Whamcloud - gitweb
339a6c5a905c52062f132b7f3a8f11762aab92ff
[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         quota_search_lqs(NULL, oqaq, qctxt, &lqs);
239         if (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 uid = 0, gid = 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                 gid = dquot->dq_id;
309         else
310                 uid = dquot->dq_id;
311
312         up(&dquot->dq_sem);
313
314         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, 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         quota_search_lqs(qdata, NULL, qctxt, &lqs);
445         if (QDATA_IS_BLK(qdata)) {
446                 if (!lqs) {
447                         CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
448                         qdata->qd_qunit  = qctxt->lqc_bunit_sz;
449                 } else {
450                         spin_lock(&lqs->lqs_lock);
451                         qdata->qd_qunit  = lqs->lqs_bunit_sz;
452                         spin_unlock(&lqs->lqs_lock);
453                 }
454                 QDATA_SET_ADJBLK(qdata);
455         } else {
456                 if (!lqs) {
457                         CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
458                         qdata->qd_qunit  = qctxt->lqc_iunit_sz;
459                 } else {
460                         spin_lock(&lqs->lqs_lock);
461                         qdata->qd_qunit  = lqs->lqs_iunit_sz;
462                         spin_unlock(&lqs->lqs_lock);
463                 }
464                 QDATA_SET_ADJINO(qdata);
465         }
466
467         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
468         if (lqs)
469                 lqs_putref(lqs);
470
471         return rc;
472 }
473
474 int mds_quota_adjust(struct obd_device *obd, unsigned int qcids[],
475                      unsigned int qpids[], int rc, int opc)
476 {
477         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
478         int rc2 = 0;
479         ENTRY;
480
481         if (rc && rc != -EDQUOT && rc != ENOLCK)
482                 RETURN(0);
483
484         switch (opc) {
485         case FSFILT_OP_SETATTR:
486                 /* release file quota on original owner */
487                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 0, 0,
488                                           NULL);
489                 /* release block quota on original owner */
490                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
491                                           NULL);
492                 /* acquire file quota on current owner */
493                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
494                                           NULL);
495                 /* acquire block quota on current owner */
496                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
497                                           NULL);
498                 break;
499         case FSFILT_OP_UNLINK_PARTIAL_CHILD:
500                 /* release file quota on child */
501                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
502                                           NULL);
503                 /* rlease block quota on child */
504                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
505                                           NULL);
506                 break;
507         case FSFILT_OP_CREATE_PARTIAL_CHILD:
508                 /* acquire file quota on child */
509                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
510                                           NULL);
511                 /* acquire block quota on child */
512                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
513                                           NULL);
514                 break;
515         case FSFILT_OP_LINK:
516                 /* acquire block quota on parent */
517                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
518                                           NULL);
519                 break;
520         case FSFILT_OP_UNLINK:
521                 /* release block quota on parent */
522                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
523                                           NULL);
524                 /* release file quota on child */
525                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
526                                           NULL);
527                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
528                         /* release block quota on child */
529                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0],
530                                                   qcids[1], 1, 0, NULL);
531                 break;
532         case FSFILT_OP_UNLINK_PARTIAL_PARENT:
533                 /* release block quota on parent */
534                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
535                                           NULL);
536                 break;
537         case FSFILT_OP_CREATE:
538                 /* acquire block quota on parent */
539                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
540                                           NULL);
541                 /* acquire file quota on child */
542                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
543                                           NULL);
544                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
545                         /* acquire block quota on child */
546                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0],
547                                                   qcids[1], 1, 0, NULL);
548                 break;
549         default:
550                 LBUG();
551                 break;
552         }
553
554         if (rc2)
555                 CDEBUG(rc2 == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
556                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
557                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
558         RETURN(0);
559 }
560
561 int filter_quota_adjust(struct obd_device *obd, unsigned int qcids[],
562                         unsigned int qpids[], int rc, int opc)
563 {
564         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
565         int rc2 = 0;
566         ENTRY;
567
568         if (rc && rc != -EDQUOT)
569                 RETURN(0);
570
571         switch (opc) {
572         case FSFILT_OP_SETATTR:
573                 /* acquire/release block quota on original & current owner */
574                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
575                                         NULL);
576                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
577                                          NULL);
578                 break;
579         case FSFILT_OP_UNLINK:
580                 /* release block quota on this owner */
581         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
582                 /* acquire block quota on this owner */
583                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
584                                         NULL);
585                 break;
586         default:
587                 LBUG();
588                 break;
589         }
590
591         if (rc || rc2) {
592                 if (!rc)
593                         rc = rc2;
594                 CDEBUG(rc == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
595                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
596                        QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
597         }
598
599         RETURN(0);
600 }
601
602 static const char prefix[] = "OBJECTS/";
603
604 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
605 {
606         struct mds_obd *mds = &obd->u.mds;
607         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
608         int rc = 0, i;
609         char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
610         char name[64];
611         struct lvfs_run_ctxt saved;
612
613         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
614
615         if (oqctl->qc_type != USRQUOTA &&
616             oqctl->qc_type != GRPQUOTA &&
617             oqctl->qc_type != UGQUOTA)
618                 return -EINVAL;
619
620         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
621
622         down(&mds->mds_qonoff_sem);
623
624         for (i = 0; i < MAXQUOTAS; i++) {
625                 struct file *fp;
626
627                 if (!Q_TYPESET(oqctl, i))
628                         continue;
629
630                 /* quota file has been opened ? */
631                 if (qinfo->qi_files[i]) {
632                         rc = -EBUSY;
633                         goto out;
634                 }
635
636                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
637                 sprintf(name, "%s%s", prefix, quotafile[i]);
638
639                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
640                 if (IS_ERR(fp)) {
641                         rc = PTR_ERR(fp);
642                         CERROR("error invalidating admin quotafile %s (rc:%d)\n",
643                                name, rc);
644                 }
645                 else
646                         filp_close(fp, 0);
647         }
648
649 out:
650         up(&mds->mds_qonoff_sem);
651
652         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
653
654         return rc;
655 }
656
657 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
658 {
659         struct mds_obd *mds = &obd->u.mds;
660         int rc;
661         struct lvfs_run_ctxt saved;
662
663         if (oqctl->qc_type != USRQUOTA &&
664             oqctl->qc_type != GRPQUOTA &&
665             oqctl->qc_type != UGQUOTA)
666                 RETURN(-EINVAL);
667
668         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
669         down(&mds->mds_qonoff_sem);
670
671         oqctl->qc_cmd = Q_FINVALIDATE;
672         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
673         if (!rc)
674                 rc = obd_quotactl(mds->mds_osc_exp, oqctl);
675
676         up(&mds->mds_qonoff_sem);
677         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
678
679         return rc;
680 }
681
682 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
683 {
684         struct mds_obd *mds = &obd->u.mds;
685         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
686         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
687         struct lvfs_run_ctxt saved;
688         char name[64];
689         int i, rc = 0;
690         ENTRY;
691
692         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
693
694         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
695
696         down(&mds->mds_qonoff_sem);
697
698         for (i = 0; i < MAXQUOTAS && !rc; i++) {
699                 struct file *fp;
700
701                 if (!Q_TYPESET(oqctl, i))
702                         continue;
703
704                 /* quota file has been opened ? */
705                 if (qinfo->qi_files[i]) {
706                         CWARN("init %s admin quotafile while quota on.\n",
707                               i == USRQUOTA ? "user" : "group");
708                         continue;
709                 }
710
711                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
712                 sprintf(name, "%s%s", prefix, quotafile[i]);
713
714                 /* check if quota file exists and is correct */
715                 fp = filp_open(name, O_RDONLY, 0);
716                 if (!IS_ERR(fp)) {
717                         /* irregular file is not the right place for quota */
718                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
719                                 CERROR("admin quota file %s is not "
720                                        "regular!", name);
721                                 filp_close(fp, 0);
722                                 rc = -EINVAL;
723                                 break;
724                         }
725                         qinfo->qi_files[i] = fp;
726                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
727                         qinfo->qi_files[i] = 0;
728                         filp_close(fp, 0);
729                 }
730                 else
731                         rc = PTR_ERR(fp);
732
733                 if (!rc)
734                         continue;
735
736                 /* -EINVAL may be returned by quotainfo for bad quota file */
737                 if (rc != -ENOENT && rc != -EINVAL) {
738                         CERROR("error opening old quota file %s (%d)\n",
739                                name, rc);
740                         break;
741                 }
742
743                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
744                        rc == -ENOENT ? "creating" : "overwriting");
745
746                 /* create quota file overwriting old if needed */
747                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
748                 if (IS_ERR(fp)) {
749                         rc = PTR_ERR(fp);
750                         CERROR("error creating admin quotafile %s (rc:%d)\n",
751                                name, rc);
752                         break;
753                 }
754
755                 qinfo->qi_files[i] = fp;
756
757                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
758                 if (rc)
759                         CERROR("error init %s admin quotafile! (rc:%d)\n",
760                                i == USRQUOTA ? "user" : "group", rc);
761
762                 filp_close(fp, 0);
763                 qinfo->qi_files[i] = NULL;
764         }
765         up(&mds->mds_qonoff_sem);
766
767         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
768         RETURN(rc);
769 }
770
771 static int close_quota_files(struct obd_quotactl *oqctl,
772                              struct lustre_quota_info *qinfo)
773 {
774         int i, rc = 0;
775         ENTRY;
776
777         for (i = 0; i < MAXQUOTAS; i++) {
778                 if (!Q_TYPESET(oqctl, i))
779                         continue;
780                 if (qinfo->qi_files[i] == NULL) {
781                         rc = -ESRCH;
782                         continue;
783                 }
784                 filp_close(qinfo->qi_files[i], 0);
785                 qinfo->qi_files[i] = NULL;
786         }
787         RETURN(rc);
788 }
789
790 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
791 {
792         struct mds_obd *mds = &obd->u.mds;
793         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
794         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
795         char name[64];
796         int i, rc = 0;
797         ENTRY;
798
799         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
800
801         /* open admin quota files and read quotafile info */
802         for (i = 0; i < MAXQUOTAS; i++) {
803                 struct file *fp;
804
805                 if (!Q_TYPESET(oqctl, i))
806                         continue;
807
808                 LASSERT(strlen(quotafile[i])
809                         + sizeof(prefix) <= sizeof(name));
810                 sprintf(name, "%s%s", prefix, quotafile[i]);
811
812                 if (qinfo->qi_files[i] != NULL) {
813                         rc = -EBUSY;
814                         break;
815                 }
816
817                 fp = filp_open(name, O_RDWR, 0);
818                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
819                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
820                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
821                         break;
822                 }
823                 qinfo->qi_files[i] = fp;
824
825                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
826                 if (rc) {
827                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
828                         break;
829                 }
830
831                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
832                 if (rc) {
833                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
834                                rc);
835                         break;
836                 }
837         }
838
839         if (rc && rc != -EBUSY)
840                 close_quota_files(oqctl, qinfo);
841
842         RETURN(rc);
843 }
844
845 int mds_admin_quota_off(struct obd_device *obd,
846                         struct obd_quotactl *oqctl)
847 {
848         struct mds_obd *mds = &obd->u.mds;
849         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
850         int rc;
851         ENTRY;
852
853         /* close admin quota files */
854         rc = close_quota_files(oqctl, qinfo);
855         RETURN(rc);
856 }
857
858 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
859 {
860         struct mds_obd *mds = &obd->u.mds;
861         struct obd_device_target *obt = &obd->u.obt;
862         struct lvfs_run_ctxt saved;
863         int rc;
864         ENTRY;
865
866         if (oqctl->qc_type != USRQUOTA &&
867             oqctl->qc_type != GRPQUOTA &&
868             oqctl->qc_type != UGQUOTA)
869                 RETURN(-EINVAL);
870
871         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
872                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
873                 atomic_inc(&obt->obt_quotachecking);
874                 RETURN(-EBUSY);
875         }
876
877         down(&mds->mds_qonoff_sem);
878         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
879         rc = mds_admin_quota_on(obd, oqctl);
880         if (rc)
881                 GOTO(out, rc);
882
883         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
884         if (!rc)
885                 obt->obt_qctxt.lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
886         else
887                 GOTO(out, rc);
888
889         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
890
891 out:
892         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
893         up(&mds->mds_qonoff_sem);
894         atomic_inc(&obt->obt_quotachecking);
895         RETURN(rc);
896 }
897
898 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
899 {
900         struct mds_obd *mds = &obd->u.mds;
901         struct obd_device_target *obt = &obd->u.obt;
902         struct lvfs_run_ctxt saved;
903         int rc, rc2;
904         ENTRY;
905
906         if (oqctl->qc_type != USRQUOTA &&
907             oqctl->qc_type != GRPQUOTA &&
908             oqctl->qc_type != UGQUOTA)
909                 RETURN(-EINVAL);
910
911         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
912                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
913                 atomic_inc(&obt->obt_quotachecking);
914                 RETURN(-EBUSY);
915         }
916
917         down(&mds->mds_qonoff_sem);
918         /* close admin quota files */
919         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
920         mds_admin_quota_off(obd, oqctl);
921
922         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
923         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
924         if (!rc2)
925                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
926
927         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
928         up(&mds->mds_qonoff_sem);
929         atomic_inc(&obt->obt_quotachecking);
930
931         RETURN(rc ?: rc2);
932 }
933
934 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
935 {
936         struct mds_obd *mds = &obd->u.mds;
937         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
938         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
939         int rc;
940         ENTRY;
941
942         if (oqctl->qc_type != USRQUOTA &&
943             oqctl->qc_type != GRPQUOTA)
944                 RETURN(-EINVAL);
945
946         down(&mds->mds_qonoff_sem);
947         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
948                 rc = -ESRCH;
949                 goto out;
950         }
951
952         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
953         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
954         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
955
956         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
957
958 out:
959         up(&mds->mds_qonoff_sem);
960         RETURN(rc);
961 }
962
963 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
964 {
965         struct mds_obd *mds = &obd->u.mds;
966         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
967         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
968         int rc = 0;
969         ENTRY;
970
971         if (oqctl->qc_type != USRQUOTA &&
972             oqctl->qc_type != GRPQUOTA)
973                 RETURN(-EINVAL);
974
975         down(&mds->mds_qonoff_sem);
976         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
977                 rc = -ESRCH;
978                 goto out;
979         }
980
981         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
982         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
983         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
984
985 out:
986         up(&mds->mds_qonoff_sem);
987         RETURN(rc);
988 }
989
990 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
991                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
992                       int type, struct quota_adjust_qunit *oqaq)
993 {
994         __u64 bunit_curr_o, iunit_curr_o;
995         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
996         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
997         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
998                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
999         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1000                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1001         int rc = 0;
1002         ENTRY;
1003
1004         if (!dquot || !oqaq)
1005                 RETURN(-EINVAL);
1006         LASSERT_SEM_LOCKED(&dquot->dq_sem);
1007         LASSERT(oqaq->qaq_iunit_sz);
1008         LASSERT(oqaq->qaq_bunit_sz);
1009
1010         /* don't change qunit size */
1011         if (!qctxt->lqc_switch_qs)
1012                 RETURN(rc);
1013
1014         bunit_curr_o = oqaq->qaq_bunit_sz;
1015         iunit_curr_o = oqaq->qaq_iunit_sz;
1016
1017         if (dquot->dq_type == GRPQUOTA)
1018                 QAQ_SET_GRP(oqaq);
1019
1020         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1021                 __u64 b_limitation =
1022                         oqaq->qaq_bunit_sz * ost_num * shrink_qunit_limit;
1023                 /* enlarge block qunit size */
1024                 while (blimit >
1025                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1026                         oqaq->qaq_bunit_sz =
1027                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1028                                 << QUOTABLOCK_BITS;
1029                         b_limitation = oqaq->qaq_bunit_sz * ost_num *
1030                                 shrink_qunit_limit;
1031                 }
1032
1033                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1034                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1035
1036                 /* shrink block qunit size */
1037                 while (blimit <
1038                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1039                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1040                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1041                                 QUOTABLOCK_BITS;
1042                         b_limitation = oqaq->qaq_bunit_sz * ost_num *
1043                                 shrink_qunit_limit;
1044                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1045                                 break;
1046                 }
1047
1048                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1049                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1050
1051                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1052                         QAQ_SET_ADJBLK(oqaq);
1053
1054         }
1055
1056         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1057                 __u64 i_limitation =
1058                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1059                 /* enlarge file qunit size */
1060                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1061                        + 2 * i_limitation) {
1062                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1063                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1064                                 shrink_qunit_limit;
1065                 }
1066
1067                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1068                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1069
1070                 /* shrink file qunit size */
1071                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1072                        + i_limitation) {
1073                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1074                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1075                                        shrink_qunit_limit;
1076                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1077                                 break;
1078                 }
1079
1080                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1081                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1082
1083                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1084                         QAQ_SET_ADJINO(oqaq);
1085
1086         }
1087
1088         if (!dquot->dq_dqb.dqb_bhardlimit && !dquot->dq_dqb.dqb_bsoftlimit &&
1089             !dquot->dq_dqb.dqb_ihardlimit && !dquot->dq_dqb.dqb_isoftlimit) {
1090                 oqaq->qaq_bunit_sz = 0;
1091                 oqaq->qaq_iunit_sz = 0;
1092                 QAQ_SET_ADJBLK(oqaq);
1093                 QAQ_SET_ADJINO(oqaq);
1094         }
1095
1096         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1097
1098         RETURN(rc);
1099 }
1100
1101 static int mds_init_slave_ilimits(struct obd_device *obd,
1102                                   struct obd_quotactl *oqctl, int set,
1103                                   struct quota_adjust_qunit *oqaq)
1104 {
1105         /* XXX: for file limits only adjust local now */
1106         struct obd_device_target *obt = &obd->u.obt;
1107         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1108         unsigned int uid = 0, gid = 0;
1109         struct obd_quotactl *ioqc = NULL;
1110         int flag;
1111         int rc;
1112         ENTRY;
1113
1114         /* if we are going to set zero limit, needn't init slaves */
1115         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1116             !set)
1117                 RETURN(0);
1118
1119         OBD_ALLOC_PTR(ioqc);
1120         if (!ioqc)
1121                 RETURN(-ENOMEM);
1122
1123         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1124                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1125         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1126         ioqc->qc_id = oqctl->qc_id;
1127         ioqc->qc_type = oqctl->qc_type;
1128         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1129         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1130
1131         if (QAQ_IS_ADJINO(oqaq)) {
1132                 /* adjust the mds slave's inode qunit size */
1133                 rc = quota_adjust_slave_lqs(oqaq, qctxt);
1134                 if (rc < 0)
1135                         CDEBUG(D_ERROR, "adjust mds slave's inode qunit size \
1136                                failed! (rc:%d)\n", rc);
1137         }
1138
1139         /* set local limit to MIN_QLIMIT */
1140         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1141         if (rc)
1142                 GOTO(out, rc);
1143
1144         /* trigger local qunit pre-acquire */
1145         if (oqctl->qc_type == USRQUOTA)
1146                 uid = oqctl->qc_id;
1147         else
1148                 gid = oqctl->qc_id;
1149
1150         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 0, 0,
1151                                 NULL);
1152         if (rc == -EDQUOT || rc == -EBUSY) {
1153                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1154                 rc = 0;
1155         }
1156         if (rc) {
1157                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1158                        rc);
1159                 GOTO(out, rc);
1160         }
1161         /* FIXME initialize all slaves in CMD */
1162         EXIT;
1163 out:
1164         if (ioqc)
1165                 OBD_FREE_PTR(ioqc);
1166         return rc;
1167 }
1168
1169 static int mds_init_slave_blimits(struct obd_device *obd,
1170                                   struct obd_quotactl *oqctl, int set,
1171                                   struct quota_adjust_qunit *oqaq)
1172 {
1173         struct obd_device_target *obt = &obd->u.obt;
1174         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1175         struct mds_obd *mds = &obd->u.mds;
1176         struct obd_quotactl *ioqc;
1177         unsigned int uid = 0, gid = 0;
1178         int rc, rc1 = 0;
1179         int flag;
1180         ENTRY;
1181
1182         /* if we are going to set zero limit, needn't init slaves */
1183         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1184             !set)
1185                 RETURN(0);
1186
1187         OBD_ALLOC_PTR(ioqc);
1188         if (!ioqc)
1189                 RETURN(-ENOMEM);
1190
1191         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1192                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1193         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1194         ioqc->qc_id = oqctl->qc_id;
1195         ioqc->qc_type = oqctl->qc_type;
1196         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1197         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1198         if (QAQ_IS_ADJBLK(oqaq)) {
1199                 /* adjust the mds slave's block qunit size */
1200                 rc1 = quota_adjust_slave_lqs(oqaq, qctxt);
1201                 if (rc1 < 0)
1202                         CERROR("adjust mds slave's block qunit size failed!"
1203                                "(rc:%d)\n", rc1);
1204         }
1205
1206         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1207         if (rc)
1208                 GOTO(out, rc);
1209
1210         /* trigger local qunit pre-acquire */
1211         if (oqctl->qc_type == USRQUOTA)
1212                 uid = oqctl->qc_id;
1213         else
1214                 gid = oqctl->qc_id;
1215
1216         /* initialize all slave's limit */
1217         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
1218
1219         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 1, 0,
1220                                 NULL);
1221         if (rc == -EDQUOT || rc == -EBUSY) {
1222                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1223                 rc = 0;
1224         }
1225         if (rc) {
1226                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1227                 GOTO(out, rc);
1228         }
1229
1230         /* adjust all slave's qunit size when setting quota
1231          * this is will create a lqs for every ost, which will present
1232          * certain uid/gid is set quota or not */
1233         QAQ_SET_ADJBLK(oqaq);
1234         rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
1235
1236         EXIT;
1237 out:
1238         OBD_FREE_PTR(ioqc);
1239         return rc;
1240 }
1241
1242 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1243 {
1244         struct mds_obd *mds = &obd->u.mds;
1245         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1246         struct obd_device *lov_obd = class_exp2obd(mds->mds_osc_exp);
1247         struct lov_obd *lov = &lov_obd->u.lov;
1248         struct quota_adjust_qunit *oqaq = NULL;
1249         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1250         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1251         time_t btime, itime;
1252         struct lustre_dquot *dquot;
1253         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1254         /* orig_set means if quota was set before; now_set means we are
1255          * setting/cancelling quota */
1256         int orig_set, now_set;
1257         int rc, rc2 = 0, flag = 0;
1258         ENTRY;
1259
1260         if (oqctl->qc_type != USRQUOTA &&
1261             oqctl->qc_type != GRPQUOTA)
1262                 RETURN(-EINVAL);
1263
1264         OBD_ALLOC_PTR(oqaq);
1265         if (!oqaq)
1266                 RETURN(-ENOMEM);
1267         down(&mds->mds_qonoff_sem);
1268         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1269
1270         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1271                 GOTO(out_sem, rc = -ESRCH);
1272
1273         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1274         if (IS_ERR(dquot))
1275                 GOTO(out_sem, rc = PTR_ERR(dquot));
1276         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1277         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1278
1279         down(&dquot->dq_sem);
1280
1281         if (dquot->dq_status) {
1282                 up(&dquot->dq_sem);
1283                 lustre_dqput(dquot);
1284                 GOTO(out_sem, rc = -EBUSY);
1285         }
1286         dquot->dq_status |= DQ_STATUS_SET;
1287
1288         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1289         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1290         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1291         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1292         btime = dquot->dq_dqb.dqb_btime;
1293         itime = dquot->dq_dqb.dqb_itime;
1294
1295         if (dqblk->dqb_valid & QIF_BTIME)
1296                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1297         if (dqblk->dqb_valid & QIF_ITIME)
1298                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1299
1300         if (dqblk->dqb_valid & QIF_BLIMITS) {
1301                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1302                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1303                 /* clear usage (limit pool) */
1304                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1305                     !dquot->dq_dqb.dqb_bsoftlimit)
1306                         dquot->dq_dqb.dqb_curspace = 0;
1307
1308                 /* clear grace time */
1309                 if (!dqblk->dqb_bsoftlimit ||
1310                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1311                         dquot->dq_dqb.dqb_btime = 0;
1312                 /* set grace only if user hasn't provided his own */
1313                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1314                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1315                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1316
1317                 flag |= LQUOTA_FLAGS_ADJBLK;
1318         }
1319
1320         if (dqblk->dqb_valid & QIF_ILIMITS) {
1321                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1322                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1323                 /* clear usage (limit pool) */
1324                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1325                     !dquot->dq_dqb.dqb_isoftlimit)
1326                         dquot->dq_dqb.dqb_curinodes = 0;
1327
1328                 if (!dqblk->dqb_isoftlimit ||
1329                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1330                         dquot->dq_dqb.dqb_itime = 0;
1331                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1332                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1333                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1334
1335                 flag |= LQUOTA_FLAGS_ADJINO;
1336         }
1337         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1338         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1339                                flag, oqaq);
1340         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1341         if (rc < 0)
1342                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1343
1344
1345         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1346
1347         up(&dquot->dq_sem);
1348
1349         if (rc) {
1350                 CERROR("set limit failed! (rc:%d)\n", rc);
1351                 goto out;
1352         }
1353
1354         up(&mds->mds_qonoff_sem);
1355         orig_set = ihardlimit || isoftlimit;
1356         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1357         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1358                 down(&dquot->dq_sem);
1359                 dquot->dq_dqb.dqb_curinodes = 0;
1360                 up(&dquot->dq_sem);
1361                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set, oqaq);
1362                 if (rc) {
1363                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1364                         goto revoke_out;
1365                 }
1366         }
1367
1368         orig_set = bhardlimit || bsoftlimit;
1369         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1370         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1371                 down(&dquot->dq_sem);
1372                 dquot->dq_dqb.dqb_curspace = 0;
1373                 up(&dquot->dq_sem);
1374                 rc = mds_init_slave_blimits(obd, oqctl, orig_set, oqaq);
1375                 if (rc) {
1376                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1377                         goto revoke_out;
1378                 }
1379         }
1380
1381 revoke_out:
1382         down(&mds->mds_qonoff_sem);
1383         down(&dquot->dq_sem);
1384         if (rc) {
1385                 /* cancel previous setting */
1386                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1387                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1388                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1389                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1390                 dquot->dq_dqb.dqb_btime = btime;
1391                 dquot->dq_dqb.dqb_itime = itime;
1392         }
1393         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1394         up(&dquot->dq_sem);
1395
1396 out:
1397         down(&dquot->dq_sem);
1398         dquot->dq_status &= ~DQ_STATUS_SET;
1399         up(&dquot->dq_sem);
1400         lustre_dqput(dquot);
1401         EXIT;
1402 out_sem:
1403         up(&mds->mds_qonoff_sem);
1404
1405         if (oqaq)
1406                 OBD_FREE_PTR(oqaq);
1407
1408         return rc ? rc : rc2;
1409 }
1410
1411 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1412 {
1413         struct obd_quotactl *soqc;
1414         struct lvfs_run_ctxt saved;
1415         int rc, rc1;
1416         ENTRY;
1417
1418         OBD_ALLOC_PTR(soqc);
1419         if (!soqc)
1420                 RETURN(-ENOMEM);
1421
1422         soqc->qc_cmd = Q_GETOQUOTA;
1423         soqc->qc_id = oqctl->qc_id;
1424         soqc->qc_type = oqctl->qc_type;
1425
1426         /* get block usage from OSS */
1427         soqc->qc_dqblk.dqb_curspace = 0;
1428         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
1429         if (!rc) {
1430                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1431                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1432         }
1433
1434         /* get block/inode usage from MDS */
1435         soqc->qc_dqblk.dqb_curspace = 0;
1436         soqc->qc_dqblk.dqb_curinodes = 0;
1437         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1438         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1439         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1440         if (!rc1) {
1441                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1442                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1443                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1444         }
1445
1446         OBD_FREE_PTR(soqc);
1447
1448         RETURN(rc ? : rc1);
1449 }
1450
1451 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1452 {
1453         struct mds_obd *mds = &obd->u.mds;
1454         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1455         struct lustre_dquot *dquot;
1456         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1457         int rc;
1458         ENTRY;
1459
1460         if (oqctl->qc_type != USRQUOTA &&
1461             oqctl->qc_type != GRPQUOTA)
1462                 RETURN(-EINVAL);
1463
1464         down(&mds->mds_qonoff_sem);
1465         dqblk->dqb_valid = 0;
1466         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1467                 GOTO(out, rc = -ESRCH);
1468
1469         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1470         if (IS_ERR(dquot))
1471                 GOTO(out, rc = PTR_ERR(dquot));
1472
1473         down(&dquot->dq_sem);
1474         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1475         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1476         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1477         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1478         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1479         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1480         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1481         up(&dquot->dq_sem);
1482
1483         lustre_dqput(dquot);
1484
1485         /* the usages in admin quota file is inaccurate */
1486         dqblk->dqb_curinodes = 0;
1487         dqblk->dqb_curspace = 0;
1488         rc = mds_get_space(obd, oqctl);
1489         EXIT;
1490 out:
1491         up(&mds->mds_qonoff_sem);
1492         return rc;
1493 }
1494
1495 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1496 {
1497         struct lvfs_run_ctxt saved;
1498         int rc;
1499         ENTRY;
1500
1501         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1502         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1503         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1504
1505         RETURN(rc);
1506 }
1507
1508
1509 /* FIXME we only recovery block limit by now, need recovery inode
1510  * limits also after CMD involved in */
1511 static int 
1512 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1513 {
1514         struct mds_obd *mds = &obd->u.mds;
1515         struct lustre_quota_info *qinfo= &mds->mds_quota_info;
1516         struct lustre_dquot *dquot;
1517         struct obd_quotactl *qctl;
1518         __u64 total_limits = 0;
1519         int rc;
1520         ENTRY;
1521
1522         OBD_ALLOC_PTR(qctl);
1523         if (qctl == NULL)
1524                 RETURN(-ENOMEM);
1525
1526         dquot = lustre_dqget(obd, qinfo, id, type);
1527         if (IS_ERR(dquot)) {
1528                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1529                 OBD_FREE_PTR(qctl);
1530                 RETURN(PTR_ERR(dquot));
1531         }
1532
1533         down(&dquot->dq_sem);
1534
1535         /* don't recovery the dquot without limits or under setting */
1536         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1537             dquot->dq_status)
1538                 GOTO(skip, rc = 0);
1539         dquot->dq_status |= DQ_STATUS_RECOVERY;
1540
1541         up(&dquot->dq_sem);
1542
1543         /* get real bhardlimit from all slaves. */
1544         qctl->qc_cmd = Q_GETOQUOTA;
1545         qctl->qc_type = type;
1546         qctl->qc_id = id;
1547         qctl->qc_stat = QUOTA_RECOVERING;
1548         rc = obd_quotactl(mds->mds_osc_exp, qctl);
1549         if (rc)
1550                 GOTO(out, rc);
1551         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1552
1553         /* get real bhardlimit from master */
1554         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1555         if (rc)
1556                 GOTO(out, rc);
1557         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1558
1559         /* amend the usage of the administrative quotafile */
1560         down(&mds->mds_qonoff_sem);
1561         down(&dquot->dq_sem);
1562
1563         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1564
1565         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1566         if (rc)
1567                 CERROR("write dquot failed! (rc:%d)\n", rc);
1568
1569         up(&dquot->dq_sem);
1570         up(&mds->mds_qonoff_sem);
1571         EXIT;
1572 out:
1573         down(&dquot->dq_sem);
1574         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1575 skip:
1576         up(&dquot->dq_sem);
1577
1578         lustre_dqput(dquot);
1579         OBD_FREE_PTR(qctl);
1580         return rc;
1581 }
1582
1583 struct qmaster_recov_thread_data {
1584         struct obd_device *obd;
1585         struct completion comp;
1586 };
1587
1588 static int qmaster_recovery_main(void *arg)
1589 {
1590         struct qmaster_recov_thread_data *data = arg;
1591         struct obd_device *obd = data->obd;
1592         struct mds_obd *mds = &obd->u.mds;
1593         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1594         int rc = 0;
1595         unsigned short type;
1596         ENTRY;
1597
1598         ptlrpc_daemonize("qmaster_recovd");
1599
1600         /* for mds */
1601         class_incref(obd, "qmaster_recovd_mds", obd);
1602         /* for lov */
1603         class_incref(mds->mds_osc_obd, "qmaster_recovd_lov", mds->mds_osc_obd);
1604
1605         complete(&data->comp);
1606
1607         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1608                 struct list_head id_list;
1609                 struct dquot_id *dqid, *tmp;
1610
1611                 down(&mds->mds_qonoff_sem);
1612                 if (qinfo->qi_files[type] == NULL) {
1613                         up(&mds->mds_qonoff_sem);
1614                         continue;
1615                 }
1616                 CFS_INIT_LIST_HEAD(&id_list);
1617                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1618                                  &id_list);
1619                 up(&mds->mds_qonoff_sem);
1620
1621                 if (rc)
1622                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1623
1624                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1625                         list_del_init(&dqid->di_link);
1626                         if (rc)
1627                                 goto free;
1628
1629                         rc = dquot_recovery(obd, dqid->di_id, type);
1630                         if (rc)
1631                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1632                                        " rc:%d)\n", dqid->di_id, type, rc);
1633 free:
1634                         kfree(dqid);
1635                 }
1636         }
1637         class_decref(mds->mds_osc_obd, "qmaster_recovd_lov", mds->mds_osc_obd);
1638         class_decref(obd, "qmaster_recovd_mds", obd);
1639         RETURN(rc);
1640 }
1641
1642 int mds_quota_recovery(struct obd_device *obd)
1643 {
1644         struct mds_obd *mds = &obd->u.mds;
1645         struct qmaster_recov_thread_data data;
1646         int rc = 0;
1647         ENTRY;
1648
1649         if (unlikely(!mds->mds_quota || obd->obd_stopping))
1650                 RETURN(rc);
1651
1652         mutex_down(&obd->obd_dev_sem);
1653         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1654                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1655                       mds->mds_lov_desc.ld_active_tgt_count,
1656                       mds->mds_lov_objid_count);
1657                 mutex_up(&obd->obd_dev_sem);
1658                 RETURN(rc);
1659         }
1660         mutex_up(&obd->obd_dev_sem);
1661
1662         data.obd = obd;
1663         init_completion(&data.comp);
1664
1665         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1666         if (rc < 0)
1667                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1668
1669         wait_for_completion(&data.comp);
1670         RETURN(rc);
1671 }
1672
1673 #endif /* HAVE_QUOTA_SUPPORT */