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 (c) 2007, 2010, Oracle and/or its affiliates. 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 > lqs->lqs_lock */
68 static cfs_list_t lustre_dquot_hash[NR_DQHASH];
69 static cfs_rwlock_t dquot_hash_lock = CFS_RW_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 cfs_list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
130 if (dquot->dq_info == lqi &&
131 dquot->dq_id == id && dquot->dq_type == type)
137 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
140 struct lustre_dquot *dquot = NULL;
143 OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
147 CFS_INIT_LIST_HEAD(&dquot->dq_hash);
148 cfs_init_mutex_locked(&dquot->dq_sem);
149 cfs_atomic_set(&dquot->dq_refcnt, 1);
150 dquot->dq_info = lqi;
152 dquot->dq_type = type;
157 static void free_dquot(struct lustre_dquot *dquot)
159 OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
162 static void insert_dquot_nolock(struct lustre_dquot *dquot)
164 cfs_list_t *head = lustre_dquot_hash +
165 dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
166 LASSERT(cfs_list_empty(&dquot->dq_hash));
167 cfs_list_add(&dquot->dq_hash, head);
170 static void remove_dquot_nolock(struct lustre_dquot *dquot)
172 LASSERT(!cfs_list_empty(&dquot->dq_hash));
173 cfs_list_del_init(&dquot->dq_hash);
176 static void lustre_dqput(struct lustre_dquot *dquot)
179 cfs_write_lock(&dquot_hash_lock);
180 LASSERT(cfs_atomic_read(&dquot->dq_refcnt));
181 cfs_atomic_dec(&dquot->dq_refcnt);
182 if (cfs_atomic_read(&dquot->dq_refcnt) == 0) {
183 remove_dquot_nolock(dquot);
186 cfs_write_unlock(&dquot_hash_lock);
190 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
191 struct lustre_quota_info *lqi,
192 qid_t id, int type, int can_fake)
194 unsigned int hashent = dquot_hashfn(lqi, id, type);
195 struct lustre_dquot *dquot, *empty;
199 if ((empty = alloc_dquot(lqi, id, type)) == NULL)
200 RETURN(ERR_PTR(-ENOMEM));
202 cfs_read_lock(&dquot_hash_lock);
203 if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
204 cfs_atomic_inc(&dquot->dq_refcnt);
205 cfs_read_unlock(&dquot_hash_lock);
210 cfs_read_unlock(&dquot_hash_lock);
213 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
214 cfs_up(&dquot->dq_sem);
216 CERROR("can't read dquot from admin quotafile! "
221 cfs_write_lock(&dquot_hash_lock);
222 if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
223 cfs_atomic_inc(&dquot->dq_refcnt);
227 insert_dquot_nolock(dquot);
229 cfs_write_unlock(&dquot_hash_lock);
235 if (!can_fake && cfs_test_bit(DQ_FAKE_B, &dquot->dq_flags)) {
236 DQUOT_DEBUG(dquot, "It is a fake dquot: unexpected!\n");
238 dquot = ERR_PTR(-ENOENT);
248 static void init_oqaq(struct quota_adjust_qunit *oqaq,
249 struct lustre_quota_ctxt *qctxt,
252 struct lustre_qunit_size *lqs = NULL;
255 oqaq->qaq_flags = type;
256 lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 0);
257 if (lqs && !IS_ERR(lqs)) {
258 cfs_spin_lock(&lqs->lqs_lock);
259 oqaq->qaq_bunit_sz = lqs->lqs_bunit_sz;
260 oqaq->qaq_iunit_sz = lqs->lqs_iunit_sz;
261 oqaq->qaq_flags = lqs->lqs_flags;
262 cfs_spin_unlock(&lqs->lqs_lock);
265 CDEBUG(D_QUOTA, "Can't find the lustre qunit size!\n");
266 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
267 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
271 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
274 struct mds_obd *mds = &obd->u.mds;
275 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
276 __u32 ost_num = mds->mds_lov_objid_count, mdt_num = 1;
277 struct quota_adjust_qunit *oqaq = NULL;
278 unsigned int qid[MAXQUOTAS] = { 0, 0 };
279 struct lustre_quota_info *info = &mds->mds_quota_info;
280 struct lustre_dquot *dquot = NULL;
286 cfs_down_read(&mds->mds_qonoff_sem);
287 dquot = lustre_dqget(obd, info, id, type, 0);
289 RETURN(PTR_ERR(dquot));
291 cfs_up_read(&mds->mds_qonoff_sem);
294 GOTO(out, rc = -ENOMEM);
296 cfs_down(&dquot->dq_sem);
297 init_oqaq(oqaq, qctxt, id, type);
299 rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
300 is_blk ? LQUOTA_FLAGS_ADJBLK :
301 LQUOTA_FLAGS_ADJINO, oqaq);
304 CERROR("create oqaq failed! (rc:%d)\n", rc);
307 QAQ_DEBUG(oqaq, "show oqaq.\n")
309 if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
312 /* adjust the mds slave qunit size */
313 adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
314 if (adjust_res <= 0) {
315 if (adjust_res < 0) {
317 CERROR("adjust mds slave's qunit size failed! "
320 CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
326 qid[GRPQUOTA] = dquot->dq_id;
328 qid[USRQUOTA] = dquot->dq_id;
330 cfs_up(&dquot->dq_sem);
332 rc = qctxt_adjust_qunit(obd, qctxt, qid, is_blk, 0, NULL);
333 if (rc == -EDQUOT || rc == -EBUSY) {
334 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
338 CERROR("%s: mds fail to adjust file quota! (rc:%d)\n",
343 /* only when block qunit is reduced, boardcast to osts */
344 if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
345 rc = obd_quota_adjust_qunit(mds->mds_lov_exp, oqaq, qctxt, NULL);
354 cfs_up(&dquot->dq_sem);
358 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
360 struct mds_obd *mds = &obd->u.mds;
361 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
362 struct lustre_quota_info *info = &mds->mds_quota_info;
363 struct lustre_dquot *dquot = NULL;
365 __u64 hlimit = 0, slimit = 0;
367 unsigned int grace = 0;
368 struct lustre_qunit_size *lqs = NULL;
372 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
375 if (!ll_sb_has_quota_active(qctxt->lqc_sb,
376 QDATA_IS_GRP(qdata) ? GRPQUOTA : USRQUOTA))
379 lqs = quota_search_lqs(LQS_KEY(QDATA_IS_GRP(qdata), qdata->qd_id),
388 cfs_spin_lock(&lqs->lqs_lock);
389 if (LQS_IS_RECOVERY(lqs)) {
390 cfs_spin_unlock(&lqs->lqs_lock);
391 LQS_DEBUG(lqs, "this lqs is under recovery\n");
392 GOTO(skip, rc = -EBUSY);
394 cfs_spin_unlock(&lqs->lqs_lock);
396 cfs_down_write(&mds->mds_qonoff_sem);
397 dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata), 0);
399 cfs_up_write(&mds->mds_qonoff_sem);
400 GOTO(skip, rc = PTR_ERR(dquot));
403 DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
404 QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
406 cfs_down(&dquot->dq_sem);
408 if (QDATA_IS_BLK(qdata)) {
409 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
410 usage = &dquot->dq_dqb.dqb_curspace;
411 hlimit = dquot->dq_dqb.dqb_bhardlimit;
412 slimit = dquot->dq_dqb.dqb_bsoftlimit;
413 time = &dquot->dq_dqb.dqb_btime;
415 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
416 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
417 hlimit = dquot->dq_dqb.dqb_ihardlimit;
418 slimit = dquot->dq_dqb.dqb_isoftlimit;
419 time = &dquot->dq_dqb.dqb_itime;
422 /* if the quota limit in admin quotafile is zero, we just inform
423 * slave to clear quota limit with zero qd_count */
424 if (hlimit == 0 && slimit == 0) {
432 QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
434 if (QDATA_IS_CHANGE_QS(qdata) &&
435 QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
436 qdata->qd_count = (hlimit -
437 QUSG(*usage, QDATA_IS_BLK(qdata)))
438 * (QDATA_IS_BLK(qdata) ?
439 QUOTABLOCK_SIZE : 1);
441 GOTO(out, rc = -EDQUOT);
445 QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
446 if (*time && cfs_time_current_sec() >= *time)
447 GOTO(out, rc = -EDQUOT);
449 *time = cfs_time_current_sec() + grace;
452 *usage += qdata->qd_count;
455 /* The usage in administrative file might be incorrect before
457 if (*usage - qdata->qd_count < 0)
460 *usage -= qdata->qd_count;
462 /* (usage <= soft limit) but not (usage < soft limit) */
463 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
470 rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
473 cfs_up(&dquot->dq_sem);
474 cfs_up_write(&mds->mds_qonoff_sem);
477 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
478 QDATA_IS_BLK(qdata));
480 cfs_spin_lock(&lqs->lqs_lock);
481 qdata->qd_qunit = QDATA_IS_BLK(qdata) ? lqs->lqs_bunit_sz :
483 cfs_spin_unlock(&lqs->lqs_lock);
485 if (QDATA_IS_BLK(qdata))
486 QDATA_SET_ADJBLK(qdata);
488 QDATA_SET_ADJINO(qdata);
490 QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
497 int mds_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
498 const unsigned int qpids[], int rc, int opc)
500 struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
504 if (rc && rc != -EDQUOT && rc != ENOLCK)
508 case FSFILT_OP_SETATTR:
509 /* release file quota on original owner */
510 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 0, 0, NULL);
511 /* release block quota on original owner */
512 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
513 /* acquire file quota on current owner */
514 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
515 /* acquire block quota on current owner */
516 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
518 case FSFILT_OP_UNLINK_PARTIAL_CHILD:
519 /* release file quota on child */
520 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
521 /* rlease block quota on child */
522 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
524 case FSFILT_OP_CREATE_PARTIAL_CHILD:
525 /* acquire file quota on child */
526 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
527 /* acquire block quota on child */
528 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
531 /* acquire block quota on parent */
532 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
534 case FSFILT_OP_UNLINK:
535 /* release block quota on parent */
536 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
537 /* release file quota on child */
538 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
539 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
540 /* release block quota on child */
541 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
544 case FSFILT_OP_UNLINK_PARTIAL_PARENT:
545 /* release block quota on parent */
546 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
548 case FSFILT_OP_CREATE:
549 /* acquire block quota on parent */
550 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
551 /* acquire file quota on child */
552 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
553 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
554 /* acquire block quota on child */
555 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
565 "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
566 rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
570 int filter_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
571 const unsigned int qpids[], int rc, int opc)
573 struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
577 if (rc && rc != -EDQUOT)
581 case FSFILT_OP_SETATTR:
582 /* acquire/release block quota on original & current owner */
583 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
584 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
586 case FSFILT_OP_UNLINK:
587 /* release block quota on this owner */
588 case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
589 /* acquire block quota on this owner */
590 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
601 "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
602 rc == QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
608 static const char prefix[] = "OBJECTS/";
610 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
612 struct mds_obd *mds = &obd->u.mds;
613 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
614 struct obd_device_target *obt = &obd->u.obt;
615 int rc = 0, i, rc1 = 0;
616 char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
618 struct lvfs_run_ctxt saved;
621 LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
623 if (oqctl->qc_type != USRQUOTA &&
624 oqctl->qc_type != GRPQUOTA &&
625 oqctl->qc_type != UGQUOTA)
628 cfs_down(&obt->obt_quotachecking);
629 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
630 cfs_down_write(&mds->mds_qonoff_sem);
632 for (i = 0; i < MAXQUOTAS; i++) {
635 if (!Q_TYPESET(oqctl, i))
638 /* quota file has been opened ? */
639 if (qinfo->qi_files[i]) {
640 CWARN("quota[%d] is on yet\n", i);
645 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
646 sprintf(name, "%s%s", prefix, quotafile[i]);
648 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
651 CERROR("%s: error invalidating admin quotafile %s (rc:%d)\n",
652 obd->obd_name, name, rc);
658 cfs_up_write(&mds->mds_qonoff_sem);
659 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
660 cfs_up(&obt->obt_quotachecking);
664 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
666 struct mds_obd *mds = &obd->u.mds;
667 struct obd_device_target *obt = &obd->u.obt;
669 struct lvfs_run_ctxt saved;
672 if (oqctl->qc_type != USRQUOTA &&
673 oqctl->qc_type != GRPQUOTA &&
674 oqctl->qc_type != UGQUOTA)
677 cfs_down(&obt->obt_quotachecking);
678 if (obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))
679 GOTO(out, rc = -EBUSY);
680 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
681 cfs_down_write(&mds->mds_qonoff_sem);
683 oqctl->qc_cmd = Q_FINVALIDATE;
684 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
686 rc = obd_quotactl(mds->mds_lov_exp, oqctl);
688 cfs_up_write(&mds->mds_qonoff_sem);
689 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
691 cfs_up(&obt->obt_quotachecking);
695 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
697 struct mds_obd *mds = &obd->u.mds;
698 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
699 const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
700 struct lvfs_run_ctxt saved;
705 LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
707 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
708 cfs_down_write(&mds->mds_qonoff_sem);
710 for (i = 0; i < MAXQUOTAS && !rc; i++) {
713 if (!Q_TYPESET(oqctl, i))
716 /* quota file has been opened ? */
717 if (qinfo->qi_files[i]) {
718 CWARN("init %s admin quotafile while quota on.\n",
719 i == USRQUOTA ? "user" : "group");
723 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
724 sprintf(name, "%s%s", prefix, quotafile[i]);
726 /* check if quota file exists and is correct */
727 fp = filp_open(name, O_RDONLY, 0);
729 /* irregular file is not the right place for quota */
730 if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
731 CERROR("admin quota file %s is not "
737 qinfo->qi_files[i] = fp;
738 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
739 qinfo->qi_files[i] = NULL;
748 /* -EINVAL may be returned by quotainfo for bad quota file */
749 if (rc != -ENOENT && rc != -EINVAL) {
750 CERROR("%s: error opening old quota file %s (%d)\n",
751 obd->obd_name, name, rc);
755 CDEBUG(D_INFO, "%s new quota file %s\n", name,
756 rc == -ENOENT ? "creating" : "overwriting");
758 /* create quota file overwriting old if needed */
759 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
762 CERROR("%s: error creating admin quotafile %s (rc:%d)\n",
763 obd->obd_name, name, rc);
767 qinfo->qi_files[i] = fp;
769 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
771 CERROR("error init %s admin quotafile! (rc:%d)\n",
772 i == USRQUOTA ? "user" : "group", rc);
775 qinfo->qi_files[i] = NULL;
778 cfs_up_write(&mds->mds_qonoff_sem);
779 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
783 static int close_quota_files(struct obd_quotactl *oqctl,
784 struct lustre_quota_info *qinfo)
789 for (i = 0; i < MAXQUOTAS; i++) {
790 if (!Q_TYPESET(oqctl, i))
792 if (qinfo->qi_files[i] == NULL) {
793 CWARN("quota[%d] is off already\n", i);
797 filp_close(qinfo->qi_files[i], 0);
798 qinfo->qi_files[i] = NULL;
803 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
805 struct mds_obd *mds = &obd->u.mds;
806 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
807 const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
809 int i, rc = 0, rc1 = 0;
812 LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
814 /* open admin quota files and read quotafile info */
815 for (i = 0; i < MAXQUOTAS; i++) {
818 if (!Q_TYPESET(oqctl, i) || qinfo->qi_files[i] != NULL)
821 LASSERT(strlen(quotafile[i])
822 + sizeof(prefix) <= sizeof(name));
823 sprintf(name, "%s%s", prefix, quotafile[i]);
824 fp = filp_open(name, O_RDWR, 0);
825 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
826 rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
827 CERROR("error open/create %s! (rc:%d)\n", name, rc);
830 qinfo->qi_files[i] = fp;
832 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
834 CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
838 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
840 CERROR("error read quotainfo of %s! (rc:%d)\n", name,
846 if (rc && rc1 != -EALREADY)
847 close_quota_files(oqctl, qinfo);
852 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
857 if (oqctl->qc_type != USRQUOTA &&
858 oqctl->qc_type != GRPQUOTA &&
859 oqctl->qc_type != UGQUOTA)
862 rc = generic_quota_on(obd, oqctl, 1);
868 int mds_admin_quota_off(struct obd_device *obd,
869 struct obd_quotactl *oqctl)
871 struct mds_obd *mds = &obd->u.mds;
872 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
876 /* close admin quota files */
877 rc = close_quota_files(oqctl, qinfo);
882 /* with obt->obt_quotachecking held */
883 int do_mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
885 struct mds_obd *mds = &obd->u.mds;
886 struct obd_device_target *obt = &obd->u.obt;
887 struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
888 struct lvfs_run_ctxt saved;
889 int rc = 0, rc1 = 0, rc2 = 0;
892 LASSERT_SEM_LOCKED(&obt->obt_quotachecking);
894 if (oqctl->qc_type != USRQUOTA &&
895 oqctl->qc_type != GRPQUOTA &&
896 oqctl->qc_type != UGQUOTA)
899 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
900 cfs_down_write(&mds->mds_qonoff_sem);
901 /* close admin quota files */
902 rc2 = mds_admin_quota_off(obd, oqctl);
903 if (rc2 && rc2 != -EALREADY) {
904 CWARN("mds quota[%d] is failed to be off for %d\n", oqctl->qc_type, rc2);
908 rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
910 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
911 } else if (quota_is_off(qctxt, oqctl)) {
912 CWARN("mds local quota[%d] is off already\n", oqctl->qc_type);
915 if (rc2 != -EALREADY) {
916 CWARN("mds local quota[%d] is failed to be off for %d\n",
917 oqctl->qc_type, rc1);
918 oqctl->qc_cmd = Q_QUOTAON;
919 mds_admin_quota_on(obd, oqctl);
920 oqctl->qc_cmd = Q_QUOTAOFF;
925 rc = obd_quotactl(mds->mds_lov_exp, oqctl);
926 if (rc && rc != -EALREADY) {
927 CWARN("mds remote quota[%d] is failed to be off for %d\n",
929 oqctl->qc_cmd = Q_QUOTAON;
930 if (rc2 != -EALREADY)
931 mds_admin_quota_on(obd, oqctl);
932 if (rc1 != -EALREADY) {
933 fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
934 qctxt->lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
936 oqctl->qc_cmd = Q_QUOTAOFF;
941 CDEBUG(D_QUOTA, "%s: quotaoff type:flags:rc %u:%lu:%d\n",
942 obd->obd_name, oqctl->qc_type, qctxt->lqc_flags, rc);
943 cfs_up_write(&mds->mds_qonoff_sem);
944 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
945 return rc ? : (rc1 ? : rc2);
948 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
950 struct obd_device_target *obt = &obd->u.obt;
954 cfs_down(&obt->obt_quotachecking);
955 rc = do_mds_quota_off(obd, oqctl);
956 cfs_up(&obt->obt_quotachecking);
960 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
962 struct mds_obd *mds = &obd->u.mds;
963 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
964 struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
968 if (oqctl->qc_type != USRQUOTA &&
969 oqctl->qc_type != GRPQUOTA)
972 cfs_down_write(&mds->mds_qonoff_sem);
973 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
974 CWARN("quota[%u] is off\n", oqctl->qc_type);
975 GOTO(out, rc = -ESRCH);
978 qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
979 qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
980 qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
982 rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
986 cfs_up_write(&mds->mds_qonoff_sem);
990 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
992 struct mds_obd *mds = &obd->u.mds;
993 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
994 struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
998 if (oqctl->qc_type != USRQUOTA &&
999 oqctl->qc_type != GRPQUOTA)
1002 cfs_down_read(&mds->mds_qonoff_sem);
1003 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1004 CWARN("quota[%u] is off\n", oqctl->qc_type);
1005 GOTO(out, rc = -ESRCH);
1008 dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
1009 dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
1010 dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
1014 cfs_up_read(&mds->mds_qonoff_sem);
1018 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
1019 struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
1020 int type, struct quota_adjust_qunit *oqaq)
1022 __u64 bunit_curr_o, iunit_curr_o;
1023 unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
1024 unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
1025 __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
1026 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
1027 __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1028 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1032 if (!dquot || !oqaq)
1034 LASSERT_SEM_LOCKED(&dquot->dq_sem);
1035 LASSERT(oqaq->qaq_iunit_sz);
1036 LASSERT(oqaq->qaq_bunit_sz);
1038 /* don't change qunit size */
1039 if (!qctxt->lqc_switch_qs)
1042 bunit_curr_o = oqaq->qaq_bunit_sz;
1043 iunit_curr_o = oqaq->qaq_iunit_sz;
1045 if (dquot->dq_type == GRPQUOTA)
1048 if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1049 __u64 b_limitation =
1050 oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1051 /* enlarge block qunit size */
1053 QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1054 oqaq->qaq_bunit_sz =
1055 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1057 b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1061 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1062 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1064 /* shrink block qunit size */
1066 QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1067 do_div(oqaq->qaq_bunit_sz , cqs_factor);
1068 oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1070 b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1072 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1076 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1077 oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1079 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1080 QAQ_SET_ADJBLK(oqaq);
1084 if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1085 __u64 i_limitation =
1086 oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1087 /* enlarge file qunit size */
1088 while (ilimit > dquot->dq_dqb.dqb_curinodes
1089 + 2 * i_limitation) {
1090 oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1091 i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1095 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1096 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1098 /* shrink file qunit size */
1099 while (ilimit < dquot->dq_dqb.dqb_curinodes
1101 do_div(oqaq->qaq_iunit_sz, cqs_factor);
1102 i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1104 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1108 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1109 oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1111 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1112 QAQ_SET_ADJINO(oqaq);
1116 QAQ_DEBUG(oqaq, "the oqaq computed\n");
1121 static int mds_init_slave_ilimits(struct obd_device *obd,
1122 struct obd_quotactl *oqctl, int set)
1124 /* XXX: for file limits only adjust local now */
1125 struct obd_device_target *obt = &obd->u.obt;
1126 struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1127 unsigned int id[MAXQUOTAS] = { 0, 0 };
1128 struct obd_quotactl *ioqc = NULL;
1129 struct lustre_qunit_size *lqs;
1134 /* if we are going to set zero limit, needn't init slaves */
1135 if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1139 OBD_ALLOC_PTR(ioqc);
1143 flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1144 oqctl->qc_dqblk.dqb_isoftlimit || !set;
1145 ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1146 ioqc->qc_id = oqctl->qc_id;
1147 ioqc->qc_type = oqctl->qc_type;
1148 ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1149 ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1151 /* build lqs for mds */
1152 lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1153 qctxt, flag ? 1 : 0);
1154 if (lqs && !IS_ERR(lqs)) {
1156 lqs->lqs_flags |= QI_SET;
1158 lqs->lqs_flags &= ~QI_SET;
1161 CERROR("fail to %s lqs for inode(%s id: %u)!\n",
1162 flag ? "create" : "search",
1163 oqctl->qc_type ? "group" : "user",
1165 GOTO(out, rc = PTR_ERR(lqs));
1168 /* set local limit to MIN_QLIMIT */
1169 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1173 /* trigger local qunit pre-acquire */
1174 if (oqctl->qc_type == USRQUOTA)
1175 id[USRQUOTA] = oqctl->qc_id;
1177 id[GRPQUOTA] = oqctl->qc_id;
1179 rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1180 if (rc == -EDQUOT || rc == -EBUSY) {
1181 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1185 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1189 /* FIXME initialize all slaves in CMD */
1197 static int mds_init_slave_blimits(struct obd_device *obd,
1198 struct obd_quotactl *oqctl, int set)
1200 struct obd_device_target *obt = &obd->u.obt;
1201 struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1202 struct mds_obd *mds = &obd->u.mds;
1203 struct obd_quotactl *ioqc;
1204 struct lustre_qunit_size *lqs;
1205 unsigned int id[MAXQUOTAS] = { 0, 0 };
1210 /* if we are going to set zero limit, needn't init slaves */
1211 if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1215 OBD_ALLOC_PTR(ioqc);
1219 flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1220 oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1221 ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1222 ioqc->qc_id = oqctl->qc_id;
1223 ioqc->qc_type = oqctl->qc_type;
1224 ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1225 ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1227 /* build lqs for mds */
1228 lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1229 qctxt, flag ? 1 : 0);
1230 if (lqs && !IS_ERR(lqs)) {
1232 lqs->lqs_flags |= QB_SET;
1234 lqs->lqs_flags &= ~QB_SET;
1237 CERROR("fail to %s lqs for block(%s id: %u)!\n",
1238 flag ? "create" : "search",
1239 oqctl->qc_type ? "group" : "user",
1241 GOTO(out, rc = PTR_ERR(lqs));
1244 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1248 /* trigger local qunit pre-acquire */
1249 if (oqctl->qc_type == USRQUOTA)
1250 id[USRQUOTA] = oqctl->qc_id;
1252 id[GRPQUOTA] = oqctl->qc_id;
1254 /* initialize all slave's limit */
1255 rc = obd_quotactl(mds->mds_lov_exp, ioqc);
1257 rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1258 if (rc == -EDQUOT || rc == -EBUSY) {
1259 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1263 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1273 static void adjust_lqs(struct obd_device *obd, struct quota_adjust_qunit *qaq)
1275 struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
1278 QAQ_SET_CREATE_LQS(qaq);
1279 /* adjust local lqs */
1280 rc = quota_adjust_slave_lqs(qaq, qctxt);
1282 CERROR("adjust master's qunit size failed!(rc=%d)\n", rc);
1284 /* adjust remote lqs */
1285 if (QAQ_IS_ADJBLK(qaq)) {
1286 rc = obd_quota_adjust_qunit(obd->u.mds.mds_lov_exp, qaq, qctxt, NULL);
1288 CERROR("adjust slaves' qunit size failed!(rc=%d)\n", rc);
1293 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1295 struct mds_obd *mds = &obd->u.mds;
1296 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1297 struct obd_device *lov_obd = class_exp2obd(mds->mds_lov_exp);
1298 struct lov_obd *lov = &lov_obd->u.lov;
1299 struct quota_adjust_qunit *oqaq = NULL;
1300 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1301 __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1302 time_t btime, itime;
1303 struct lustre_dquot *dquot;
1304 struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1305 /* orig_set means if quota was set before; now_set means we are
1306 * setting/cancelling quota */
1307 int orig_set, now_set;
1308 struct lustre_qunit_size *lqs;
1309 int rc = 0, rc2 = 0, flag = 0;
1312 if (oqctl->qc_type != USRQUOTA &&
1313 oqctl->qc_type != GRPQUOTA)
1316 OBD_ALLOC_PTR(oqaq);
1320 cfs_down_write(&mds->mds_qonoff_sem);
1321 init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1323 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1324 CWARN("quota[%u] is off\n", oqctl->qc_type);
1325 GOTO(out_sem, rc = -ESRCH);
1328 dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1330 GOTO(out_sem, rc = PTR_ERR(dquot));
1331 DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1332 QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1334 lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id), qctxt, 1);
1342 cfs_down(&dquot->dq_sem);
1343 cfs_spin_lock(&lqs->lqs_lock);
1344 if (LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1345 cfs_spin_unlock(&lqs->lqs_lock);
1346 cfs_up(&dquot->dq_sem);
1347 GOTO(skip, rc = -EBUSY);
1349 LQS_SET_SETQUOTA(lqs);
1350 cfs_spin_unlock(&lqs->lqs_lock);
1352 ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1353 isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1354 bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1355 bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1356 btime = dquot->dq_dqb.dqb_btime;
1357 itime = dquot->dq_dqb.dqb_itime;
1359 if (dqblk->dqb_valid & QIF_BTIME)
1360 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1361 if (dqblk->dqb_valid & QIF_ITIME)
1362 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1364 if (dqblk->dqb_valid & QIF_BLIMITS) {
1365 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1366 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1367 /* clear usage (limit pool) */
1368 if (!dquot->dq_dqb.dqb_bhardlimit &&
1369 !dquot->dq_dqb.dqb_bsoftlimit)
1370 dquot->dq_dqb.dqb_curspace = 0;
1372 /* clear grace time */
1373 if (!dqblk->dqb_bsoftlimit ||
1374 toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1375 dquot->dq_dqb.dqb_btime = 0;
1376 /* set grace only if user hasn't provided his own */
1377 else if (!(dqblk->dqb_valid & QIF_BTIME))
1378 dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1379 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1381 flag |= LQUOTA_FLAGS_ADJBLK;
1384 if (dqblk->dqb_valid & QIF_ILIMITS) {
1385 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1386 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1387 /* clear usage (limit pool) */
1388 if (!dquot->dq_dqb.dqb_ihardlimit &&
1389 !dquot->dq_dqb.dqb_isoftlimit)
1390 dquot->dq_dqb.dqb_curinodes = 0;
1392 if (!dqblk->dqb_isoftlimit ||
1393 dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1394 dquot->dq_dqb.dqb_itime = 0;
1395 else if (!(dqblk->dqb_valid & QIF_ITIME))
1396 dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1397 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1399 flag |= LQUOTA_FLAGS_ADJINO;
1401 QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1402 rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1404 QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1406 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1409 rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1411 cfs_up(&dquot->dq_sem);
1414 CERROR("set limit failed! (rc:%d)\n", rc);
1415 GOTO(update_fail, rc);
1418 cfs_up_write(&mds->mds_qonoff_sem);
1419 adjust_lqs(obd, oqaq);
1421 orig_set = ihardlimit || isoftlimit;
1422 now_set = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1423 if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1424 cfs_down(&dquot->dq_sem);
1425 dquot->dq_dqb.dqb_curinodes = 0;
1426 cfs_up(&dquot->dq_sem);
1427 rc = mds_init_slave_ilimits(obd, oqctl, orig_set);
1429 CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1434 orig_set = bhardlimit || bsoftlimit;
1435 now_set = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1436 if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1437 cfs_down(&dquot->dq_sem);
1438 dquot->dq_dqb.dqb_curspace = 0;
1439 cfs_up(&dquot->dq_sem);
1440 rc = mds_init_slave_blimits(obd, oqctl, orig_set);
1442 CERROR("init slave blimits failed! (rc:%d)\n", rc);
1448 cfs_down_write(&mds->mds_qonoff_sem);
1449 cfs_down(&dquot->dq_sem);
1451 /* cancel previous setting */
1452 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1453 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1454 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1455 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1456 dquot->dq_dqb.dqb_btime = btime;
1457 dquot->dq_dqb.dqb_itime = itime;
1459 rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1460 cfs_up(&dquot->dq_sem);
1462 cfs_spin_lock(&lqs->lqs_lock);
1463 LQS_CLEAR_SETQUOTA(lqs);
1464 cfs_spin_unlock(&lqs->lqs_lock);
1468 lustre_dqput(dquot);
1471 cfs_up_write(&mds->mds_qonoff_sem);
1476 return rc ? rc : rc2;
1479 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1481 struct obd_quotactl *soqc;
1482 struct lvfs_run_ctxt saved;
1486 OBD_ALLOC_PTR(soqc);
1490 soqc->qc_cmd = Q_GETOQUOTA;
1491 soqc->qc_id = oqctl->qc_id;
1492 soqc->qc_type = oqctl->qc_type;
1494 /* get block usage from OSS */
1495 soqc->qc_dqblk.dqb_curspace = 0;
1496 rc = obd_quotactl(obd->u.mds.mds_lov_exp, soqc);
1497 if (!rc || rc == -EREMOTEIO) {
1498 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1499 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1502 /* get block/inode usage from MDS */
1503 soqc->qc_dqblk.dqb_curspace = 0;
1504 soqc->qc_dqblk.dqb_curinodes = 0;
1505 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1506 rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1507 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1509 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1510 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1511 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1519 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1521 struct mds_obd *mds = &obd->u.mds;
1522 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1523 struct lustre_dquot *dquot;
1524 struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1528 if (oqctl->qc_type != USRQUOTA &&
1529 oqctl->qc_type != GRPQUOTA)
1532 cfs_down_read(&mds->mds_qonoff_sem);
1533 dqblk->dqb_valid = 0;
1534 if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1535 CWARN("quota[%u] is off\n", oqctl->qc_type);
1536 GOTO(out, rc = -ESRCH);
1539 dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1541 GOTO(out, rc = PTR_ERR(dquot));
1543 cfs_down(&dquot->dq_sem);
1544 dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1545 dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1546 dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1547 dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1548 dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1549 dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1550 dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1551 cfs_up(&dquot->dq_sem);
1553 lustre_dqput(dquot);
1554 cfs_up_read(&mds->mds_qonoff_sem);
1556 /* the usages in admin quota file is inaccurate */
1557 dqblk->dqb_curinodes = 0;
1558 dqblk->dqb_curspace = 0;
1559 rc = mds_get_space(obd, oqctl);
1562 * Querying of curinodes and/or curspace may have failed, administrative
1563 * quota data are likely to be better approximation to the real usage in
1566 if (!(dqblk->dqb_valid & QIF_INODES) && dquot->dq_dqb.dqb_curinodes > 0)
1567 dqblk->dqb_curinodes = dquot->dq_dqb.dqb_curinodes;
1569 if (!(dqblk->dqb_valid & QIF_SPACE) && dquot->dq_dqb.dqb_curspace > 0)
1570 dqblk->dqb_curspace = dquot->dq_dqb.dqb_curspace;
1575 cfs_up_read(&mds->mds_qonoff_sem);
1579 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1581 struct lvfs_run_ctxt saved;
1585 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1586 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1587 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1593 /* FIXME we only recovery block limit by now, need recovery inode
1594 * limits also after CMD involved in */
1596 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1598 struct mds_obd *mds = &obd->u.mds;
1599 struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1600 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1601 struct lustre_qunit_size *lqs;
1602 struct lustre_dquot *dquot;
1603 struct obd_quotactl *qctl;
1604 __u64 total_limits = 0;
1608 OBD_ALLOC_PTR(qctl);
1612 dquot = lustre_dqget(obd, qinfo, id, type, 0);
1613 if (IS_ERR(dquot)) {
1614 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1616 RETURN(PTR_ERR(dquot));
1619 lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 1);
1627 cfs_down(&dquot->dq_sem);
1629 /* don't recover the dquot without limits or quota is setting or
1630 * another recovery is already going on */
1631 if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1632 LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1633 cfs_up(&dquot->dq_sem);
1634 GOTO(skip1, rc = 0);
1637 cfs_spin_lock(&lqs->lqs_lock);
1638 LQS_SET_RECOVERY(lqs);
1639 cfs_spin_unlock(&lqs->lqs_lock);
1640 cfs_up(&dquot->dq_sem);
1642 /* release mds_qonoff_sem during obd_quotactl ops here */
1643 cfs_up_write(&mds->mds_qonoff_sem);
1645 /* get real bhardlimit from all slaves. */
1646 qctl->qc_cmd = Q_GETOQUOTA;
1647 qctl->qc_type = type;
1649 qctl->qc_stat = QUOTA_RECOVERING;
1650 rc = obd_quotactl(mds->mds_lov_exp, qctl);
1651 cfs_down_write(&mds->mds_qonoff_sem);
1654 total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1656 /* get real bhardlimit from master */
1657 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1660 total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1662 /* amend the usage of the administrative quotafile */
1663 cfs_down(&dquot->dq_sem);
1665 dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1667 rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1669 CERROR("write dquot failed! (rc:%d)\n", rc);
1671 cfs_up(&dquot->dq_sem);
1674 cfs_spin_lock(&lqs->lqs_lock);
1675 LQS_CLEAR_RECOVERY(lqs);
1676 cfs_spin_unlock(&lqs->lqs_lock);
1680 lustre_dqput(dquot);
1685 struct qmaster_recov_thread_data {
1686 struct obd_device *obd;
1687 cfs_completion_t comp;
1690 static int qmaster_recovery_main(void *arg)
1692 struct qmaster_recov_thread_data *data = arg;
1693 struct obd_device *obd = data->obd;
1694 struct mds_obd *mds = &obd->u.mds;
1695 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1697 unsigned short type;
1700 cfs_daemonize_ctxt("qmaster_recovd");
1703 class_incref(obd, "qmaster_recovd_mds", obd);
1705 class_incref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1707 cfs_complete(&data->comp);
1709 cfs_down_write(&mds->mds_qonoff_sem);
1710 for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1712 struct dquot_id *dqid, *tmp;
1714 if (qinfo->qi_files[type] == NULL)
1717 CFS_INIT_LIST_HEAD(&id_list);
1718 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1721 CERROR("error get ids from admin quotafile.(%d)\n", rc);
1723 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1724 cfs_list_del_init(&dqid->di_link);
1728 rc = dquot_recovery(obd, dqid->di_id, type);
1730 CERROR("%s: qmaster recovery failed for %sid %d"
1731 " rc:%d)\n", obd->obd_name,
1732 type ? "g" : "u", dqid->di_id, rc);
1737 cfs_up_write(&mds->mds_qonoff_sem);
1738 class_decref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1739 class_decref(obd, "qmaster_recovd_mds", obd);
1743 int mds_quota_recovery(struct obd_device *obd)
1745 struct mds_obd *mds = &obd->u.mds;
1746 struct qmaster_recov_thread_data data;
1750 if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
1753 if (unlikely(!mds->mds_quota || obd->obd_stopping))
1756 cfs_mutex_down(&obd->obd_dev_sem);
1757 if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1758 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1759 mds->mds_lov_desc.ld_active_tgt_count,
1760 mds->mds_lov_objid_count);
1761 cfs_mutex_up(&obd->obd_dev_sem);
1764 cfs_mutex_up(&obd->obd_dev_sem);
1767 cfs_init_completion(&data.comp);
1769 rc = cfs_kernel_thread(qmaster_recovery_main, &data,
1770 CLONE_VM|CLONE_FILES);
1772 CERROR("%s: cannot start quota recovery thread: rc %d\n",
1775 cfs_wait_for_completion(&data.comp);
1779 #endif /* HAVE_QUOTA_SUPPORT */