 /* environment passed by ptlrpcd is mostly used by CLIO and hasn't the
332 rc = lu_env_refill_by_tags((struct lu_env *)env, LCT_DT_THREAD, 0);
334 LQUOTA_ERROR(lqe, "failed to refill environmnent %d", rc);
336 /* can't afford to adjust quota space with no suitable lu_env */
337 GOTO(out_noadjust, rc);
342 LQUOTA_DEBUG(lqe, "DQACQ returned %d, flags:0x%x", ret,
345 /* despite -EDQUOT & -EINPROGRESS errors, the master might still
346 * grant us back quota space to adjust quota overrun */
347 if (ret != 0 && ret != -EDQUOT && ret != -EINPROGRESS) {
348 if (ret != -ETIMEDOUT && ret != -ENOTCONN &&
349 ret != -ESHUTDOWN && ret != -EAGAIN)
350 /* print errors only if return code is unexpected */
351 LQUOTA_ERROR(lqe, "DQACQ failed with %d, flags:0x%x",
352 ret, reqbody->qb_flags);
356 /* Set the lqe_lockh */
357 if (lustre_handle_is_used(lockh) &&
358 !lustre_handle_equal(lockh, &lqe->lqe_lockh))
359 lustre_handle_copy(&lqe->lqe_lockh, lockh);
361 /* If the replied qb_count is zero, it means master didn't process
362 * the DQACQ since the limit for this ID has been removed, so we
363 * should not update quota entry & slave index copy neither. */
364 if (repbody != NULL && repbody->qb_count != 0) {
365 LQUOTA_DEBUG(lqe, "DQACQ qb_count:%llu", repbody->qb_count);
367 if (lqe->lqe_is_reset) {
368 lqe->lqe_granted = 0;
369 } else if (req_is_rel(reqbody->qb_flags)) {
370 if (lqe->lqe_granted < repbody->qb_count) {
371 LQUOTA_ERROR(lqe, "can't release more space "
372 "than owned %llu<%llu",
375 lqe->lqe_granted = 0;
377 lqe->lqe_granted -= repbody->qb_count;
379 /* Cancel the per-ID lock initiatively when there
380 * isn't any usage & grant, which can avoid master
381 * sending glimpse unnecessarily to this slave on
383 if (!lqe->lqe_pending_write && !lqe->lqe_granted &&
384 !lqe->lqe_waiting_write && !lqe->lqe_usage)
387 lqe->lqe_granted += repbody->qb_count;
389 qti->qti_rec.lqr_slv_rec.qsr_granted = lqe->lqe_granted;
390 lqe_write_unlock(lqe);
392 /* Update the slave index file in the dedicated thread. So far,
393 * We don't update the version of slave index copy on DQACQ.
394 * No locking is necessary since nobody can change
395 * lqe->lqe_granted while lqe->lqe_pending_req > 0 */
396 if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_GRANT))
397 qti->qti_rec.lqr_slv_rec.qsr_granted =
398 0xFFFFFFFFFFDEC80CULL;
399 qsd_upd_schedule(qqi, lqe, &lqe->lqe_id, &qti->qti_rec, 0,
404 /* extract information from lvb */
405 if (ret == 0 && lvb != NULL) {
406 if (lvb->lvb_id_qunit != 0)
407 qsd_set_qunit(lqe, lvb->lvb_id_qunit);
408 qsd_set_edquot(lqe, !!(lvb->lvb_flags & LQUOTA_FL_EDQUOT));
409 } else if (repbody != NULL && repbody->qb_qunit != 0) {
410 qsd_set_qunit(lqe, repbody->qb_qunit);
413 /* turn off pre-acquire if it failed with -EDQUOT. This is done to avoid
414 * flooding the master with acquire request. Pre-acquire will be turned
415 * on again as soon as qunit is modified */
416 if (req_is_preacq(reqbody->qb_flags) && ret == -EDQUOT)
417 lqe->lqe_nopreacq = true;
419 adjust = qsd_adjust_needed(lqe);
420 if (reqbody && req_is_acq(reqbody->qb_flags) && ret != -EDQUOT) {
421 lqe->lqe_acq_rc = ret;
422 lqe->lqe_acq_time = ktime_get_seconds();
425 qsd_request_exit(lqe);
426 lqe_write_unlock(lqe);
428 /* release reference on per-ID lock */
429 if (lustre_handle_is_used(lockh))
430 ldlm_lock_decref(lockh, qsd_id_einfo.ei_mode);
433 qsd_adjust_schedule(lqe, false, true);
435 if (!ret || ret == -EDQUOT)
436 qsd_adjust_schedule(lqe, false, false);
438 qsd_adjust_schedule(lqe, true, false);
448 * Try to consume local quota space.
450 * \param lqe - is the qid entry to be processed
451 * \param space - is the amount of quota space needed to complete the operation
453 * \retval 0 - success
454 * \retval -EDQUOT - out of quota
455 * \retval -EAGAIN - need to acquire space from master
457 static int qsd_acquire_local(struct lquota_entry *lqe, __u64 space)
463 if (!lqe->lqe_enforced)
464 /* not enforced any more, we are good */
468 /* use latest usage */
469 usage = lqe->lqe_usage;
470 /* take pending write into account */
471 usage += lqe->lqe_pending_write;
473 if (space + usage <= lqe->lqe_granted - lqe->lqe_pending_rel) {
474 /* Yay! we got enough space */
475 lqe->lqe_pending_write += space;
476 lqe->lqe_waiting_write -= space;
478 /* lqe_edquot flag is used to avoid flooding dqacq requests when
479 * the user is over quota, however, the lqe_edquot could be stale
480 * sometimes due to the race reply of dqacq vs. id lock glimpse
481 * (see LU-4505), so we revalidate it every 5 seconds. */
482 } else if (lqe->lqe_edquot &&
483 (lqe->lqe_edquot_time > ktime_get_seconds() - 5)) {
488 lqe_write_unlock(lqe);
494 * Compute how much quota space should be acquire from the master based
495 * on how much is currently granted to this slave and pending/waiting
498 * \param lqe - is the lquota entry for which we would like to adjust quota
500 * \param qbody - is the quota body of the acquire request to fill
502 * \retval true - space acquisition is needed and qbody is filled
503 * \retval false - no space acquisition required
505 static inline bool qsd_calc_acquire(struct lquota_entry *lqe,
506 struct quota_body *qbody)
508 __u64 usage, granted;
510 usage = lqe->lqe_usage;
511 usage += lqe->lqe_pending_write + lqe->lqe_waiting_write;
512 granted = lqe->lqe_granted;
516 /* if we overconsumed quota space, we report usage in request so that
517 * master can adjust it unconditionally */
518 if (lqe->lqe_usage > lqe->lqe_granted) {
519 qbody->qb_usage = lqe->lqe_usage;
520 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
521 granted = lqe->lqe_usage;
524 /* acquire as much as needed, but not more */
525 if (usage > granted) {
526 qbody->qb_count = usage - granted;
527 qbody->qb_flags |= QUOTA_DQACQ_FL_ACQ;
530 return qbody->qb_flags != 0;
534 * Acquire quota space from master.
535 * There are at most 1 in-flight dqacq/dqrel.
537 * \param env - the environment passed by the caller
538 * \param lqe - is the qid entry to be processed
540 * \retval 0 - success
541 * \retval -EDQUOT - out of quota
542 * \retval -EINPROGRESS - inform client to retry write/create
543 * \retval -EBUSY - already a quota request in flight
544 * \retval -ve - other appropriate errors
546 static int qsd_acquire_remote(const struct lu_env *env,
547 struct lquota_entry *lqe)
549 struct qsd_thread_info *qti = qsd_info(env);
550 struct quota_body *qbody = &qti->qti_body;
551 struct qsd_instance *qsd;
552 struct qsd_qtype_info *qqi;
556 memset(qbody, 0, sizeof(*qbody));
557 rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
566 /* is quota really enforced for this id? */
567 if (!lqe->lqe_enforced) {
568 lqe_write_unlock(lqe);
569 LQUOTA_DEBUG(lqe, "quota not enforced any more");
573 /* fill qb_count & qb_flags */
574 if (!qsd_calc_acquire(lqe, qbody)) {
575 lqe_write_unlock(lqe);
576 LQUOTA_DEBUG(lqe, "No acquire required");
580 /* check whether an acquire request completed recently */
581 if (lqe->lqe_acq_rc != 0 &&
582 lqe->lqe_acq_time > ktime_get_seconds() - 1) {
583 lqe_write_unlock(lqe);
584 LQUOTA_DEBUG(lqe, "using cached return code %d", lqe->lqe_acq_rc);
585 RETURN(lqe->lqe_acq_rc);
588 /* only 1 quota request in flight for a given ID is allowed */
589 rc = qsd_request_enter(lqe);
591 lqe_write_unlock(lqe);
595 lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
596 lqe_write_unlock(lqe);
598 /* hold a refcount until completion */
601 /* fill other quota body fields */
602 qbody->qb_fid = qqi->qqi_fid;
603 qbody->qb_id = lqe->lqe_id;
605 /* check whether we already own a valid lock for this ID */
606 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
608 struct lquota_lvb *lvb;
613 qsd_req_completion(env, qqi, qbody, NULL,
614 &qti->qti_lockh, NULL, lqe, rc);
617 /* no lock found, should use intent */
618 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, true,
619 IT_QUOTA_DQACQ, qsd_req_completion,
620 qqi, lvb, (void *)lqe);
622 /* lock found, should use regular dqacq */
623 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, true,
624 qsd_req_completion, qqi, &qti->qti_lockh,
628 /* the completion function will be called by qsd_send_dqacq or
634 * Acquire \a space of quota space in order to complete an operation.
635 * Try to consume local quota space first and send acquire request to quota
636 * master if required.
638 * \param env - the environment passed by the caller
639 * \param lqe - is the qid entry to be processed
640 * \param space - is the amount of quota required for the operation
641 * \param ret - is the return code (-EDQUOT, -EINPROGRESS, ...)
643 * \retval true - stop waiting in wait_event_idle_timeout,
644 * and real return value in \a ret
645 * \retval false - continue waiting
647 static bool qsd_acquire(const struct lu_env *env, struct lquota_entry *lqe,
648 long long space, int *ret)
651 int wait_pending = 0;
652 struct qsd_qtype_info *qqi = lqe2qqi(lqe);
656 for (count = 0; rc == 0; count++) {
657 LQUOTA_DEBUG(lqe, "acquiring:%lld count=%d", space, count);
659 if (lqe2qqi(lqe)->qqi_qsd->qsd_stopping) {
664 /* refresh disk usage */
665 rc = qsd_refresh_usage(env, lqe);
669 /* try to consume local quota space first */
670 rc = qsd_acquire_local(lqe, space);
672 /* rc == 0, Wouhou! enough local quota space
673 * rc < 0, something bad happened */
676 * There might be a window that commit transaction
677 * have updated usage but pending write doesn't change
678 * wait for it before acquiring remotely.
680 if (lqe->lqe_pending_write >= space && !wait_pending) {
682 dt_sync(env, qqi->qqi_qsd->qsd_dev);
686 /* if we have gotten some quota and stil wait more quota,
687 * it's better to give QMT some time to reclaim from clients */
689 schedule_timeout_interruptible(cfs_time_seconds(1));
691 /* need to acquire more quota space from master */
692 rc = qsd_acquire_remote(env, lqe);
696 /* already a request in flight, continue waiting */
703 * Quota enforcement handler. If local quota can satisfy this operation,
704 * return success, otherwise, acquire more quota from master.
705 * (for write operation, if master isn't available at this moment, return
706 * -EINPROGRESS to inform client to retry the write)
708 * \param env - the environment passed by the caller
709 * \param qsd - is the qsd instance associated with the device in charge
711 * \param qid - is the qid information attached in the transaction handle
712 * \param space - is the space required by the operation
713 * \param flags - if the operation is write, return caller no user/group
714 * and sync commit flags
716 * \retval 0 - success
717 * \retval -EDQUOT - out of quota
718 * \retval -EINPROGRESS - inform client to retry write
719 * \retval -ve - other appropriate errors
721 static int qsd_op_begin0(const struct lu_env *env, struct qsd_qtype_info *qqi,
722 struct lquota_id_info *qid, long long space,
723 enum osd_quota_local_flags *local_flags)
725 struct lquota_entry *lqe;
726 enum osd_quota_local_flags qtype_flag = 0;
727 int rc, ret = -EINPROGRESS;
730 if (qid->lqi_qentry != NULL) {
731 /* we already had to deal with this id for this transaction */
732 lqe = qid->lqi_qentry;
733 if (!lqe->lqe_enforced)
736 /* look up lquota entry associated with qid */
737 lqe = lqe_locate(env, qqi->qqi_site, &qid->lqi_id);
739 RETURN(PTR_ERR(lqe));
740 if (!lqe->lqe_enforced) {
744 qid->lqi_qentry = lqe;
745 /* lqe will be released in qsd_op_end() */
749 /* when space is negative or null, we don't need to consume
750 * quota space. That said, we still want to perform space
751 * adjustments in qsd_op_end, so we return here, but with
752 * a reference on the lqe */
753 if (local_flags != NULL) {
754 rc = qsd_refresh_usage(env, lqe);
760 LQUOTA_DEBUG(lqe, "op_begin space:%lld", space);
763 lqe->lqe_waiting_write += space;
764 lqe_write_unlock(lqe);
766 /* acquire quota space for the operation, cap overall wait time to
767 * prevent a service thread from being stuck for too long */
768 rc = wait_event_idle_timeout(
769 lqe->lqe_waiters, qsd_acquire(env, lqe, space, &ret),
770 cfs_time_seconds(qsd_wait_timeout(qqi->qqi_qsd)));
772 if (rc > 0 && ret == 0) {
773 qid->lqi_space += space;
781 LQUOTA_DEBUG(lqe, "acquire quota failed:%d", rc);
784 lqe->lqe_waiting_write -= space;
786 if (local_flags && lqe->lqe_pending_write != 0)
787 /* Inform OSD layer that there are pending writes.
788 * It might want to retry after a sync if appropriate */
789 *local_flags |= QUOTA_FL_SYNC;
790 lqe_write_unlock(lqe);
792 /* convert recoverable error into -EINPROGRESS, client will
794 if (rc == -ETIMEDOUT || rc == -ENOTCONN || rc == -ENOLCK ||
795 rc == -EAGAIN || rc == -EINTR) {
797 } else if (rc == -ESRCH) {
799 LQUOTA_ERROR(lqe, "ID isn't enforced on master, it "
800 "probably due to a legeal race, if this "
801 "message is showing up constantly, there "
802 "could be some inconsistence between "
803 "master & slave, and quota reintegration "
804 "needs be re-triggered.");
808 if (local_flags != NULL) {
810 LASSERT(qid->lqi_is_blk);
812 *local_flags |= lquota_over_fl(qqi->qqi_qtype);
817 usage = lqe->lqe_pending_write;
818 usage += lqe->lqe_waiting_write;
819 /* There is a chance to successfully grant more quota
820 * but get edquot flag through glimpse. */
821 if (lqe->lqe_edquot || (lqe->lqe_qunit != 0 &&
822 (usage % lqe->lqe_qunit >
823 qqi->qqi_qsd->qsd_sync_threshold)))
824 usage += qqi->qqi_qsd->qsd_sync_threshold;
826 usage += lqe->lqe_usage;
828 qtype_flag = lquota_over_fl(qqi->qqi_qtype);
829 /* if we should notify client to start sync write */
830 if (usage >= lqe->lqe_granted - lqe->lqe_pending_rel)
831 *local_flags |= qtype_flag;
833 *local_flags &= ~qtype_flag;
834 lqe_read_unlock(lqe);
841 * helper function comparing two lquota_id_info structures
843 static inline bool qid_equal(struct lquota_id_info *q1,
844 struct lquota_id_info *q2)
846 if (q1->lqi_is_blk != q2->lqi_is_blk || q1->lqi_type != q2->lqi_type)
848 return (q1->lqi_id.qid_uid == q2->lqi_id.qid_uid) ? true : false;
852 * Enforce quota, it's called in the declaration of each operation.
853 * qsd_op_end() will then be called later once all the operations have been
854 * completed in order to release/adjust the quota space.
856 * \param env - the environment passed by the caller
857 * \param qsd - is the qsd instance associated with the device in charge of
859 * \param trans - is the quota transaction information
860 * \param qi - qid & space required by current operation
861 * \param flags - if the operation is write, return caller no user/group and
864 * \retval 0 - success
865 * \retval -EDQUOT - out of quota
866 * \retval -EINPROGRESS - inform client to retry write
867 * \retval -ve - other appropriate errors
869 int qsd_op_begin(const struct lu_env *env, struct qsd_instance *qsd,
870 struct lquota_trans *trans, struct lquota_id_info *qi,
871 enum osd_quota_local_flags *local_flags)
877 /* fast path, ignore quota enforcement request for root owned files */
878 if (qi->lqi_id.qid_uid == 0)
881 if (unlikely(qsd == NULL))
884 if (qsd->qsd_dev->dd_rdonly)
887 /* We don't enforce quota until the qsd_instance is started */
888 read_lock(&qsd->qsd_lock);
889 if (!qsd->qsd_started) {
890 read_unlock(&qsd->qsd_lock);
893 read_unlock(&qsd->qsd_lock);
895 /* ignore block quota on MDTs, ignore inode quota on OSTs */
896 if ((!qsd->qsd_is_md && !qi->lqi_is_blk) ||
897 (qsd->qsd_is_md && qi->lqi_is_blk))
900 /* ignore quota enforcement request when:
901 * - quota isn't enforced for this quota type
902 * or - the user/group is root
903 * or - quota accounting isn't enabled */
904 if (!qsd_type_enabled(qsd, qi->lqi_type) ||
905 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed)
908 if (local_flags && qi->lqi_id.qid_projid && qsd->qsd_root_prj_enable)
909 *local_flags |= QUOTA_FL_ROOT_PRJQUOTA;
911 LASSERTF(trans->lqt_id_cnt <= QUOTA_MAX_TRANSIDS, "id_cnt=%d\n",
913 /* check whether we already allocated a slot for this id */
914 for (i = 0; i < trans->lqt_id_cnt; i++) {
915 if (qid_equal(qi, &trans->lqt_ids[i])) {
922 if (unlikely(i >= QUOTA_MAX_TRANSIDS)) {
923 CERROR("%s: more than %d qids enforced for a "
924 "transaction?\n", qsd->qsd_svname, i);
929 trans->lqt_ids[i].lqi_id = qi->lqi_id;
930 trans->lqt_ids[i].lqi_type = qi->lqi_type;
931 trans->lqt_ids[i].lqi_is_blk = qi->lqi_is_blk;
935 /* manage quota enforcement for this ID */
936 rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type],
937 &trans->lqt_ids[i], qi->lqi_space, local_flags);
940 EXPORT_SYMBOL(qsd_op_begin);
943 * Adjust quota space (by acquiring or releasing) hold by the quota slave.
944 * This function is called after each quota request completion and during
945 * reintegration in order to report usage or re-acquire quota locks.
946 * Space adjustment is aborted if there is already a quota request in flight
949 * \param env - the environment passed by the caller
950 * \param lqe - is the qid entry to be processed
952 * \retval 0 on success, appropriate errors on failure
954 int qsd_adjust(const struct lu_env *env, struct lquota_entry *lqe)
956 struct qsd_thread_info *qti = qsd_info(env);
957 struct quota_body *qbody = &qti->qti_body;
958 struct qsd_instance *qsd;
959 struct qsd_qtype_info *qqi;
964 memset(qbody, 0, sizeof(*qbody));
965 rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
967 /* add to adjust list again to trigger adjustment later when
969 LQUOTA_DEBUG(lqe, "delaying adjustment since qsd isn't ready");
970 qsd_adjust_schedule(lqe, true, false);
977 if (qsd->qsd_dev->dd_rdonly)
982 /* fill qb_count & qb_flags */
983 if (!qsd_calc_adjust(lqe, qbody)) {
984 lqe_write_unlock(lqe);
985 LQUOTA_DEBUG(lqe, "no adjustment required");
989 /* only 1 quota request in flight for a given ID is allowed */
990 rc = qsd_request_enter(lqe);
992 /* already a request in flight, space adjustment will be run
993 * again on request completion */
994 lqe_write_unlock(lqe);
998 if (req_is_rel(qbody->qb_flags))
999 lqe->lqe_pending_rel = qbody->qb_count;
1000 lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
1001 lqe_write_unlock(lqe);
1003 /* hold a refcount until completion */
1006 /* fill other quota body fields */
1007 qbody->qb_fid = qqi->qqi_fid;
1008 qbody->qb_id = lqe->lqe_id;
1010 if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
1011 /* check whether we own a valid lock for this ID */
1012 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
1014 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
1015 if (req_is_preacq(qbody->qb_flags)) {
1016 if (req_has_rep(qbody->qb_flags))
1017 /* still want to report usage */
1018 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
1020 /* no pre-acquire if no per-ID lock */
1021 GOTO(out, rc = -ENOLCK);
1023 /* no lock found, should use intent */
1026 } else if (req_is_acq(qbody->qb_flags) &&
1027 qbody->qb_count == 0) {
1028 /* found cached lock, no need to acquire */
1032 /* release and report don't need a per-ID lock */
1033 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
1037 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, false,
1038 qsd_req_completion, qqi, &qti->qti_lockh,
1041 struct lquota_lvb *lvb;
1045 GOTO(out, rc = -ENOMEM);
1047 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, false,
1048 IT_QUOTA_DQACQ, qsd_req_completion,
1049 qqi, lvb, (void *)lqe);
1051 /* the completion function will be called by qsd_send_dqacq or
1052 * qsd_intent_lock */
1055 qsd_req_completion(env, qqi, qbody, NULL, &qti->qti_lockh, NULL, lqe,
1061 * Post quota operation, pre-acquire/release quota from master.
1063 * \param env - the environment passed by the caller
1064 * \param qsd - is the qsd instance attached to the OSD device which
1065 * is handling the operation.
1066 * \param qqi - is the qsd_qtype_info structure associated with the quota ID
1067 * subject to the operation
1068 * \param qid - stores information related to his ID for the operation
1069 * which has just completed
1071 * \retval 0 - success
1072 * \retval -ve - failure
1074 static void qsd_op_end0(const struct lu_env *env, struct qsd_qtype_info *qqi,
1075 struct lquota_id_info *qid)
1077 struct lquota_entry *lqe;
1081 lqe = qid->lqi_qentry;
1084 qid->lqi_qentry = NULL;
1086 /* refresh cached usage if a suitable environment is passed */
1088 qsd_refresh_usage(env, lqe);
1090 lqe_write_lock(lqe);
1091 if (qid->lqi_space > 0) {
1092 if (lqe->lqe_pending_write < qid->lqi_space) {
1094 "More pending write quota to reduce (pending %llu, space %lld)\n",
1095 lqe->lqe_pending_write, qid->lqi_space);
1096 lqe->lqe_pending_write = 0;
1098 lqe->lqe_pending_write -= qid->lqi_space;
1102 adjust = qsd_adjust_needed(lqe);
1105 lqe_write_unlock(lqe);
1108 /* pre-acquire/release quota space is needed */
1110 qsd_adjust(env, lqe);
1112 /* no suitable environment, handle adjustment in
1113 * separate thread context */
1114 qsd_adjust_schedule(lqe, false, false);
1121 * Post quota operation. It's called after each operation transaction stopped.
1123 * \param env - the environment passed by the caller
1124 * \param qsd - is the qsd instance associated with device which is handling
1126 * \param qids - all qids information attached in the transaction handle
1127 * \param count - is the number of qid entries in the qids array.
1129 * \retval 0 - success
1130 * \retval -ve - failure
1132 void qsd_op_end(const struct lu_env *env, struct qsd_instance *qsd,
1133 struct lquota_trans *trans)
1138 if (unlikely(qsd == NULL))
1141 if (qsd->qsd_dev->dd_rdonly)
1144 /* We don't enforce quota until the qsd_instance is started */
1145 read_lock(&qsd->qsd_lock);
1146 if (!qsd->qsd_started) {
1147 read_unlock(&qsd->qsd_lock);
1150 read_unlock(&qsd->qsd_lock);
1152 LASSERT(trans != NULL);
1154 for (i = 0; i < trans->lqt_id_cnt; i++) {
1155 struct qsd_qtype_info *qqi;
1157 if (trans->lqt_ids[i].lqi_qentry == NULL)
1160 qqi = qsd->qsd_type_array[trans->lqt_ids[i].lqi_type];
1161 qsd_op_end0(env, qqi, &trans->lqt_ids[i]);
1164 /* reset id_count to 0 so that a second accidental call to qsd_op_end()
1165 * does not result in failure */
1166 trans->lqt_id_cnt = 0;
1169 EXPORT_SYMBOL(qsd_op_end);
1171 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1172 * setattr needs. As a reminder, only the root user can change ownership of
1173 * a file, that's why EDQUOT & EINPROGRESS errors are discarded
1175 int qsd_transfer(const struct lu_env *env, struct qsd_instance *qsd,
1176 struct lquota_trans *trans, unsigned int qtype,
1177 u64 orig_id, u64 new_id, u64 bspace,
1178 struct lquota_id_info *qi)
1185 LASSERT(qtype < LL_MAXQUOTAS);
1186 if (qtype == PRJQUOTA)
1187 if (!projid_valid(make_kprojid(&init_user_ns, new_id)))
1190 qi->lqi_type = qtype;
1192 /* inode accounting */
1193 qi->lqi_is_blk = false;
1195 /* one more inode for the new owner ... */
1196 qi->lqi_id.qid_uid = new_id;
1198 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1199 if (rc == -EDQUOT || rc == -EINPROGRESS)
1204 /* and one less inode for the current id */
1205 qi->lqi_id.qid_uid = orig_id;
1207 /* can't get EDQUOT when reducing usage */
1208 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1209 if (rc == -EINPROGRESS)
1214 /* block accounting */
1215 qi->lqi_is_blk = true;
1217 /* more blocks for the new owner ... */
1218 qi->lqi_id.qid_uid = new_id;
1219 qi->lqi_space = bspace;
1220 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1221 if (rc == -EDQUOT || rc == -EINPROGRESS)
1226 /* and finally less blocks for the current owner */
1227 qi->lqi_id.qid_uid = orig_id;
1228 qi->lqi_space = -bspace;
1229 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1230 /* can't get EDQUOT when reducing usage */
1231 if (rc == -EINPROGRESS)
1235 EXPORT_SYMBOL(qsd_transfer);
1238 * Trigger pre-acquire/release if necessary.
1239 * It's only used by ldiskfs osd so far. When unlink a file in ldiskfs, the
1240 * quota accounting isn't updated when the transaction stopped. Instead, it'll
1241 * be updated on the final iput, so qsd_op_adjust() will be called then (in
1242 * osd_object_delete()) to trigger quota release if necessary.
1244 * \param env - the environment passed by the caller
1245 * \param qsd - is the qsd instance associated with the device in charge
1247 * \param qid - is the lquota ID of the user/group for which to trigger
1248 * quota space adjustment
1249 * \param qtype - is the quota type (USRQUOTA or GRPQUOTA)
1251 void qsd_op_adjust(const struct lu_env *env, struct qsd_instance *qsd,
1252 union lquota_id *qid, int qtype)
1254 struct lquota_entry *lqe;
1255 struct qsd_qtype_info *qqi;
1259 if (unlikely(qsd == NULL))
1262 /* We don't enforce quota until the qsd_instance is started */
1263 read_lock(&qsd->qsd_lock);
1264 if (!qsd->qsd_started) {
1265 read_unlock(&qsd->qsd_lock);
1268 read_unlock(&qsd->qsd_lock);
1270 qqi = qsd->qsd_type_array[qtype];
1273 if (!qsd_type_enabled(qsd, qtype) || qqi->qqi_acct_obj == NULL ||
1274 qid->qid_uid == 0 || qsd->qsd_dev->dd_rdonly)
1277 read_lock(&qsd->qsd_lock);
1278 if (!qsd->qsd_started) {
1279 read_unlock(&qsd->qsd_lock);
1282 read_unlock(&qsd->qsd_lock);
1284 lqe = lqe_locate(env, qqi->qqi_site, qid);
1286 CERROR("%s: fail to locate lqe for id:%llu, type:%d\n",
1287 qsd->qsd_svname, qid->qid_uid, qtype);
1291 qsd_refresh_usage(env, lqe);
1294 adjust = qsd_adjust_needed(lqe);
1295 lqe_read_unlock(lqe);
1298 qsd_adjust(env, lqe);
1303 EXPORT_SYMBOL(qsd_op_adjust);
1306 * Reserve or free quota.
1308 * Currently, It's used to reserve quota space before changing the file's group
1309 * for normal user and free the reserved quota after the group change.
1311 * \param env - the environment passed by the caller
1312 * \param qsd - is the qsd instance associated with the device in charge of
1314 * \param qi - qid & space required by current operation
1316 * \retval 0 - success
1317 * \retval -EDQUOT - out of quota
1318 * \retval -EINPROGRESS - inform client to retry write
1319 * \retval -ve - other appropriate errors
1321 int qsd_reserve_or_free_quota(const struct lu_env *env,
1322 struct qsd_instance *qsd,
1323 struct lquota_id_info *qi)
1325 struct qsd_qtype_info *qqi;
1326 bool is_free = qi->lqi_space < 0;
1331 if (unlikely(qsd == NULL))
1334 if (qsd->qsd_dev->dd_rdonly)
1338 qi->lqi_space *= -1;
1340 /* We don't enforce quota until the qsd_instance is started */
1341 read_lock(&qsd->qsd_lock);
1342 if (!qsd->qsd_started) {
1343 read_unlock(&qsd->qsd_lock);
1346 read_unlock(&qsd->qsd_lock);
1348 qqi = qsd->qsd_type_array[qi->lqi_type];
1351 CDEBUG(D_QUOTA, "type %s, acct %s, free %d, count %llu\n",
1352 qsd_type_enabled(qsd, qi->lqi_type) ? "enabled" : "disabled",
1353 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed ? "failed" :
1354 "succeed", is_free, qi->lqi_space);
1356 /* ignore quota enforcement request when:
1357 * - quota isn't enforced for this quota type
1358 * or - the user/group is root
1359 * or - quota accounting isn't enabled
1362 (!qsd_type_enabled(qsd, qi->lqi_type) || qi->lqi_id.qid_uid == 0 ||
1363 (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed))
1367 qsd_op_end0(env, qsd->qsd_type_array[qi->lqi_type], qi);
1369 long long qspace = qi->lqi_space;
1371 /* the acquired quota will add to lqi_space in qsd_op_begin0 */
1373 rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type], qi,
1375 if (rc && qi->lqi_qentry) {
1376 lqe_putref(qi->lqi_qentry);
1377 qi->lqi_qentry = NULL;
1381 CDEBUG(D_QUOTA, "%s quota: type %i, uid %llu, gid %llu, space %llu\n",
1382 is_free ? "Free" : "Reserve", qi->lqi_type, qi->lqi_id.qid_uid,
1383 qi->lqi_id.qid_gid, qi->lqi_space);
1387 EXPORT_SYMBOL(qsd_reserve_or_free_quota);