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]);
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 qmt_glb_lock_notify(env, lqe, ver);
287 if (lqe->lqe_id.qid_uid == 0) {
288 struct qmt_entry_iter_data iter_data;
290 LQUOTA_DEBUG(lqe, "notify all lqe with default quota");
291 iter_data.qeid_env = env;
292 iter_data.qeid_qmt = qmt;
293 cfs_hash_for_each(lqe->lqe_site->lqs_hash,
294 qmt_entry_iter_cb, &iter_data);
295 /* Always notify slaves with default values. Don't
296 * care about overhead as will be sent only not changed
297 * values(see qmt_id_lock_cb for details).*/
298 need_id_notify = true;
300 /* qti_lqes_inited > 0 means we came here from another
301 * qmt_pool_lqes_lookup(qmt_dqacq, intent_policy ...). Thus
302 * we can't init and add new lqes to don't overwrite already
305 if (!qti_lqes_inited(env) && need_id_notify)
306 qmt_set_id_notify(env, qmt, lqe);
313 * Update quota settings for a given identifier.
315 * \param env - is the environment passed by the caller
316 * \param qmt - is the quota master target
317 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or
318 * inode (i.e. LQUOTA_RES_MD)
319 * \param qtype - is the quota type
320 * \param id - is the quota indentifier for which we want to modify
322 * \param hard - is the new hard limit
323 * \param soft - is the new soft limit
324 * \param time - is the new grace time
325 * \param valid - is the list of settings to change
326 * \param is_default - true for default quota setting
327 * \param is_updated - true if the lqe is updated and no need to write back
329 static int qmt_set(const struct lu_env *env, struct qmt_device *qmt,
330 __u8 restype, __u8 qtype, union lquota_id *id,
331 __u64 hard, __u64 soft, __u64 time, __u32 valid,
332 bool is_default, bool is_updated, char *pool_name)
334 struct lquota_entry *lqe;
338 if (pool_name && !strnlen(pool_name, LOV_MAXPOOLNAME))
341 /* look-up quota entry associated with this ID */
342 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, id, pool_name);
344 RETURN(PTR_ERR(lqe));
346 lqe->lqe_is_deleted = 0;
347 lqe->lqe_is_reset = 0;
348 rc = qmt_set_with_lqe(env, qmt, lqe, hard, soft, time, valid,
349 is_default, is_updated);
351 lqe->lqe_is_deleted = 0;
358 * Delete the quota setting of the specified quota ID
360 * \param env - is the environment passed by the caller
361 * \param qmt - is the quota master target
362 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or
363 * inode (i.e. LQUOTA_RES_MD)
364 * \param qtype - is the quota type
365 * \param qid - is the quota indentifier for which we want to delete its
368 static int qmt_delete_qid(const struct lu_env *env, struct qmt_device *qmt,
369 __u8 restype, __u8 qtype, __u64 qid)
371 struct qmt_thread_info *qti = qmt_info(env);
372 union lquota_id *quota_id = &qti->qti_id;
373 struct thandle *th = NULL;
374 struct qmt_pool_info *qpi = NULL;
375 struct lquota_entry *lqe = NULL;
381 quota_id->qid_uid = qid;
382 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, quota_id, NULL);
384 RETURN(PTR_ERR(lqe));
388 qpi = qmt_pool_lookup_glb(env, qmt, restype);
390 GOTO(out, rc = -ENOMEM);
392 th = qmt_trans_start(env, lqe);
394 GOTO(out, rc = PTR_ERR(th));
396 rc = lquota_disk_delete(env, th,
397 qpi->qpi_glb_obj[qtype], qid, &ver);
399 dt_trans_stop(env, qmt->qmt_child, th);
402 lqe_set_deleted(lqe);
403 qmt_glb_lock_notify(env, lqe, ver);
404 } else if (rc == -ENOENT) {
409 if (!IS_ERR_OR_NULL(qpi))
410 qpi_putref(env, qpi);
412 lqe_write_unlock(lqe);
418 static int qmt_reset_slv_cb(const struct lu_env *env, struct lu_fid *glb_fid,
419 char *slv_name, struct lu_fid *slv_fid, void *arg)
421 struct qmt_device *qmt = (struct qmt_device *)arg;
422 struct qmt_thread_info *qti = qmt_info(env);
423 struct dt_object *slv_obj = NULL;
424 struct lquota_slv_rec rec;
425 struct thandle *th = NULL;
428 slv_obj = dt_locate(env, qmt->qmt_child, slv_fid);
430 GOTO(out, rc = PTR_ERR(slv_obj));
432 if (slv_obj->do_index_ops == NULL) {
433 rc = slv_obj->do_ops->do_index_try(env, slv_obj,
434 &dt_quota_slv_features);
436 CERROR("%s: fail to setup slave idx for %s: rc = %d\n",
437 qmt->qmt_child->dd_lu_dev.ld_obd->obd_name,
443 th = qmt_trans_start(env, qti_lqes(env)[0]);
445 GOTO(out, rc = PTR_ERR(th));
448 rc = lquota_disk_write(env, th, slv_obj, &qti->qti_id,
449 (struct dt_rec *)&rec, 0, NULL);
451 CERROR("%s: failed to reset slave grant for %s: rc = %d\n",
452 qmt->qmt_child->dd_lu_dev.ld_obd->obd_name, slv_name,
455 if (!IS_ERR_OR_NULL(th))
456 dt_trans_stop(env, qmt->qmt_child, th);
459 dt_object_put(env, slv_obj);
464 * Reset the quota of the quota ID, it will reset the soft/hard limit and grant
466 * \param env - is the environment passed by the caller
467 * \param qmt - is the quota master target
468 * \param restype - is the pool type, either block (i.e. LQUOTA_RES_DT) or
469 * inode (i.e. LQUOTA_RES_MD)
470 * \param qtype - is the quota type
471 * \param qid - is the quota indentifier for which we want to delete its
474 static int qmt_reset_qid(const struct lu_env *env, struct qmt_device *qmt,
475 __u8 restype, __u8 qtype, __u64 qid)
477 struct qmt_thread_info *qti = qmt_info(env);
478 union lquota_id *quota_id = &qti->qti_id;
479 struct qmt_pool_info *qpi = NULL;
480 struct lquota_entry *lqe = NULL;
481 struct thandle *th = NULL;
482 __u64 softlimit = 0, hardlimit = 0;
488 quota_id->qid_uid = qid;
489 lqe = qmt_pool_lqe_lookup(env, qmt, restype, qtype, quota_id, NULL);
491 RETURN(PTR_ERR(lqe));
495 qpi = qmt_pool_lookup_glb(env, qmt, restype);
497 GOTO(out, rc = -ENOMEM);
499 th = qmt_trans_start(env, lqe);
501 GOTO(out, rc = PTR_ERR(th));
503 softlimit = lqe->lqe_softlimit;
504 hardlimit = lqe->lqe_hardlimit;
506 lqe->lqe_softlimit = 0;
507 lqe->lqe_hardlimit = 0;
508 lqe->lqe_granted = 0;
511 lqe->lqe_is_default = 0;
512 lqe->lqe_is_deleted = 0;
513 lqe->lqe_is_reset = 1;
514 rc = qmt_glb_write(env, th, lqe, LQUOTA_BUMP_VER, &ver);
516 LQUOTA_ERROR(lqe, "failed to write quota global rec\n");
517 dt_trans_stop(env, qmt->qmt_child, th);
521 lquota_generate_fid(&qti->qti_fid, restype, qtype);
522 qti_lqes(env)[0] = lqe;
523 lquota_disk_for_each_slv(env, qpi->qpi_root, &qti->qti_fid,
524 qmt_reset_slv_cb, qmt);
526 qmt_glb_lock_notify(env, lqe, ver);
531 lqe->lqe_softlimit = softlimit;
533 lqe->lqe_hardlimit = hardlimit;
534 lqe->lqe_is_reset = 0;
537 if (!IS_ERR_OR_NULL(qpi))
538 qpi_putref(env, qpi);
540 lqe_write_unlock(lqe);
546 * Handle quotactl request.
548 * \param env - is the environment passed by the caller
549 * \param ld - is the lu device associated with the qmt
550 * \param oqctl - is the quotactl request
552 static int qmt_quotactl(const struct lu_env *env, struct lu_device *ld,
553 struct obd_quotactl *oqctl)
555 struct qmt_thread_info *qti = qmt_info(env);
556 union lquota_id *id = &qti->qti_id;
557 struct qmt_device *qmt = lu2qmt_dev(ld);
558 struct obd_dqblk *dqb = &oqctl->qc_dqblk;
561 bool is_default = false;
564 LASSERT(qmt != NULL);
566 if (oqctl->qc_type >= LL_MAXQUOTAS)
567 /* invalid quota type */
570 poolname = LUSTRE_Q_CMD_IS_POOL(oqctl->qc_cmd) ?
571 oqctl->qc_poolname : NULL;
573 switch (oqctl->qc_cmd) {
575 case Q_GETINFO: /* read grace times */
576 case LUSTRE_Q_GETINFOPOOL:
577 /* Global grace time is stored in quota settings of ID 0. */
580 /* read inode grace time */
581 rc = qmt_get(env, qmt, LQUOTA_RES_MD, oqctl->qc_type, id, NULL,
582 NULL, &oqctl->qc_dqinfo.dqi_igrace,
584 /* There could be no MD pool, so try to find DT pool */
585 if (rc && rc != -ENOENT)
588 /* read block grace time */
589 rc = qmt_get(env, qmt, LQUOTA_RES_DT, oqctl->qc_type, id, NULL,
590 NULL, &oqctl->qc_dqinfo.dqi_bgrace,
594 case Q_SETINFO: /* modify grace times */
595 case LUSTRE_Q_SETINFOPOOL:
596 /* setinfo should be using dqi->dqi_valid, but lfs incorrectly
597 * sets the valid flags in dqb->dqb_valid instead, try to live
600 /* Global grace time is stored in quota settings of ID 0. */
603 if ((dqb->dqb_valid & QIF_ITIME) != 0) {
604 /* set inode grace time */
605 rc = qmt_set(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
606 id, 0, 0, oqctl->qc_dqinfo.dqi_igrace,
607 QIF_TIMES, false, false,
613 if ((dqb->dqb_valid & QIF_BTIME) != 0)
614 /* set block grace time */
615 rc = qmt_set(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
616 id, 0, 0, oqctl->qc_dqinfo.dqi_bgrace,
617 QIF_TIMES, false, false,
621 case LUSTRE_Q_GETDEFAULT:
622 case LUSTRE_Q_GETDEFAULT_POOL:
626 case Q_GETQUOTA: /* consult quota limit */
627 case LUSTRE_Q_GETQUOTAPOOL:
628 /* extract quota ID from quotactl request */
629 id->qid_uid = oqctl->qc_id;
631 /* look-up inode quota settings */
632 rc = qmt_get(env, qmt, LQUOTA_RES_MD, oqctl->qc_type, id,
633 &dqb->dqb_ihardlimit, &dqb->dqb_isoftlimit,
634 &dqb->dqb_itime, is_default, poolname);
635 /* There could be no MD pool, so try to find DT pool */
636 if (rc && rc != -ENOENT)
639 dqb->dqb_valid |= QIF_ILIMITS | QIF_ITIME;
641 /* master isn't aware of actual inode usage */
642 dqb->dqb_curinodes = 0;
644 /* look-up block quota settings */
645 rc = qmt_get(env, qmt, LQUOTA_RES_DT, oqctl->qc_type, id,
646 &dqb->dqb_bhardlimit, &dqb->dqb_bsoftlimit,
647 &dqb->dqb_btime, is_default, poolname);
651 dqb->dqb_valid |= QIF_BLIMITS | QIF_BTIME;
652 /* master doesn't know the actual block usage */
653 dqb->dqb_curspace = 0;
656 case LUSTRE_Q_SETDEFAULT:
657 case LUSTRE_Q_SETDEFAULT_POOL:
661 case Q_SETQUOTA: /* change quota limits */
662 case LUSTRE_Q_SETQUOTAPOOL:
663 /* extract quota ID from quotactl request */
664 id->qid_uid = oqctl->qc_id;
666 if ((dqb->dqb_valid & QIF_IFLAGS) != 0) {
667 /* update inode quota settings */
668 rc = qmt_set(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
669 id, dqb->dqb_ihardlimit,
670 dqb->dqb_isoftlimit, dqb->dqb_itime,
671 dqb->dqb_valid & QIF_IFLAGS, is_default,
677 if ((dqb->dqb_valid & QIF_BFLAGS) != 0)
678 /* update block quota settings */
679 rc = qmt_set(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
680 id, dqb->dqb_bhardlimit,
681 dqb->dqb_bsoftlimit, dqb->dqb_btime,
682 dqb->dqb_valid & QIF_BFLAGS, is_default,
686 case LUSTRE_Q_DELETEQID:
687 rc = qmt_delete_qid(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
692 rc = qmt_delete_qid(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
696 case LUSTRE_Q_RESETQID:
697 if (oqctl->qc_id == 0)
700 id->qid_uid = oqctl->qc_id;
701 /* save the quota setting before resetting */
702 rc = qmt_get(env, qmt, LQUOTA_RES_MD, oqctl->qc_type, id,
703 &dqb->dqb_ihardlimit, &dqb->dqb_isoftlimit,
704 &dqb->dqb_itime, false, NULL);
708 dqb->dqb_valid |= QIF_ILIMITS | QIF_ITIME;
710 rc = qmt_get(env, qmt, LQUOTA_RES_DT, oqctl->qc_type, id,
711 &dqb->dqb_bhardlimit, &dqb->dqb_bsoftlimit,
712 &dqb->dqb_btime, false, NULL);
716 dqb->dqb_valid |= QIF_BLIMITS | QIF_BTIME;
717 dqb->dqb_curinodes = 0;
718 dqb->dqb_curspace = 0;
720 /* reset the corresponding quota ID */
721 rc = qmt_reset_qid(env, qmt, LQUOTA_RES_MD, oqctl->qc_type,
726 rc = qmt_reset_qid(env, qmt, LQUOTA_RES_DT, oqctl->qc_type,
732 CERROR("%s: unsupported quotactl command: %d\n",
733 qmt->qmt_svname, oqctl->qc_cmd);
741 void qmt_grant_lqes(const struct lu_env *env, __u64 *slv, __u64 cnt)
745 for (i = 0; i < qti_lqes_cnt(env); i++)
746 qti_lqe_granted(env, i) += cnt;
751 static inline bool qmt_lqes_can_rel(const struct lu_env *env, __u64 cnt)
753 bool can_release = true;
756 for (i = 0; i < qti_lqes_cnt(env); i++) {
757 if (cnt > qti_lqe_granted(env, i)) {
758 LQUOTA_ERROR(qti_lqes(env)[i],
759 "Can't release %llu that is larger than lqe_granted.\n",
767 static inline void qmt_rel_lqes(const struct lu_env *env, __u64 *slv, __u64 cnt,
772 for (i = 0; i < qti_lqes_cnt(env); i++) {
774 qti_lqe_granted(env, i) = 0;
776 qti_lqe_granted(env, i) -= cnt;
785 static inline bool qmt_lqes_cannot_grant(const struct lu_env *env, __u64 cnt)
787 bool cannot_grant = false;
790 for (i = 0; i < qti_lqes_cnt(env); i++) {
791 if (qti_lqe_hard(env, i) != 0 &&
792 qti_lqe_granted(env, i) + cnt > qti_lqe_hard(env, i)) {
800 static inline __u64 qmt_lqes_grant_some_quota(const struct lu_env *env)
802 __u64 min_count, tmp;
806 for (i = 0, min_count = 0; i < qti_lqes_cnt(env); i++) {
807 if (!qti_lqes(env)[i]->lqe_enforced &&
808 !qti_lqes(env)[i]->lqe_is_global)
811 tmp = qti_lqe_hard(env, i) - qti_lqe_granted(env, i);
813 min_count = tmp < min_count ? tmp : min_count;
822 static inline __u64 qmt_lqes_alloc_expand(const struct lu_env *env,
823 __u64 slv_granted, __u64 spare)
825 __u64 min_count, tmp;
829 for (i = 0, min_count = 0; i < qti_lqes_cnt(env); i++) {
830 /* Don't take into account not enforced lqes that belong
831 * to non global pool. These lqes present in array to
832 * support actual lqe_granted even for lqes without limits. */
833 if (!qti_lqes(env)[i]->lqe_enforced &&
834 !qti_lqes(env)[i]->lqe_is_global)
837 tmp = qmt_alloc_expand(qti_lqes(env)[i], slv_granted, spare);
839 min_count = tmp < min_count ? tmp : min_count;
848 static inline void qmt_lqes_tune_grace(const struct lu_env *env, __u64 now)
852 for (i = 0; i < qti_lqes_cnt(env); i++) {
853 struct lquota_entry *lqe;
855 lqe = qti_lqes(env)[i];
856 if (lqe->lqe_softlimit != 0) {
857 if (lqe->lqe_granted > lqe->lqe_softlimit &&
858 lqe->lqe_gracetime == 0) {
859 /* First time over soft limit, let's start grace
861 lqe->lqe_gracetime = now + qmt_lqe_grace(lqe);
862 } else if (lqe->lqe_granted <= lqe->lqe_softlimit &&
863 lqe->lqe_gracetime != 0) {
864 /* Clear grace timer */
865 lqe->lqe_gracetime = 0;
872 * Helper function to handle quota request from slave.
874 * \param env - is the environment passed by the caller
875 * \param qmt - is the master device
876 * \param uuid - is the uuid associated with the slave
877 * \param qb_flags - are the quota request flags as packed in the quota_body
878 * \param qb_count - is the amount of quota space the slave wants to
880 * \param qb_usage - is the current space usage on the slave
881 * \param repbody - is the quota_body of reply
883 * \retval 0 : success
884 * \retval -EDQUOT : out of quota
885 * -EINPROGRESS : inform client to retry write/create
886 * -ve : other appropriate errors
888 int qmt_dqacq0(const struct lu_env *env, struct qmt_device *qmt,
889 struct obd_uuid *uuid, __u32 qb_flags, __u64 qb_count,
890 __u64 qb_usage, struct quota_body *repbody)
892 __u64 now, count = 0;
893 struct dt_object *slv_obj = NULL;
894 __u64 slv_granted, slv_granted_bck;
895 struct thandle *th = NULL;
897 struct lquota_entry *lqe = qti_lqes_glbl(env);
900 LASSERT(uuid != NULL);
902 /* initialize reply */
903 memset(repbody, 0, sizeof(*repbody));
904 memcpy(&repbody->qb_id, &lqe->lqe_id, sizeof(repbody->qb_id));
906 if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_RECOVERABLE_ERR))
907 RETURN(-cfs_fail_val);
909 if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_PREACQ) &&
910 (req_is_preacq(qb_flags) || req_is_rel(qb_flags)))
913 if (qti_lqes_restore_init(env))
916 /* look-up index file associated with acquiring slave */
917 slv_obj = lquota_disk_slv_find(env, qmt->qmt_child, LQE_ROOT(lqe),
918 lu_object_fid(&LQE_GLB_OBJ(lqe)->do_lu),
921 GOTO(out, rc = PTR_ERR(slv_obj));
923 /* pack slave fid in reply just for sanity check */
924 memcpy(&repbody->qb_slv_fid, lu_object_fid(&slv_obj->do_lu),
925 sizeof(struct lu_fid));
927 /* allocate & start transaction with enough credits to update
928 * global & slave indexes */
929 th = qmt_trans_start_with_slv(env, NULL, slv_obj, false);
931 GOTO(out, rc = PTR_ERR(th));
933 qti_lqes_write_lock(env);
935 LQUOTA_DEBUG_LQES(env, "dqacq starts uuid:%s flags:0x%x wanted:%llu"
936 " usage:%llu", obd_uuid2str(uuid), qb_flags, qb_count,
939 /* Legal race, limits have been removed on master, but slave didn't
940 * receive the change yet. Just return EINPROGRESS until the slave gets
942 if (!lqe->lqe_enforced && !req_is_rel(qb_flags))
943 GOTO(out_locked, rc = -ESRCH);
945 /* recompute qunit in case it was never initialized */
946 qmt_revalidate_lqes(env, qmt, qb_flags);
948 /* slave just wants to acquire per-ID lock */
949 if (req_is_acq(qb_flags) && qb_count == 0)
950 GOTO(out_locked, rc = 0);
952 if (lqe->lqe_is_reset) {
953 lqe->lqe_granted = 0;
954 repbody->qb_count = qb_count;
955 qmt_rel_lqes(env, &slv_granted, qb_count, lqe->lqe_is_reset);
956 GOTO(out_locked, rc = 0);
959 /* fetch how much quota space is already granted to this slave */
960 rc = qmt_slv_read(env, &lqe->lqe_id, slv_obj, &slv_granted);
962 LQUOTA_ERROR(lqe, "Failed to get granted for slave %s, rc=%d",
963 obd_uuid2str(uuid), rc);
964 GOTO(out_locked, rc);
966 /* recall how much space this slave currently owns in order to restore
967 * it in case of failure */
968 slv_granted_bck = slv_granted;
970 /* record current time for soft limit & grace time management */
971 now = ktime_get_real_seconds();
973 if (req_is_rel(qb_flags)) {
974 /* Slave would like to release quota space */
975 if (slv_granted < qb_count ||
976 !qmt_lqes_can_rel(env, qb_count)) {
977 /* can't release more than granted */
978 LQUOTA_ERROR_LQES(env,
979 "Release too much! uuid:%s release: %llu granted:%llu, total:%llu",
980 obd_uuid2str(uuid), qb_count,
981 slv_granted, lqe->lqe_granted);
982 GOTO(out_locked, rc = -EINVAL);
985 repbody->qb_count = qb_count;
986 /* put released space back to global pool */
987 qmt_rel_lqes(env, &slv_granted, qb_count, lqe->lqe_is_reset);
988 GOTO(out_write, rc = 0);
991 if (req_has_rep(qb_flags) && slv_granted < qb_usage) {
992 /* Slave is reporting space usage in quota request and it turns
993 * out to be using more quota space than owned, so we adjust
994 * granted space regardless of the current state of affairs */
995 repbody->qb_count = qb_usage - slv_granted;
996 qmt_grant_lqes(env, &slv_granted, repbody->qb_count);
999 if (!req_is_acq(qb_flags) && !req_is_preacq(qb_flags))
1000 GOTO(out_write, rc = 0);
1002 qmt_adjust_edquot_notify(env, qmt, now, qb_flags);
1003 if (qti_lqes_edquot(env))
1004 /* no hope to claim further space back */
1005 GOTO(out_write, rc = -EDQUOT);
1007 if (qmt_space_exhausted_lqes(env, now)) {
1008 /* might have some free space once rebalancing is completed */
1009 rc = req_is_acq(qb_flags) ? -EINPROGRESS : -EDQUOT;
1010 GOTO(out_write, rc);
1013 if (req_is_preacq(qb_flags)) {
1014 /* slave would like to pre-acquire quota space. To do so, it
1015 * reports in qb_count how much spare quota space it owns and we
1016 * can grant back quota space which is consistent with qunit
1018 if (qb_count >= qti_lqes_min_qunit(env))
1019 /* slave already own the maximum it should */
1020 GOTO(out_write, rc = 0);
1022 count = qmt_lqes_alloc_expand(env, slv_granted, qb_count);
1024 GOTO(out_write, rc = -EDQUOT);
1026 repbody->qb_count += count;
1027 qmt_grant_lqes(env, &slv_granted, count);
1028 GOTO(out_write, rc = 0);
1031 /* processing acquire request with clients waiting */
1032 if (qmt_lqes_cannot_grant(env, qb_count)) {
1033 /* cannot grant as much as asked, but can still afford to grant
1034 * some quota space back */
1035 count = qmt_lqes_grant_some_quota(env);
1036 repbody->qb_count += count;
1037 qmt_grant_lqes(env, &slv_granted, count);
1038 GOTO(out_write, rc = 0);
1041 /* Whouhou! we can satisfy the slave request! */
1042 repbody->qb_count += qb_count;
1043 qmt_grant_lqes(env, &slv_granted, qb_count);
1045 /* Try to expand the acquired count for DQACQ */
1046 count = qmt_lqes_alloc_expand(env, slv_granted, 0);
1048 /* can even grant more than asked, it is like xmas ... */
1049 repbody->qb_count += count;
1050 qmt_grant_lqes(env, &slv_granted, count);
1051 GOTO(out_write, rc = 0);
1054 GOTO(out_write, rc = 0);
1056 if (repbody->qb_count == 0)
1057 GOTO(out_locked, rc);
1059 /* start/stop grace timer if required */
1060 qmt_lqes_tune_grace(env, now);
1062 if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_GRANT))
1063 slv_granted = 0xFFFFFFFFFFF00000;
1065 /* Update slave index first since it is easier to roll back */
1066 ret = qmt_slv_write(env, th, lqe, slv_obj, LQUOTA_BUMP_VER,
1067 &repbody->qb_slv_ver, slv_granted);
1069 /* restore initial quota settings */
1070 qmt_restore_lqes(env);
1071 /* reset qb_count */
1072 repbody->qb_count = 0;
1073 GOTO(out_locked, rc = ret);
1076 /* Update global index, no version bump needed */
1077 ret = qmt_glb_write_lqes(env, th, 0, NULL);
1080 /* restore initial quota settings */
1081 qmt_restore_lqes(env);
1082 /* reset qb_count */
1083 repbody->qb_count = 0;
1085 /* restore previous granted value */
1086 ret = qmt_slv_write(env, th, lqe, slv_obj, 0, NULL,
1089 LQUOTA_ERROR(lqe, "failed to restore initial slave "
1090 "value rc:%d ret%d", rc, ret);
1093 qmt_adjust_edquot_notify(env, qmt, now, qb_flags);
1094 GOTO(out_locked, rc);
1097 /* Total granted has been changed, let's try to adjust the qunit
1098 * size according to the total granted & limits. */
1100 /* clear/set edquot flag and notify slaves via glimpse if needed */
1101 qmt_adjust_and_notify(env, qmt, now, qb_flags);
1103 LQUOTA_DEBUG_LQES(env, "dqacq ends count:%llu ver:%llu rc:%d",
1104 repbody->qb_count, repbody->qb_slv_ver, rc);
1105 qti_lqes_write_unlock(env);
1107 qti_lqes_restore_fini(env);
1109 if (th != NULL && !IS_ERR(th))
1110 dt_trans_stop(env, qmt->qmt_child, th);
1112 if (slv_obj != NULL && !IS_ERR(slv_obj))
1113 dt_object_put(env, slv_obj);
1115 if ((req_is_acq(qb_flags) || req_is_preacq(qb_flags)) &&
1116 OBD_FAIL_CHECK(OBD_FAIL_QUOTA_EDQUOT)) {
1117 /* introduce inconsistency between granted value in slave index
1118 * and slave index copy of slave */
1119 repbody->qb_count = 0;
1127 * Extract index from uuid or quota index file name.
1129 * \param[in] uuid uuid or quota index name(0x1020000-OST0001_UUID)
1130 * \param[out] idx pointer to save index
1132 * \retval slave type(QMT_STYPE_MDT or QMT_STYPE_OST)
1133 * \retval -EINVAL wrong uuid
1135 int qmt_uuid2idx(struct obd_uuid *uuid, int *idx)
1137 char *uuid_str, *name, *dash;
1140 uuid_str = (char *)uuid->uuid;
1142 if (strnlen(uuid_str, UUID_MAX) >= UUID_MAX) {
1143 CERROR("quota: UUID '%.*s' missing trailing NUL: rc = %d\n",
1144 UUID_MAX, uuid_str, rc);
1148 dash = strrchr(uuid_str, '-');
1150 /* Going to get index from MDTXXXX/OSTXXXX. Thus uuid should
1151 * have at least 8 bytes after '-': 3 for MDT/OST, 4 for index
1152 * and 1 byte for null character. */
1153 if (*dash != '-' || ((uuid_str + UUID_MAX - name) < 8)) {
1154 CERROR("quota: wrong UUID format '%s': rc = %d\n",
1159 rc = target_name2index(name, idx, NULL);
1161 case LDD_F_SV_TYPE_MDT:
1164 case LDD_F_SV_TYPE_OST:
1168 CERROR("quota: wrong UUID type '%s': rc = %d\n", uuid_str, rc);
1176 * Handle quota request from slave.
1178 * \param env - is the environment passed by the caller
1179 * \param ld - is the lu device associated with the qmt
1180 * \param req - is the quota acquire request
1182 static int qmt_dqacq(const struct lu_env *env, struct lu_device *ld,
1183 struct ptlrpc_request *req)
1185 struct qmt_device *qmt = lu2qmt_dev(ld);
1186 struct quota_body *qbody, *repbody;
1187 struct obd_uuid *uuid;
1188 struct ldlm_lock *lock;
1193 qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
1195 RETURN(err_serious(-EPROTO));
1197 repbody = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_BODY);
1198 if (repbody == NULL)
1199 RETURN(err_serious(-EFAULT));
1201 /* verify if global lock is stale */
1202 if (!lustre_handle_is_used(&qbody->qb_glb_lockh))
1205 lock = ldlm_handle2lock(&qbody->qb_glb_lockh);
1208 LDLM_LOCK_PUT(lock);
1210 uuid = &req->rq_export->exp_client_uuid;
1211 stype = qmt_uuid2idx(uuid, &idx);
1215 if (req_is_rel(qbody->qb_flags) + req_is_acq(qbody->qb_flags) +
1216 req_is_preacq(qbody->qb_flags) > 1) {
1217 CERROR("%s: malformed quota request with conflicting flags set "
1218 "(%x) from slave %s\n", qmt->qmt_svname,
1219 qbody->qb_flags, obd_uuid2str(uuid));
1223 if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
1224 /* acquire and pre-acquire should use a valid ID lock */
1226 if (!lustre_handle_is_used(&qbody->qb_lockh))
1229 lock = ldlm_handle2lock(&qbody->qb_lockh);
1231 /* no lock associated with this handle */
1234 LDLM_DEBUG(lock, "%sacquire request",
1235 req_is_preacq(qbody->qb_flags) ? "pre" : "");
1237 if (!obd_uuid_equals(&lock->l_export->exp_client_uuid, uuid)) {
1238 /* sorry, no way to cheat ... */
1239 LDLM_LOCK_PUT(lock);
1243 if (ldlm_is_ast_sent(lock)) {
1244 struct ptlrpc_service_part *svc;
1247 svc = req->rq_rqbd->rqbd_svcpt;
1248 timeout = at_est2timeout(at_get(&svc->scp_at_estimate));
1249 timeout += (ldlm_bl_timeout(lock) >> 1);
1251 /* lock is being cancelled, prolong timeout */
1252 ldlm_refresh_waiting_lock(lock, timeout);
1254 LDLM_LOCK_PUT(lock);
1257 /* extract quota information from global index FID packed in the
1259 rc = lquota_extract_fid(&qbody->qb_fid, &rtype, &qtype);
1263 /* Find the quota entry associated with the quota id */
1264 rc = qmt_pool_lqes_lookup(env, qmt, rtype, stype, qtype,
1265 &qbody->qb_id, NULL, idx);
1269 rc = qmt_dqacq0(env, qmt, uuid, qbody->qb_flags,
1270 qbody->qb_count, qbody->qb_usage, repbody);
1272 if (lustre_handle_is_used(&qbody->qb_lockh))
1273 /* return current qunit value only to slaves owning an per-ID
1274 * quota lock. For enqueue, the qunit value will be returned in
1276 repbody->qb_qunit = qti_lqes_min_qunit(env);
1277 CDEBUG(D_QUOTA, "qmt_dqacq return qb_qunit %llu qb_count %llu\n",
1278 repbody->qb_qunit, repbody->qb_count);
1283 /* Vector of quota request handlers. This vector is used by the MDT to forward
1284 * requests to the quota master. */
1285 struct qmt_handlers qmt_hdls = {
1286 /* quota request handlers */
1287 .qmth_quotactl = qmt_quotactl,
1288 .qmth_dqacq = qmt_dqacq,
1291 .qmth_intent_policy = qmt_intent_policy,
1292 .qmth_lvbo_init = qmt_lvbo_init,
1293 .qmth_lvbo_update = qmt_lvbo_update,
1294 .qmth_lvbo_size = qmt_lvbo_size,
1295 .qmth_lvbo_fill = qmt_lvbo_fill,
1296 .qmth_lvbo_free = qmt_lvbo_free,
1298 EXPORT_SYMBOL(qmt_hdls);