Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / quota / quota_master.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
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.
11  *
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).
17  *
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
21  *
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
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
30  * Use is subject to license terms.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/quota/quota_master.c
37  *
38  * Lustre Quota Master request handler
39  *
40  * Author: Niu YaWei <niu@clusterfs.com>
41  */
42
43 #ifndef EXPORT_SYMTAB
44 # define EXPORT_SYMTAB
45 #endif
46
47 #define DEBUG_SUBSYSTEM S_LQUOTA
48
49 #include <linux/version.h>
50 #include <linux/fs.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>
57
58 #include <obd_class.h>
59 #include <lustre_quota.h>
60 #include <lustre_fsfilt.h>
61 #include <lustre_mds.h>
62
63 #include "quota_internal.h"
64
65 #ifdef HAVE_QUOTA_SUPPORT
66
67 /* lock ordering: mds->mds_qonoff_sem > dquot->dq_sem */
68 static struct list_head lustre_dquot_hash[NR_DQHASH];
69 static spinlock_t dquot_hash_lock = SPIN_LOCK_UNLOCKED;
70
71 cfs_mem_cache_t *lustre_dquot_cachep;
72
73 int lustre_dquot_init(void)
74 {
75         int i;
76         ENTRY;
77
78         LASSERT(lustre_dquot_cachep == NULL);
79         lustre_dquot_cachep = cfs_mem_cache_create("lustre_dquot_cache",
80                                                    sizeof(struct lustre_dquot),
81                                                    0, 0);
82         if (!lustre_dquot_cachep)
83                 return (-ENOMEM);
84
85         for (i = 0; i < NR_DQHASH; i++) {
86                 CFS_INIT_LIST_HEAD(lustre_dquot_hash + i);
87         }
88         RETURN(0);
89 }
90
91 void lustre_dquot_exit(void)
92 {
93         int i;
94         ENTRY;
95         /* FIXME cleanup work ?? */
96
97         for (i = 0; i < NR_DQHASH; i++) {
98                 LASSERT(list_empty(lustre_dquot_hash + i));
99         }
100         if (lustre_dquot_cachep) {
101                 int rc;
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;
105         }
106         EXIT;
107 }
108
109 static inline int
110 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
111              __attribute__((__const__));
112
113 static inline int
114 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
115 {
116         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
117         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
118         return tmp;
119 }
120
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,
124                                        int type)
125 {
126         struct lustre_dquot *dquot;
127         ENTRY;
128
129         LASSERT_SPIN_LOCKED(&dquot_hash_lock);
130         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)
133                         RETURN(dquot);
134         }
135         RETURN(NULL);
136 }
137
138 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
139                                         qid_t id, int type)
140 {
141         struct lustre_dquot *dquot = NULL;
142         ENTRY;
143
144         OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
145         if (dquot == NULL)
146                 RETURN(NULL);
147
148         CFS_INIT_LIST_HEAD(&dquot->dq_hash);
149         init_mutex_locked(&dquot->dq_sem);
150         dquot->dq_refcnt = 1;
151         dquot->dq_info = lqi;
152         dquot->dq_id = id;
153         dquot->dq_type = type;
154         dquot->dq_status = DQ_STATUS_AVAIL;
155
156         RETURN(dquot);
157 }
158
159 static void free_dquot(struct lustre_dquot *dquot)
160 {
161         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
162 }
163
164 static void insert_dquot_nolock(struct lustre_dquot *dquot)
165 {
166         struct list_head *head = lustre_dquot_hash +
167             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
168         LASSERT(list_empty(&dquot->dq_hash));
169         list_add(&dquot->dq_hash, head);
170 }
171
172 static void remove_dquot_nolock(struct lustre_dquot *dquot)
173 {
174         LASSERT(!list_empty(&dquot->dq_hash));
175         list_del_init(&dquot->dq_hash);
176 }
177
178 static void lustre_dqput(struct lustre_dquot *dquot)
179 {
180         ENTRY;
181         spin_lock(&dquot_hash_lock);
182         LASSERT(dquot->dq_refcnt);
183         dquot->dq_refcnt--;
184         if (!dquot->dq_refcnt) {
185                 remove_dquot_nolock(dquot);
186                 free_dquot(dquot);
187         }
188         spin_unlock(&dquot_hash_lock);
189         EXIT;
190 }
191
192 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
193                                          struct lustre_quota_info *lqi,
194                                          qid_t id, int type)
195 {
196         unsigned int hashent = dquot_hashfn(lqi, id, type);
197         struct lustre_dquot *dquot, *empty;
198         ENTRY;
199
200         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
201                 RETURN(ERR_PTR(-ENOMEM));
202
203         spin_lock(&dquot_hash_lock);
204         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
205                 dquot->dq_refcnt++;
206                 spin_unlock(&dquot_hash_lock);
207                 free_dquot(empty);
208         } else {
209                 int rc;
210
211                 dquot = empty;
212                 insert_dquot_nolock(dquot);
213                 spin_unlock(&dquot_hash_lock);
214
215                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
216                 up(&dquot->dq_sem);
217                 if (rc) {
218                         CERROR("can't read dquot from admin quotafile! "
219                                "(rc:%d)\n", rc);
220                         lustre_dqput(dquot);
221                         RETURN(ERR_PTR(rc));
222                 }
223
224         }
225
226         LASSERT(dquot);
227         RETURN(dquot);
228 }
229
230 static void init_oqaq(struct quota_adjust_qunit *oqaq,
231                       struct lustre_quota_ctxt *qctxt,
232                       qid_t id, int type)
233 {
234         struct lustre_qunit_size *lqs = NULL;
235
236         oqaq->qaq_id = id;
237         oqaq->qaq_flags = type;
238         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 0);
239         if (lqs && !IS_ERR(lqs)) {
240                 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                 spin_unlock(&lqs->lqs_lock);
245                 lqs_putref(lqs);
246         } else {
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;
250         }
251 }
252
253 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
254                           __u32 is_blk)
255 {
256         struct mds_obd *mds = &obd->u.mds;
257         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
258         struct obd_device *lov_mds_obd = class_exp2obd(mds->mds_osc_exp);
259         struct lov_obd *lov = &lov_mds_obd->u.lov;
260         __u32 ost_num = lov->desc.ld_tgt_count, mdt_num = 1;
261         struct quota_adjust_qunit *oqaq = NULL;
262         unsigned int qid[MAXQUOTAS] = { 0, 0 };
263         struct lustre_quota_info *info = &mds->mds_quota_info;
264         struct lustre_dquot *dquot = NULL;
265         int adjust_res = 0;
266         int rc = 0;
267         ENTRY;
268
269         LASSERT(mds);
270         dquot = lustre_dqget(obd, info, id, type);
271         if (IS_ERR(dquot))
272                 RETURN(PTR_ERR(dquot));
273
274         OBD_ALLOC_PTR(oqaq);
275         if (!oqaq)
276                 GOTO(out, rc = -ENOMEM);
277
278         down(&dquot->dq_sem);
279         init_oqaq(oqaq, qctxt, id, type);
280
281         rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
282                                is_blk ? LQUOTA_FLAGS_ADJBLK :
283                                LQUOTA_FLAGS_ADJINO, oqaq);
284
285         if (rc < 0) {
286                 CDEBUG(D_ERROR, "create oqaq failed! (rc:%d)\n", rc);
287                 GOTO(out_sem, rc);
288         }
289         QAQ_DEBUG(oqaq, "show oqaq.\n")
290
291         if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
292                 GOTO(out_sem, rc);
293
294         /* adjust the mds slave qunit size */
295         adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
296         if (adjust_res <= 0) {
297                 if (adjust_res < 0) {
298                         rc = adjust_res;
299                         CDEBUG(D_ERROR, "adjust mds slave's qunit size failed! \
300                                (rc:%d)\n", rc);
301                 } else {
302                         CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
303                 }
304                 GOTO(out_sem, rc);
305         }
306
307         if (type)
308                 qid[GRPQUOTA] = dquot->dq_id;
309         else
310                 qid[USRQUOTA] = dquot->dq_id;
311
312         up(&dquot->dq_sem);
313
314         rc = qctxt_adjust_qunit(obd, qctxt, qid, is_blk, 0, NULL);
315         if (rc == -EDQUOT || rc == -EBUSY) {
316                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
317                 rc = 0;
318         }
319         if (rc) {
320                 CDEBUG(D_ERROR, "mds fail to adjust file quota! \
321                                (rc:%d)\n", rc);
322                 GOTO(out, rc);
323         }
324
325         /* only when block qunit is reduced, boardcast to osts */
326         if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
327                 rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
328
329 out:
330         lustre_dqput(dquot);
331         if (oqaq)
332                 OBD_FREE_PTR(oqaq);
333
334         RETURN(rc);
335 out_sem:
336         up(&dquot->dq_sem);
337         goto out;
338 }
339
340 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
341 {
342         struct mds_obd *mds = &obd->u.mds;
343         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
344         struct lustre_quota_info *info = &mds->mds_quota_info;
345         struct lustre_dquot *dquot = NULL;
346         __u64 *usage = NULL;
347         __u64 hlimit = 0, slimit = 0;
348         time_t *time = NULL;
349         unsigned int grace = 0;
350         struct lustre_qunit_size *lqs = NULL;
351         int rc = 0;
352         ENTRY;
353
354         if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
355                 RETURN(-EIO);
356
357         dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata));
358         if (IS_ERR(dquot))
359                 RETURN(PTR_ERR(dquot));
360
361         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
362         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
363
364         down(&mds->mds_qonoff_sem);
365         down(&dquot->dq_sem);
366
367         if (dquot->dq_status & DQ_STATUS_RECOVERY) {
368                 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
369                 GOTO(out, rc = -EBUSY);
370         }
371
372         if (QDATA_IS_BLK(qdata)) {
373                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
374                 usage = &dquot->dq_dqb.dqb_curspace;
375                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
376                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
377                 time = &dquot->dq_dqb.dqb_btime;
378         } else {
379                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
380                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
381                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
382                 slimit = dquot->dq_dqb.dqb_isoftlimit;
383                 time = &dquot->dq_dqb.dqb_itime;
384         }
385
386         /* if the quota limit in admin quotafile is zero, we just inform
387          * slave to clear quota limit with zero qd_count */
388         if (hlimit == 0 && slimit == 0) {
389                 qdata->qd_count = 0;
390                 GOTO(out, rc);
391         }
392
393         switch (opc) {
394         case QUOTA_DQACQ:
395                 if (hlimit &&
396                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
397                 {
398                         if (QDATA_IS_CHANGE_QS(qdata) &&
399                             QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
400                                 qdata->qd_count = (hlimit -
401                                         QUSG(*usage, QDATA_IS_BLK(qdata)))
402                                         * (QDATA_IS_BLK(qdata) ?
403                                            QUOTABLOCK_SIZE : 1);
404                         else
405                                 GOTO(out, rc = -EDQUOT);
406                 }
407
408                 if (slimit &&
409                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
410                         if (*time && cfs_time_current_sec() >= *time)
411                                 GOTO(out, rc = -EDQUOT);
412                         else if (!*time)
413                                 *time = cfs_time_current_sec() + grace;
414                 }
415
416                 *usage += qdata->qd_count;
417                 break;
418         case QUOTA_DQREL:
419                 /* The usage in administrative file might be incorrect before
420                  * recovery done */
421                 if (*usage - qdata->qd_count < 0)
422                         *usage = 0;
423                 else
424                         *usage -= qdata->qd_count;
425
426                 /* (usage <= soft limit) but not (usage < soft limit) */
427                 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
428                         *time = 0;
429                 break;
430         default:
431                 LBUG();
432         }
433
434         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
435         EXIT;
436 out:
437         up(&dquot->dq_sem);
438         up(&mds->mds_qonoff_sem);
439         lustre_dqput(dquot);
440         if (rc != -EDQUOT)
441                 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
442                                       QDATA_IS_BLK(qdata));
443
444         lqs = quota_search_lqs(LQS_KEY(QDATA_IS_GRP(qdata), qdata->qd_id),
445                                qctxt, 0);
446         if (lqs == NULL || IS_ERR(lqs)) {
447                 CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
448                 qdata->qd_qunit  = QDATA_IS_BLK(qdata) ? qctxt->lqc_bunit_sz :
449                                                          qctxt->lqc_iunit_sz;
450         } else {
451                 spin_lock(&lqs->lqs_lock);
452                 qdata->qd_qunit  = QDATA_IS_BLK(qdata) ? lqs->lqs_bunit_sz :
453                                                          lqs->lqs_iunit_sz;
454                 spin_unlock(&lqs->lqs_lock);
455         }
456
457         if (QDATA_IS_BLK(qdata))
458                 QDATA_SET_ADJBLK(qdata);
459         else
460                 QDATA_SET_ADJINO(qdata);
461
462         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
463         if (lqs)
464                 lqs_putref(lqs);
465
466         return rc;
467 }
468
469 int mds_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
470                      const unsigned int qpids[], int rc, int opc)
471 {
472         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
473         int rc2 = 0;
474         ENTRY;
475
476         if (rc && rc != -EDQUOT && rc != ENOLCK)
477                 RETURN(0);
478
479         switch (opc) {
480         case FSFILT_OP_SETATTR:
481                 /* release file quota on original owner */
482                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 0, 0, NULL);
483                 /* release block quota on original owner */
484                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
485                 /* acquire file quota on current owner */
486                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
487                 /* acquire block quota on current owner */
488                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
489                 break;
490         case FSFILT_OP_UNLINK_PARTIAL_CHILD:
491                 /* release file quota on child */
492                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
493                 /* rlease block quota on child */
494                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
495                 break;
496         case FSFILT_OP_CREATE_PARTIAL_CHILD:
497                 /* acquire file quota on child */
498                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
499                 /* acquire block quota on child */
500                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
501                 break;
502         case FSFILT_OP_LINK:
503                 /* acquire block quota on parent */
504                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
505                 break;
506         case FSFILT_OP_UNLINK:
507                 /* release block quota on parent */
508                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
509                 /* release file quota on child */
510                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
511                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
512                         /* release block quota on child */
513                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
514                                                   NULL);
515                 break;
516         case FSFILT_OP_UNLINK_PARTIAL_PARENT:
517                 /* release block quota on parent */
518                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
519                 break;
520         case FSFILT_OP_CREATE:
521                 /* acquire block quota on parent */
522                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
523                 /* acquire file quota on child */
524                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
525                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
526                         /* acquire block quota on child */
527                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
528                                                   NULL);
529                 break;
530         default:
531                 LBUG();
532                 break;
533         }
534
535         if (rc2)
536                 CDEBUG(rc2 == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
537                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
538                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
539         RETURN(0);
540 }
541
542 int filter_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
543                         const unsigned int qpids[], int rc, int opc)
544 {
545         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
546         int rc2 = 0;
547         ENTRY;
548
549         if (rc && rc != -EDQUOT)
550                 RETURN(0);
551
552         switch (opc) {
553         case FSFILT_OP_SETATTR:
554                 /* acquire/release block quota on original & current owner */
555                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
556                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
557                 break;
558         case FSFILT_OP_UNLINK:
559                 /* release block quota on this owner */
560         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
561                 /* acquire block quota on this owner */
562                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
563                 break;
564         default:
565                 LBUG();
566                 break;
567         }
568
569         if (rc || rc2) {
570                 if (!rc)
571                         rc = rc2;
572                 CDEBUG(rc == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
573                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
574                        QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
575         }
576
577         RETURN(0);
578 }
579
580 static const char prefix[] = "OBJECTS/";
581
582 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
583 {
584         struct mds_obd *mds = &obd->u.mds;
585         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
586         int rc = 0, i;
587         char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
588         char name[64];
589         struct lvfs_run_ctxt saved;
590
591         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
592
593         if (oqctl->qc_type != USRQUOTA &&
594             oqctl->qc_type != GRPQUOTA &&
595             oqctl->qc_type != UGQUOTA)
596                 return -EINVAL;
597
598         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
599
600         down(&mds->mds_qonoff_sem);
601
602         for (i = 0; i < MAXQUOTAS; i++) {
603                 struct file *fp;
604
605                 if (!Q_TYPESET(oqctl, i))
606                         continue;
607
608                 /* quota file has been opened ? */
609                 if (qinfo->qi_files[i]) {
610                         rc = -EBUSY;
611                         goto out;
612                 }
613
614                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
615                 sprintf(name, "%s%s", prefix, quotafile[i]);
616
617                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
618                 if (IS_ERR(fp)) {
619                         rc = PTR_ERR(fp);
620                         CERROR("error invalidating admin quotafile %s (rc:%d)\n",
621                                name, rc);
622                 }
623                 else
624                         filp_close(fp, 0);
625         }
626
627 out:
628         up(&mds->mds_qonoff_sem);
629
630         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
631
632         return rc;
633 }
634
635 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
636 {
637         struct mds_obd *mds = &obd->u.mds;
638         int rc;
639         struct lvfs_run_ctxt saved;
640
641         if (oqctl->qc_type != USRQUOTA &&
642             oqctl->qc_type != GRPQUOTA &&
643             oqctl->qc_type != UGQUOTA)
644                 RETURN(-EINVAL);
645
646         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
647         down(&mds->mds_qonoff_sem);
648
649         oqctl->qc_cmd = Q_FINVALIDATE;
650         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
651         if (!rc)
652                 rc = obd_quotactl(mds->mds_osc_exp, oqctl);
653
654         up(&mds->mds_qonoff_sem);
655         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
656
657         return rc;
658 }
659
660 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
661 {
662         struct mds_obd *mds = &obd->u.mds;
663         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
664         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
665         struct lvfs_run_ctxt saved;
666         char name[64];
667         int i, rc = 0;
668         ENTRY;
669
670         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
671
672         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
673
674         down(&mds->mds_qonoff_sem);
675
676         for (i = 0; i < MAXQUOTAS && !rc; i++) {
677                 struct file *fp;
678
679                 if (!Q_TYPESET(oqctl, i))
680                         continue;
681
682                 /* quota file has been opened ? */
683                 if (qinfo->qi_files[i]) {
684                         CWARN("init %s admin quotafile while quota on.\n",
685                               i == USRQUOTA ? "user" : "group");
686                         continue;
687                 }
688
689                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
690                 sprintf(name, "%s%s", prefix, quotafile[i]);
691
692                 /* check if quota file exists and is correct */
693                 fp = filp_open(name, O_RDONLY, 0);
694                 if (!IS_ERR(fp)) {
695                         /* irregular file is not the right place for quota */
696                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
697                                 CERROR("admin quota file %s is not "
698                                        "regular!", name);
699                                 filp_close(fp, 0);
700                                 rc = -EINVAL;
701                                 break;
702                         }
703                         qinfo->qi_files[i] = fp;
704                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
705                         qinfo->qi_files[i] = 0;
706                         filp_close(fp, 0);
707                 }
708                 else
709                         rc = PTR_ERR(fp);
710
711                 if (!rc)
712                         continue;
713
714                 /* -EINVAL may be returned by quotainfo for bad quota file */
715                 if (rc != -ENOENT && rc != -EINVAL) {
716                         CERROR("error opening old quota file %s (%d)\n",
717                                name, rc);
718                         break;
719                 }
720
721                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
722                        rc == -ENOENT ? "creating" : "overwriting");
723
724                 /* create quota file overwriting old if needed */
725                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
726                 if (IS_ERR(fp)) {
727                         rc = PTR_ERR(fp);
728                         CERROR("error creating admin quotafile %s (rc:%d)\n",
729                                name, rc);
730                         break;
731                 }
732
733                 qinfo->qi_files[i] = fp;
734
735                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
736                 if (rc)
737                         CERROR("error init %s admin quotafile! (rc:%d)\n",
738                                i == USRQUOTA ? "user" : "group", rc);
739
740                 filp_close(fp, 0);
741                 qinfo->qi_files[i] = NULL;
742         }
743         up(&mds->mds_qonoff_sem);
744
745         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
746         RETURN(rc);
747 }
748
749 static int close_quota_files(struct obd_quotactl *oqctl,
750                              struct lustre_quota_info *qinfo)
751 {
752         int i, rc = 0;
753         ENTRY;
754
755         for (i = 0; i < MAXQUOTAS; i++) {
756                 if (!Q_TYPESET(oqctl, i))
757                         continue;
758                 if (qinfo->qi_files[i] == NULL) {
759                         rc = -ESRCH;
760                         continue;
761                 }
762                 filp_close(qinfo->qi_files[i], 0);
763                 qinfo->qi_files[i] = NULL;
764         }
765         RETURN(rc);
766 }
767
768 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
769 {
770         struct mds_obd *mds = &obd->u.mds;
771         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
772         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
773         char name[64];
774         int i, rc = 0;
775         ENTRY;
776
777         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
778
779         /* open admin quota files and read quotafile info */
780         for (i = 0; i < MAXQUOTAS; i++) {
781                 struct file *fp;
782
783                 if (!Q_TYPESET(oqctl, i))
784                         continue;
785
786                 LASSERT(strlen(quotafile[i])
787                         + sizeof(prefix) <= sizeof(name));
788                 sprintf(name, "%s%s", prefix, quotafile[i]);
789
790                 if (qinfo->qi_files[i] != NULL) {
791                         rc = -EBUSY;
792                         break;
793                 }
794
795                 fp = filp_open(name, O_RDWR, 0);
796                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
797                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
798                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
799                         break;
800                 }
801                 qinfo->qi_files[i] = fp;
802
803                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
804                 if (rc) {
805                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
806                         break;
807                 }
808
809                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
810                 if (rc) {
811                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
812                                rc);
813                         break;
814                 }
815         }
816
817         if (rc && rc != -EBUSY)
818                 close_quota_files(oqctl, qinfo);
819
820         RETURN(rc);
821 }
822
823 int mds_admin_quota_off(struct obd_device *obd,
824                         struct obd_quotactl *oqctl)
825 {
826         struct mds_obd *mds = &obd->u.mds;
827         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
828         int rc;
829         ENTRY;
830
831         /* close admin quota files */
832         rc = close_quota_files(oqctl, qinfo);
833         RETURN(rc);
834 }
835
836 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
837 {
838         struct mds_obd *mds = &obd->u.mds;
839         struct obd_device_target *obt = &obd->u.obt;
840         struct lvfs_run_ctxt saved;
841         int rc;
842         ENTRY;
843
844         if (oqctl->qc_type != USRQUOTA &&
845             oqctl->qc_type != GRPQUOTA &&
846             oqctl->qc_type != UGQUOTA)
847                 RETURN(-EINVAL);
848
849         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
850                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
851                 atomic_inc(&obt->obt_quotachecking);
852                 RETURN(-EBUSY);
853         }
854
855         LASSERT(!obt->obt_qctxt.lqc_immutable);
856         down(&mds->mds_qonoff_sem);
857         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
858         rc = mds_admin_quota_on(obd, oqctl);
859         if (rc)
860                 GOTO(out, rc);
861
862         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
863         if (!rc)
864                 obt->obt_qctxt.lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
865         else
866                 GOTO(out, rc);
867
868         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
869
870 out:
871         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
872         up(&mds->mds_qonoff_sem);
873         atomic_inc(&obt->obt_quotachecking);
874         RETURN(rc);
875 }
876
877 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
878 {
879         struct mds_obd *mds = &obd->u.mds;
880         struct obd_device_target *obt = &obd->u.obt;
881         struct lvfs_run_ctxt saved;
882         int rc, rc2, imm;
883         ENTRY;
884
885         imm = oqctl->qc_type & IMMQUOTA;
886         oqctl->qc_type &= ~IMMQUOTA;
887
888         if (oqctl->qc_type != USRQUOTA &&
889             oqctl->qc_type != GRPQUOTA &&
890             oqctl->qc_type != UGQUOTA)
891                 RETURN(-EINVAL);
892
893         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
894                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
895                 atomic_inc(&obt->obt_quotachecking);
896                 RETURN(-EBUSY);
897         }
898
899         down(&mds->mds_qonoff_sem);
900         /* close admin quota files */
901         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
902         mds_admin_quota_off(obd, oqctl);
903
904         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
905         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
906         if (!rc2) {
907                 if (imm)
908                         obt->obt_qctxt.lqc_immutable = 1;
909                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
910         }
911         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
912         up(&mds->mds_qonoff_sem);
913         atomic_inc(&obt->obt_quotachecking);
914
915         RETURN(rc ?: rc2);
916 }
917
918 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
919 {
920         struct mds_obd *mds = &obd->u.mds;
921         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
922         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
923         int rc;
924         ENTRY;
925
926         if (oqctl->qc_type != USRQUOTA &&
927             oqctl->qc_type != GRPQUOTA)
928                 RETURN(-EINVAL);
929
930         down(&mds->mds_qonoff_sem);
931         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
932                 rc = -ESRCH;
933                 goto out;
934         }
935
936         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
937         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
938         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
939
940         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
941
942 out:
943         up(&mds->mds_qonoff_sem);
944         RETURN(rc);
945 }
946
947 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
948 {
949         struct mds_obd *mds = &obd->u.mds;
950         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
951         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
952         int rc = 0;
953         ENTRY;
954
955         if (oqctl->qc_type != USRQUOTA &&
956             oqctl->qc_type != GRPQUOTA)
957                 RETURN(-EINVAL);
958
959         down(&mds->mds_qonoff_sem);
960         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
961                 rc = -ESRCH;
962                 goto out;
963         }
964
965         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
966         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
967         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
968
969 out:
970         up(&mds->mds_qonoff_sem);
971         RETURN(rc);
972 }
973
974 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
975                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
976                       int type, struct quota_adjust_qunit *oqaq)
977 {
978         __u64 bunit_curr_o, iunit_curr_o;
979         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
980         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
981         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
982                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
983         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
984                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
985         int rc = 0;
986         ENTRY;
987
988         if (!dquot || !oqaq)
989                 RETURN(-EINVAL);
990         LASSERT_SEM_LOCKED(&dquot->dq_sem);
991         LASSERT(oqaq->qaq_iunit_sz);
992         LASSERT(oqaq->qaq_bunit_sz);
993
994         /* don't change qunit size */
995         if (!qctxt->lqc_switch_qs)
996                 RETURN(rc);
997
998         bunit_curr_o = oqaq->qaq_bunit_sz;
999         iunit_curr_o = oqaq->qaq_iunit_sz;
1000
1001         if (dquot->dq_type == GRPQUOTA)
1002                 QAQ_SET_GRP(oqaq);
1003
1004         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1005                 __u64 b_limitation =
1006                         oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1007                 /* enlarge block qunit size */
1008                 while (blimit >
1009                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1010                         oqaq->qaq_bunit_sz =
1011                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1012                                 << QUOTABLOCK_BITS;
1013                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1014                                 shrink_qunit_limit;
1015                 }
1016
1017                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1018                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1019
1020                 /* shrink block qunit size */
1021                 while (blimit <
1022                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1023                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1024                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1025                                 QUOTABLOCK_BITS;
1026                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1027                                 shrink_qunit_limit;
1028                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1029                                 break;
1030                 }
1031
1032                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1033                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1034
1035                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1036                         QAQ_SET_ADJBLK(oqaq);
1037
1038         }
1039
1040         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1041                 __u64 i_limitation =
1042                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1043                 /* enlarge file qunit size */
1044                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1045                        + 2 * i_limitation) {
1046                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1047                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1048                                 shrink_qunit_limit;
1049                 }
1050
1051                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1052                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1053
1054                 /* shrink file qunit size */
1055                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1056                        + i_limitation) {
1057                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1058                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1059                                        shrink_qunit_limit;
1060                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1061                                 break;
1062                 }
1063
1064                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1065                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1066
1067                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1068                         QAQ_SET_ADJINO(oqaq);
1069
1070         }
1071
1072         if (!dquot->dq_dqb.dqb_bhardlimit && !dquot->dq_dqb.dqb_bsoftlimit &&
1073             !dquot->dq_dqb.dqb_ihardlimit && !dquot->dq_dqb.dqb_isoftlimit) {
1074                 oqaq->qaq_bunit_sz = 0;
1075                 oqaq->qaq_iunit_sz = 0;
1076                 QAQ_SET_ADJBLK(oqaq);
1077                 QAQ_SET_ADJINO(oqaq);
1078         }
1079
1080         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1081
1082         RETURN(rc);
1083 }
1084
1085 static int mds_init_slave_ilimits(struct obd_device *obd,
1086                                   struct obd_quotactl *oqctl, int set,
1087                                   struct quota_adjust_qunit *oqaq)
1088 {
1089         /* XXX: for file limits only adjust local now */
1090         struct obd_device_target *obt = &obd->u.obt;
1091         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1092         unsigned int id[MAXQUOTAS] = { 0, 0 };
1093         struct obd_quotactl *ioqc = NULL;
1094         int flag;
1095         int rc;
1096         ENTRY;
1097
1098         /* if we are going to set zero limit, needn't init slaves */
1099         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1100             !set)
1101                 RETURN(0);
1102
1103         OBD_ALLOC_PTR(ioqc);
1104         if (!ioqc)
1105                 RETURN(-ENOMEM);
1106
1107         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1108                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1109         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1110         ioqc->qc_id = oqctl->qc_id;
1111         ioqc->qc_type = oqctl->qc_type;
1112         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1113         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1114
1115         if (QAQ_IS_ADJINO(oqaq)) {
1116                 /* adjust the mds slave's inode qunit size */
1117                 rc = quota_adjust_slave_lqs(oqaq, qctxt);
1118                 if (rc < 0)
1119                         CDEBUG(D_ERROR, "adjust mds slave's inode qunit size \
1120                                failed! (rc:%d)\n", rc);
1121         }
1122
1123         /* set local limit to MIN_QLIMIT */
1124         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1125         if (rc)
1126                 GOTO(out, rc);
1127
1128         /* trigger local qunit pre-acquire */
1129         if (oqctl->qc_type == USRQUOTA)
1130                 id[USRQUOTA] = oqctl->qc_id;
1131         else
1132                 id[GRPQUOTA] = oqctl->qc_id;
1133
1134         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1135         if (rc == -EDQUOT || rc == -EBUSY) {
1136                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1137                 rc = 0;
1138         }
1139         if (rc) {
1140                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1141                        rc);
1142                 GOTO(out, rc);
1143         }
1144         /* FIXME initialize all slaves in CMD */
1145         EXIT;
1146 out:
1147         if (ioqc)
1148                 OBD_FREE_PTR(ioqc);
1149         return rc;
1150 }
1151
1152 static int mds_init_slave_blimits(struct obd_device *obd,
1153                                   struct obd_quotactl *oqctl, int set,
1154                                   struct quota_adjust_qunit *oqaq)
1155 {
1156         struct obd_device_target *obt = &obd->u.obt;
1157         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1158         struct mds_obd *mds = &obd->u.mds;
1159         struct obd_quotactl *ioqc;
1160         unsigned int id[MAXQUOTAS] = { 0, 0 };
1161         int rc, rc1 = 0;
1162         int flag;
1163         ENTRY;
1164
1165         /* if we are going to set zero limit, needn't init slaves */
1166         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1167             !set)
1168                 RETURN(0);
1169
1170         OBD_ALLOC_PTR(ioqc);
1171         if (!ioqc)
1172                 RETURN(-ENOMEM);
1173
1174         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1175                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1176         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1177         ioqc->qc_id = oqctl->qc_id;
1178         ioqc->qc_type = oqctl->qc_type;
1179         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1180         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1181         if (QAQ_IS_ADJBLK(oqaq)) {
1182                 /* adjust the mds slave's block qunit size */
1183                 rc1 = quota_adjust_slave_lqs(oqaq, qctxt);
1184                 if (rc1 < 0)
1185                         CERROR("adjust mds slave's block qunit size failed!"
1186                                "(rc:%d)\n", rc1);
1187         }
1188
1189         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1190         if (rc)
1191                 GOTO(out, rc);
1192
1193         /* trigger local qunit pre-acquire */
1194         if (oqctl->qc_type == USRQUOTA)
1195                 id[USRQUOTA] = oqctl->qc_id;
1196         else
1197                 id[GRPQUOTA] = oqctl->qc_id;
1198
1199         /* initialize all slave's limit */
1200         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
1201
1202         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1203         if (rc == -EDQUOT || rc == -EBUSY) {
1204                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1205                 rc = 0;
1206         }
1207         if (rc) {
1208                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1209                 GOTO(out, rc);
1210         }
1211
1212         /* adjust all slave's qunit size when setting quota
1213          * this is will create a lqs for every ost, which will present
1214          * certain uid/gid is set quota or not */
1215         QAQ_SET_ADJBLK(oqaq);
1216         rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
1217
1218         EXIT;
1219 out:
1220         OBD_FREE_PTR(ioqc);
1221         return rc;
1222 }
1223
1224 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1225 {
1226         struct mds_obd *mds = &obd->u.mds;
1227         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1228         struct obd_device *lov_obd = class_exp2obd(mds->mds_osc_exp);
1229         struct lov_obd *lov = &lov_obd->u.lov;
1230         struct quota_adjust_qunit *oqaq = NULL;
1231         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1232         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1233         time_t btime, itime;
1234         struct lustre_dquot *dquot;
1235         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1236         /* orig_set means if quota was set before; now_set means we are
1237          * setting/cancelling quota */
1238         int orig_set, now_set;
1239         int rc, rc2 = 0, flag = 0;
1240         ENTRY;
1241
1242         if (oqctl->qc_type != USRQUOTA &&
1243             oqctl->qc_type != GRPQUOTA)
1244                 RETURN(-EINVAL);
1245
1246         OBD_ALLOC_PTR(oqaq);
1247         if (!oqaq)
1248                 RETURN(-ENOMEM);
1249         down(&mds->mds_qonoff_sem);
1250         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1251
1252         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1253                 GOTO(out_sem, rc = -ESRCH);
1254
1255         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1256         if (IS_ERR(dquot))
1257                 GOTO(out_sem, rc = PTR_ERR(dquot));
1258         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1259         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1260
1261         down(&dquot->dq_sem);
1262
1263         if (dquot->dq_status) {
1264                 up(&dquot->dq_sem);
1265                 lustre_dqput(dquot);
1266                 GOTO(out_sem, rc = -EBUSY);
1267         }
1268         dquot->dq_status |= DQ_STATUS_SET;
1269
1270         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1271         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1272         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1273         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1274         btime = dquot->dq_dqb.dqb_btime;
1275         itime = dquot->dq_dqb.dqb_itime;
1276
1277         if (dqblk->dqb_valid & QIF_BTIME)
1278                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1279         if (dqblk->dqb_valid & QIF_ITIME)
1280                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1281
1282         if (dqblk->dqb_valid & QIF_BLIMITS) {
1283                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1284                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1285                 /* clear usage (limit pool) */
1286                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1287                     !dquot->dq_dqb.dqb_bsoftlimit)
1288                         dquot->dq_dqb.dqb_curspace = 0;
1289
1290                 /* clear grace time */
1291                 if (!dqblk->dqb_bsoftlimit ||
1292                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1293                         dquot->dq_dqb.dqb_btime = 0;
1294                 /* set grace only if user hasn't provided his own */
1295                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1296                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1297                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1298
1299                 flag |= LQUOTA_FLAGS_ADJBLK;
1300         }
1301
1302         if (dqblk->dqb_valid & QIF_ILIMITS) {
1303                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1304                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1305                 /* clear usage (limit pool) */
1306                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1307                     !dquot->dq_dqb.dqb_isoftlimit)
1308                         dquot->dq_dqb.dqb_curinodes = 0;
1309
1310                 if (!dqblk->dqb_isoftlimit ||
1311                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1312                         dquot->dq_dqb.dqb_itime = 0;
1313                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1314                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1315                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1316
1317                 flag |= LQUOTA_FLAGS_ADJINO;
1318         }
1319         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1320         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1321                                flag, oqaq);
1322         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1323         if (rc < 0)
1324                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1325
1326
1327         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1328
1329         up(&dquot->dq_sem);
1330
1331         if (rc) {
1332                 CERROR("set limit failed! (rc:%d)\n", rc);
1333                 goto out;
1334         }
1335
1336         up(&mds->mds_qonoff_sem);
1337         orig_set = ihardlimit || isoftlimit;
1338         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1339         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1340                 down(&dquot->dq_sem);
1341                 dquot->dq_dqb.dqb_curinodes = 0;
1342                 up(&dquot->dq_sem);
1343                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set, oqaq);
1344                 if (rc) {
1345                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1346                         goto revoke_out;
1347                 }
1348         }
1349
1350         orig_set = bhardlimit || bsoftlimit;
1351         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1352         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1353                 down(&dquot->dq_sem);
1354                 dquot->dq_dqb.dqb_curspace = 0;
1355                 up(&dquot->dq_sem);
1356                 rc = mds_init_slave_blimits(obd, oqctl, orig_set, oqaq);
1357                 if (rc) {
1358                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1359                         goto revoke_out;
1360                 }
1361         }
1362
1363 revoke_out:
1364         down(&mds->mds_qonoff_sem);
1365         down(&dquot->dq_sem);
1366         if (rc) {
1367                 /* cancel previous setting */
1368                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1369                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1370                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1371                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1372                 dquot->dq_dqb.dqb_btime = btime;
1373                 dquot->dq_dqb.dqb_itime = itime;
1374         }
1375         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1376         up(&dquot->dq_sem);
1377
1378 out:
1379         down(&dquot->dq_sem);
1380         dquot->dq_status &= ~DQ_STATUS_SET;
1381         up(&dquot->dq_sem);
1382         lustre_dqput(dquot);
1383         EXIT;
1384 out_sem:
1385         up(&mds->mds_qonoff_sem);
1386
1387         if (oqaq)
1388                 OBD_FREE_PTR(oqaq);
1389
1390         return rc ? rc : rc2;
1391 }
1392
1393 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1394 {
1395         struct obd_quotactl *soqc;
1396         struct lvfs_run_ctxt saved;
1397         int rc, rc1;
1398         ENTRY;
1399
1400         OBD_ALLOC_PTR(soqc);
1401         if (!soqc)
1402                 RETURN(-ENOMEM);
1403
1404         soqc->qc_cmd = Q_GETOQUOTA;
1405         soqc->qc_id = oqctl->qc_id;
1406         soqc->qc_type = oqctl->qc_type;
1407
1408         /* get block usage from OSS */
1409         soqc->qc_dqblk.dqb_curspace = 0;
1410         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
1411         if (!rc || rc == -EREMOTEIO) {
1412                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1413                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1414         }
1415
1416         /* get block/inode usage from MDS */
1417         soqc->qc_dqblk.dqb_curspace = 0;
1418         soqc->qc_dqblk.dqb_curinodes = 0;
1419         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1420         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1421         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1422         if (!rc1) {
1423                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1424                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1425                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1426         }
1427
1428         OBD_FREE_PTR(soqc);
1429
1430         RETURN(rc ? : rc1);
1431 }
1432
1433 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1434 {
1435         struct mds_obd *mds = &obd->u.mds;
1436         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1437         struct lustre_dquot *dquot;
1438         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1439         int rc;
1440         ENTRY;
1441
1442         if (oqctl->qc_type != USRQUOTA &&
1443             oqctl->qc_type != GRPQUOTA)
1444                 RETURN(-EINVAL);
1445
1446         down(&mds->mds_qonoff_sem);
1447         dqblk->dqb_valid = 0;
1448         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1449                 GOTO(out, rc = -ESRCH);
1450
1451         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1452         if (IS_ERR(dquot))
1453                 GOTO(out, rc = PTR_ERR(dquot));
1454
1455         down(&dquot->dq_sem);
1456         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1457         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1458         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1459         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1460         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1461         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1462         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1463         up(&dquot->dq_sem);
1464
1465         lustre_dqput(dquot);
1466
1467         /* the usages in admin quota file is inaccurate */
1468         dqblk->dqb_curinodes = 0;
1469         dqblk->dqb_curspace = 0;
1470         rc = mds_get_space(obd, oqctl);
1471         EXIT;
1472 out:
1473         up(&mds->mds_qonoff_sem);
1474         return rc;
1475 }
1476
1477 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1478 {
1479         struct lvfs_run_ctxt saved;
1480         int rc;
1481         ENTRY;
1482
1483         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1484         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1485         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1486
1487         RETURN(rc);
1488 }
1489
1490
1491 /* FIXME we only recovery block limit by now, need recovery inode
1492  * limits also after CMD involved in */
1493 static int 
1494 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1495 {
1496         struct mds_obd *mds = &obd->u.mds;
1497         struct lustre_quota_info *qinfo= &mds->mds_quota_info;
1498         struct lustre_dquot *dquot;
1499         struct obd_quotactl *qctl;
1500         __u64 total_limits = 0;
1501         int rc;
1502         ENTRY;
1503
1504         OBD_ALLOC_PTR(qctl);
1505         if (qctl == NULL)
1506                 RETURN(-ENOMEM);
1507
1508         dquot = lustre_dqget(obd, qinfo, id, type);
1509         if (IS_ERR(dquot)) {
1510                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1511                 OBD_FREE_PTR(qctl);
1512                 RETURN(PTR_ERR(dquot));
1513         }
1514
1515         down(&dquot->dq_sem);
1516
1517         /* don't recovery the dquot without limits or under setting */
1518         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1519             dquot->dq_status)
1520                 GOTO(skip, rc = 0);
1521         dquot->dq_status |= DQ_STATUS_RECOVERY;
1522
1523         up(&dquot->dq_sem);
1524
1525         /* get real bhardlimit from all slaves. */
1526         qctl->qc_cmd = Q_GETOQUOTA;
1527         qctl->qc_type = type;
1528         qctl->qc_id = id;
1529         qctl->qc_stat = QUOTA_RECOVERING;
1530         rc = obd_quotactl(mds->mds_osc_exp, qctl);
1531         if (rc)
1532                 GOTO(out, rc);
1533         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1534
1535         /* get real bhardlimit from master */
1536         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1537         if (rc)
1538                 GOTO(out, rc);
1539         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1540
1541         /* amend the usage of the administrative quotafile */
1542         down(&mds->mds_qonoff_sem);
1543         down(&dquot->dq_sem);
1544
1545         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1546
1547         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1548         if (rc)
1549                 CERROR("write dquot failed! (rc:%d)\n", rc);
1550
1551         up(&dquot->dq_sem);
1552         up(&mds->mds_qonoff_sem);
1553         EXIT;
1554 out:
1555         down(&dquot->dq_sem);
1556         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1557 skip:
1558         up(&dquot->dq_sem);
1559
1560         lustre_dqput(dquot);
1561         OBD_FREE_PTR(qctl);
1562         return rc;
1563 }
1564
1565 struct qmaster_recov_thread_data {
1566         struct obd_device *obd;
1567         struct completion comp;
1568 };
1569
1570 static int qmaster_recovery_main(void *arg)
1571 {
1572         struct qmaster_recov_thread_data *data = arg;
1573         struct obd_device *obd = data->obd;
1574         struct mds_obd *mds = &obd->u.mds;
1575         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1576         int rc = 0;
1577         unsigned short type;
1578         ENTRY;
1579
1580         ptlrpc_daemonize("qmaster_recovd");
1581
1582         /* for mds */
1583         class_incref(obd, "qmaster_recovd_mds", obd);
1584         /* for lov */
1585         class_incref(mds->mds_osc_obd, "qmaster_recovd_lov", mds->mds_osc_obd);
1586
1587         complete(&data->comp);
1588
1589         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1590                 struct list_head id_list;
1591                 struct dquot_id *dqid, *tmp;
1592
1593                 down(&mds->mds_qonoff_sem);
1594                 if (qinfo->qi_files[type] == NULL) {
1595                         up(&mds->mds_qonoff_sem);
1596                         continue;
1597                 }
1598                 CFS_INIT_LIST_HEAD(&id_list);
1599                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1600                                  &id_list);
1601                 up(&mds->mds_qonoff_sem);
1602
1603                 if (rc)
1604                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1605
1606                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1607                         list_del_init(&dqid->di_link);
1608                         if (rc)
1609                                 goto free;
1610
1611                         rc = dquot_recovery(obd, dqid->di_id, type);
1612                         if (rc)
1613                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1614                                        " rc:%d)\n", dqid->di_id, type, rc);
1615 free:
1616                         kfree(dqid);
1617                 }
1618         }
1619         class_decref(mds->mds_osc_obd, "qmaster_recovd_lov", mds->mds_osc_obd);
1620         class_decref(obd, "qmaster_recovd_mds", obd);
1621         RETURN(rc);
1622 }
1623
1624 int mds_quota_recovery(struct obd_device *obd)
1625 {
1626         struct mds_obd *mds = &obd->u.mds;
1627         struct qmaster_recov_thread_data data;
1628         int rc = 0;
1629         ENTRY;
1630
1631         if (unlikely(!mds->mds_quota || obd->obd_stopping))
1632                 RETURN(rc);
1633
1634         mutex_down(&obd->obd_dev_sem);
1635         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1636                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1637                       mds->mds_lov_desc.ld_active_tgt_count,
1638                       mds->mds_lov_objid_count);
1639                 mutex_up(&obd->obd_dev_sem);
1640                 RETURN(rc);
1641         }
1642         mutex_up(&obd->obd_dev_sem);
1643
1644         data.obd = obd;
1645         init_completion(&data.comp);
1646
1647         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1648         if (rc < 0)
1649                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1650
1651         wait_for_completion(&data.comp);
1652         RETURN(rc);
1653 }
1654
1655 #endif /* HAVE_QUOTA_SUPPORT */