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 (c) 2007, 2010, Oracle and/or its affiliates. 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 > lqs->lqs_lock */
  68 static cfs_list_t lustre_dquot_hash[NR_DQHASH];
  69 static cfs_rwlock_t dquot_hash_lock = CFS_RW_LOCK_UNLOCKED;
  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(cfs_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         cfs_list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
 130                 if (dquot->dq_info == lqi &&
 131                     dquot->dq_id == id && dquot->dq_type == type)
 132                         RETURN(dquot);
 133         }
 134         RETURN(NULL);
 135 }
 136
 137 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 138                                         qid_t id, int type)
 139 {
 140         struct lustre_dquot *dquot = NULL;
 141         ENTRY;
 142
 143         OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
 144         if (dquot == NULL)
 145                 RETURN(NULL);
 146
 147         CFS_INIT_LIST_HEAD(&dquot->dq_hash);
 148         cfs_init_mutex_locked(&dquot->dq_sem);
 149         cfs_atomic_set(&dquot->dq_refcnt, 1);
 150         dquot->dq_info = lqi;
 151         dquot->dq_id = id;
 152         dquot->dq_type = type;
 153
 154         RETURN(dquot);
 155 }
 156
 157 static void free_dquot(struct lustre_dquot *dquot)
 158 {
 159         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 160 }
 161
 162 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 163 {
 164         cfs_list_t *head = lustre_dquot_hash +
 165             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 166         LASSERT(cfs_list_empty(&dquot->dq_hash));
 167         cfs_list_add(&dquot->dq_hash, head);
 168 }
 169
 170 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 171 {
 172         LASSERT(!cfs_list_empty(&dquot->dq_hash));
 173         cfs_list_del_init(&dquot->dq_hash);
 174 }
 175
 176 static void lustre_dqput(struct lustre_dquot *dquot)
 177 {
 178         ENTRY;
 179         cfs_write_lock(&dquot_hash_lock);
 180         LASSERT(cfs_atomic_read(&dquot->dq_refcnt));
 181         cfs_atomic_dec(&dquot->dq_refcnt);
 182         if (cfs_atomic_read(&dquot->dq_refcnt) == 0) {
 183                 remove_dquot_nolock(dquot);
 184                 free_dquot(dquot);
 185         }
 186         cfs_write_unlock(&dquot_hash_lock);
 187         EXIT;
 188 }
 189
 190 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 191                                          struct lustre_quota_info *lqi,
 192                                          qid_t id, int type, int can_fake)
 193 {
 194         unsigned int hashent = dquot_hashfn(lqi, id, type);
 195         struct lustre_dquot *dquot, *empty;
 196         int free_dq = 0;
 197         ENTRY;
 198
 199         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 200                 RETURN(ERR_PTR(-ENOMEM));
 201
 202         cfs_read_lock(&dquot_hash_lock);
 203         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 204                 cfs_atomic_inc(&dquot->dq_refcnt);
 205                 cfs_read_unlock(&dquot_hash_lock);
 206                 free_dq = 1;
 207         } else {
 208                 int rc;
 209
 210                 cfs_read_unlock(&dquot_hash_lock);
 211
 212                 dquot = empty;
 213                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
 214                 cfs_up(&dquot->dq_sem);
 215                 if (rc) {
 216                         CERROR("can't read dquot from admin quotafile! "
 217                                "(rc:%d)\n", rc);
 218                         lustre_dqput(dquot);
 219                         RETURN(ERR_PTR(rc));
 220                 } else {
 221                         cfs_write_lock(&dquot_hash_lock);
 222                         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 223                                 cfs_atomic_inc(&dquot->dq_refcnt);
 224                                 free_dq = 1;
 225                         } else {
 226                                 dquot = empty;
 227                                 insert_dquot_nolock(dquot);
 228                         }
 229                         cfs_write_unlock(&dquot_hash_lock);
 230                 }
 231
 232         }
 233
 234         LASSERT(dquot);
 235         if (!can_fake && cfs_test_bit(DQ_FAKE_B, &dquot->dq_flags)) {
 236                 DQUOT_DEBUG(dquot, "It is a fake dquot: unexpected!\n");
 237                 lustre_dqput(dquot);
 238                 dquot = ERR_PTR(-ENOENT);
 239         }
 240
 241         if (free_dq)
 242                 free_dquot(empty);
 243
 244
 245         RETURN(dquot);
 246 }
 247
 248 static void init_oqaq(struct quota_adjust_qunit *oqaq,
 249                       struct lustre_quota_ctxt *qctxt,
 250                       qid_t id, int type)
 251 {
 252         struct lustre_qunit_size *lqs = NULL;
 253
 254         oqaq->qaq_id = id;
 255         oqaq->qaq_flags = type;
 256         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 0);
 257         if (lqs && !IS_ERR(lqs)) {
 258                 cfs_spin_lock(&lqs->lqs_lock);
 259                 oqaq->qaq_bunit_sz = lqs->lqs_bunit_sz;
 260                 oqaq->qaq_iunit_sz = lqs->lqs_iunit_sz;
 261                 oqaq->qaq_flags    = lqs->lqs_flags;
 262                 cfs_spin_unlock(&lqs->lqs_lock);
 263                 lqs_putref(lqs);
 264         } else {
 265                 CDEBUG(D_QUOTA, "Can't find the lustre qunit size!\n");
 266                 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
 267                 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
 268         }
 269 }
 270
 271 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
 272                           __u32 is_blk)
 273 {
 274         struct mds_obd *mds = &obd->u.mds;
 275         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 276         __u32 ost_num = mds->mds_lov_objid_count, mdt_num = 1;
 277         struct quota_adjust_qunit *oqaq = NULL;
 278         unsigned int qid[MAXQUOTAS] = { 0, 0 };
 279         struct lustre_quota_info *info = &mds->mds_quota_info;
 280         struct lustre_dquot *dquot = NULL;
 281         int adjust_res = 0;
 282         int rc = 0;
 283         ENTRY;
 284
 285         LASSERT(mds);
 286         cfs_down_read(&mds->mds_qonoff_sem);
 287         dquot = lustre_dqget(obd, info, id, type, 0);
 288         if (IS_ERR(dquot))
 289                 RETURN(PTR_ERR(dquot));
 290
 291         cfs_up_read(&mds->mds_qonoff_sem);
 292         OBD_ALLOC_PTR(oqaq);
 293         if (!oqaq)
 294                 GOTO(out, rc = -ENOMEM);
 295
 296         cfs_down(&dquot->dq_sem);
 297         init_oqaq(oqaq, qctxt, id, type);
 298
 299         rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
 300                                is_blk ? LQUOTA_FLAGS_ADJBLK :
 301                                LQUOTA_FLAGS_ADJINO, oqaq);
 302
 303         if (rc < 0) {
 304                 CERROR("create oqaq failed! (rc:%d)\n", rc);
 305                 GOTO(out_sem, rc);
 306         }
 307         QAQ_DEBUG(oqaq, "show oqaq.\n")
 308
 309         if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
 310                 GOTO(out_sem, rc);
 311
 312         /* adjust the mds slave qunit size */
 313         adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
 314         if (adjust_res <= 0) {
 315                 if (adjust_res < 0) {
 316                         rc = adjust_res;
 317                         CERROR("adjust mds slave's qunit size failed! "
 318                                "(rc:%d)\n", rc);
 319                 } else {
 320                         CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
 321                 }
 322                 GOTO(out_sem, rc);
 323         }
 324
 325         if (type)
 326                 qid[GRPQUOTA] = dquot->dq_id;
 327         else
 328                 qid[USRQUOTA] = dquot->dq_id;
 329
 330         cfs_up(&dquot->dq_sem);
 331
 332         rc = qctxt_adjust_qunit(obd, qctxt, qid, is_blk, 0, NULL);
 333         if (rc == -EDQUOT || rc == -EBUSY) {
 334                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 335                 rc = 0;
 336         }
 337         if (rc) {
 338                 CERROR("%s: mds fail to adjust file quota! (rc:%d)\n",
 339                        obd->obd_name, rc);
 340                 GOTO(out, rc);
 341         }
 342
 343         /* only when block qunit is reduced, boardcast to osts */
 344         if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
 345                 rc = obd_quota_adjust_qunit(mds->mds_lov_exp, oqaq, qctxt, NULL);
 346
 347 out:
 348         lustre_dqput(dquot);
 349         if (oqaq)
 350                 OBD_FREE_PTR(oqaq);
 351
 352         RETURN(rc);
 353 out_sem:
 354         cfs_up(&dquot->dq_sem);
 355         goto out;
 356 }
 357
 358 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
 359 {
 360         struct mds_obd *mds = &obd->u.mds;
 361         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 362         struct lustre_quota_info *info = &mds->mds_quota_info;
 363         struct lustre_dquot *dquot = NULL;
 364         __u64 *usage = NULL;
 365         __u64 hlimit = 0, slimit = 0;
 366         time_t *time = NULL;
 367         unsigned int grace = 0;
 368         struct lustre_qunit_size *lqs = NULL;
 369         int rc = 0;
 370         ENTRY;
 371
 372         if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
 373                 RETURN(-EIO);
 374
 375         lqs = quota_search_lqs(LQS_KEY(QDATA_IS_GRP(qdata), qdata->qd_id),
 376                                qctxt, 0);
 377         if (lqs == NULL)
 378                 rc = -ENOENT;
 379         if (IS_ERR(lqs))
 380                 rc = PTR_ERR(lqs);
 381         if (rc)
 382                 RETURN(rc);
 383
 384         cfs_spin_lock(&lqs->lqs_lock);
 385         if (LQS_IS_RECOVERY(lqs)) {
 386                 cfs_spin_unlock(&lqs->lqs_lock);
 387                 LQS_DEBUG(lqs, "this lqs is under recovery\n");
 388                 GOTO(skip, rc = -EBUSY);
 389         }
 390         cfs_spin_unlock(&lqs->lqs_lock);
 391
 392         cfs_down_write(&mds->mds_qonoff_sem);
 393         dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata), 0);
 394         if (IS_ERR(dquot)) {
 395                 cfs_up_write(&mds->mds_qonoff_sem);
 396                 GOTO(skip, rc = PTR_ERR(dquot));
 397         }
 398
 399         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 400         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 401
 402         cfs_down(&dquot->dq_sem);
 403
 404         if (QDATA_IS_BLK(qdata)) {
 405                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
 406                 usage = &dquot->dq_dqb.dqb_curspace;
 407                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 408                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 409                 time = &dquot->dq_dqb.dqb_btime;
 410         } else {
 411                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
 412                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 413                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 414                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 415                 time = &dquot->dq_dqb.dqb_itime;
 416         }
 417
 418         /* if the quota limit in admin quotafile is zero, we just inform
 419          * slave to clear quota limit with zero qd_count */
 420         if (hlimit == 0 && slimit == 0) {
 421                 qdata->qd_count = 0;
 422                 GOTO(out, rc);
 423         }
 424
 425         switch (opc) {
 426         case QUOTA_DQACQ:
 427                 if (hlimit &&
 428                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
 429                 {
 430                         if (QDATA_IS_CHANGE_QS(qdata) &&
 431                             QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
 432                                 qdata->qd_count = (hlimit -
 433                                         QUSG(*usage, QDATA_IS_BLK(qdata)))
 434                                         * (QDATA_IS_BLK(qdata) ?
 435                                            QUOTABLOCK_SIZE : 1);
 436                         else
 437                                 GOTO(out, rc = -EDQUOT);
 438                 }
 439
 440                 if (slimit &&
 441                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
 442                         if (*time && cfs_time_current_sec() >= *time)
 443                                 GOTO(out, rc = -EDQUOT);
 444                         else if (!*time)
 445                                 *time = cfs_time_current_sec() + grace;
 446                 }
 447
 448                 *usage += qdata->qd_count;
 449                 break;
 450         case QUOTA_DQREL:
 451                 /* The usage in administrative file might be incorrect before
 452                  * recovery done */
 453                 if (*usage - qdata->qd_count < 0)
 454                         *usage = 0;
 455                 else
 456                         *usage -= qdata->qd_count;
 457
 458                 /* (usage <= soft limit) but not (usage < soft limit) */
 459                 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
 460                         *time = 0;
 461                 break;
 462         default:
 463                 LBUG();
 464         }
 465
 466         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 467         EXIT;
 468 out:
 469         cfs_up(&dquot->dq_sem);
 470         cfs_up_write(&mds->mds_qonoff_sem);
 471         lustre_dqput(dquot);
 472         if (rc != -EDQUOT)
 473                 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
 474                                       QDATA_IS_BLK(qdata));
 475
 476         cfs_spin_lock(&lqs->lqs_lock);
 477         qdata->qd_qunit  = QDATA_IS_BLK(qdata) ? lqs->lqs_bunit_sz :
 478                 lqs->lqs_iunit_sz;
 479         cfs_spin_unlock(&lqs->lqs_lock);
 480
 481         if (QDATA_IS_BLK(qdata))
 482                 QDATA_SET_ADJBLK(qdata);
 483         else
 484                 QDATA_SET_ADJINO(qdata);
 485
 486         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
 487 skip:
 488         lqs_putref(lqs);
 489
 490         return rc;
 491 }
 492
 493 int mds_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
 494                      const unsigned int qpids[], int rc, int opc)
 495 {
 496         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 497         int rc2 = 0;
 498         ENTRY;
 499
 500         if (rc && rc != -EDQUOT && rc != ENOLCK)
 501                 RETURN(0);
 502
 503         switch (opc) {
 504         case FSFILT_OP_SETATTR:
 505                 /* release file quota on original owner */
 506                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 0, 0, NULL);
 507                 /* release block quota on original owner */
 508                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 509                 /* acquire file quota on current owner */
 510                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 511                 /* acquire block quota on current owner */
 512                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 513                 break;
 514         case FSFILT_OP_UNLINK_PARTIAL_CHILD:
 515                 /* release file quota on child */
 516                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 517                 /* rlease block quota on child */
 518                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 519                 break;
 520         case FSFILT_OP_CREATE_PARTIAL_CHILD:
 521                 /* acquire file quota on child */
 522                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 523                 /* acquire block quota on child */
 524                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 525                 break;
 526         case FSFILT_OP_LINK:
 527                 /* acquire block quota on parent */
 528                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 529                 break;
 530         case FSFILT_OP_UNLINK:
 531                 /* release block quota on parent */
 532                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 533                 /* release file quota on child */
 534                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 535                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 536                         /* release block quota on child */
 537                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
 538                                                   NULL);
 539                 break;
 540         case FSFILT_OP_UNLINK_PARTIAL_PARENT:
 541                 /* release block quota on parent */
 542                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 543                 break;
 544         case FSFILT_OP_CREATE:
 545                 /* acquire block quota on parent */
 546                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 547                 /* acquire file quota on child */
 548                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 549                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 550                         /* acquire block quota on child */
 551                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
 552                                                   NULL);
 553                 break;
 554         default:
 555                 LBUG();
 556                 break;
 557         }
 558
 559         if (rc2)
 560                 CDEBUG(D_QUOTA,
 561                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
 562                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
 563         RETURN(0);
 564 }
 565
 566 int filter_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
 567                         const unsigned int qpids[], int rc, int opc)
 568 {
 569         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 570         int rc2 = 0;
 571         ENTRY;
 572
 573         if (rc && rc != -EDQUOT)
 574                 RETURN(0);
 575
 576         switch (opc) {
 577         case FSFILT_OP_SETATTR:
 578                 /* acquire/release block quota on original & current owner */
 579                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 580                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 581                 break;
 582         case FSFILT_OP_UNLINK:
 583                 /* release block quota on this owner */
 584         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 585                 /* acquire block quota on this owner */
 586                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 587                 break;
 588         default:
 589                 LBUG();
 590                 break;
 591         }
 592
 593         if (rc || rc2) {
 594                 if (!rc)
 595                         rc = rc2;
 596                 CDEBUG(D_QUOTA,
 597                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
 598                        rc == QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
 599         }
 600
 601         RETURN(0);
 602 }
 603
 604 static const char prefix[] = "OBJECTS/";
 605
 606 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 607 {
 608         struct mds_obd *mds = &obd->u.mds;
 609         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 610         struct obd_device_target *obt = &obd->u.obt;
 611         int rc = 0, i, rc1 = 0;
 612         char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 613         char name[64];
 614         struct lvfs_run_ctxt saved;
 615         ENTRY;
 616
 617         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 618
 619         if (oqctl->qc_type != USRQUOTA &&
 620             oqctl->qc_type != GRPQUOTA &&
 621             oqctl->qc_type != UGQUOTA)
 622                 RETURN(-EINVAL);
 623
 624         cfs_down(&obt->obt_quotachecking);
 625         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 626         cfs_down_write(&mds->mds_qonoff_sem);
 627
 628         for (i = 0; i < MAXQUOTAS; i++) {
 629                 struct file *fp;
 630
 631                 if (!Q_TYPESET(oqctl, i))
 632                         continue;
 633
 634                 /* quota file has been opened ? */
 635                 if (qinfo->qi_files[i]) {
 636                         CWARN("quota[%d] is on yet\n", i);
 637                         rc1 = -EBUSY;
 638                         continue;
 639                 }
 640
 641                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 642                 sprintf(name, "%s%s", prefix, quotafile[i]);
 643
 644                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 645                 if (IS_ERR(fp)) {
 646                         rc = PTR_ERR(fp);
 647                         CERROR("%s: error invalidating admin quotafile %s (rc:%d)\n",
 648                                obd->obd_name, name, rc);
 649                 }
 650                 else
 651                         filp_close(fp, 0);
 652         }
 653
 654         cfs_up_write(&mds->mds_qonoff_sem);
 655         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 656         cfs_up(&obt->obt_quotachecking);
 657         RETURN(rc ? : rc1);
 658 }
 659
 660 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 661 {
 662         struct mds_obd *mds = &obd->u.mds;
 663         struct obd_device_target *obt = &obd->u.obt;
 664         int rc;
 665         struct lvfs_run_ctxt saved;
 666         ENTRY;
 667
 668         if (oqctl->qc_type != USRQUOTA &&
 669             oqctl->qc_type != GRPQUOTA &&
 670             oqctl->qc_type != UGQUOTA)
 671                 RETURN(-EINVAL);
 672
 673         cfs_down(&obt->obt_quotachecking);
 674         if (obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))
 675                 GOTO(out, rc = -EBUSY);
 676         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 677         cfs_down_write(&mds->mds_qonoff_sem);
 678
 679         oqctl->qc_cmd = Q_FINVALIDATE;
 680         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 681         if (!rc)
 682                 rc = obd_quotactl(mds->mds_lov_exp, oqctl);
 683
 684         cfs_up_write(&mds->mds_qonoff_sem);
 685         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 686 out:
 687         cfs_up(&obt->obt_quotachecking);
 688         RETURN(rc);
 689 }
 690
 691 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 692 {
 693         struct mds_obd *mds = &obd->u.mds;
 694         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 695         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 696         struct lvfs_run_ctxt saved;
 697         char name[64];
 698         int i, rc = 0;
 699         ENTRY;
 700
 701         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 702
 703         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 704         cfs_down_write(&mds->mds_qonoff_sem);
 705
 706         for (i = 0; i < MAXQUOTAS && !rc; i++) {
 707                 struct file *fp;
 708
 709                 if (!Q_TYPESET(oqctl, i))
 710                         continue;
 711
 712                 /* quota file has been opened ? */
 713                 if (qinfo->qi_files[i]) {
 714                         CWARN("init %s admin quotafile while quota on.\n",
 715                               i == USRQUOTA ? "user" : "group");
 716                         continue;
 717                 }
 718
 719                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 720                 sprintf(name, "%s%s", prefix, quotafile[i]);
 721
 722                 /* check if quota file exists and is correct */
 723                 fp = filp_open(name, O_RDONLY, 0);
 724                 if (!IS_ERR(fp)) {
 725                         /* irregular file is not the right place for quota */
 726                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 727                                 CERROR("admin quota file %s is not "
 728                                        "regular!", name);
 729                                 filp_close(fp, 0);
 730                                 rc = -EINVAL;
 731                                 break;
 732                         }
 733                         qinfo->qi_files[i] = fp;
 734                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 735                         qinfo->qi_files[i] = NULL;
 736                         filp_close(fp, 0);
 737                 }
 738                 else
 739                         rc = PTR_ERR(fp);
 740
 741                 if (!rc)
 742                         continue;
 743
 744                 /* -EINVAL may be returned by quotainfo for bad quota file */
 745                 if (rc != -ENOENT && rc != -EINVAL) {
 746                         CERROR("%s: error opening old quota file %s (%d)\n",
 747                                obd->obd_name, name, rc);
 748                         break;
 749                 }
 750
 751                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
 752                        rc == -ENOENT ? "creating" : "overwriting");
 753
 754                 /* create quota file overwriting old if needed */
 755                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 756                 if (IS_ERR(fp)) {
 757                         rc = PTR_ERR(fp);
 758                         CERROR("%s: error creating admin quotafile %s (rc:%d)\n",
 759                                obd->obd_name, name, rc);
 760                         break;
 761                 }
 762
 763                 qinfo->qi_files[i] = fp;
 764
 765                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 766                 if (rc)
 767                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 768                                i == USRQUOTA ? "user" : "group", rc);
 769
 770                 filp_close(fp, 0);
 771                 qinfo->qi_files[i] = NULL;
 772         }
 773
 774         cfs_up_write(&mds->mds_qonoff_sem);
 775         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 776         RETURN(rc);
 777 }
 778
 779 static int close_quota_files(struct obd_quotactl *oqctl,
 780                              struct lustre_quota_info *qinfo)
 781 {
 782         int i, rc = 0;
 783         ENTRY;
 784
 785         for (i = 0; i < MAXQUOTAS; i++) {
 786                 if (!Q_TYPESET(oqctl, i))
 787                         continue;
 788                 if (qinfo->qi_files[i] == NULL) {
 789                         CWARN("quota[%d] is off already\n", i);
 790                         rc = -EALREADY;
 791                         continue;
 792                 }
 793                 filp_close(qinfo->qi_files[i], 0);
 794                 qinfo->qi_files[i] = NULL;
 795         }
 796         RETURN(rc);
 797 }
 798
 799 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 800 {
 801         struct mds_obd *mds = &obd->u.mds;
 802         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 803         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 804         char name[64];
 805         int i, rc = 0, rc1 = 0;
 806         ENTRY;
 807
 808         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 809
 810         /* open admin quota files and read quotafile info */
 811         for (i = 0; i < MAXQUOTAS; i++) {
 812                 struct file *fp;
 813
 814                 if (!Q_TYPESET(oqctl, i) || qinfo->qi_files[i] != NULL)
 815                         continue;
 816
 817                 LASSERT(strlen(quotafile[i])
 818                         + sizeof(prefix) <= sizeof(name));
 819                 sprintf(name, "%s%s", prefix, quotafile[i]);
 820                 fp = filp_open(name, O_RDWR, 0);
 821                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 822                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
 823                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
 824                         break;
 825                 }
 826                 qinfo->qi_files[i] = fp;
 827
 828                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 829                 if (rc) {
 830                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
 831                         break;
 832                 }
 833
 834                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 835                 if (rc) {
 836                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
 837                                rc);
 838                         break;
 839                 }
 840         }
 841
 842         if (rc && rc1 != -EALREADY)
 843                 close_quota_files(oqctl, qinfo);
 844
 845         RETURN(rc ? : rc1);
 846 }
 847
 848 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 849 {
 850         int rc;
 851         ENTRY;
 852
 853         if (oqctl->qc_type != USRQUOTA &&
 854             oqctl->qc_type != GRPQUOTA &&
 855             oqctl->qc_type != UGQUOTA)
 856                 RETURN(-EINVAL);
 857
 858         rc = generic_quota_on(obd, oqctl, 1);
 859
 860         RETURN(rc);
 861 }
 862
 863
 864 int mds_admin_quota_off(struct obd_device *obd,
 865                         struct obd_quotactl *oqctl)
 866 {
 867         struct mds_obd *mds = &obd->u.mds;
 868         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 869         int rc;
 870         ENTRY;
 871
 872         /* close admin quota files */
 873         rc = close_quota_files(oqctl, qinfo);
 874         RETURN(rc);
 875 }
 876
 877
 878 /* with obt->obt_quotachecking held */
 879 int do_mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 880 {
 881         struct mds_obd *mds = &obd->u.mds;
 882         struct obd_device_target *obt = &obd->u.obt;
 883         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 884         struct lvfs_run_ctxt saved;
 885         int rc = 0, rc1 = 0, rc2 = 0;
 886         ENTRY;
 887
 888         LASSERT_SEM_LOCKED(&obt->obt_quotachecking);
 889
 890         if (oqctl->qc_type != USRQUOTA &&
 891             oqctl->qc_type != GRPQUOTA &&
 892             oqctl->qc_type != UGQUOTA)
 893                 RETURN(-EINVAL);
 894
 895         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 896         cfs_down_write(&mds->mds_qonoff_sem);
 897         /* close admin quota files */
 898         rc2 = mds_admin_quota_off(obd, oqctl);
 899         if (rc2 && rc2 != -EALREADY) {
 900                 CWARN("mds quota[%d] is failed to be off for %d\n", oqctl->qc_type, rc2);
 901                 GOTO(out, rc2);
 902         }
 903
 904         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 905         if (!rc1) {
 906                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
 907         } else if (quota_is_off(qctxt, oqctl)) {
 908                 CWARN("mds local quota[%d] is off already\n", oqctl->qc_type);
 909                 rc1 = -EALREADY;
 910         } else {
 911                 if (rc2 != -EALREADY) {
 912                         CWARN("mds local quota[%d] is failed to be off for %d\n",
 913                               oqctl->qc_type, rc1);
 914                         oqctl->qc_cmd = Q_QUOTAON;
 915                         mds_admin_quota_on(obd, oqctl);
 916                         oqctl->qc_cmd = Q_QUOTAOFF;
 917                 }
 918                 GOTO(out, rc1);
 919         }
 920
 921         rc = obd_quotactl(mds->mds_lov_exp, oqctl);
 922         if (rc && rc != -EALREADY) {
 923                 CWARN("mds remote quota[%d] is failed to be off for %d\n",
 924                       oqctl->qc_type, rc);
 925                 oqctl->qc_cmd = Q_QUOTAON;
 926                 if (rc2 != -EALREADY)
 927                         mds_admin_quota_on(obd, oqctl);
 928                 if (rc1 != -EALREADY) {
 929                         fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 930                         qctxt->lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
 931                 }
 932                 oqctl->qc_cmd = Q_QUOTAOFF;
 933         }
 934         EXIT;
 935
 936 out:
 937         cfs_up_write(&mds->mds_qonoff_sem);
 938         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 939         return rc ? : (rc1 ? : rc2);
 940 }
 941
 942 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 943 {
 944         struct obd_device_target *obt = &obd->u.obt;
 945         int rc;
 946         ENTRY;
 947
 948         cfs_down(&obt->obt_quotachecking);
 949         rc = do_mds_quota_off(obd, oqctl);
 950         cfs_up(&obt->obt_quotachecking);
 951         RETURN(rc);
 952 }
 953
 954 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 955 {
 956         struct mds_obd *mds = &obd->u.mds;
 957         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 958         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 959         int rc;
 960         ENTRY;
 961
 962         if (oqctl->qc_type != USRQUOTA &&
 963             oqctl->qc_type != GRPQUOTA)
 964                 RETURN(-EINVAL);
 965
 966         cfs_down_write(&mds->mds_qonoff_sem);
 967         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 968                 CWARN("quota[%u] is off\n", oqctl->qc_type);
 969                 GOTO(out, rc = -ESRCH);
 970         }
 971
 972         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 973         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 974         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 975
 976         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 977         EXIT;
 978
 979 out:
 980         cfs_up_write(&mds->mds_qonoff_sem);
 981         return rc;
 982 }
 983
 984 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 985 {
 986         struct mds_obd *mds = &obd->u.mds;
 987         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 988         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 989         int rc = 0;
 990         ENTRY;
 991
 992         if (oqctl->qc_type != USRQUOTA &&
 993             oqctl->qc_type != GRPQUOTA)
 994                 RETURN(-EINVAL);
 995
 996         cfs_down_read(&mds->mds_qonoff_sem);
 997         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 998                 CWARN("quota[%u] is off\n", oqctl->qc_type);
 999                 GOTO(out, rc = -ESRCH);
1000         }
1001
1002         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
1003         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
1004         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
1005         EXIT;
1006
1007 out:
1008         cfs_up_read(&mds->mds_qonoff_sem);
1009         return rc;
1010 }
1011
1012 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
1013                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
1014                       int type, struct quota_adjust_qunit *oqaq)
1015 {
1016         __u64 bunit_curr_o, iunit_curr_o;
1017         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
1018         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
1019         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
1020                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
1021         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1022                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1023         int rc = 0;
1024         ENTRY;
1025
1026         if (!dquot || !oqaq)
1027                 RETURN(-EINVAL);
1028         LASSERT_SEM_LOCKED(&dquot->dq_sem);
1029         LASSERT(oqaq->qaq_iunit_sz);
1030         LASSERT(oqaq->qaq_bunit_sz);
1031
1032         /* don't change qunit size */
1033         if (!qctxt->lqc_switch_qs)
1034                 RETURN(rc);
1035
1036         bunit_curr_o = oqaq->qaq_bunit_sz;
1037         iunit_curr_o = oqaq->qaq_iunit_sz;
1038
1039         if (dquot->dq_type == GRPQUOTA)
1040                 QAQ_SET_GRP(oqaq);
1041
1042         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1043                 __u64 b_limitation =
1044                         oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1045                 /* enlarge block qunit size */
1046                 while (blimit >
1047                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1048                         oqaq->qaq_bunit_sz =
1049                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1050                                 << QUOTABLOCK_BITS;
1051                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1052                                 shrink_qunit_limit;
1053                 }
1054
1055                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1056                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1057
1058                 /* shrink block qunit size */
1059                 while (blimit <
1060                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1061                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1062                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1063                                 QUOTABLOCK_BITS;
1064                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1065                                 shrink_qunit_limit;
1066                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1067                                 break;
1068                 }
1069
1070                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1071                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1072
1073                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1074                         QAQ_SET_ADJBLK(oqaq);
1075
1076         }
1077
1078         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1079                 __u64 i_limitation =
1080                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1081                 /* enlarge file qunit size */
1082                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1083                        + 2 * i_limitation) {
1084                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1085                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1086                                 shrink_qunit_limit;
1087                 }
1088
1089                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1090                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1091
1092                 /* shrink file qunit size */
1093                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1094                        + i_limitation) {
1095                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1096                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1097                                        shrink_qunit_limit;
1098                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1099                                 break;
1100                 }
1101
1102                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1103                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1104
1105                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1106                         QAQ_SET_ADJINO(oqaq);
1107
1108         }
1109
1110         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1111
1112         RETURN(rc);
1113 }
1114
1115 static int mds_init_slave_ilimits(struct obd_device *obd,
1116                                   struct obd_quotactl *oqctl, int set)
1117 {
1118         /* XXX: for file limits only adjust local now */
1119         struct obd_device_target *obt = &obd->u.obt;
1120         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1121         unsigned int id[MAXQUOTAS] = { 0, 0 };
1122         struct obd_quotactl *ioqc = NULL;
1123         struct lustre_qunit_size *lqs;
1124         int flag;
1125         int rc;
1126         ENTRY;
1127
1128         /* if we are going to set zero limit, needn't init slaves */
1129         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1130             !set)
1131                 RETURN(0);
1132
1133         OBD_ALLOC_PTR(ioqc);
1134         if (!ioqc)
1135                 RETURN(-ENOMEM);
1136
1137         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1138                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1139         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1140         ioqc->qc_id = oqctl->qc_id;
1141         ioqc->qc_type = oqctl->qc_type;
1142         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1143         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1144
1145         /* build lqs for mds */
1146         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1147                                qctxt, flag ? 1 : 0);
1148         if (lqs && !IS_ERR(lqs)) {
1149                 if (flag)
1150                         lqs->lqs_flags |= QI_SET;
1151                 else
1152                         lqs->lqs_flags &= ~QI_SET;
1153                 lqs_putref(lqs);
1154         } else {
1155                 CERROR("fail to %s lqs for inode(%s id: %u)!\n",
1156                        flag ? "create" : "search",
1157                        oqctl->qc_type ? "group" : "user",
1158                        oqctl->qc_id);
1159                 GOTO(out, rc = PTR_ERR(lqs));
1160         }
1161
1162         /* set local limit to MIN_QLIMIT */
1163         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1164         if (rc)
1165                 GOTO(out, rc);
1166
1167         /* trigger local qunit pre-acquire */
1168         if (oqctl->qc_type == USRQUOTA)
1169                 id[USRQUOTA] = oqctl->qc_id;
1170         else
1171                 id[GRPQUOTA] = oqctl->qc_id;
1172
1173         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1174         if (rc == -EDQUOT || rc == -EBUSY) {
1175                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1176                 rc = 0;
1177         }
1178         if (rc) {
1179                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1180                        rc);
1181                 GOTO(out, rc);
1182         }
1183         /* FIXME initialize all slaves in CMD */
1184         EXIT;
1185 out:
1186         if (ioqc)
1187                 OBD_FREE_PTR(ioqc);
1188         return rc;
1189 }
1190
1191 static int mds_init_slave_blimits(struct obd_device *obd,
1192                                   struct obd_quotactl *oqctl, int set)
1193 {
1194         struct obd_device_target *obt = &obd->u.obt;
1195         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1196         struct mds_obd *mds = &obd->u.mds;
1197         struct obd_quotactl *ioqc;
1198         struct lustre_qunit_size *lqs;
1199         unsigned int id[MAXQUOTAS] = { 0, 0 };
1200         int rc;
1201         int flag;
1202         ENTRY;
1203
1204         /* if we are going to set zero limit, needn't init slaves */
1205         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1206             !set)
1207                 RETURN(0);
1208
1209         OBD_ALLOC_PTR(ioqc);
1210         if (!ioqc)
1211                 RETURN(-ENOMEM);
1212
1213         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1214                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1215         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1216         ioqc->qc_id = oqctl->qc_id;
1217         ioqc->qc_type = oqctl->qc_type;
1218         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1219         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1220
1221         /* build lqs for mds */
1222         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1223                                qctxt, flag ? 1 : 0);
1224         if (lqs && !IS_ERR(lqs)) {
1225                 if (flag)
1226                         lqs->lqs_flags |= QB_SET;
1227                 else
1228                         lqs->lqs_flags &= ~QB_SET;
1229                 lqs_putref(lqs);
1230         } else {
1231                 CERROR("fail to %s lqs for block(%s id: %u)!\n",
1232                        flag ? "create" : "search",
1233                        oqctl->qc_type ? "group" : "user",
1234                        oqctl->qc_id);
1235                 GOTO(out, rc = PTR_ERR(lqs));
1236         }
1237
1238         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1239         if (rc)
1240                 GOTO(out, rc);
1241
1242         /* trigger local qunit pre-acquire */
1243         if (oqctl->qc_type == USRQUOTA)
1244                 id[USRQUOTA] = oqctl->qc_id;
1245         else
1246                 id[GRPQUOTA] = oqctl->qc_id;
1247
1248         /* initialize all slave's limit */
1249         rc = obd_quotactl(mds->mds_lov_exp, ioqc);
1250
1251         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1252         if (rc == -EDQUOT || rc == -EBUSY) {
1253                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1254                 rc = 0;
1255         }
1256         if (rc) {
1257                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1258                 GOTO(out, rc);
1259         }
1260
1261         EXIT;
1262 out:
1263         OBD_FREE_PTR(ioqc);
1264         return rc;
1265 }
1266
1267 static void adjust_lqs(struct obd_device *obd, struct quota_adjust_qunit *qaq)
1268 {
1269         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
1270         int rc = 0;
1271
1272         QAQ_SET_CREATE_LQS(qaq);
1273         /* adjust local lqs */
1274         rc = quota_adjust_slave_lqs(qaq, qctxt);
1275         if (rc < 0)
1276                 CERROR("adjust master's qunit size failed!(rc=%d)\n", rc);
1277
1278         /* adjust remote lqs */
1279         if (QAQ_IS_ADJBLK(qaq)) {
1280                 rc = obd_quota_adjust_qunit(obd->u.mds.mds_lov_exp, qaq, qctxt, NULL);
1281                 if (rc < 0)
1282                         CERROR("adjust slaves' qunit size failed!(rc=%d)\n", rc);
1283
1284         }
1285 }
1286
1287 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1288 {
1289         struct mds_obd *mds = &obd->u.mds;
1290         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1291         struct obd_device *lov_obd = class_exp2obd(mds->mds_lov_exp);
1292         struct lov_obd *lov = &lov_obd->u.lov;
1293         struct quota_adjust_qunit *oqaq = NULL;
1294         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1295         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1296         time_t btime, itime;
1297         struct lustre_dquot *dquot;
1298         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1299         /* orig_set means if quota was set before; now_set means we are
1300          * setting/cancelling quota */
1301         int orig_set, now_set;
1302         struct lustre_qunit_size *lqs;
1303         int rc = 0, rc2 = 0, flag = 0;
1304         ENTRY;
1305
1306         if (oqctl->qc_type != USRQUOTA &&
1307             oqctl->qc_type != GRPQUOTA)
1308                 RETURN(-EINVAL);
1309
1310         OBD_ALLOC_PTR(oqaq);
1311         if (!oqaq)
1312                 RETURN(-ENOMEM);
1313
1314         cfs_down_write(&mds->mds_qonoff_sem);
1315         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1316
1317         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1318                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1319                 GOTO(out_sem, rc = -ESRCH);
1320         }
1321
1322         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1323         if (IS_ERR(dquot))
1324                 GOTO(out_sem, rc = PTR_ERR(dquot));
1325         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1326         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1327
1328         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id), qctxt, 1);
1329         if (lqs == NULL)
1330                 rc = -ENOENT;
1331         if (IS_ERR(lqs))
1332                 rc = PTR_ERR(lqs);
1333         if (rc)
1334                 GOTO(out, rc);
1335
1336         cfs_down(&dquot->dq_sem);
1337         cfs_spin_lock(&lqs->lqs_lock);
1338         if (LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1339                 cfs_spin_unlock(&lqs->lqs_lock);
1340                 cfs_up(&dquot->dq_sem);
1341                 GOTO(skip, rc = -EBUSY);
1342         }
1343         LQS_SET_SETQUOTA(lqs);
1344         cfs_spin_unlock(&lqs->lqs_lock);
1345
1346         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1347         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1348         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1349         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1350         btime = dquot->dq_dqb.dqb_btime;
1351         itime = dquot->dq_dqb.dqb_itime;
1352
1353         if (dqblk->dqb_valid & QIF_BTIME)
1354                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1355         if (dqblk->dqb_valid & QIF_ITIME)
1356                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1357
1358         if (dqblk->dqb_valid & QIF_BLIMITS) {
1359                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1360                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1361                 /* clear usage (limit pool) */
1362                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1363                     !dquot->dq_dqb.dqb_bsoftlimit)
1364                         dquot->dq_dqb.dqb_curspace = 0;
1365
1366                 /* clear grace time */
1367                 if (!dqblk->dqb_bsoftlimit ||
1368                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1369                         dquot->dq_dqb.dqb_btime = 0;
1370                 /* set grace only if user hasn't provided his own */
1371                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1372                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1373                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1374
1375                 flag |= LQUOTA_FLAGS_ADJBLK;
1376         }
1377
1378         if (dqblk->dqb_valid & QIF_ILIMITS) {
1379                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1380                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1381                 /* clear usage (limit pool) */
1382                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1383                     !dquot->dq_dqb.dqb_isoftlimit)
1384                         dquot->dq_dqb.dqb_curinodes = 0;
1385
1386                 if (!dqblk->dqb_isoftlimit ||
1387                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1388                         dquot->dq_dqb.dqb_itime = 0;
1389                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1390                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1391                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1392
1393                 flag |= LQUOTA_FLAGS_ADJINO;
1394         }
1395         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1396         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1397                                flag, oqaq);
1398         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1399         if (rc < 0)
1400                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1401
1402
1403         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1404
1405         cfs_up(&dquot->dq_sem);
1406
1407         if (rc) {
1408                 CERROR("set limit failed! (rc:%d)\n", rc);
1409                 GOTO(update_fail, rc);
1410         }
1411
1412         cfs_up_write(&mds->mds_qonoff_sem);
1413         adjust_lqs(obd, oqaq);
1414
1415         orig_set = ihardlimit || isoftlimit;
1416         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1417         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1418                 cfs_down(&dquot->dq_sem);
1419                 dquot->dq_dqb.dqb_curinodes = 0;
1420                 cfs_up(&dquot->dq_sem);
1421                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set);
1422                 if (rc) {
1423                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1424                         goto revoke_out;
1425                 }
1426         }
1427
1428         orig_set = bhardlimit || bsoftlimit;
1429         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1430         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1431                 cfs_down(&dquot->dq_sem);
1432                 dquot->dq_dqb.dqb_curspace = 0;
1433                 cfs_up(&dquot->dq_sem);
1434                 rc = mds_init_slave_blimits(obd, oqctl, orig_set);
1435                 if (rc) {
1436                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1437                         goto revoke_out;
1438                 }
1439         }
1440
1441 revoke_out:
1442         cfs_down_write(&mds->mds_qonoff_sem);
1443         cfs_down(&dquot->dq_sem);
1444         if (rc) {
1445                 /* cancel previous setting */
1446                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1447                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1448                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1449                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1450                 dquot->dq_dqb.dqb_btime = btime;
1451                 dquot->dq_dqb.dqb_itime = itime;
1452         }
1453         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1454         cfs_up(&dquot->dq_sem);
1455 update_fail:
1456         cfs_spin_lock(&lqs->lqs_lock);
1457         LQS_CLEAR_SETQUOTA(lqs);
1458         cfs_spin_unlock(&lqs->lqs_lock);
1459 skip:
1460         lqs_putref(lqs);
1461 out:
1462         lustre_dqput(dquot);
1463         EXIT;
1464 out_sem:
1465         cfs_up_write(&mds->mds_qonoff_sem);
1466
1467         if (oqaq)
1468                 OBD_FREE_PTR(oqaq);
1469
1470         return rc ? rc : rc2;
1471 }
1472
1473 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1474 {
1475         struct obd_quotactl *soqc;
1476         struct lvfs_run_ctxt saved;
1477         int rc, rc1;
1478         ENTRY;
1479
1480         OBD_ALLOC_PTR(soqc);
1481         if (!soqc)
1482                 RETURN(-ENOMEM);
1483
1484         soqc->qc_cmd = Q_GETOQUOTA;
1485         soqc->qc_id = oqctl->qc_id;
1486         soqc->qc_type = oqctl->qc_type;
1487
1488         /* get block usage from OSS */
1489         soqc->qc_dqblk.dqb_curspace = 0;
1490         rc = obd_quotactl(obd->u.mds.mds_lov_exp, soqc);
1491         if (!rc || rc == -EREMOTEIO) {
1492                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1493                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1494         }
1495
1496         /* get block/inode usage from MDS */
1497         soqc->qc_dqblk.dqb_curspace = 0;
1498         soqc->qc_dqblk.dqb_curinodes = 0;
1499         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1500         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1501         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1502         if (!rc1) {
1503                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1504                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1505                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1506         }
1507
1508         OBD_FREE_PTR(soqc);
1509
1510         RETURN(rc ? : rc1);
1511 }
1512
1513 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1514 {
1515         struct mds_obd *mds = &obd->u.mds;
1516         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1517         struct lustre_dquot *dquot;
1518         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1519         int rc;
1520         ENTRY;
1521
1522         if (oqctl->qc_type != USRQUOTA &&
1523             oqctl->qc_type != GRPQUOTA)
1524                 RETURN(-EINVAL);
1525
1526         cfs_down_read(&mds->mds_qonoff_sem);
1527         dqblk->dqb_valid = 0;
1528         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1529                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1530                 GOTO(out, rc = -ESRCH);
1531         }
1532
1533         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1534         if (IS_ERR(dquot))
1535                 GOTO(out, rc = PTR_ERR(dquot));
1536
1537         cfs_down(&dquot->dq_sem);
1538         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1539         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1540         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1541         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1542         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1543         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1544         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1545         cfs_up(&dquot->dq_sem);
1546
1547         lustre_dqput(dquot);
1548         cfs_up_read(&mds->mds_qonoff_sem);
1549
1550         /* the usages in admin quota file is inaccurate */
1551         dqblk->dqb_curinodes = 0;
1552         dqblk->dqb_curspace = 0;
1553         rc = mds_get_space(obd, oqctl);
1554
1555         /*
1556          * Querying of curinodes and/or curspace may have failed, administrative
1557          * quota data are likely to be better approximation to the real usage in
1558          * this case.
1559          */
1560         if (!(dqblk->dqb_valid & QIF_INODES) && dquot->dq_dqb.dqb_curinodes > 0)
1561                 dqblk->dqb_curinodes = dquot->dq_dqb.dqb_curinodes;
1562
1563         if (!(dqblk->dqb_valid & QIF_SPACE) && dquot->dq_dqb.dqb_curspace > 0)
1564                 dqblk->dqb_curspace = dquot->dq_dqb.dqb_curspace;
1565
1566         RETURN(rc);
1567
1568 out:
1569         cfs_up_read(&mds->mds_qonoff_sem);
1570         return rc;
1571 }
1572
1573 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1574 {
1575         struct lvfs_run_ctxt saved;
1576         int rc;
1577         ENTRY;
1578
1579         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1580         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1581         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1582
1583         RETURN(rc);
1584 }
1585
1586
1587 /* FIXME we only recovery block limit by now, need recovery inode
1588  * limits also after CMD involved in */
1589 static int
1590 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1591 {
1592         struct mds_obd *mds = &obd->u.mds;
1593         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1594         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1595         struct lustre_qunit_size *lqs;
1596         struct lustre_dquot *dquot;
1597         struct obd_quotactl *qctl;
1598         __u64 total_limits = 0;
1599         int rc = 0;
1600         ENTRY;
1601
1602         OBD_ALLOC_PTR(qctl);
1603         if (qctl == NULL)
1604                 RETURN(-ENOMEM);
1605
1606         dquot = lustre_dqget(obd, qinfo, id, type, 0);
1607         if (IS_ERR(dquot)) {
1608                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1609                 OBD_FREE_PTR(qctl);
1610                 RETURN(PTR_ERR(dquot));
1611         }
1612
1613         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 1);
1614         if (lqs == NULL)
1615                 rc = -ENOENT;
1616         if (IS_ERR(lqs))
1617                 rc = PTR_ERR(lqs);
1618         if (rc)
1619                 GOTO(skip, rc);
1620
1621         cfs_down(&dquot->dq_sem);
1622
1623         /* don't recover the dquot without limits or quota is setting or
1624          * another recovery is already going on */
1625         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1626             LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1627                 cfs_up(&dquot->dq_sem);
1628                 GOTO(skip1, rc = 0);
1629         }
1630
1631         cfs_spin_lock(&lqs->lqs_lock);
1632         LQS_SET_RECOVERY(lqs);
1633         cfs_spin_unlock(&lqs->lqs_lock);
1634         cfs_up(&dquot->dq_sem);
1635
1636         /* release mds_qonoff_sem during obd_quotactl ops here */
1637         cfs_up_write(&mds->mds_qonoff_sem);
1638
1639         /* get real bhardlimit from all slaves. */
1640         qctl->qc_cmd = Q_GETOQUOTA;
1641         qctl->qc_type = type;
1642         qctl->qc_id = id;
1643         qctl->qc_stat = QUOTA_RECOVERING;
1644         rc = obd_quotactl(mds->mds_lov_exp, qctl);
1645         cfs_down_write(&mds->mds_qonoff_sem);
1646         if (rc)
1647                 GOTO(out, rc);
1648         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1649
1650         /* get real bhardlimit from master */
1651         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1652         if (rc)
1653                 GOTO(out, rc);
1654         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1655
1656         /* amend the usage of the administrative quotafile */
1657         cfs_down(&dquot->dq_sem);
1658
1659         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1660
1661         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1662         if (rc)
1663                 CERROR("write dquot failed! (rc:%d)\n", rc);
1664
1665         cfs_up(&dquot->dq_sem);
1666         EXIT;
1667 out:
1668         cfs_spin_lock(&lqs->lqs_lock);
1669         LQS_CLEAR_RECOVERY(lqs);
1670         cfs_spin_unlock(&lqs->lqs_lock);
1671 skip1:
1672         lqs_putref(lqs);
1673 skip:
1674         lustre_dqput(dquot);
1675         OBD_FREE_PTR(qctl);
1676         return rc;
1677 }
1678
1679 struct qmaster_recov_thread_data {
1680         struct obd_device *obd;
1681         cfs_completion_t comp;
1682 };
1683
1684 static int qmaster_recovery_main(void *arg)
1685 {
1686         struct qmaster_recov_thread_data *data = arg;
1687         struct obd_device *obd = data->obd;
1688         struct mds_obd *mds = &obd->u.mds;
1689         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1690         int rc = 0;
1691         unsigned short type;
1692         ENTRY;
1693
1694         cfs_daemonize_ctxt("qmaster_recovd");
1695
1696         /* for mds */
1697         class_incref(obd, "qmaster_recovd_mds", obd);
1698         /* for lov */
1699         class_incref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1700
1701         cfs_complete(&data->comp);
1702
1703         cfs_down_write(&mds->mds_qonoff_sem);
1704         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1705                 cfs_list_t id_list;
1706                 struct dquot_id *dqid, *tmp;
1707
1708                 if (qinfo->qi_files[type] == NULL)
1709                         continue;
1710
1711                 CFS_INIT_LIST_HEAD(&id_list);
1712                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1713                                  &id_list);
1714                 if (rc)
1715                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1716
1717                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1718                         cfs_list_del_init(&dqid->di_link);
1719                         if (rc)
1720                                 goto free;
1721
1722                         rc = dquot_recovery(obd, dqid->di_id, type);
1723                         if (rc)
1724                                 CERROR("%s: qmaster recovery failed for %sid %d"
1725                                        " rc:%d)\n", obd->obd_name,
1726                                        type ? "g" : "u", dqid->di_id, rc);
1727 free:
1728                         OBD_FREE_PTR(dqid);
1729                 }
1730         }
1731         cfs_up_write(&mds->mds_qonoff_sem);
1732         class_decref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1733         class_decref(obd, "qmaster_recovd_mds", obd);
1734         RETURN(rc);
1735 }
1736
1737 int mds_quota_recovery(struct obd_device *obd)
1738 {
1739         struct mds_obd *mds = &obd->u.mds;
1740         struct qmaster_recov_thread_data data;
1741         int rc = 0;
1742         ENTRY;
1743
1744         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
1745                 RETURN(0);
1746
1747         if (unlikely(!mds->mds_quota || obd->obd_stopping))
1748                 RETURN(rc);
1749
1750         cfs_mutex_down(&obd->obd_dev_sem);
1751         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1752                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1753                       mds->mds_lov_desc.ld_active_tgt_count,
1754                       mds->mds_lov_objid_count);
1755                 cfs_mutex_up(&obd->obd_dev_sem);
1756                 RETURN(rc);
1757         }
1758         cfs_mutex_up(&obd->obd_dev_sem);
1759
1760         data.obd = obd;
1761         cfs_init_completion(&data.comp);
1762
1763         rc = cfs_kernel_thread(qmaster_recovery_main, &data,
1764                                CLONE_VM|CLONE_FILES);
1765         if (rc < 0)
1766                 CERROR("%s: cannot start quota recovery thread: rc %d\n",
1767                        obd->obd_name, rc);
1768
1769         cfs_wait_for_completion(&data.comp);
1770         RETURN(rc);
1771 }
1772
1773 #endif /* HAVE_QUOTA_SUPPORT */