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 <obd_class.h>
34 #include "qmt_internal.h"
37 * Retrieve quota settings for a given identifier.
39 * \param env - is the environment passed by the caller
40 * \param qmt - is the quota master target
41 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or inode
42 * (i.e. LQUOTA_RES_MD)
43 * \param qtype - is the quota type
44 * \param id - is the quota indentifier for which we want to acces quota
46 * \param hard - is the output variable where to copy the hard limit
47 * \param soft - is the output variable where to copy the soft limit
48 * \param time - is the output variable where to copy the grace time
50 static int qmt_get(const struct lu_env *env, struct qmt_device *qmt,
51 __u8 restype, __u8 qtype, union lquota_id *id,
52 __u64 *hard, __u64 *soft, __u64 *time, bool is_default,
55 struct lquota_entry *lqe;
58 LASSERT(!is_default || id->qid_uid == 0);
59 if (pool_name && !strnlen(pool_name, LOV_MAXPOOLNAME))
62 /* look-up lqe structure containing quota settings */
63 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, id, pool_name);
67 /* copy quota settings */
69 LQUOTA_DEBUG(lqe, "fetch settings");
71 *hard = lqe->lqe_hardlimit;
73 *soft = lqe->lqe_softlimit;
75 *time = lqe->lqe_gracetime;
76 if (lqe->lqe_is_default)
77 *time |= (__u64)LQUOTA_FLAG_DEFAULT <<
86 struct qmt_entry_iter_data {
87 const struct lu_env *qeid_env;
88 struct qmt_device *qeid_qmt;
91 static int qmt_entry_iter_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
92 struct hlist_node *hnode, void *d)
94 struct qmt_entry_iter_data *iter = (struct qmt_entry_iter_data *)d;
95 struct lquota_entry *lqe;
97 lqe = hlist_entry(hnode, struct lquota_entry, lqe_hash);
98 LASSERT(atomic_read(&lqe->lqe_ref) > 0);
100 if (lqe->lqe_id.qid_uid == 0 || !lqe->lqe_is_default)
103 return qmt_set_with_lqe(iter->qeid_env, iter->qeid_qmt, lqe, 0, 0, 0, 0,
107 static void qmt_set_id_notify(const struct lu_env *env, struct qmt_device *qmt,
108 struct lquota_entry *lqe)
110 struct lquota_entry *lqe_gl;
113 lqe_gl = lqe->lqe_is_global ? lqe : NULL;
114 rc = qmt_pool_lqes_lookup_spec(env, qmt, lqe_rtype(lqe),
115 lqe_qtype(lqe), &lqe->lqe_id);
119 if (!lqe_gl && qti_lqes_glbl(env)->lqe_is_global)
120 lqe_gl = qti_lqes_glbl(env);
125 mutex_lock(&lqe_gl->lqe_glbl_data_lock);
126 if (lqe_gl->lqe_glbl_data)
127 qmt_seed_glbe(env, lqe_gl->lqe_glbl_data);
128 mutex_unlock(&lqe_gl->lqe_glbl_data_lock);
130 /* Even if slaves haven't enqueued quota lock yet,
131 * it is needed to set lqe_revoke_time in qmt_id_lock_glimpse
132 * in case of reaching qpi_least_qunit */
133 qmt_id_lock_notify(qmt, lqe_gl);
139 * Update quota settings for a given lqe.
141 * \param env - is the environment passed by the caller
142 * \param qmt - is the quota master target
143 * \param lqe - is the lquota_entry for which we want to modify quota
145 * \param hard - is the new hard limit
146 * \param soft - is the new soft limit
147 * \param time - is the new grace time
148 * \param valid - is the list of settings to change
149 * \param is_default - true for default quota setting
150 * \param is_updated - true if the lqe is updated and no need to write back
153 int qmt_set_with_lqe(const struct lu_env *env, struct qmt_device *qmt,
154 struct lquota_entry *lqe, __u64 hard, __u64 soft,
155 __u64 time, __u32 valid, bool is_default, bool is_updated)
157 struct thandle *th = NULL;
160 bool dirtied = false;
162 bool need_id_notify = false;
165 /* need to write back to global quota file? */
167 /* By default we should have here only 1 lqe,
168 * so no allocations should be done. */
169 if (qti_lqes_restore_init(env))
170 GOTO(out_nolock, rc = -ENOMEM);
171 /* allocate & start transaction with enough credits to update
172 * quota settings in the global index file */
173 th = qmt_trans_start(env, lqe);
175 GOTO(out_nolock, rc = PTR_ERR(th));
178 now = ktime_get_real_seconds();
182 "changing quota settings valid:%x hard:%llu soft:%llu time:%llu",
183 valid, hard, soft, time);
185 if (is_default && lqe->lqe_id.qid_uid != 0) {
186 LQUOTA_DEBUG(lqe, "set qid %llu to use default quota setting",
187 lqe->lqe_id.qid_uid);
189 qmt_lqe_set_default(env, lqe->lqe_site->lqs_parent, lqe, false);
193 if ((valid & QIF_TIMES) != 0 && lqe->lqe_gracetime != time) {
194 /* change time settings */
195 lqe->lqe_gracetime = time;
199 if ((valid & QIF_LIMITS) != 0 &&
200 (lqe->lqe_hardlimit != hard || lqe->lqe_softlimit != soft)) {
201 rc = qmt_validate_limits(lqe, hard, soft);
205 /* change quota limits */
206 lqe->lqe_hardlimit = hard;
207 lqe->lqe_softlimit = soft;
210 GOTO(quota_write, 0);
214 /* recompute qunit in case it was never initialized */
215 if (qmt_revalidate(env, lqe))
216 need_id_notify = true;
218 /* clear grace time */
219 if (lqe->lqe_softlimit == 0 ||
220 lqe->lqe_granted <= lqe->lqe_softlimit)
221 /* no soft limit or below soft limit, let's clear grace
223 lqe->lqe_gracetime = 0;
224 else if ((valid & QIF_TIMES) == 0)
225 /* set grace only if user hasn't provided his own */
226 lqe->lqe_gracetime = now + qmt_lqe_grace(lqe);
228 /* change enforced status based on new parameters */
229 if (lqe->lqe_id.qid_uid == 0 || (lqe->lqe_hardlimit == 0 &&
230 lqe->lqe_softlimit == 0)) {
231 if (lqe->lqe_enforced) {
232 lqe->lqe_enforced = false;
233 /* Clear qunit and edquot as lqe_adjust_edquot
234 * does not handle not enforced lqes */
237 need_id_notify = true;
240 lqe->lqe_enforced = true;
246 if (!is_default && lqe->lqe_is_default) {
247 LQUOTA_DEBUG(lqe, "the qid %llu has been set quota"
248 " explicitly, clear the default flag",
249 lqe->lqe_id.qid_uid);
251 qmt_lqe_clear_default(lqe);
258 /* write new quota settings to disk */
259 rc = qmt_glb_write(env, th, lqe, LQUOTA_BUMP_VER, &ver);
261 /* restore initial quota settings */
262 qmt_restore(lqe, &qti_lqes_rstr(env)[0]);
265 } else if (lqe->lqe_is_global) {
266 ver = dt_version_get(env, LQE_GLB_OBJ(lqe));
269 /* compute new qunit value now that we have modified the quota
270 * settings or clear/set edquot flag if needed */
271 need_id_notify |= qmt_adjust_qunit(env, lqe);
272 need_id_notify |= qmt_adjust_edquot(lqe, now);
276 lqe_write_unlock(lqe);
280 if (th != NULL && !IS_ERR(th))
281 dt_trans_stop(env, qmt->qmt_child, th);
282 qti_lqes_restore_fini(env);
285 if (rc == 0 && dirtied) {
286 if (lqe->lqe_is_global)
287 qmt_glb_lock_notify(env, lqe, ver);
288 if (lqe->lqe_id.qid_uid == 0) {
289 struct qmt_entry_iter_data iter_data;
291 LQUOTA_DEBUG(lqe, "notify all lqe with default quota");
292 iter_data.qeid_env = env;
293 iter_data.qeid_qmt = qmt;
294 cfs_hash_for_each(lqe->lqe_site->lqs_hash,
295 qmt_entry_iter_cb, &iter_data);
296 /* Always notify slaves with default values. Don't
297 * care about overhead as will be sent only not changed
298 * values(see qmt_id_lock_cb for details).*/
299 need_id_notify = true;
301 /* qti_lqes_inited > 0 means we came here from another
302 * qmt_pool_lqes_lookup(qmt_dqacq, intent_policy ...). Thus
303 * we can't init and add new lqes to don't overwrite already
306 if (!qti_lqes_inited(env) && need_id_notify)
307 qmt_set_id_notify(env, qmt, lqe);
314 * Update quota settings for a given identifier.
316 * \param env - is the environment passed by the caller
317 * \param qmt - is the quota master target
318 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or
319 * inode (i.e. LQUOTA_RES_MD)
320 * \param qtype - is the quota type
321 * \param id - is the quota indentifier for which we want to modify
323 * \param hard - is the new hard limit
324 * \param soft - is the new soft limit
325 * \param time - is the new grace time
326 * \param valid - is the list of settings to change
327 * \param is_default - true for default quota setting
328 * \param is_updated - true if the lqe is updated and no need to write back
330 static int qmt_set(const struct lu_env *env, struct qmt_device *qmt,
331 __u8 restype, __u8 qtype, union lquota_id *id,
332 __u64 hard, __u64 soft, __u64 time, __u32 valid,
333 bool is_default, bool is_updated, char *pool_name)
335 struct lquota_entry *lqe;
339 if (pool_name && !strnlen(pool_name, LOV_MAXPOOLNAME))
342 /* look-up quota entry associated with this ID */
343 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, id, pool_name);
345 RETURN(PTR_ERR(lqe));
347 lqe->lqe_is_deleted = 0;
348 lqe->lqe_is_reset = 0;
349 rc = qmt_set_with_lqe(env, qmt, lqe, hard, soft, time, valid,
350 is_default, is_updated);
352 lqe->lqe_is_deleted = 0;
359 * Delete the quota setting of the specified quota ID
361 * \param env - is the environment passed by the caller
362 * \param qmt - is the quota master target
363 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or
364 * inode (i.e. LQUOTA_RES_MD)
365 * \param qtype - is the quota type
366 * \param qid - is the quota indentifier for which we want to delete its
369 static int qmt_delete_qid(const struct lu_env *env, struct qmt_device *qmt,
370 __u8 restype, __u8 qtype, __u64 qid)
372 struct qmt_thread_info *qti = qmt_info(env);
373 union lquota_id *quota_id = &qti->qti_id;
374 struct thandle *th = NULL;
375 struct qmt_pool_info *qpi = NULL;
376 struct lquota_entry *lqe = NULL;
382 quota_id->qid_uid = qid;
383 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, quota_id, NULL);
385 RETURN(PTR_ERR(lqe));
389 qpi = qmt_pool_lookup_glb(env, qmt, restype);
391 GOTO(out, rc = -ENOMEM);
393 th = qmt_trans_start(env, lqe);
395 GOTO(out, rc = PTR_ERR(th));
397 rc = lquota_disk_delete(env, th,
398 qpi->qpi_glb_obj[qtype], qid, &ver);
400 dt_trans_stop(env, qmt->qmt_child, th);
403 lqe_set_deleted(lqe);
404 qmt_glb_lock_notify(env, lqe, ver);
405 } else if (rc == -ENOENT) {
410 if (!IS_ERR_OR_NULL(qpi))
411 qpi_putref(env, qpi);
413 lqe_write_unlock(lqe);
419 static int qmt_reset_slv_cb(const struct lu_env *env, struct lu_fid *glb_fid,
420 char *slv_name, struct lu_fid *slv_fid, void *arg)
422 struct qmt_device *qmt = (struct qmt_device *)arg;
423 struct qmt_thread_info *qti = qmt_info(env);
424 struct dt_object *slv_obj = NULL;
425 struct lquota_slv_rec rec;
426 struct thandle *th = NULL;
429 slv_obj = dt_locate(env, qmt->qmt_child, slv_fid);
431 GOTO(out, rc = PTR_ERR(slv_obj));
433 if (slv_obj->do_index_ops == NULL) {
434 rc = slv_obj->do_ops->do_index_try(env, slv_obj,
435 &dt_quota_slv_features);
437 CERROR("%s: fail to setup slave idx for %s: rc = %d\n",
438 qmt->qmt_child->dd_lu_dev.ld_obd->obd_name,
444 th = qmt_trans_start(env, qti_lqes(env)[0]);
446 GOTO(out, rc = PTR_ERR(th));
449 rc = lquota_disk_write(env, th, slv_obj, &qti->qti_id,
450 (struct dt_rec *)&rec, 0, NULL);
452 CERROR("%s: failed to reset slave grant for %s: rc = %d\n",
453 qmt->qmt_child->dd_lu_dev.ld_obd->obd_name, slv_name,
456 if (!IS_ERR_OR_NULL(th))
457 dt_trans_stop(env, qmt->qmt_child, th);
460 dt_object_put(env, slv_obj);
465 * Reset the quota of the quota ID, it will reset the soft/hard limit and grant
467 * \param env - is the environment passed by the caller
468 * \param qmt - is the quota master target
469 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or
470 * inode (i.e. LQUOTA_RES_MD)
471 * \param qtype - is the quota type
472 * \param qid - is the quota indentifier for which we want to delete its
475 static int qmt_reset_qid(const struct lu_env *env, struct qmt_device *qmt,
476 __u8 restype, __u8 qtype, __u64 qid)
478 struct qmt_thread_info *qti = qmt_info(env);
479 union lquota_id *quota_id = &qti->qti_id;
480 struct qmt_pool_info *qpi = NULL;
481 struct lquota_entry *lqe = NULL;
482 struct thandle *th = NULL;
483 __u64 softlimit = 0, hardlimit = 0;
489 quota_id->qid_uid = qid;
490 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, quota_id, NULL);
492 RETURN(PTR_ERR(lqe));
496 qpi = qmt_pool_lookup_glb(env, qmt, restype);
498 GOTO(out, rc = -ENOMEM);
500 th = qmt_trans_start(env, lqe);
502 GOTO(out, rc = PTR_ERR(th));
504 softlimit = lqe->lqe_softlimit;
505 hardlimit = lqe->lqe_hardlimit;
507 lqe->lqe_softlimit = 0;
508 lqe->lqe_hardlimit = 0;
509 lqe->lqe_granted = 0;
512 lqe->lqe_is_default = 0;
513 lqe->lqe_is_deleted = 0;
514 lqe->lqe_is_reset = 1;
515 rc = qmt_glb_write(env, th, lqe, LQUOTA_BUMP_VER, &ver);
517 LQUOTA_ERROR(lqe, "failed to write quota global rec\n");
518 dt_trans_stop(env, qmt->qmt_child, th);
522 lquota_generate_fid(&qti->qti_fid, restype, qtype);
523 qti_lqes(env)[0] = lqe;
524 lquota_disk_for_each_slv(env, qpi->qpi_root, &qti->qti_fid,
525 qmt_reset_slv_cb, qmt);
527 qmt_glb_lock_notify(env, lqe, ver);
532 lqe->lqe_softlimit = softlimit;
534 lqe->lqe_hardlimit = hardlimit;
535 lqe->lqe_is_reset = 0;
538 if (!IS_ERR_OR_NULL(qpi))
539 qpi_putref(env, qpi);
541 lqe_write_unlock(lqe);
547 * Handle quotactl request.
549 * \param env - is the environment passed by the caller
550 * \param ld - is the lu device associated with the qmt
551 * \param oqctl - is the quotactl request
553 static int qmt_quotactl(const struct lu_env *env, struct lu_device *ld,
554 struct obd_quotactl *oqctl)
556 struct qmt_thread_info *qti = qmt_info(env);
557 union lquota_id *id = &qti->qti_id;
558 struct qmt_device *qmt = lu2qmt_dev(ld);
559 struct obd_dqblk *dqb = &oqctl->qc_dqblk;
562 bool is_default = false;
565 LASSERT(qmt != NULL);
567 if (oqctl->qc_type >= LL_MAXQUOTAS)
568 /* invalid quota type */
571 poolname = LUSTRE_Q_CMD_IS_POOL(oqctl->qc_cmd) ?
572 oqctl->qc_poolname : NULL;
574 switch (oqctl->qc_cmd) {
576 case Q_GETINFO: /* read grace times */
577 case LUSTRE_Q_GETINFOPOOL:
578 /* Global grace time is stored in quota settings of ID 0. */
581 /* read inode grace time */
582 rc = qmt_get(env, qmt, LQUOTA_RES_MD, oqctl->qc_type, id, NULL,
583 NULL, &oqctl->qc_dqinfo.dqi_igrace,
585 /* There could be no MD pool, so try to find DT pool */
586 if (rc && rc != -ENOENT)
589 /* read block grace time */
590 rc = qmt_get(env, qmt, LQUOTA_RES_DT, oqctl->qc_type, id, NULL,
591 NULL, &oqctl->qc_dqinfo.dqi_bgrace,
595 case Q_SETINFO: /* modify grace times */
596 case LUSTRE_Q_SETINFOPOOL:
597 /* setinfo should be using dqi->dqi_valid, but lfs incorrectly
598 * sets the valid flags in dqb->dqb_valid instead, try to live
601 /* Global grace time is stored in quota settings of ID 0. */
604 if ((dqb->dqb_valid & QIF_ITIME) != 0) {
605 /* set inode grace time */
606 rc = qmt_set(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
607 id, 0, 0, oqctl->qc_dqinfo.dqi_igrace,
608 QIF_TIMES, false, false,
614 if ((dqb->dqb_valid & QIF_BTIME) != 0)
615 /* set block grace time */
616 rc = qmt_set(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
617 id, 0, 0, oqctl->qc_dqinfo.dqi_bgrace,
618 QIF_TIMES, false, false,
622 case LUSTRE_Q_GETDEFAULT:
623 case LUSTRE_Q_GETDEFAULT_POOL:
627 case Q_GETQUOTA: /* consult quota limit */
628 case LUSTRE_Q_GETQUOTAPOOL:
629 /* extract quota ID from quotactl request */
630 id->qid_uid = oqctl->qc_id;
632 /* look-up inode quota settings */
633 rc = qmt_get(env, qmt, LQUOTA_RES_MD, oqctl->qc_type, id,
634 &dqb->dqb_ihardlimit, &dqb->dqb_isoftlimit,
635 &dqb->dqb_itime, is_default, poolname);
636 /* There could be no MD pool, so try to find DT pool */
637 if (rc && rc != -ENOENT)
640 dqb->dqb_valid |= QIF_ILIMITS | QIF_ITIME;
642 /* master isn't aware of actual inode usage */
643 dqb->dqb_curinodes = 0;
645 /* look-up block quota settings */
646 rc = qmt_get(env, qmt, LQUOTA_RES_DT, oqctl->qc_type, id,
647 &dqb->dqb_bhardlimit, &dqb->dqb_bsoftlimit,
648 &dqb->dqb_btime, is_default, poolname);
652 dqb->dqb_valid |= QIF_BLIMITS | QIF_BTIME;
653 /* master doesn't know the actual block usage */
654 dqb->dqb_curspace = 0;
657 case LUSTRE_Q_SETDEFAULT:
658 case LUSTRE_Q_SETDEFAULT_POOL:
662 case Q_SETQUOTA: /* change quota limits */
663 case LUSTRE_Q_SETQUOTAPOOL:
664 /* extract quota ID from quotactl request */
665 id->qid_uid = oqctl->qc_id;
667 if ((dqb->dqb_valid & QIF_IFLAGS) != 0) {
668 /* update inode quota settings */
669 rc = qmt_set(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
670 id, dqb->dqb_ihardlimit,
671 dqb->dqb_isoftlimit, dqb->dqb_itime,
672 dqb->dqb_valid & QIF_IFLAGS, is_default,
678 if ((dqb->dqb_valid & QIF_BFLAGS) != 0)
679 /* update block quota settings */
680 rc = qmt_set(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
681 id, dqb->dqb_bhardlimit,
682 dqb->dqb_bsoftlimit, dqb->dqb_btime,
683 dqb->dqb_valid & QIF_BFLAGS, is_default,
687 case LUSTRE_Q_DELETEQID:
688 rc = qmt_delete_qid(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
693 rc = qmt_delete_qid(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
697 case LUSTRE_Q_RESETQID:
698 if (oqctl->qc_id == 0)
701 id->qid_uid = oqctl->qc_id;
702 /* save the quota setting before resetting */
703 rc = qmt_get(env, qmt, LQUOTA_RES_MD, oqctl->qc_type, id,
704 &dqb->dqb_ihardlimit, &dqb->dqb_isoftlimit,
705 &dqb->dqb_itime, false, NULL);
709 dqb->dqb_valid |= QIF_ILIMITS | QIF_ITIME;
711 rc = qmt_get(env, qmt, LQUOTA_RES_DT, oqctl->qc_type, id,
712 &dqb->dqb_bhardlimit, &dqb->dqb_bsoftlimit,
713 &dqb->dqb_btime, false, NULL);
717 dqb->dqb_valid |= QIF_BLIMITS | QIF_BTIME;
718 dqb->dqb_curinodes = 0;
719 dqb->dqb_curspace = 0;
721 /* reset the corresponding quota ID */
722 rc = qmt_reset_qid(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
727 rc = qmt_reset_qid(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
733 CERROR("%s: unsupported quotactl command: %d\n",
734 qmt->qmt_svname, oqctl->qc_cmd);
742 void qmt_grant_lqes(const struct lu_env *env, __u64 *slv, __u64 cnt)
746 for (i = 0; i < qti_lqes_cnt(env); i++)
747 qti_lqe_granted(env, i) += cnt;
752 static inline bool qmt_lqes_can_rel(const struct lu_env *env, __u64 cnt)
754 bool can_release = true;
757 for (i = 0; i < qti_lqes_cnt(env); i++) {
758 if (cnt > qti_lqe_granted(env, i)) {
759 LQUOTA_ERROR(qti_lqes(env)[i],
760 "Can't release %llu that is larger than lqe_granted.\n",
768 static inline void qmt_rel_lqes(const struct lu_env *env, __u64 *slv, __u64 cnt,
773 for (i = 0; i < qti_lqes_cnt(env); i++) {
775 qti_lqe_granted(env, i) = 0;
777 qti_lqe_granted(env, i) -= cnt;
786 static inline bool qmt_lqes_cannot_grant(const struct lu_env *env, __u64 cnt)
788 bool cannot_grant = false;
791 for (i = 0; i < qti_lqes_cnt(env); i++) {
792 if (qti_lqe_hard(env, i) != 0 &&
793 qti_lqe_granted(env, i) + cnt > qti_lqe_hard(env, i)) {
801 static inline __u64 qmt_lqes_grant_some_quota(const struct lu_env *env)
803 __u64 min_count, tmp;
807 for (i = 0, min_count = 0; i < qti_lqes_cnt(env); i++) {
808 if (!qti_lqes(env)[i]->lqe_enforced &&
809 !qti_lqes(env)[i]->lqe_is_global)
812 tmp = qti_lqe_hard(env, i) - qti_lqe_granted(env, i);
814 min_count = tmp < min_count ? tmp : min_count;
823 static inline __u64 qmt_lqes_alloc_expand(const struct lu_env *env,
824 __u64 slv_granted, __u64 spare)
826 __u64 min_count, tmp;
830 for (i = 0, min_count = 0; i < qti_lqes_cnt(env); i++) {
831 /* Don't take into account not enforced lqes that belong
832 * to non global pool. These lqes present in array to
833 * support actual lqe_granted even for lqes without limits. */
834 if (!qti_lqes(env)[i]->lqe_enforced &&
835 !qti_lqes(env)[i]->lqe_is_global)
838 tmp = qmt_alloc_expand(qti_lqes(env)[i], slv_granted, spare);
840 min_count = tmp < min_count ? tmp : min_count;
849 static inline void qmt_lqes_tune_grace(const struct lu_env *env, __u64 now)
853 for (i = 0; i < qti_lqes_cnt(env); i++) {
854 struct lquota_entry *lqe;
856 lqe = qti_lqes(env)[i];
857 if (lqe->lqe_softlimit != 0) {
858 if (lqe->lqe_granted > lqe->lqe_softlimit &&
859 lqe->lqe_gracetime == 0) {
860 /* First time over soft limit, let's start grace
862 lqe->lqe_gracetime = now + qmt_lqe_grace(lqe);
863 } else if (lqe->lqe_granted <= lqe->lqe_softlimit &&
864 lqe->lqe_gracetime != 0) {
865 /* Clear grace timer */
866 lqe->lqe_gracetime = 0;
873 * Helper function to handle quota request from slave.
875 * \param env - is the environment passed by the caller
876 * \param qmt - is the master device
877 * \param uuid - is the uuid associated with the slave
878 * \param qb_flags - are the quota request flags as packed in the quota_body
879 * \param qb_count - is the amount of quota space the slave wants to
881 * \param qb_usage - is the current space usage on the slave
882 * \param repbody - is the quota_body of reply
883 * \param idx - is the index of a slave target
885 * \retval 0 : success
886 * \retval -EDQUOT : out of quota
887 * -EINPROGRESS : inform client to retry write/create
888 * -ve : other appropriate errors
890 int qmt_dqacq0(const struct lu_env *env, struct qmt_device *qmt,
891 struct obd_uuid *uuid, __u32 qb_flags, __u64 qb_count,
892 __u64 qb_usage, struct quota_body *repbody, int idx)
894 __u64 now, count = 0;
895 struct dt_object *slv_obj = NULL;
896 __u64 slv_granted, slv_granted_bck;
897 struct thandle *th = NULL;
899 struct lquota_entry *lqe = qti_lqes_glbl(env);
902 LASSERT(uuid != NULL);
904 /* initialize reply */
905 memset(repbody, 0, sizeof(*repbody));
906 memcpy(&repbody->qb_id, &lqe->lqe_id, sizeof(repbody->qb_id));
908 if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_RECOVERABLE_ERR))
909 RETURN(-cfs_fail_val);
911 if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_PREACQ) &&
912 (req_is_preacq(qb_flags) || req_is_rel(qb_flags)))
915 if (qti_lqes_restore_init(env))
918 /* look-up index file associated with acquiring slave */
919 slv_obj = lquota_disk_slv_find(env, qmt->qmt_child, LQE_ROOT(lqe),
920 lu_object_fid(&LQE_GLB_OBJ(lqe)->do_lu),
923 GOTO(out, rc = PTR_ERR(slv_obj));
925 /* pack slave fid in reply just for sanity check */
926 memcpy(&repbody->qb_slv_fid, lu_object_fid(&slv_obj->do_lu),
927 sizeof(struct lu_fid));
929 /* allocate & start transaction with enough credits to update
930 * global & slave indexes */
931 th = qmt_trans_start_with_slv(env, NULL, slv_obj, false);
933 GOTO(out, rc = PTR_ERR(th));
935 qti_lqes_write_lock(env);
937 LQUOTA_DEBUG_LQES(env, "dqacq starts uuid:%s flags:0x%x wanted:%llu"
938 " usage:%llu", obd_uuid2str(uuid), qb_flags, qb_count,
941 /* Legal race, limits have been removed on master, but slave didn't
942 * receive the change yet. Just return EINPROGRESS until the slave gets
944 if (!lqe->lqe_enforced && !req_is_rel(qb_flags))
945 GOTO(out_locked, rc = -ESRCH);
947 /* recompute qunit in case it was never initialized */
948 qmt_revalidate_lqes(env, qmt, qb_flags);
950 /* slave just wants to acquire per-ID lock */
951 if (req_is_acq(qb_flags) && qb_count == 0)
952 GOTO(out_locked, rc = 0);
954 if (lqe->lqe_is_reset) {
955 lqe->lqe_granted = 0;
956 repbody->qb_count = qb_count;
957 qmt_rel_lqes(env, &slv_granted, qb_count, lqe->lqe_is_reset);
958 GOTO(out_locked, rc = 0);
961 /* fetch how much quota space is already granted to this slave */
962 rc = qmt_slv_read(env, &lqe->lqe_id, slv_obj, &slv_granted);
964 LQUOTA_ERROR(lqe, "Failed to get granted for slave %s, rc=%d",
965 obd_uuid2str(uuid), rc);
966 GOTO(out_locked, rc);
968 /* recall how much space this slave currently owns in order to restore
969 * it in case of failure */
970 slv_granted_bck = slv_granted;
972 /* record current time for soft limit & grace time management */
973 now = ktime_get_real_seconds();
975 if (req_is_rel(qb_flags)) {
976 /* Slave would like to release quota space */
977 if (slv_granted < qb_count ||
978 !qmt_lqes_can_rel(env, qb_count)) {
979 /* can't release more than granted */
980 LQUOTA_ERROR_LQES(env,
981 "Release too much! uuid:%s release: %llu granted:%llu, total:%llu",
982 obd_uuid2str(uuid), qb_count,
983 slv_granted, lqe->lqe_granted);
984 GOTO(out_locked, rc = -EINVAL);
987 repbody->qb_count = qb_count;
988 /* put released space back to global pool */
989 qmt_rel_lqes(env, &slv_granted, qb_count, lqe->lqe_is_reset);
990 GOTO(out_write, rc = 0);
993 if (req_has_rep(qb_flags) && slv_granted < qb_usage) {
994 /* Slave is reporting space usage in quota request and it turns
995 * out to be using more quota space than owned, so we adjust
996 * granted space regardless of the current state of affairs */
997 repbody->qb_count = qb_usage - slv_granted;
998 qmt_grant_lqes(env, &slv_granted, repbody->qb_count);
1001 if (!req_is_acq(qb_flags) && !req_is_preacq(qb_flags))
1002 GOTO(out_write, rc = 0);
1004 qmt_adjust_edquot_notify(env, qmt, now, qb_flags);
1005 if (qti_lqes_edquot(env))
1006 /* no hope to claim further space back */
1007 GOTO(out_write, rc = -EDQUOT);
1009 if (qmt_space_exhausted_lqes(env, now)) {
1010 /* might have some free space once rebalancing is completed */
1011 rc = req_is_acq(qb_flags) ? -EINPROGRESS : -EDQUOT;
1012 GOTO(out_write, rc);
1015 if (req_is_preacq(qb_flags)) {
1016 /* slave would like to pre-acquire quota space. To do so, it
1017 * reports in qb_count how much spare quota space it owns and we
1018 * can grant back quota space which is consistent with qunit
1020 if (qb_count >= qti_lqes_min_qunit(env))
1021 /* slave already own the maximum it should */
1022 GOTO(out_write, rc = 0);
1024 count = qmt_lqes_alloc_expand(env, slv_granted, qb_count);
1026 GOTO(out_write, rc = -EDQUOT);
1028 repbody->qb_count += count;
1029 qmt_grant_lqes(env, &slv_granted, count);
1030 GOTO(out_write, rc = 0);
1033 /* processing acquire request with clients waiting */
1034 if (qmt_lqes_cannot_grant(env, qb_count)) {
1035 /* cannot grant as much as asked, but can still afford to grant
1036 * some quota space back */
1037 count = qmt_lqes_grant_some_quota(env);
1038 repbody->qb_count += count;
1039 qmt_grant_lqes(env, &slv_granted, count);
1040 GOTO(out_write, rc = 0);
1043 /* Whouhou! we can satisfy the slave request! */
1044 repbody->qb_count += qb_count;
1045 qmt_grant_lqes(env, &slv_granted, qb_count);
1047 /* Try to expand the acquired count for DQACQ */
1048 count = qmt_lqes_alloc_expand(env, slv_granted, 0);
1050 /* can even grant more than asked, it is like xmas ... */
1051 repbody->qb_count += count;
1052 qmt_grant_lqes(env, &slv_granted, count);
1053 GOTO(out_write, rc = 0);
1056 GOTO(out_write, rc = 0);
1058 if (repbody->qb_count == 0)
1059 GOTO(out_locked, rc);
1061 /* start/stop grace timer if required */
1062 qmt_lqes_tune_grace(env, now);
1064 if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_GRANT))
1065 slv_granted = 0xFFFFFFFFFFF00000;
1067 /* Update slave index first since it is easier to roll back */
1068 ret = qmt_slv_write(env, th, lqe, slv_obj, LQUOTA_BUMP_VER,
1069 &repbody->qb_slv_ver, slv_granted);
1071 /* restore initial quota settings */
1072 qmt_restore_lqes(env);
1073 /* reset qb_count */
1074 repbody->qb_count = 0;
1075 GOTO(out_locked, rc = ret);
1078 /* Update global index, no version bump needed */
1079 ret = qmt_glb_write_lqes(env, th, 0, NULL);
1082 /* restore initial quota settings */
1083 qmt_restore_lqes(env);
1084 /* reset qb_count */
1085 repbody->qb_count = 0;
1087 /* restore previous granted value */
1088 ret = qmt_slv_write(env, th, lqe, slv_obj, 0, NULL,
1091 LQUOTA_ERROR(lqe, "failed to restore initial slave "
1092 "value rc:%d ret%d", rc, ret);
1095 qmt_adjust_edquot_notify(env, qmt, now, qb_flags);
1096 GOTO(out_locked, rc);
1099 /* Total granted has been changed, let's try to adjust the qunit
1100 * size according to the total granted & limits. */
1102 /* clear/set edquot flag and notify slaves via glimpse if needed */
1103 qmt_adjust_notify_nu(env, qmt, now, qb_flags, idx);
1105 LQUOTA_DEBUG_LQES(env, "dqacq ends count:%llu ver:%llu rc:%d",
1106 repbody->qb_count, repbody->qb_slv_ver, rc);
1107 qti_lqes_write_unlock(env);
1109 qti_lqes_restore_fini(env);
1111 if (th != NULL && !IS_ERR(th))
1112 dt_trans_stop(env, qmt->qmt_child, th);
1114 if (slv_obj != NULL && !IS_ERR(slv_obj))
1115 dt_object_put(env, slv_obj);
1117 if ((req_is_acq(qb_flags) || req_is_preacq(qb_flags)) &&
1118 CFS_FAIL_CHECK(OBD_FAIL_QUOTA_EDQUOT)) {
1119 /* introduce inconsistency between granted value in slave index
1120 * and slave index copy of slave */
1121 repbody->qb_count = 0;
1129 * Extract index from uuid or quota index file name.
1131 * \param[in] uuid uuid or quota index name(0x1020000-OST0001_UUID)
1132 * \param[out] idx pointer to save index
1134 * \retval slave type(QMT_STYPE_MDT or QMT_STYPE_OST)
1135 * \retval -EINVAL wrong uuid
1137 int qmt_uuid2idx(struct obd_uuid *uuid, int *idx)
1139 char *uuid_str, *name, *dash;
1142 uuid_str = (char *)uuid->uuid;
1144 if (strnlen(uuid_str, UUID_MAX) >= UUID_MAX) {
1145 CERROR("quota: UUID '%.*s' missing trailing NUL: rc = %d\n",
1146 UUID_MAX, uuid_str, rc);
1150 dash = strrchr(uuid_str, '-');
1152 /* Going to get index from MDTXXXX/OSTXXXX. Thus uuid should
1153 * have at least 8 bytes after '-': 3 for MDT/OST, 4 for index
1154 * and 1 byte for null character. */
1155 if (*dash != '-' || ((uuid_str + UUID_MAX - name) < 8)) {
1156 CERROR("quota: wrong UUID format '%s': rc = %d\n",
1161 rc = target_name2index(name, idx, NULL);
1163 case LDD_F_SV_TYPE_MDT:
1166 case LDD_F_SV_TYPE_OST:
1170 CERROR("quota: wrong UUID type '%s': rc = %d\n", uuid_str, rc);
1178 * Handle quota request from slave.
1180 * \param env - is the environment passed by the caller
1181 * \param ld - is the lu device associated with the qmt
1182 * \param req - is the quota acquire request
1184 static int qmt_dqacq(const struct lu_env *env, struct lu_device *ld,
1185 struct ptlrpc_request *req)
1187 struct qmt_device *qmt = lu2qmt_dev(ld);
1188 struct quota_body *qbody, *repbody;
1189 struct obd_uuid *uuid;
1190 struct ldlm_lock *lock;
1195 qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
1197 RETURN(err_serious(-EPROTO));
1199 repbody = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_BODY);
1200 if (repbody == NULL)
1201 RETURN(err_serious(-EFAULT));
1203 /* verify if global lock is stale */
1204 if (!lustre_handle_is_used(&qbody->qb_glb_lockh))
1207 lock = ldlm_handle2lock(&qbody->qb_glb_lockh);
1210 LDLM_LOCK_PUT(lock);
1212 uuid = &req->rq_export->exp_client_uuid;
1213 stype = qmt_uuid2idx(uuid, &idx);
1217 if (req_is_rel(qbody->qb_flags) + req_is_acq(qbody->qb_flags) +
1218 req_is_preacq(qbody->qb_flags) > 1) {
1219 CERROR("%s: malformed quota request with conflicting flags set "
1220 "(%x) from slave %s\n", qmt->qmt_svname,
1221 qbody->qb_flags, obd_uuid2str(uuid));
1225 if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
1226 /* acquire and pre-acquire should use a valid ID lock */
1228 if (!lustre_handle_is_used(&qbody->qb_lockh))
1231 lock = ldlm_handle2lock(&qbody->qb_lockh);
1233 /* no lock associated with this handle */
1236 LDLM_DEBUG(lock, "%sacquire request",
1237 req_is_preacq(qbody->qb_flags) ? "pre" : "");
1239 if (!obd_uuid_equals(&lock->l_export->exp_client_uuid, uuid)) {
1240 /* sorry, no way to cheat ... */
1241 LDLM_LOCK_PUT(lock);
1245 if (ldlm_is_ast_sent(lock)) {
1246 struct ptlrpc_service_part *svc;
1249 svc = req->rq_rqbd->rqbd_svcpt;
1250 timeout = at_est2timeout(at_get(&svc->scp_at_estimate));
1251 timeout += (ldlm_bl_timeout(lock) >> 1);
1253 /* lock is being cancelled, prolong timeout */
1254 ldlm_refresh_waiting_lock(lock, timeout);
1256 LDLM_LOCK_PUT(lock);
1259 /* extract quota information from global index FID packed in the
1261 rc = lquota_extract_fid(&qbody->qb_fid, &rtype, &qtype);
1265 /* Find the quota entry associated with the quota id */
1266 rc = qmt_pool_lqes_lookup(env, qmt, rtype, stype, qtype,
1267 &qbody->qb_id, NULL, idx);
1271 rc = qmt_dqacq0(env, qmt, uuid, qbody->qb_flags,
1272 qbody->qb_count, qbody->qb_usage, repbody,
1273 qmt_dom(rtype, stype) ? -1 : idx);
1275 if (lustre_handle_is_used(&qbody->qb_lockh))
1276 /* return current qunit value only to slaves owning an per-ID
1277 * quota lock. For enqueue, the qunit value will be returned in
1279 repbody->qb_qunit = qti_lqes_min_qunit(env);
1280 CDEBUG(D_QUOTA, "qmt_dqacq return qb_qunit %llu qb_count %llu\n",
1281 repbody->qb_qunit, repbody->qb_count);
1286 /* Vector of quota request handlers. This vector is used by the MDT to forward
1287 * requests to the quota master. */
1288 struct qmt_handlers qmt_hdls = {
1289 /* quota request handlers */
1290 .qmth_quotactl = qmt_quotactl,
1291 .qmth_dqacq = qmt_dqacq,
1294 .qmth_intent_policy = qmt_intent_policy,
1295 .qmth_lvbo_init = qmt_lvbo_init,
1296 .qmth_lvbo_update = qmt_lvbo_update,
1297 .qmth_lvbo_size = qmt_lvbo_size,
1298 .qmth_lvbo_fill = qmt_lvbo_fill,
1299 .qmth_lvbo_free = qmt_lvbo_free,
1301 EXPORT_SYMBOL(qmt_hdls);