1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/quota/quota_master.c
38 * Lustre Quota Master request handler
40 * Author: Niu YaWei <niu@clusterfs.com>
44 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_LQUOTA
49 #include <linux/version.h>
51 #include <asm/unistd.h>
52 #include <linux/slab.h>
53 #include <linux/quotaops.h>
54 #include <linux/module.h>
55 #include <linux/init.h>
56 #include <linux/quota.h>
58 #include <obd_class.h>
59 #include <lustre_quota.h>
60 #include <lustre_fsfilt.h>
61 #include <lustre_mds.h>
63 #include "quota_internal.h"
65 #ifdef HAVE_QUOTA_SUPPORT
67 /* lock ordering: mds->mds_qonoff_sem > dquot->dq_sem */
68 static cfs_list_t lustre_dquot_hash[NR_DQHASH];
69 static cfs_spinlock_t dquot_hash_lock = CFS_SPIN_LOCK_UNLOCKED;
71 cfs_mem_cache_t *lustre_dquot_cachep;
73 int lustre_dquot_init(void)
78 LASSERT(lustre_dquot_cachep == NULL);
79 lustre_dquot_cachep = cfs_mem_cache_create("lustre_dquot_cache",
80 sizeof(struct lustre_dquot),
82 if (!lustre_dquot_cachep)
85 for (i = 0; i < NR_DQHASH; i++) {
86 CFS_INIT_LIST_HEAD(lustre_dquot_hash + i);
91 void lustre_dquot_exit(void)
95 /* FIXME cleanup work ?? */
97 for (i = 0; i < NR_DQHASH; i++) {
98 LASSERT(cfs_list_empty(lustre_dquot_hash + i));
100 if (lustre_dquot_cachep) {
102 rc = cfs_mem_cache_destroy(lustre_dquot_cachep);
103 LASSERTF(rc == 0,"couldn't destroy lustre_dquot_cachep slab\n");
104 lustre_dquot_cachep = NULL;
110 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
111 __attribute__((__const__));
114 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
116 unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
117 tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
121 /* caller must hold dquot_hash_lock */
122 static struct lustre_dquot *find_dquot(int hashent,
123 struct lustre_quota_info *lqi, qid_t id,
126 struct lustre_dquot *dquot;
129 LASSERT_SPIN_LOCKED(&dquot_hash_lock);
130 cfs_list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
131 if (dquot->dq_info == lqi &&
132 dquot->dq_id == id && dquot->dq_type == type)
138 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
141 struct lustre_dquot *dquot = NULL;
144 OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
148 CFS_INIT_LIST_HEAD(&dquot->dq_hash);
149 cfs_init_mutex_locked(&dquot->dq_sem);
150 dquot->dq_refcnt = 1;
151 dquot->dq_info = lqi;
153 dquot->dq_type = type;
154 dquot->dq_status = DQ_STATUS_AVAIL;
159 static void free_dquot(struct lustre_dquot *dquot)
161 OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
164 static void insert_dquot_nolock(struct lustre_dquot *dquot)
166 cfs_list_t *head = lustre_dquot_hash +
167 dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
168 LASSERT(cfs_list_empty(&dquot->dq_hash));
169 cfs_list_add(&dquot->dq_hash, head);
172 static void remove_dquot_nolock(struct lustre_dquot *dquot)
174 LASSERT(!cfs_list_empty(&dquot->dq_hash));
175 cfs_list_del_init(&dquot->dq_hash);
178 static void lustre_dqput(struct lustre_dquot *dquot)
181 cfs_spin_lock(&dquot_hash_lock);
182 LASSERT(dquot->dq_refcnt);
184 if (!dquot->dq_refcnt) {
185 remove_dquot_nolock(dquot);
188 cfs_spin_unlock(&dquot_hash_lock);
192 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
193 struct lustre_quota_info *lqi,
196 unsigned int hashent = dquot_hashfn(lqi, id, type);
197 struct lustre_dquot *dquot, *empty;
200 if ((empty = alloc_dquot(lqi, id, type)) == NULL)
201 RETURN(ERR_PTR(-ENOMEM));
203 cfs_spin_lock(&dquot_hash_lock);
204 if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
206 cfs_spin_unlock(&dquot_hash_lock);
212 insert_dquot_nolock(dquot);
213 cfs_spin_unlock(&dquot_hash_lock);
215 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
216 cfs_up(&dquot->dq_sem);
218 CERROR("can't read dquot from admin quotafile! "
230 static void init_oqaq(struct quota_adjust_qunit *oqaq,
231 struct lustre_quota_ctxt *qctxt,
234 struct lustre_qunit_size *lqs = NULL;
237 oqaq->qaq_flags = type;
238 lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 0);
239 if (lqs && !IS_ERR(lqs)) {
240 cfs_spin_lock(&lqs->lqs_lock);
241 oqaq->qaq_bunit_sz = lqs->lqs_bunit_sz;
242 oqaq->qaq_iunit_sz = lqs->lqs_iunit_sz;
243 oqaq->qaq_flags = lqs->lqs_flags;
244 cfs_spin_unlock(&lqs->lqs_lock);
247 CDEBUG(D_QUOTA, "Can't find the lustre qunit size!\n");
248 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
249 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
253 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
256 struct mds_obd *mds = &obd->u.mds;
257 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
258 __u32 ost_num = mds->mds_lov_objid_count, mdt_num = 1;
259 struct quota_adjust_qunit *oqaq = NULL;
260 unsigned int qid[MAXQUOTAS] = { 0, 0 };
261 struct lustre_quota_info *info = &mds->mds_quota_info;
262 struct lustre_dquot *dquot = NULL;
268 dquot = lustre_dqget(obd, info, id, type);
270 RETURN(PTR_ERR(dquot));
274 GOTO(out, rc = -ENOMEM);
276 cfs_down(&dquot->dq_sem);
277 init_oqaq(oqaq, qctxt, id, type);
279 rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
280 is_blk ? LQUOTA_FLAGS_ADJBLK :
281 LQUOTA_FLAGS_ADJINO, oqaq);
284 CERROR("create oqaq failed! (rc:%d)\n", rc);
287 QAQ_DEBUG(oqaq, "show oqaq.\n")
289 if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
292 /* adjust the mds slave qunit size */
293 adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
294 if (adjust_res <= 0) {
295 if (adjust_res < 0) {
297 CERROR("adjust mds slave's qunit size failed! "
300 CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
306 qid[GRPQUOTA] = dquot->dq_id;
308 qid[USRQUOTA] = dquot->dq_id;
310 cfs_up(&dquot->dq_sem);
312 rc = qctxt_adjust_qunit(obd, qctxt, qid, is_blk, 0, NULL);
313 if (rc == -EDQUOT || rc == -EBUSY) {
314 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
318 CERROR("%s: mds fail to adjust file quota! (rc:%d)\n",
323 /* only when block qunit is reduced, boardcast to osts */
324 if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
325 rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
334 cfs_up(&dquot->dq_sem);
338 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
340 struct mds_obd *mds = &obd->u.mds;
341 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
342 struct lustre_quota_info *info = &mds->mds_quota_info;
343 struct lustre_dquot *dquot = NULL;
345 __u64 hlimit = 0, slimit = 0;
347 unsigned int grace = 0;
348 struct lustre_qunit_size *lqs = NULL;
352 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
355 dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata));
357 RETURN(PTR_ERR(dquot));
359 DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
360 QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
362 cfs_down(&mds->mds_qonoff_sem);
363 cfs_down(&dquot->dq_sem);
365 if (dquot->dq_status & DQ_STATUS_RECOVERY) {
366 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
367 GOTO(out, rc = -EBUSY);
370 if (QDATA_IS_BLK(qdata)) {
371 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
372 usage = &dquot->dq_dqb.dqb_curspace;
373 hlimit = dquot->dq_dqb.dqb_bhardlimit;
374 slimit = dquot->dq_dqb.dqb_bsoftlimit;
375 time = &dquot->dq_dqb.dqb_btime;
377 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
378 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
379 hlimit = dquot->dq_dqb.dqb_ihardlimit;
380 slimit = dquot->dq_dqb.dqb_isoftlimit;
381 time = &dquot->dq_dqb.dqb_itime;
384 /* if the quota limit in admin quotafile is zero, we just inform
385 * slave to clear quota limit with zero qd_count */
386 if (hlimit == 0 && slimit == 0) {
394 QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
396 if (QDATA_IS_CHANGE_QS(qdata) &&
397 QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
398 qdata->qd_count = (hlimit -
399 QUSG(*usage, QDATA_IS_BLK(qdata)))
400 * (QDATA_IS_BLK(qdata) ?
401 QUOTABLOCK_SIZE : 1);
403 GOTO(out, rc = -EDQUOT);
407 QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
408 if (*time && cfs_time_current_sec() >= *time)
409 GOTO(out, rc = -EDQUOT);
411 *time = cfs_time_current_sec() + grace;
414 *usage += qdata->qd_count;
417 /* The usage in administrative file might be incorrect before
419 if (*usage - qdata->qd_count < 0)
422 *usage -= qdata->qd_count;
424 /* (usage <= soft limit) but not (usage < soft limit) */
425 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
432 rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
435 cfs_up(&dquot->dq_sem);
436 cfs_up(&mds->mds_qonoff_sem);
439 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
440 QDATA_IS_BLK(qdata));
442 lqs = quota_search_lqs(LQS_KEY(QDATA_IS_GRP(qdata), qdata->qd_id),
444 if (lqs == NULL || IS_ERR(lqs)) {
445 CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
446 qdata->qd_qunit = QDATA_IS_BLK(qdata) ? qctxt->lqc_bunit_sz :
449 cfs_spin_lock(&lqs->lqs_lock);
450 qdata->qd_qunit = QDATA_IS_BLK(qdata) ? lqs->lqs_bunit_sz :
452 cfs_spin_unlock(&lqs->lqs_lock);
455 if (QDATA_IS_BLK(qdata))
456 QDATA_SET_ADJBLK(qdata);
458 QDATA_SET_ADJINO(qdata);
460 QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
467 int mds_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
468 const unsigned int qpids[], int rc, int opc)
470 struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
474 if (rc && rc != -EDQUOT && rc != ENOLCK)
478 case FSFILT_OP_SETATTR:
479 /* release file quota on original owner */
480 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 0, 0, NULL);
481 /* release block quota on original owner */
482 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
483 /* acquire file quota on current owner */
484 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
485 /* acquire block quota on current owner */
486 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
488 case FSFILT_OP_UNLINK_PARTIAL_CHILD:
489 /* release file quota on child */
490 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
491 /* rlease block quota on child */
492 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
494 case FSFILT_OP_CREATE_PARTIAL_CHILD:
495 /* acquire file quota on child */
496 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
497 /* acquire block quota on child */
498 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
501 /* acquire block quota on parent */
502 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
504 case FSFILT_OP_UNLINK:
505 /* release block quota on parent */
506 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
507 /* release file quota on child */
508 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
509 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
510 /* release block quota on child */
511 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
514 case FSFILT_OP_UNLINK_PARTIAL_PARENT:
515 /* release block quota on parent */
516 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
518 case FSFILT_OP_CREATE:
519 /* acquire block quota on parent */
520 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
521 /* acquire file quota on child */
522 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
523 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
524 /* acquire block quota on child */
525 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
535 "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
536 rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
540 int filter_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
541 const unsigned int qpids[], int rc, int opc)
543 struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
547 if (rc && rc != -EDQUOT)
551 case FSFILT_OP_SETATTR:
552 /* acquire/release block quota on original & current owner */
553 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
554 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
556 case FSFILT_OP_UNLINK:
557 /* release block quota on this owner */
558 case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
559 /* acquire block quota on this owner */
560 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
571 "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
572 rc == QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
578 static const char prefix[] = "OBJECTS/";
580 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
582 struct mds_obd *mds = &obd->u.mds;
583 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
584 struct obd_device_target *obt = &obd->u.obt;
585 int rc = 0, i, rc1 = 0;
586 char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
588 struct lvfs_run_ctxt saved;
591 LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
593 if (oqctl->qc_type != USRQUOTA &&
594 oqctl->qc_type != GRPQUOTA &&
595 oqctl->qc_type != UGQUOTA)
598 cfs_down(&obt->obt_quotachecking);
599 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
600 cfs_down(&mds->mds_qonoff_sem);
602 for (i = 0; i < MAXQUOTAS; i++) {
605 if (!Q_TYPESET(oqctl, i))
608 /* quota file has been opened ? */
609 if (qinfo->qi_files[i]) {
610 CWARN("quota[%d] is on yet\n", i);
615 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
616 sprintf(name, "%s%s", prefix, quotafile[i]);
618 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
621 CERROR("%s: error invalidating admin quotafile %s (rc:%d)\n",
622 obd->obd_name, name, rc);
628 cfs_up(&mds->mds_qonoff_sem);
629 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
630 cfs_up(&obt->obt_quotachecking);
634 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
636 struct mds_obd *mds = &obd->u.mds;
637 struct obd_device_target *obt = &obd->u.obt;
639 struct lvfs_run_ctxt saved;
642 if (oqctl->qc_type != USRQUOTA &&
643 oqctl->qc_type != GRPQUOTA &&
644 oqctl->qc_type != UGQUOTA)
647 cfs_down(&obt->obt_quotachecking);
648 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
649 cfs_down(&mds->mds_qonoff_sem);
651 oqctl->qc_cmd = Q_FINVALIDATE;
652 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
654 rc = obd_quotactl(mds->mds_osc_exp, oqctl);
656 cfs_up(&mds->mds_qonoff_sem);
657 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
658 cfs_up(&obt->obt_quotachecking);
662 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
664 struct mds_obd *mds = &obd->u.mds;
665 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
666 const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
667 struct lvfs_run_ctxt saved;
672 LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
674 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
676 cfs_down(&mds->mds_qonoff_sem);
678 for (i = 0; i < MAXQUOTAS && !rc; i++) {
681 if (!Q_TYPESET(oqctl, i))
684 /* quota file has been opened ? */
685 if (qinfo->qi_files[i]) {
686 CWARN("init %s admin quotafile while quota on.\n",
687 i == USRQUOTA ? "user" : "group");
691 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
692 sprintf(name, "%s%s", prefix, quotafile[i]);
694 /* check if quota file exists and is correct */
695 fp = filp_open(name, O_RDONLY, 0);
697 /* irregular file is not the right place for quota */
698 if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
699 CERROR("admin quota file %s is not "
705 qinfo->qi_files[i] = fp;
706 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
707 qinfo->qi_files[i] = 0;
716 /* -EINVAL may be returned by quotainfo for bad quota file */
717 if (rc != -ENOENT && rc != -EINVAL) {
718 CERROR("%s: error opening old quota file %s (%d)\n",
719 obd->obd_name, name, rc);
723 CDEBUG(D_INFO, "%s new quota file %s\n", name,
724 rc == -ENOENT ? "creating" : "overwriting");
726 /* create quota file overwriting old if needed */
727 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
730 CERROR("%s: error creating admin quotafile %s (rc:%d)\n",
731 obd->obd_name, name, rc);
735 qinfo->qi_files[i] = fp;
737 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
739 CERROR("error init %s admin quotafile! (rc:%d)\n",
740 i == USRQUOTA ? "user" : "group", rc);
743 qinfo->qi_files[i] = NULL;
745 cfs_up(&mds->mds_qonoff_sem);
747 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
751 static int close_quota_files(struct obd_quotactl *oqctl,
752 struct lustre_quota_info *qinfo)
757 for (i = 0; i < MAXQUOTAS; i++) {
758 if (!Q_TYPESET(oqctl, i))
760 if (qinfo->qi_files[i] == NULL) {
761 CWARN("quota[%d] is off already\n", i);
765 filp_close(qinfo->qi_files[i], 0);
766 qinfo->qi_files[i] = NULL;
771 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
773 struct mds_obd *mds = &obd->u.mds;
774 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
775 const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
777 int i, rc = 0, rc1 = 0;
780 LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
782 /* open admin quota files and read quotafile info */
783 for (i = 0; i < MAXQUOTAS; i++) {
786 if (!Q_TYPESET(oqctl, i) || qinfo->qi_files[i] != NULL)
789 LASSERT(strlen(quotafile[i])
790 + sizeof(prefix) <= sizeof(name));
791 sprintf(name, "%s%s", prefix, quotafile[i]);
792 fp = filp_open(name, O_RDWR, 0);
793 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
794 rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
795 CERROR("error open/create %s! (rc:%d)\n", name, rc);
798 qinfo->qi_files[i] = fp;
800 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
802 CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
806 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
808 CERROR("error read quotainfo of %s! (rc:%d)\n", name,
814 if (rc && rc1 != -EALREADY)
815 close_quota_files(oqctl, qinfo);
820 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
825 if (oqctl->qc_type != USRQUOTA &&
826 oqctl->qc_type != GRPQUOTA &&
827 oqctl->qc_type != UGQUOTA)
830 rc = generic_quota_on(obd, oqctl, 1);
836 int mds_admin_quota_off(struct obd_device *obd,
837 struct obd_quotactl *oqctl)
839 struct mds_obd *mds = &obd->u.mds;
840 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
844 /* close admin quota files */
845 rc = close_quota_files(oqctl, qinfo);
850 /* with obt->obt_quotachecking held */
851 int do_mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
853 struct mds_obd *mds = &obd->u.mds;
854 struct obd_device_target *obt = &obd->u.obt;
855 struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
856 struct lvfs_run_ctxt saved;
857 int rc = 0, rc1 = 0, rc2 = 0;
860 LASSERT_SEM_LOCKED(&obt->obt_quotachecking);
862 if (oqctl->qc_type != USRQUOTA &&
863 oqctl->qc_type != GRPQUOTA &&
864 oqctl->qc_type != UGQUOTA)
867 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
868 cfs_down(&mds->mds_qonoff_sem);
869 /* close admin quota files */
870 rc2 = mds_admin_quota_off(obd, oqctl);
871 if (rc2 && rc2 != -EALREADY) {
872 CWARN("mds quota[%d] is failed to be off for %d\n", oqctl->qc_type, rc2);
876 rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
878 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
879 } else if (quota_is_off(qctxt, oqctl)) {
880 CWARN("mds local quota[%d] is off already\n", oqctl->qc_type);
883 if (rc2 != -EALREADY) {
884 CWARN("mds local quota[%d] is failed to be off for %d\n",
885 oqctl->qc_type, rc1);
886 oqctl->qc_cmd = Q_QUOTAON;
887 mds_admin_quota_on(obd, oqctl);
888 oqctl->qc_cmd = Q_QUOTAOFF;
893 rc = obd_quotactl(mds->mds_osc_exp, oqctl);
894 if (rc && rc != -EALREADY) {
895 CWARN("mds remote quota[%d] is failed to be off for %d\n",
897 oqctl->qc_cmd = Q_QUOTAON;
898 if (rc2 != -EALREADY)
899 mds_admin_quota_on(obd, oqctl);
900 if (rc1 != -EALREADY) {
901 fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
902 qctxt->lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
904 oqctl->qc_cmd = Q_QUOTAOFF;
909 cfs_up(&mds->mds_qonoff_sem);
910 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
911 return rc ? : (rc1 ? : rc2);
914 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
916 struct obd_device_target *obt = &obd->u.obt;
920 cfs_down(&obt->obt_quotachecking);
921 rc = do_mds_quota_off(obd, oqctl);
922 cfs_up(&obt->obt_quotachecking);
926 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
928 struct mds_obd *mds = &obd->u.mds;
929 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
930 struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
934 if (oqctl->qc_type != USRQUOTA &&
935 oqctl->qc_type != GRPQUOTA)
938 cfs_down(&mds->mds_qonoff_sem);
939 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
940 CWARN("quota[%u] is off\n", oqctl->qc_type);
941 GOTO(out, rc = -ESRCH);
944 qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
945 qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
946 qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
948 rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
952 cfs_up(&mds->mds_qonoff_sem);
956 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
958 struct mds_obd *mds = &obd->u.mds;
959 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
960 struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
964 if (oqctl->qc_type != USRQUOTA &&
965 oqctl->qc_type != GRPQUOTA)
968 cfs_down(&mds->mds_qonoff_sem);
969 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
970 CWARN("quota[%u] is off\n", oqctl->qc_type);
971 GOTO(out, rc = -ESRCH);
974 dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
975 dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
976 dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
980 cfs_up(&mds->mds_qonoff_sem);
984 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
985 struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
986 int type, struct quota_adjust_qunit *oqaq)
988 __u64 bunit_curr_o, iunit_curr_o;
989 unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
990 unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
991 __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
992 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
993 __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
994 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1000 LASSERT_SEM_LOCKED(&dquot->dq_sem);
1001 LASSERT(oqaq->qaq_iunit_sz);
1002 LASSERT(oqaq->qaq_bunit_sz);
1004 /* don't change qunit size */
1005 if (!qctxt->lqc_switch_qs)
1008 bunit_curr_o = oqaq->qaq_bunit_sz;
1009 iunit_curr_o = oqaq->qaq_iunit_sz;
1011 if (dquot->dq_type == GRPQUOTA)
1014 if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1015 __u64 b_limitation =
1016 oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1017 /* enlarge block qunit size */
1019 QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1020 oqaq->qaq_bunit_sz =
1021 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1023 b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1027 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1028 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1030 /* shrink block qunit size */
1032 QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1033 do_div(oqaq->qaq_bunit_sz , cqs_factor);
1034 oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1036 b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1038 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1042 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1043 oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1045 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1046 QAQ_SET_ADJBLK(oqaq);
1050 if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1051 __u64 i_limitation =
1052 oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1053 /* enlarge file qunit size */
1054 while (ilimit > dquot->dq_dqb.dqb_curinodes
1055 + 2 * i_limitation) {
1056 oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1057 i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1061 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1062 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1064 /* shrink file qunit size */
1065 while (ilimit < dquot->dq_dqb.dqb_curinodes
1067 do_div(oqaq->qaq_iunit_sz, cqs_factor);
1068 i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1070 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1074 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1075 oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1077 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1078 QAQ_SET_ADJINO(oqaq);
1082 QAQ_DEBUG(oqaq, "the oqaq computed\n");
1087 static int mds_init_slave_ilimits(struct obd_device *obd,
1088 struct obd_quotactl *oqctl, int set)
1090 /* XXX: for file limits only adjust local now */
1091 struct obd_device_target *obt = &obd->u.obt;
1092 struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1093 unsigned int id[MAXQUOTAS] = { 0, 0 };
1094 struct obd_quotactl *ioqc = NULL;
1095 struct lustre_qunit_size *lqs;
1100 /* if we are going to set zero limit, needn't init slaves */
1101 if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1105 OBD_ALLOC_PTR(ioqc);
1109 flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1110 oqctl->qc_dqblk.dqb_isoftlimit || !set;
1111 ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1112 ioqc->qc_id = oqctl->qc_id;
1113 ioqc->qc_type = oqctl->qc_type;
1114 ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1115 ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1117 /* build lqs for mds */
1118 lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1119 qctxt, flag ? 1 : 0);
1120 if (lqs && !IS_ERR(lqs)) {
1122 lqs->lqs_flags |= QI_SET;
1124 lqs->lqs_flags &= ~QI_SET;
1127 CERROR("fail to %s lqs for inode(%s id: %u)!\n",
1128 flag ? "create" : "search",
1129 oqctl->qc_type ? "group" : "user",
1131 GOTO(out, rc = PTR_ERR(lqs));
1134 /* set local limit to MIN_QLIMIT */
1135 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1139 /* trigger local qunit pre-acquire */
1140 if (oqctl->qc_type == USRQUOTA)
1141 id[USRQUOTA] = oqctl->qc_id;
1143 id[GRPQUOTA] = oqctl->qc_id;
1145 rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1146 if (rc == -EDQUOT || rc == -EBUSY) {
1147 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1151 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1155 /* FIXME initialize all slaves in CMD */
1163 static int mds_init_slave_blimits(struct obd_device *obd,
1164 struct obd_quotactl *oqctl, int set)
1166 struct obd_device_target *obt = &obd->u.obt;
1167 struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1168 struct mds_obd *mds = &obd->u.mds;
1169 struct obd_quotactl *ioqc;
1170 struct lustre_qunit_size *lqs;
1171 unsigned int id[MAXQUOTAS] = { 0, 0 };
1176 /* if we are going to set zero limit, needn't init slaves */
1177 if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1181 OBD_ALLOC_PTR(ioqc);
1185 flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1186 oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1187 ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1188 ioqc->qc_id = oqctl->qc_id;
1189 ioqc->qc_type = oqctl->qc_type;
1190 ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1191 ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1193 /* build lqs for mds */
1194 lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1195 qctxt, flag ? 1 : 0);
1196 if (lqs && !IS_ERR(lqs)) {
1198 lqs->lqs_flags |= QB_SET;
1200 lqs->lqs_flags &= ~QB_SET;
1203 CERROR("fail to %s lqs for block(%s id: %u)!\n",
1204 flag ? "create" : "search",
1205 oqctl->qc_type ? "group" : "user",
1207 GOTO(out, rc = PTR_ERR(lqs));
1210 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1214 /* trigger local qunit pre-acquire */
1215 if (oqctl->qc_type == USRQUOTA)
1216 id[USRQUOTA] = oqctl->qc_id;
1218 id[GRPQUOTA] = oqctl->qc_id;
1220 /* initialize all slave's limit */
1221 rc = obd_quotactl(mds->mds_osc_exp, ioqc);
1223 rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1224 if (rc == -EDQUOT || rc == -EBUSY) {
1225 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1229 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1239 static void adjust_lqs(struct obd_device *obd, struct quota_adjust_qunit *qaq)
1241 struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
1244 QAQ_SET_CREATE_LQS(qaq);
1245 /* adjust local lqs */
1246 rc = quota_adjust_slave_lqs(qaq, qctxt);
1248 CERROR("adjust master's qunit size failed!(rc=%d)\n", rc);
1250 /* adjust remote lqs */
1251 if (QAQ_IS_ADJBLK(qaq)) {
1252 rc = obd_quota_adjust_qunit(obd->u.mds.mds_osc_exp, qaq, qctxt);
1254 CERROR("adjust slaves' qunit size failed!(rc=%d)\n", rc);
1259 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1261 struct mds_obd *mds = &obd->u.mds;
1262 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1263 struct obd_device *lov_obd = class_exp2obd(mds->mds_osc_exp);
1264 struct lov_obd *lov = &lov_obd->u.lov;
1265 struct quota_adjust_qunit *oqaq = NULL;
1266 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1267 __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1268 time_t btime, itime;
1269 struct lustre_dquot *dquot;
1270 struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1271 /* orig_set means if quota was set before; now_set means we are
1272 * setting/cancelling quota */
1273 int orig_set, now_set;
1274 int rc, rc2 = 0, flag = 0;
1277 if (oqctl->qc_type != USRQUOTA &&
1278 oqctl->qc_type != GRPQUOTA)
1281 OBD_ALLOC_PTR(oqaq);
1284 cfs_down(&mds->mds_qonoff_sem);
1285 init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1287 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1288 CWARN("quota[%u] is off\n", oqctl->qc_type);
1289 GOTO(out_sem, rc = -ESRCH);
1292 dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1294 GOTO(out_sem, rc = PTR_ERR(dquot));
1295 DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1296 QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1298 cfs_down(&dquot->dq_sem);
1300 if (dquot->dq_status) {
1301 cfs_up(&dquot->dq_sem);
1302 lustre_dqput(dquot);
1303 GOTO(out_sem, rc = -EBUSY);
1305 dquot->dq_status |= DQ_STATUS_SET;
1307 ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1308 isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1309 bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1310 bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1311 btime = dquot->dq_dqb.dqb_btime;
1312 itime = dquot->dq_dqb.dqb_itime;
1314 if (dqblk->dqb_valid & QIF_BTIME)
1315 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1316 if (dqblk->dqb_valid & QIF_ITIME)
1317 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1319 if (dqblk->dqb_valid & QIF_BLIMITS) {
1320 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1321 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1322 /* clear usage (limit pool) */
1323 if (!dquot->dq_dqb.dqb_bhardlimit &&
1324 !dquot->dq_dqb.dqb_bsoftlimit)
1325 dquot->dq_dqb.dqb_curspace = 0;
1327 /* clear grace time */
1328 if (!dqblk->dqb_bsoftlimit ||
1329 toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1330 dquot->dq_dqb.dqb_btime = 0;
1331 /* set grace only if user hasn't provided his own */
1332 else if (!(dqblk->dqb_valid & QIF_BTIME))
1333 dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1334 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1336 flag |= LQUOTA_FLAGS_ADJBLK;
1339 if (dqblk->dqb_valid & QIF_ILIMITS) {
1340 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1341 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1342 /* clear usage (limit pool) */
1343 if (!dquot->dq_dqb.dqb_ihardlimit &&
1344 !dquot->dq_dqb.dqb_isoftlimit)
1345 dquot->dq_dqb.dqb_curinodes = 0;
1347 if (!dqblk->dqb_isoftlimit ||
1348 dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1349 dquot->dq_dqb.dqb_itime = 0;
1350 else if (!(dqblk->dqb_valid & QIF_ITIME))
1351 dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1352 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1354 flag |= LQUOTA_FLAGS_ADJINO;
1356 QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1357 rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1359 QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1361 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1364 rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1366 cfs_up(&dquot->dq_sem);
1369 CERROR("set limit failed! (rc:%d)\n", rc);
1373 cfs_up(&mds->mds_qonoff_sem);
1375 adjust_lqs(obd, oqaq);
1377 orig_set = ihardlimit || isoftlimit;
1378 now_set = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1379 if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1380 cfs_down(&dquot->dq_sem);
1381 dquot->dq_dqb.dqb_curinodes = 0;
1382 cfs_up(&dquot->dq_sem);
1383 rc = mds_init_slave_ilimits(obd, oqctl, orig_set);
1385 CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1390 orig_set = bhardlimit || bsoftlimit;
1391 now_set = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1392 if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1393 cfs_down(&dquot->dq_sem);
1394 dquot->dq_dqb.dqb_curspace = 0;
1395 cfs_up(&dquot->dq_sem);
1396 rc = mds_init_slave_blimits(obd, oqctl, orig_set);
1398 CERROR("init slave blimits failed! (rc:%d)\n", rc);
1404 cfs_down(&mds->mds_qonoff_sem);
1405 cfs_down(&dquot->dq_sem);
1407 /* cancel previous setting */
1408 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1409 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1410 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1411 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1412 dquot->dq_dqb.dqb_btime = btime;
1413 dquot->dq_dqb.dqb_itime = itime;
1415 rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1416 cfs_up(&dquot->dq_sem);
1419 cfs_down(&dquot->dq_sem);
1420 dquot->dq_status &= ~DQ_STATUS_SET;
1421 cfs_up(&dquot->dq_sem);
1422 lustre_dqput(dquot);
1425 cfs_up(&mds->mds_qonoff_sem);
1430 return rc ? rc : rc2;
1433 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1435 struct obd_quotactl *soqc;
1436 struct lvfs_run_ctxt saved;
1440 OBD_ALLOC_PTR(soqc);
1444 soqc->qc_cmd = Q_GETOQUOTA;
1445 soqc->qc_id = oqctl->qc_id;
1446 soqc->qc_type = oqctl->qc_type;
1448 /* get block usage from OSS */
1449 soqc->qc_dqblk.dqb_curspace = 0;
1450 rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
1451 if (!rc || rc == -EREMOTEIO) {
1452 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1453 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1456 /* get block/inode usage from MDS */
1457 soqc->qc_dqblk.dqb_curspace = 0;
1458 soqc->qc_dqblk.dqb_curinodes = 0;
1459 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1460 rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1461 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1463 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1464 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1465 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1473 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1475 struct mds_obd *mds = &obd->u.mds;
1476 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1477 struct lustre_dquot *dquot;
1478 struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1482 if (oqctl->qc_type != USRQUOTA &&
1483 oqctl->qc_type != GRPQUOTA)
1486 cfs_down(&mds->mds_qonoff_sem);
1487 dqblk->dqb_valid = 0;
1488 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1489 CWARN("quota[%u] is off\n", oqctl->qc_type);
1490 GOTO(out, rc = -ESRCH);
1493 dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1495 GOTO(out, rc = PTR_ERR(dquot));
1497 cfs_down(&dquot->dq_sem);
1498 dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1499 dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1500 dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1501 dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1502 dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1503 dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1504 dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1505 cfs_up(&dquot->dq_sem);
1507 lustre_dqput(dquot);
1508 cfs_up(&mds->mds_qonoff_sem);
1510 /* the usages in admin quota file is inaccurate */
1511 dqblk->dqb_curinodes = 0;
1512 dqblk->dqb_curspace = 0;
1513 rc = mds_get_space(obd, oqctl);
1517 cfs_up(&mds->mds_qonoff_sem);
1521 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1523 struct lvfs_run_ctxt saved;
1527 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1528 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1529 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1535 /* FIXME we only recovery block limit by now, need recovery inode
1536 * limits also after CMD involved in */
1538 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1540 struct mds_obd *mds = &obd->u.mds;
1541 struct lustre_quota_info *qinfo= &mds->mds_quota_info;
1542 struct lustre_dquot *dquot;
1543 struct obd_quotactl *qctl;
1544 __u64 total_limits = 0;
1548 OBD_ALLOC_PTR(qctl);
1552 dquot = lustre_dqget(obd, qinfo, id, type);
1553 if (IS_ERR(dquot)) {
1554 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1556 RETURN(PTR_ERR(dquot));
1559 cfs_down(&dquot->dq_sem);
1561 /* don't recovery the dquot without limits or under setting */
1562 if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1565 dquot->dq_status |= DQ_STATUS_RECOVERY;
1567 cfs_up(&dquot->dq_sem);
1569 /* get real bhardlimit from all slaves. */
1570 qctl->qc_cmd = Q_GETOQUOTA;
1571 qctl->qc_type = type;
1573 qctl->qc_stat = QUOTA_RECOVERING;
1574 rc = obd_quotactl(mds->mds_osc_exp, qctl);
1577 total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1579 /* get real bhardlimit from master */
1580 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1583 total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1585 /* amend the usage of the administrative quotafile */
1586 cfs_down(&mds->mds_qonoff_sem);
1587 cfs_down(&dquot->dq_sem);
1589 dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1591 rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1593 CERROR("write dquot failed! (rc:%d)\n", rc);
1595 cfs_up(&dquot->dq_sem);
1596 cfs_up(&mds->mds_qonoff_sem);
1599 cfs_down(&dquot->dq_sem);
1600 dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1602 cfs_up(&dquot->dq_sem);
1604 lustre_dqput(dquot);
1609 struct qmaster_recov_thread_data {
1610 struct obd_device *obd;
1611 cfs_completion_t comp;
1614 static int qmaster_recovery_main(void *arg)
1616 struct qmaster_recov_thread_data *data = arg;
1617 struct obd_device *obd = data->obd;
1618 struct mds_obd *mds = &obd->u.mds;
1619 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1621 unsigned short type;
1624 cfs_daemonize_ctxt("qmaster_recovd");
1627 class_incref(obd, "qmaster_recovd_mds", obd);
1629 class_incref(mds->mds_osc_obd, "qmaster_recovd_lov", mds->mds_osc_obd);
1631 cfs_complete(&data->comp);
1633 for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1635 struct dquot_id *dqid, *tmp;
1637 cfs_down(&mds->mds_qonoff_sem);
1638 if (qinfo->qi_files[type] == NULL) {
1639 cfs_up(&mds->mds_qonoff_sem);
1642 CFS_INIT_LIST_HEAD(&id_list);
1643 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1645 cfs_up(&mds->mds_qonoff_sem);
1648 CERROR("error get ids from admin quotafile.(%d)\n", rc);
1650 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1651 cfs_list_del_init(&dqid->di_link);
1655 rc = dquot_recovery(obd, dqid->di_id, type);
1657 CERROR("%s: qmaster recovery failed for %sid %d"
1658 " rc:%d)\n", obd->obd_name,
1659 type ? "g" : "u", dqid->di_id, rc);
1664 class_decref(mds->mds_osc_obd, "qmaster_recovd_lov", mds->mds_osc_obd);
1665 class_decref(obd, "qmaster_recovd_mds", obd);
1669 int mds_quota_recovery(struct obd_device *obd)
1671 struct mds_obd *mds = &obd->u.mds;
1672 struct qmaster_recov_thread_data data;
1676 if (unlikely(!mds->mds_quota || obd->obd_stopping))
1679 cfs_mutex_down(&obd->obd_dev_sem);
1680 if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1681 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1682 mds->mds_lov_desc.ld_active_tgt_count,
1683 mds->mds_lov_objid_count);
1684 cfs_mutex_up(&obd->obd_dev_sem);
1687 cfs_mutex_up(&obd->obd_dev_sem);
1690 cfs_init_completion(&data.comp);
1692 rc = cfs_kernel_thread(qmaster_recovery_main, &data,
1693 CLONE_VM|CLONE_FILES);
1695 CERROR("%s: cannot start quota recovery thread: rc %d\n",
1698 cfs_wait_for_completion(&data.comp);
1702 #endif /* HAVE_QUOTA_SUPPORT */