4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2012, 2016, Intel Corporation.
25 * Use is subject to license terms.
27 * Author: Johann Lombardi <johann.lombardi@intel.com>
28 * Author: Niu Yawei <yawei.niu@intel.com>
31 #define DEBUG_SUBSYSTEM S_LQUOTA
33 #include "qmt_internal.h"
36 * Initialize qmt-specific fields of quota entry.
38 * \param lqe - is the quota entry to initialize
39 * \param arg - is the pointer to the qmt_pool_info structure
41 static void qmt_lqe_init(struct lquota_entry *lqe, void *arg)
43 LASSERT(lqe_is_master(lqe));
45 lqe->lqe_revoke_time = 0;
46 init_rwsem(&lqe->lqe_sem);
47 mutex_init(&lqe->lqe_glbl_data_lock);
50 /* Apply the default quota setting to the specified quota entry
52 * \param env - is the environment passed by the caller
53 * \param pool - is the quota pool of the quota entry
54 * \param lqe - is the lquota_entry object to apply default quota on
55 * \param create_record - if true, an global quota record will be created and
59 * \retval -ve : other appropriate errors
61 int qmt_lqe_set_default(const struct lu_env *env, struct qmt_pool_info *pool,
62 struct lquota_entry *lqe, bool create_record)
64 struct lquota_entry *lqe_def;
69 if (lqe->lqe_id.qid_uid == 0)
72 lqe_def = pool->qpi_grace_lqe[lqe_qtype(lqe)];
74 LQUOTA_DEBUG(lqe, "inherit default quota");
76 lqe->lqe_is_default = true;
77 lqe->lqe_hardlimit = lqe_def->lqe_hardlimit;
78 lqe->lqe_softlimit = lqe_def->lqe_softlimit;
81 lqe->lqe_uptodate = true;
82 rc = qmt_set_with_lqe(env, pool->qpi_qmt, lqe, 0, 0,
83 LQUOTA_GRACE_FLAG(0, LQUOTA_FLAG_DEFAULT),
84 QIF_TIMES, true, false);
87 LQUOTA_ERROR(lqe, "failed to create the global quota"
91 if (lqe->lqe_hardlimit == 0 && lqe->lqe_softlimit == 0)
92 lqe->lqe_enforced = false;
94 lqe->lqe_enforced = true;
100 * Update a lquota entry. This is done by reading quota settings from the global
101 * index. The lquota entry must be write locked.
103 * \param env - the environment passed by the caller
104 * \param lqe - is the quota entry to refresh
105 * \param arg - is the pointer to the qmt_pool_info structure
106 * \param find - don't create lqe on disk in case of ENOENT if true
108 static int qmt_lqe_read(const struct lu_env *env, struct lquota_entry *lqe,
109 void *arg, bool find)
111 struct qmt_thread_info *qti = qmt_info(env);
112 struct qmt_pool_info *pool = (struct qmt_pool_info *)arg;
116 LASSERT(lqe_is_master(lqe));
118 /* read record from disk */
119 rc = lquota_disk_read(env, pool->qpi_glb_obj[lqe->lqe_site->lqs_qtype],
120 &lqe->lqe_id, (struct dt_rec *)&qti->qti_glb_rec);
126 qmt_lqe_set_default(env, pool, lqe, true);
129 /* copy quota settings from on-disk record */
130 lqe->lqe_granted = qti->qti_glb_rec.qbr_granted;
131 lqe->lqe_hardlimit = qti->qti_glb_rec.qbr_hardlimit;
132 lqe->lqe_softlimit = qti->qti_glb_rec.qbr_softlimit;
133 lqe->lqe_gracetime = LQUOTA_GRACE(qti->qti_glb_rec.qbr_time);
135 if (lqe->lqe_hardlimit == 0 && lqe->lqe_softlimit == 0 &&
136 (LQUOTA_FLAG(qti->qti_glb_rec.qbr_time) &
137 LQUOTA_FLAG_DEFAULT))
138 qmt_lqe_set_default(env, pool, lqe, false);
139 else if (LQUOTA_FLAG(qti->qti_glb_rec.qbr_time) &
141 lqe->lqe_is_reset = true;
144 LQUOTA_ERROR(lqe, "failed to read quota entry from disk, rc:%d",
149 if (lqe->lqe_id.qid_uid == 0 ||
150 (lqe->lqe_hardlimit == 0 && lqe->lqe_softlimit == 0))
151 /* {hard,soft}limit=0 means no quota enforced */
152 lqe->lqe_enforced = false;
154 lqe->lqe_enforced = true;
156 if (qmt_pool_global(pool))
157 lqe->lqe_is_global = 1;
159 LQUOTA_DEBUG(lqe, "read");
164 * Print lqe information for debugging.
166 * \param lqe - is the quota entry to debug
167 * \param arg - is the pointer to the qmt_pool_info structure
168 * \param msgdata - debug message
169 * \param fmt - format of debug message
171 static void qmt_lqe_debug(struct lquota_entry *lqe, void *arg,
172 struct libcfs_debug_msg_data *msgdata,
173 struct va_format *vaf)
175 struct qmt_pool_info *pool = (struct qmt_pool_info *)arg;
177 libcfs_debug_msg(msgdata,
178 "%pV qmt:%s pool:%s-%s id:%llu enforced:%d hard:%llu soft:%llu granted:%llu time:%llu qunit: %llu edquot:%d may_rel:%llu revoke:%lld default:%s\n",
179 vaf, pool->qpi_qmt->qmt_svname,
180 RES_NAME(pool->qpi_rtype),
182 lqe->lqe_id.qid_uid, lqe->lqe_enforced,
183 lqe->lqe_hardlimit, lqe->lqe_softlimit,
184 lqe->lqe_granted, lqe->lqe_gracetime,
185 lqe->lqe_qunit, lqe->lqe_edquot, lqe->lqe_may_rel,
186 lqe->lqe_revoke_time,
187 lqe->lqe_is_default ? "yes" : "no");
191 * Vector of quota entry operations supported on the master
193 const struct lquota_entry_operations qmt_lqe_ops = {
194 .lqe_init = qmt_lqe_init,
195 .lqe_read = qmt_lqe_read,
196 .lqe_debug = qmt_lqe_debug,
200 * Reserve enough credits to update records in both the global index and
201 * the slave index identified by \slv_obj
203 * \param env - is the environment passed by the caller
204 * \param lqe - is the quota entry associated with the identifier
205 * subject to the change. If it is NULL lqes array is
206 * taken from env with qti_lqes_env(env).
207 * \param slv_obj - is the dt_object associated with the index file
208 * \param sync - make transaction sync if true
210 struct thandle *qmt_trans_start_with_slv(const struct lu_env *env,
211 struct lquota_entry *lqe,
212 struct dt_object *slv_obj,
215 struct qmt_device *qmt;
217 struct lquota_entry **lqes;
218 struct qmt_lqe_restore *restore;
222 restore = qti_lqes_rstr(env);
224 lqes_cnt = qti_lqes_cnt(env);
225 lqes = qti_lqes(env);
231 /* qmt is the same for all lqes, so take it from the 1st */
232 qmt = lqe2qpi(lqes[0])->qpi_qmt;
235 LQUOTA_DEBUG(lqes[0], "declare write for slv "DFID,
236 PFID(lu_object_fid(&slv_obj->do_lu)));
238 /* start transaction */
239 th = dt_trans_create(env, qmt->qmt_child);
244 /* quota settings on master are updated synchronously for the
248 /* reserve credits for global index update */
249 for (i = 0; i < lqes_cnt; i++) {
250 rc = lquota_disk_declare_write(env, th,
251 LQE_GLB_OBJ(lqes[i]),
257 if (slv_obj != NULL) {
258 /* reserve credits for slave index update */
259 rc = lquota_disk_declare_write(env, th, slv_obj,
265 /* start transaction */
266 rc = dt_trans_start_local(env, qmt->qmt_child, th);
273 dt_trans_stop(env, qmt->qmt_child, th);
275 LQUOTA_ERROR(lqes[0], "failed to slv declare write for "DFID
276 ", rc:%d", PFID(lu_object_fid(&slv_obj->do_lu)),
279 for (i = 0; i < lqes_cnt; i++) {
280 restore[i].qlr_hardlimit = lqes[i]->lqe_hardlimit;
281 restore[i].qlr_softlimit = lqes[i]->lqe_softlimit;
282 restore[i].qlr_gracetime = lqes[i]->lqe_gracetime;
283 restore[i].qlr_granted = lqes[i]->lqe_granted;
284 restore[i].qlr_qunit = lqes[i]->lqe_qunit;
291 * Reserve enough credits to update a record in the global index
293 * \param env - is the environment passed by the caller
294 * \param lqe - is the quota entry to be modified in the global index
295 * \param restore - is a temporary storage for current quota settings which will
296 * be restored if something goes wrong at index update time.
298 struct thandle *qmt_trans_start(const struct lu_env *env,
299 struct lquota_entry *lqe)
301 LQUOTA_DEBUG(lqe, "declare write");
302 return qmt_trans_start_with_slv(env, lqe, NULL, true);
305 int qmt_glb_write_lqes(const struct lu_env *env, struct thandle *th,
306 __u32 flags, __u64 *ver)
310 for (i = 0; i < qti_lqes_cnt(env); i++) {
311 rc = qmt_glb_write(env, th, qti_lqes(env)[i], flags, ver);
319 * Update record associated with a quota entry in the global index.
320 * If LQUOTA_BUMP_VER is set, then the global index version must also be
322 * The entry must be at least read locked, dirty and up-to-date.
324 * \param env - the environment passed by the caller
325 * \param th - is the transaction handle to be used for the disk writes
326 * \param lqe - is the quota entry to udpate
327 * \param obj - is the dt_object associated with the index file
328 * \param flags - can be LQUOTA_BUMP_VER or LQUOTA_SET_VER.
329 * \param ver - is used to return the new version of the index.
331 * \retval - 0 on success and lqe dirty flag cleared,
332 * appropriate error on failure and uptodate flag cleared.
334 int qmt_glb_write(const struct lu_env *env, struct thandle *th,
335 struct lquota_entry *lqe, __u32 flags, __u64 *ver)
337 struct qmt_thread_info *qti = qmt_info(env);
338 struct lquota_glb_rec *rec;
342 LASSERT(lqe != NULL);
343 LASSERT(lqe_is_master(lqe));
344 LASSERT(lqe_is_locked(lqe));
345 LASSERT(lqe->lqe_uptodate);
346 LASSERT((flags & ~(LQUOTA_BUMP_VER | LQUOTA_SET_VER)) == 0);
348 LQUOTA_DEBUG(lqe, "write glb");
350 /* never delete the entry even when the id isn't enforced and
351 * no any guota granted, otherwise, this entry will not be
352 * synced to slave during the reintegration. */
353 rec = &qti->qti_glb_rec;
355 /* fill global index with updated quota settings */
356 rec->qbr_granted = lqe->lqe_granted;
357 if (lqe->lqe_is_default) {
358 rec->qbr_hardlimit = 0;
359 rec->qbr_softlimit = 0;
360 rec->qbr_time = LQUOTA_GRACE_FLAG(lqe->lqe_gracetime,
361 LQUOTA_FLAG_DEFAULT);
362 } else if (lqe->lqe_is_reset) {
363 rec->qbr_hardlimit = 0;
364 rec->qbr_softlimit = 0;
365 rec->qbr_granted = 0;
366 rec->qbr_time = LQUOTA_GRACE_FLAG(lqe->lqe_gracetime,
369 rec->qbr_hardlimit = lqe->lqe_hardlimit;
370 rec->qbr_softlimit = lqe->lqe_softlimit;
371 rec->qbr_time = lqe->lqe_gracetime;
374 /* write new quota settings */
375 rc = lquota_disk_write(env, th, LQE_GLB_OBJ(lqe), &lqe->lqe_id,
376 (struct dt_rec *)rec, flags, ver);
378 /* we failed to write the new quota settings to disk, report
379 * error to caller who will restore the initial value */
380 LQUOTA_ERROR(lqe, "failed to update global index, rc:%d", rc);
386 * Read from disk how much quota space is allocated to a slave.
387 * This is done by reading records from the dedicated slave index file.
388 * Return in \granted how much quota space is currently allocated to the
390 * The entry must be at least read locked.
392 * \param env - the environment passed by the caller
393 * \param lqe_id - is the quota id associated with the identifier to look-up
395 * \param slv_obj - is the dt_object associated with the slave index
396 * \param granted - is the output parameter where to return how much space
397 * is granted to the slave.
399 * \retval - 0 on success, appropriate error on failure
401 int qmt_slv_read(const struct lu_env *env, union lquota_id *qid,
402 struct dt_object *slv_obj, __u64 *granted)
404 struct qmt_thread_info *qti = qmt_info(env);
405 struct lquota_slv_rec *slv_rec = &qti->qti_slv_rec;
409 CDEBUG(D_QUOTA, "read id:%llu form slv "DFID"\n",
410 qid->qid_uid, PFID(lu_object_fid(&slv_obj->do_lu)));
412 /* read slave record from disk */
413 rc = lquota_disk_read(env, slv_obj, qid,
414 (struct dt_rec *)slv_rec);
420 /* extract granted from on-disk record */
421 *granted = slv_rec->qsr_granted;
424 CERROR("Failed to read slave record for %llu from "DFID"\n",
425 qid->qid_uid, PFID(lu_object_fid(&slv_obj->do_lu)));
429 CDEBUG(D_QUOTA, "Successful slv read %llu\n", *granted);
435 * Update record in slave index file.
436 * The entry must be at least read locked.
438 * \param env - the environment passed by the caller
439 * \param th - is the transaction handle to be used for the disk writes
440 * \param lqe - is the dirty quota entry which will be updated at the same time
442 * \param slv_obj - is the dt_object associated with the slave index
443 * \param flags - can be LQUOTA_BUMP_VER or LQUOTA_SET_VER.
444 * \param ver - is used to return the new version of the index.
445 * \param granted - is the new amount of quota space owned by the slave
447 * \retval - 0 on success, appropriate error on failure
449 int qmt_slv_write(const struct lu_env *env, struct thandle *th,
450 struct lquota_entry *lqe, struct dt_object *slv_obj,
451 __u32 flags, __u64 *ver, __u64 granted)
453 struct qmt_thread_info *qti = qmt_info(env);
454 struct lquota_slv_rec *rec;
458 LASSERT(lqe != NULL);
459 LASSERT(lqe_is_master(lqe));
460 LASSERT(lqe_is_locked(lqe));
462 LQUOTA_DEBUG(lqe, "write slv "DFID" granted:%llu",
463 PFID(lu_object_fid(&slv_obj->do_lu)), granted);
465 /* never delete the entry, otherwise, it'll not be transferred
466 * to slave during reintegration. */
467 rec = &qti->qti_slv_rec;
469 /* updated space granted to this slave */
470 rec->qsr_granted = granted;
472 /* write new granted space */
473 rc = lquota_disk_write(env, th, slv_obj, &lqe->lqe_id,
474 (struct dt_rec *)rec, flags, ver);
477 "failed to update slave index "DFID" granted:%llu",
478 PFID(lu_object_fid(&slv_obj->do_lu)),
487 * Check whether new limits are valid for this pool
489 * \param lqe - is the quota entry subject to the setquota
490 * \param hard - is the new hard limit
491 * \param soft - is the new soft limit
493 int qmt_validate_limits(struct lquota_entry *lqe, __u64 hard, __u64 soft)
497 if (hard != 0 && soft > hard)
498 /* soft limit must be less than hard limit */
504 * Set/clear edquot flag after quota space allocation/release or settings
505 * change. Slaves will be notified of changes via glimpse on per-ID lock
507 * \param lqe - is the quota entry to check
508 * \param now - is the current time in second used for grace time managment
510 bool qmt_adjust_edquot(struct lquota_entry *lqe, __u64 now)
512 struct qmt_pool_info *pool = lqe2qpi(lqe);
515 if (!lqe->lqe_enforced || lqe->lqe_id.qid_uid == 0)
518 if (!lqe->lqe_edquot) {
519 /* space exhausted flag not set, let's check whether it is time
522 if (!qmt_space_exhausted(lqe, now))
523 /* the qmt still has available space */
526 /* See comment in qmt_adjust_qunit(). LU-4139 */
527 if (qmt_hard_exhausted(lqe) ||
528 pool->qpi_rtype != LQUOTA_RES_DT) {
531 /* we haven't reached the minimal qunit yet so there is
532 * still hope that the rebalancing process might free
533 * up some quota space */
534 if (lqe->lqe_qunit != pool->qpi_least_qunit)
537 /* least qunit value not sent to all slaves yet */
538 if (lqe->lqe_revoke_time == 0)
541 /* Let's give more time to slave to release space */
542 lapse = ktime_get_seconds() - QMT_REBA_TIMEOUT;
543 if (lqe->lqe_may_rel != 0 && lqe->lqe_revoke_time > lapse)
546 if (lqe->lqe_qunit > pool->qpi_soft_least_qunit)
550 /* set edquot flag */
551 lqe->lqe_edquot = true;
553 /* space exhausted flag set, let's check whether it is time to
556 if (qmt_space_exhausted(lqe, now))
557 /* the qmt still has not space */
560 if (lqe->lqe_hardlimit != 0 &&
561 lqe->lqe_granted + pool->qpi_least_qunit >
563 /* we clear the flag only once at least one least qunit
567 /* clear edquot flag */
568 lqe->lqe_edquot = false;
571 LQUOTA_DEBUG(lqe, "changing edquot flag");
573 /* let's notify slave by issuing glimpse on per-ID lock.
574 * the rebalance thread will take care of this */
578 /* Using least_qunit when over block softlimit will seriously impact the
579 * write performance, we need to do some special tweaking on that. */
580 static __u64 qmt_calc_softlimit(struct lquota_entry *lqe, bool *oversoft)
582 struct qmt_pool_info *pool = lqe2qpi(lqe);
584 LASSERT(lqe->lqe_softlimit != 0);
586 /* No need to do special tweaking for inode limit */
587 if (pool->qpi_rtype != LQUOTA_RES_DT)
588 return lqe->lqe_softlimit;
590 if (lqe->lqe_granted <= lqe->lqe_softlimit +
591 pool->qpi_soft_least_qunit) {
592 return lqe->lqe_softlimit;
593 } else if (lqe->lqe_hardlimit != 0) {
595 return lqe->lqe_hardlimit;
603 * Try to grant more quota space back to slave.
605 * \param lqe - is the quota entry for which we would like to allocate more
607 * \param granted - is how much was already granted as part of the request
609 * \param spare - is how much unused quota space the slave already owns
611 * \retval return how additional space can be granted to the slave
613 __u64 qmt_alloc_expand(struct lquota_entry *lqe, __u64 granted, __u64 spare)
615 struct qmt_pool_info *pool = lqe2qpi(lqe);
616 __u64 remaining, qunit;
619 LASSERT(lqe->lqe_enforced && lqe->lqe_qunit != 0);
621 slv_cnt = qpi_slv_nr(lqe2qpi(lqe), lqe_qtype(lqe));
622 qunit = lqe->lqe_qunit;
624 /* See comment in qmt_adjust_qunit(). LU-4139. */
625 if (lqe->lqe_softlimit != 0) {
627 remaining = qmt_calc_softlimit(lqe, &oversoft);
629 remaining = lqe->lqe_granted +
630 pool->qpi_soft_least_qunit;
632 remaining = lqe->lqe_hardlimit;
635 if (lqe->lqe_granted >= remaining)
638 remaining -= lqe->lqe_granted;
644 granted &= (qunit - 1);
646 if (remaining > (slv_cnt * qunit) >> 1) {
647 /* enough room to grant more space w/o additional
648 * shrinking ... at least for now */
649 remaining -= (slv_cnt * qunit) >> 1;
650 } else if (qunit != pool->qpi_least_qunit) {
655 granted &= (qunit - 1);
657 RETURN(min_t(__u64, qunit - spare, remaining));
659 RETURN(min_t(__u64, qunit - granted, remaining));
660 } while (qunit >= pool->qpi_least_qunit);
666 qmt_adjust_qunit_set_revoke(const struct lu_env *env, struct lquota_entry *lqe,
667 unsigned long least_qunit)
669 struct lquota_entry *lqe2;
673 if (qti_lqes_cnt(env) <= 1)
676 for (i = 0; i < qti_lqes_cnt(env); i++) {
677 lqe2 = qti_lqes(env)[i];
678 if ((lqe2->lqe_qunit == least_qunit) && lqe2->lqe_revoke_time) {
680 min = lqe2->lqe_revoke_time;
683 min = lqe2->lqe_revoke_time < min ?
684 lqe2->lqe_revoke_time : min;
688 lqe->lqe_revoke_time = min;
693 * Adjust qunit size according to quota limits and total granted count.
694 * The caller must have locked the lqe.
696 * \param env - the environment passed by the caller
697 * \param lqe - is the qid entry to be adjusted
698 * \retval true - need reseed glbe array
700 bool qmt_adjust_qunit(const struct lu_env *env, struct lquota_entry *lqe)
702 struct qmt_pool_info *pool = lqe2qpi(lqe);
703 bool need_reseed = false;
705 __u64 qunit, limit, qunit2 = 0;
708 LASSERT(lqe_is_locked(lqe));
710 if (!lqe->lqe_enforced || lqe->lqe_id.qid_uid == 0)
711 /* no quota limits */
714 /* record how many slaves have already registered */
715 slv_cnt = qpi_slv_nr(pool, lqe_qtype(lqe));
717 /* Pool hasn't slaves anymore. Qunit will be adjusted
718 * again when new slaves would be added. */
719 if (lqe->lqe_qunit) {
723 /* wait for at least one slave to join */
727 /* Qunit calculation is based on soft limit, if any, hard limit
728 * otherwise. This means that qunit is shrunk to the minimum when
729 * beyond the soft limit. This will impact performance, but that's the
730 * price of an accurate grace time management. */
731 if (lqe->lqe_softlimit != 0) {
733 /* As a compromise of write performance and the grace time
734 * accuracy, the block qunit size will be shrunk to
735 * qpi_soft_least_qunit when over softlimit. LU-4139. */
736 limit = qmt_calc_softlimit(lqe, &oversoft);
738 qunit2 = pool->qpi_soft_least_qunit;
740 GOTO(done, qunit = qunit2);
741 } else if (lqe->lqe_hardlimit != 0) {
742 limit = lqe->lqe_hardlimit;
744 LQUOTA_ERROR(lqe, "enforced bit set, but neither hard nor soft "
749 qunit = lqe->lqe_qunit == 0 ? pool->qpi_least_qunit : lqe->lqe_qunit;
751 /* The qunit value is computed as follows: limit / (2 * slv_cnt).
752 * Then 75% of the quota space can be granted with current qunit value.
753 * The remaining 25% are then used with reduced qunit size (by a factor
754 * of 4) which is then divided in a similar manner.
756 * |---------------------limit---------------------|
757 * |-------limit / 2-------|-limit / 4-|-limit / 4-|
758 * |qunit|qunit|qunit|qunit| | |
759 * |----slv_cnt * qunit----| | |
760 * |-grow limit-| | | |
761 * |--------------shrink limit---------| |
762 * |---space granted in qunit chunks---|-remaining-|
768 * qunit >>= 2; |qunit*slv_cnt|qunit*slv_cnt|
769 * |---space in qunit---|remain|
771 if (qunit == pool->qpi_least_qunit ||
772 limit >= lqe->lqe_granted + ((slv_cnt * qunit) >> 1)) {
773 /* current qunit value still fits, let's see if we can afford to
774 * increase qunit now ...
775 * To increase qunit again, we have to be under 25% */
776 while (qunit && limit >= lqe->lqe_granted + 6 * qunit * slv_cnt)
781 do_div(qunit, 2 * slv_cnt);
785 /* shrink qunit until we find a suitable value */
786 while (qunit > pool->qpi_least_qunit &&
787 limit < lqe->lqe_granted + ((slv_cnt * qunit) >> 1))
791 if (qunit2 && qunit > qunit2)
794 if (lqe->lqe_qunit == qunit)
795 /* keep current qunit */
798 LQUOTA_DEBUG(lqe, "%s qunit to %llu",
799 lqe->lqe_qunit < qunit ? "increasing" : "decreasing",
802 /* store new qunit value */
803 swap(lqe->lqe_qunit, qunit);
805 /* reseed glbe array and notify
806 * slave if qunit was shrinked */
808 /* reset revoke time */
809 lqe->lqe_revoke_time = 0;
811 if (lqe->lqe_qunit == pool->qpi_least_qunit) {
812 if (lqe->lqe_qunit >= qunit)
813 /* initial qunit value is the smallest one */
814 lqe->lqe_revoke_time = ktime_get_seconds();
815 /* If there are several lqes and lqe_revoke_time is set for
816 * some of them, it means appropriate OSTs have been already
817 * notified with the least qunit and there is no chance to
818 * free more space. Find an lqe with the minimum(earliest)
819 * revoke_time and set this time to the current one.
821 qmt_adjust_qunit_set_revoke(env, lqe, pool->qpi_least_qunit);
826 bool qmt_adjust_edquot_qunit_notify(const struct lu_env *env,
827 struct qmt_device *qmt,
828 __u64 now, bool edquot,
829 bool qunit, __u32 qb_flags,
832 struct lquota_entry *lqe_gl, *lqe;
833 bool need_reseed = false;
834 bool need_notify = false;
837 lqe_gl = qti_lqes_glbl(env);
839 for (i = 0; i < qti_lqes_cnt(env); i++) {
840 lqe = qti_lqes(env)[i];
842 need_reseed |= qmt_adjust_qunit(env, lqe);
844 need_reseed |= qmt_adjust_edquot(lqe, now);
848 if (!lqe_gl->lqe_glbl_data &&
849 (req_has_rep(qb_flags) || req_is_rel(qb_flags))) {
852 "%s: can not notify - lge_glbl_data is not set\n",
857 if (need_reseed || idx >= 0) {
858 mutex_lock(&lqe_gl->lqe_glbl_data_lock);
859 if (lqe_gl->lqe_glbl_data) {
860 struct lqe_glbl_data *lgd = lqe_gl->lqe_glbl_data;
863 qmt_seed_glbe_all(env, lgd, qunit, edquot);
864 } else if (idx >= 0) {
865 LASSERT(idx <= lgd->lqeg_num_used);
866 /* If there are no locks yet when
867 * lge_qunit/edquot_nu is set, slaves
868 * are still not notified with new
869 * qunit/edquot value. In a such case
870 * we need to notify them with new values to
871 * avoid endless EINPROGRESS if qunit is equal
872 * to the least qunit, but lqe_revoke_time is
875 need_notify = lgd->lqeg_arr[idx].lge_qunit_nu ||
876 lgd->lqeg_arr[idx].lge_edquot_nu;
879 mutex_unlock(&lqe_gl->lqe_glbl_data_lock);
882 if (need_reseed || need_notify)
883 qmt_id_lock_notify(qmt, lqe_gl);
890 * Adjust qunit & edquot flag in case it wasn't initialized already (e.g.
891 * limit set while no slaves were connected yet)
893 bool qmt_revalidate(const struct lu_env *env, struct lquota_entry *lqe)
895 bool need_notify = false;
897 if (lqe->lqe_qunit == 0) {
898 /* lqe was read from disk, but neither qunit, nor edquot flag
899 * were initialized */
900 need_notify = qmt_adjust_qunit(env, lqe);
901 if (lqe->lqe_qunit != 0)
902 need_notify |= qmt_adjust_edquot(lqe,
903 ktime_get_real_seconds());
909 void qmt_revalidate_lqes(const struct lu_env *env,
910 struct qmt_device *qmt, __u32 qb_flags)
912 struct lquota_entry *lqe_gl = qti_lqes_glbl(env);
913 bool need_notify = false;
916 for (i = 0; i < qti_lqes_cnt(env); i++)
917 need_notify |= qmt_revalidate(env, qti_lqes(env)[i]);
922 /* There could be no ID lock to the moment of reconciliation.
923 * As a result lqe global data is not initialised yet. It is ok
924 * for release and report requests. */
925 if (!lqe_gl->lqe_glbl_data &&
926 (req_is_rel(qb_flags) || req_has_rep(qb_flags))) {
930 mutex_lock(&lqe_gl->lqe_glbl_data_lock);
931 if (lqe_gl->lqe_glbl_data)
932 qmt_seed_glbe(env, lqe_gl->lqe_glbl_data);
933 mutex_unlock(&lqe_gl->lqe_glbl_data_lock);
935 qmt_id_lock_notify(qmt, lqe_gl);
938 void qti_lqes_init(const struct lu_env *env)
940 struct qmt_thread_info *qti = qmt_info(env);
942 qti->qti_lqes_cnt = 0;
943 qti->qti_glbl_lqe_idx = 0;
944 qti->qti_lqes_num = QMT_MAX_POOL_NUM;
947 int qti_lqes_add(const struct lu_env *env, struct lquota_entry *lqe)
949 struct qmt_thread_info *qti = qmt_info(env);
951 if (qti->qti_lqes_cnt >= qti->qti_lqes_num) {
952 struct lquota_entry **lqes;
953 lqes = qti->qti_lqes;
954 OBD_ALLOC(lqes, sizeof(lqe) * qti->qti_lqes_num * 2);
957 memcpy(lqes, qti_lqes(env), qti->qti_lqes_cnt * sizeof(lqe));
958 /* Don't need to free, if it is the very 1st allocation */
959 if (qti->qti_lqes_num > QMT_MAX_POOL_NUM)
960 OBD_FREE(qti->qti_lqes,
961 qti->qti_lqes_num * sizeof(lqe));
962 qti->qti_lqes = lqes;
963 qti->qti_lqes_num *= 2;
966 if (lqe->lqe_is_global)
967 qti->qti_glbl_lqe_idx = qti->qti_lqes_cnt;
968 qti_lqes(env)[qti->qti_lqes_cnt++] = lqe;
970 /* The pool could be accessed directly from lqe, so take
971 * extra reference that is put in qti_lqes_fini */
972 qpi_getref(lqe2qpi(lqe));
974 CDEBUG(D_QUOTA, "LQE %p %lu is added, lqe_cnt %d lqes_num %d\n",
975 lqe, (long unsigned)lqe->lqe_id.qid_uid,
976 qti->qti_lqes_cnt, qti->qti_lqes_num);
977 LASSERT(qti->qti_lqes_num != 0);
982 void qti_lqes_del(const struct lu_env *env, int index)
984 struct lquota_entry **lqes;
985 int lqes_cnt = qti_lqes_cnt(env);
986 int lqep_size = sizeof(struct lquota_entry *);
989 /* We can't handle non global lqes correctly without
990 * global lqe located at index 0. If we try to do so,
991 * something goes wrong. */
992 LQUOTA_ERROR(qti_lqes_glbl(env),
993 "quota: cannot remove lqe at index 0 as it is global");
994 LASSERT(qti_lqes_glbl(env)->lqe_is_global);
997 lqes = qti_lqes(env);
998 qpi_putref(env, lqe2qpi(lqes[index]));
999 lqe_putref(lqes[index]);
1000 memcpy((unsigned char *)lqes + index * lqep_size,
1001 (unsigned char *)lqes + (index + 1) * lqep_size,
1002 (lqes_cnt - index - 1) * lqep_size);
1003 qti_lqes_cnt(env)--;
1006 void qti_lqes_fini(const struct lu_env *env)
1008 struct qmt_thread_info *qti = qmt_info(env);
1009 struct lquota_entry **lqes = qti->qti_lqes;
1012 lqes = qti_lqes(env);
1013 for (i = 0; i < qti->qti_lqes_cnt; i++) {
1014 qpi_putref(env, lqe2qpi(lqes[i]));
1015 lqe_putref(lqes[i]);
1018 if (qti->qti_lqes_num > QMT_MAX_POOL_NUM)
1019 OBD_FREE(qti->qti_lqes,
1020 qti->qti_lqes_num * sizeof(struct lquota_entry *));
1022 qti->qti_lqes_num = 0;
1023 qti->qti_lqes_cnt = 0;
1026 __u64 qti_lqes_min_qunit(const struct lu_env *env)
1031 for (i = 1, min = qti_lqe_qunit(env, 0); i < qti_lqes_cnt(env); i++) {
1032 qunit = qti_lqe_qunit(env, i);
1033 /* if qunit is 0, lqe is not enforced and we can ignore it */
1034 if (qunit && qunit < min)
1041 int qti_lqes_edquot(const struct lu_env *env)
1045 for (i = 0; i < qti_lqes_cnt(env); i++) {
1046 if (qti_lqes(env)[i]->lqe_edquot)
1053 int qti_lqes_restore_init(const struct lu_env *env)
1057 if (qti_lqes_inited(env) && qti_lqes_cnt(env) > QMT_MAX_POOL_NUM) {
1058 OBD_ALLOC(qmt_info(env)->qti_lqes_rstr,
1059 qti_lqes_cnt(env) * sizeof(struct qmt_lqe_restore));
1060 if (!qti_lqes_rstr(env))
1067 void qti_lqes_restore_fini(const struct lu_env *env)
1069 if (qti_lqes_inited(env) && qti_lqes_cnt(env) > QMT_MAX_POOL_NUM)
1070 OBD_FREE(qmt_info(env)->qti_lqes_rstr,
1071 qti_lqes_cnt(env) * sizeof(struct qmt_lqe_restore));
1074 void qti_lqes_write_lock(const struct lu_env *env)
1078 for (i = 0; i < qti_lqes_cnt(env); i++)
1079 lqe_write_lock(qti_lqes(env)[i]);
1082 void qti_lqes_write_unlock(const struct lu_env *env)
1086 for (i = 0; i < qti_lqes_cnt(env); i++)
1087 lqe_write_unlock(qti_lqes(env)[i]);
1090 #define QMT_INIT_SLV_CNT 64
1091 struct lqe_glbl_data *qmt_alloc_lqe_gd(struct qmt_pool_info *pool, int qtype)
1093 struct lqe_glbl_data *lgd;
1094 struct lqe_glbl_entry *lqeg_arr;
1095 int slv_cnt, glbe_num;
1097 OBD_ALLOC(lgd, sizeof(struct lqe_glbl_data));
1101 slv_cnt = qpi_slv_nr_by_rtype(pool, qtype);
1103 glbe_num = slv_cnt < QMT_INIT_SLV_CNT ? QMT_INIT_SLV_CNT : slv_cnt;
1104 OBD_ALLOC(lqeg_arr, sizeof(struct lqe_glbl_entry) * glbe_num);
1106 OBD_FREE(lgd, sizeof(struct lqe_glbl_data));
1110 CDEBUG(D_QUOTA, "slv_cnt %d glbe_num %d\n", slv_cnt, glbe_num);
1112 lgd->lqeg_num_used = slv_cnt;
1113 lgd->lqeg_num_alloc = glbe_num;
1114 lgd->lqeg_arr = lqeg_arr;
1119 void qmt_free_lqe_gd(struct lqe_glbl_data *lgd)
1121 OBD_FREE(lgd->lqeg_arr,
1122 sizeof(struct lqe_glbl_entry) * lgd->lqeg_num_alloc);
1123 OBD_FREE(lgd, sizeof(struct lqe_glbl_data));
1126 void qmt_seed_glbe_all(const struct lu_env *env, struct lqe_glbl_data *lgd,
1127 bool qunit, bool edquot)
1129 struct rw_semaphore *sem = NULL;
1130 struct qmt_pool_info *qpi;
1134 if (!qti_lqes_cnt(env))
1136 /* lqes array is sorted by qunit - the first entry has minimum qunit.
1137 * Thus start seeding global qunit's array beginning from the 1st lqe
1138 * and appropriate pool. If pools overlapped, slaves from this
1139 * overlapping get minimum qunit value.
1140 * user1: pool1, pool2, pool_glbl;
1141 * pool1: OST1; user1_qunit = 10M;
1142 * pool2: OST0, OST1, OST2; user1_qunit = 30M;
1143 * pool_glbl: OST0, OST1, OST2, OST3; user1_qunit = 160M;
1144 * qunit array after seeding should be:
1145 * OST0: 30M; OST1: 10M; OST2: 30M; OST3: 160M; */
1147 /* edquot resetup algorythm works fine
1148 * with not sorted lqes */
1152 for (i = 0; i < lgd->lqeg_num_used; i++) {
1153 lgd->lqeg_arr[i].lge_qunit_set = 0;
1154 lgd->lqeg_arr[i].lge_qunit_nu = 0;
1155 lgd->lqeg_arr[i].lge_edquot_nu = 0;
1158 for (i = 0; i < qti_lqes_cnt(env); i++) {
1159 struct lquota_entry *lqe = qti_lqes(env)[i];
1162 CDEBUG(D_QUOTA, "lqes_cnt %d, i %d\n", qti_lqes_cnt(env), i);
1164 if (qmt_pool_global(qpi)) {
1165 slaves_cnt = qpi_slv_nr_by_rtype(lqe2qpi(lqe),
1168 sem = qmt_sarr_rwsem(qpi);
1170 slaves_cnt = qmt_sarr_count(qpi);
1173 for (j = 0; j < slaves_cnt; j++) {
1174 idx = qmt_sarr_get_idx(qpi, j);
1178 int lge_edquot, new_edquot, edquot_nu;
1180 lge_edquot = lgd->lqeg_arr[idx].lge_edquot;
1181 edquot_nu = lgd->lqeg_arr[idx].lge_edquot_nu;
1182 new_edquot = lqe->lqe_edquot;
1184 if (lge_edquot == new_edquot ||
1185 (edquot_nu && lge_edquot == 1))
1187 lgd->lqeg_arr[idx].lge_edquot = new_edquot;
1188 /* it is needed for the following case:
1189 * initial values for idx i -
1190 * lqe_edquot = 1, lqe_edquot_nu == 0;
1191 * 1: new_edquot == 0 ->
1192 * lqe_edquot = 0, lqe_edquot_nu = 1;
1193 * 2: new_edquot == 1 ->
1194 * lqe_edquot = 1, lqe_edquot_nu = 0;
1195 * At the 2nd iteration lge_edquot comes back
1196 * to 1, so no changes and we don't need
1197 * to notify slave. */
1198 lgd->lqeg_arr[idx].lge_edquot_nu = !edquot_nu;
1202 __u64 lge_qunit, new_qunit;
1205 "idx %d lge_qunit_set %d lge_qunit %llu new_qunit %llu\n",
1206 idx, lgd->lqeg_arr[idx].lge_qunit_set,
1207 lgd->lqeg_arr[idx].lge_qunit,
1209 /* lge for this idx is already set
1210 * on previous iteration */
1211 if (lgd->lqeg_arr[idx].lge_qunit_set)
1213 lge_qunit = lgd->lqeg_arr[idx].lge_qunit;
1214 new_qunit = lqe->lqe_qunit;
1215 /* qunit could be not set,
1216 * so use global lqe's qunit */
1220 if (lge_qunit != new_qunit)
1221 lgd->lqeg_arr[idx].lge_qunit =
1224 /* TODO: initially slaves notification was done
1225 * only for qunit shrinking. Should we always
1226 * notify slaves with new qunit ? */
1227 if (lge_qunit > new_qunit)
1228 lgd->lqeg_arr[idx].lge_qunit_nu = 1;
1229 lgd->lqeg_arr[idx].lge_qunit_set = 1;
1233 if (!qmt_pool_global(qpi))
1236 /* TODO: only for debug purposes - remove it later */
1237 for (i = 0; i < lgd->lqeg_num_used; i++)
1239 "lgd ost %d, qunit %lu nu %d; edquot %d nu %d\n",
1240 i, (long unsigned)lgd->lqeg_arr[i].lge_qunit,
1241 lgd->lqeg_arr[i].lge_qunit_nu,
1242 lgd->lqeg_arr[i].lge_edquot,
1243 lgd->lqeg_arr[i].lge_edquot_nu);
1248 void qmt_setup_lqe_gd(const struct lu_env *env, struct qmt_device *qmt,
1249 struct lquota_entry *lqe, struct lqe_glbl_data *lgd,
1256 qunit = lqe->lqe_qunit;
1257 edquot = lqe->lqe_edquot;
1259 /* Firstly set all elements in array with
1260 * qunit and edquot of global pool */
1261 for (i = 0; i < lgd->lqeg_num_used; i++) {
1262 lgd->lqeg_arr[i].lge_qunit = qunit;
1263 lgd->lqeg_arr[i].lge_edquot = edquot;
1264 /* It is the very first lvb setup - qunit and other flags
1265 * will be sent to slaves during qmt_lvbo_fill. */
1266 lgd->lqeg_arr[i].lge_qunit_nu = 0;
1267 lgd->lqeg_arr[i].lge_edquot_nu = 0;
1270 qmt_pool_lqes_lookup_spec(env, qmt, pool_type,
1271 lqe_qtype(lqe), &lqe->lqe_id);
1272 qmt_seed_glbe(env, lgd);
1274 lqe->lqe_glbl_data = lgd;
1275 qmt_id_lock_notify(qmt, lqe);