lustre/quota/quota_master.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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 /* lock ordering: mds->mds_qonoff_sem > dquot->dq_mutex > lqs->lqs_lock */
  66 static cfs_list_t lustre_dquot_hash[NR_DQHASH];
  67 static cfs_rwlock_t dquot_hash_lock = CFS_RW_LOCK_UNLOCKED;
  68
  69 cfs_mem_cache_t *lustre_dquot_cachep;
  70
  71 int lustre_dquot_init(void)
  72 {
  73         int i;
  74         ENTRY;
  75
  76         LASSERT(lustre_dquot_cachep == NULL);
  77         lustre_dquot_cachep = cfs_mem_cache_create("lustre_dquot_cache",
  78                                                    sizeof(struct lustre_dquot),
  79                                                    0, 0);
  80         if (!lustre_dquot_cachep)
  81                 return (-ENOMEM);
  82
  83         for (i = 0; i < NR_DQHASH; i++) {
  84                 CFS_INIT_LIST_HEAD(lustre_dquot_hash + i);
  85         }
  86         RETURN(0);
  87 }
  88
  89 void lustre_dquot_exit(void)
  90 {
  91         int i;
  92         ENTRY;
  93         /* FIXME cleanup work ?? */
  94
  95         for (i = 0; i < NR_DQHASH; i++) {
  96                 LASSERT(cfs_list_empty(lustre_dquot_hash + i));
  97         }
  98         if (lustre_dquot_cachep) {
  99                 int rc;
 100                 rc = cfs_mem_cache_destroy(lustre_dquot_cachep);
 101                 LASSERTF(rc == 0,"couldn't destroy lustre_dquot_cachep slab\n");
 102                 lustre_dquot_cachep = NULL;
 103         }
 104         EXIT;
 105 }
 106
 107 static inline int
 108 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 109              __attribute__((__const__));
 110
 111 static inline int
 112 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 113 {
 114         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
 115         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 116         return tmp;
 117 }
 118
 119 /* caller must hold dquot_hash_lock */
 120 static struct lustre_dquot *find_dquot(int hashent,
 121                                        struct lustre_quota_info *lqi, qid_t id,
 122                                        int type)
 123 {
 124         struct lustre_dquot *dquot;
 125         ENTRY;
 126
 127         cfs_list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
 128                 if (dquot->dq_info == lqi &&
 129                     dquot->dq_id == id && dquot->dq_type == type)
 130                         RETURN(dquot);
 131         }
 132         RETURN(NULL);
 133 }
 134
 135 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 136                                         qid_t id, int type)
 137 {
 138         struct lustre_dquot *dquot = NULL;
 139         ENTRY;
 140
 141         OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
 142         if (dquot == NULL)
 143                 RETURN(NULL);
 144
 145         CFS_INIT_LIST_HEAD(&dquot->dq_hash);
 146         cfs_mutex_init(&dquot->dq_mutex);
 147         cfs_mutex_lock(&dquot->dq_mutex);
 148         cfs_atomic_set(&dquot->dq_refcnt, 1);
 149         dquot->dq_info = lqi;
 150         dquot->dq_id = id;
 151         dquot->dq_type = type;
 152
 153         RETURN(dquot);
 154 }
 155
 156 static void free_dquot(struct lustre_dquot *dquot)
 157 {
 158         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 159 }
 160
 161 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 162 {
 163         cfs_list_t *head = lustre_dquot_hash +
 164             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 165         LASSERT(cfs_list_empty(&dquot->dq_hash));
 166         cfs_list_add(&dquot->dq_hash, head);
 167 }
 168
 169 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 170 {
 171         LASSERT(!cfs_list_empty(&dquot->dq_hash));
 172         cfs_list_del_init(&dquot->dq_hash);
 173 }
 174
 175 static void lustre_dqput(struct lustre_dquot *dquot)
 176 {
 177         ENTRY;
 178         cfs_write_lock(&dquot_hash_lock);
 179         LASSERT(cfs_atomic_read(&dquot->dq_refcnt));
 180         cfs_atomic_dec(&dquot->dq_refcnt);
 181         if (cfs_atomic_read(&dquot->dq_refcnt) == 0) {
 182                 remove_dquot_nolock(dquot);
 183                 free_dquot(dquot);
 184         }
 185         cfs_write_unlock(&dquot_hash_lock);
 186         EXIT;
 187 }
 188
 189 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 190                                          struct lustre_quota_info *lqi,
 191                                          qid_t id, int type, int can_fake)
 192 {
 193         unsigned int hashent = dquot_hashfn(lqi, id, type);
 194         struct lustre_dquot *dquot, *empty;
 195         int free_dq = 0;
 196         ENTRY;
 197
 198         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 199                 RETURN(ERR_PTR(-ENOMEM));
 200
 201         cfs_read_lock(&dquot_hash_lock);
 202         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 203                 cfs_atomic_inc(&dquot->dq_refcnt);
 204                 cfs_read_unlock(&dquot_hash_lock);
 205                 cfs_mutex_unlock(&empty->dq_mutex);
 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_mutex_unlock(&dquot->dq_mutex);
 215                 if (rc) {
 216                         CERROR("can't read dquot from admin quotafile! "
 217                                "(rc:%d)\n", rc);
 218                         free_dquot(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_mutex_lock(&dquot->dq_mutex);
 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_mutex_unlock(&dquot->dq_mutex);
 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_mutex_unlock(&dquot->dq_mutex);
 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         if (!ll_sb_has_quota_active(qctxt->lqc_sb,
 376                                     QDATA_IS_GRP(qdata) ? GRPQUOTA : USRQUOTA))
 377                 RETURN(-EIO);
 378
 379         lqs = quota_search_lqs(LQS_KEY(QDATA_IS_GRP(qdata), qdata->qd_id),
 380                                qctxt, 0);
 381         if (lqs == NULL)
 382                 rc = -ENOENT;
 383         if (IS_ERR(lqs))
 384                 rc = PTR_ERR(lqs);
 385         if (rc)
 386                 RETURN(rc);
 387
 388         cfs_spin_lock(&lqs->lqs_lock);
 389         if (LQS_IS_RECOVERY(lqs)) {
 390                 cfs_spin_unlock(&lqs->lqs_lock);
 391                 LQS_DEBUG(lqs, "this lqs is under recovery\n");
 392                 GOTO(skip, rc = -EBUSY);
 393         }
 394         cfs_spin_unlock(&lqs->lqs_lock);
 395
 396         cfs_down_write(&mds->mds_qonoff_sem);
 397         dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata), 0);
 398         if (IS_ERR(dquot)) {
 399                 cfs_up_write(&mds->mds_qonoff_sem);
 400                 GOTO(skip, rc = PTR_ERR(dquot));
 401         }
 402
 403         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 404         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 405
 406         cfs_mutex_lock(&dquot->dq_mutex);
 407
 408         if (QDATA_IS_BLK(qdata)) {
 409                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
 410                 usage = &dquot->dq_dqb.dqb_curspace;
 411                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 412                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 413                 time = &dquot->dq_dqb.dqb_btime;
 414         } else {
 415                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
 416                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 417                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 418                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 419                 time = &dquot->dq_dqb.dqb_itime;
 420         }
 421
 422         /* if the quota limit in admin quotafile is zero, we just inform
 423          * slave to clear quota limit with zero qd_count */
 424         if (hlimit == 0 && slimit == 0) {
 425                 qdata->qd_count = 0;
 426                 GOTO(out, rc);
 427         }
 428
 429         switch (opc) {
 430         case QUOTA_DQACQ:
 431                 if (hlimit &&
 432                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
 433                 {
 434                         if (QDATA_IS_CHANGE_QS(qdata) &&
 435                             QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
 436                                 qdata->qd_count = (hlimit -
 437                                         QUSG(*usage, QDATA_IS_BLK(qdata)))
 438                                         * (QDATA_IS_BLK(qdata) ?
 439                                            QUOTABLOCK_SIZE : 1);
 440                         else
 441                                 GOTO(out, rc = -EDQUOT);
 442                 }
 443
 444                 if (slimit &&
 445                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
 446                         if (*time && cfs_time_current_sec() >= *time)
 447                                 GOTO(out, rc = -EDQUOT);
 448                         else if (!*time)
 449                                 *time = cfs_time_current_sec() + grace;
 450                 }
 451
 452                 *usage += qdata->qd_count;
 453                 break;
 454         case QUOTA_DQREL:
 455                 /* The usage in administrative file might be incorrect before
 456                  * recovery done */
 457                 if (*usage < qdata->qd_count)
 458                         *usage = 0;
 459                 else
 460                         *usage -= qdata->qd_count;
 461
 462                 /* (usage <= soft limit) but not (usage < soft limit) */
 463                 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
 464                         *time = 0;
 465                 break;
 466         default:
 467                 LBUG();
 468         }
 469
 470         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 471         EXIT;
 472 out:
 473         cfs_mutex_unlock(&dquot->dq_mutex);
 474         cfs_up_write(&mds->mds_qonoff_sem);
 475         lustre_dqput(dquot);
 476         if (rc != -EDQUOT)
 477                 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
 478                                       QDATA_IS_BLK(qdata));
 479
 480         cfs_spin_lock(&lqs->lqs_lock);
 481         qdata->qd_qunit  = QDATA_IS_BLK(qdata) ? lqs->lqs_bunit_sz :
 482                 lqs->lqs_iunit_sz;
 483         cfs_spin_unlock(&lqs->lqs_lock);
 484
 485         if (QDATA_IS_BLK(qdata))
 486                 QDATA_SET_ADJBLK(qdata);
 487         else
 488                 QDATA_SET_ADJINO(qdata);
 489
 490         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
 491 skip:
 492         lqs_putref(lqs);
 493
 494         return rc;
 495 }
 496
 497 int mds_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
 498                      const unsigned int qpids[], int rc, int opc)
 499 {
 500         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 501         int rc2 = 0;
 502         ENTRY;
 503
 504         if (rc && rc != -EDQUOT && rc != ENOLCK)
 505                 RETURN(0);
 506
 507         switch (opc) {
 508         case FSFILT_OP_SETATTR:
 509                 /* release file quota on original owner */
 510                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 0, 0, NULL);
 511                 /* release block quota on original owner */
 512                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 513                 /* acquire file quota on current owner */
 514                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 515                 /* acquire block quota on current owner */
 516                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 517                 break;
 518         case FSFILT_OP_UNLINK_PARTIAL_CHILD:
 519                 /* release file quota on child */
 520                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 521                 /* rlease block quota on child */
 522                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 523                 break;
 524         case FSFILT_OP_CREATE_PARTIAL_CHILD:
 525                 /* acquire file quota on child */
 526                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 527                 /* acquire block quota on child */
 528                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 529                 break;
 530         case FSFILT_OP_LINK:
 531                 /* acquire block quota on parent */
 532                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 533                 break;
 534         case FSFILT_OP_UNLINK:
 535                 /* release block quota on parent */
 536                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 537                 /* release file quota on child */
 538                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 539                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 540                         /* release block quota on child */
 541                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
 542                                                   NULL);
 543                 break;
 544         case FSFILT_OP_UNLINK_PARTIAL_PARENT:
 545                 /* release block quota on parent */
 546                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 547                 break;
 548         case FSFILT_OP_CREATE:
 549                 /* acquire block quota on parent */
 550                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 551                 /* acquire file quota on child */
 552                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 553                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 554                         /* acquire block quota on child */
 555                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
 556                                                   NULL);
 557                 break;
 558         default:
 559                 LBUG();
 560                 break;
 561         }
 562
 563         if (rc2)
 564                 CDEBUG(D_QUOTA,
 565                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
 566                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
 567         RETURN(0);
 568 }
 569
 570 int filter_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
 571                         const unsigned int qpids[], int rc, int opc)
 572 {
 573         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 574         int rc2 = 0;
 575         ENTRY;
 576
 577         if (rc && rc != -EDQUOT)
 578                 RETURN(0);
 579
 580         switch (opc) {
 581         case FSFILT_OP_SETATTR:
 582                 /* acquire/release block quota on original & current owner */
 583                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 584                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 585                 break;
 586         case FSFILT_OP_UNLINK:
 587                 /* release block quota on this owner */
 588         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 589                 /* acquire block quota on this owner */
 590                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 591                 break;
 592         default:
 593                 LBUG();
 594                 break;
 595         }
 596
 597         if (rc || rc2) {
 598                 if (!rc)
 599                         rc = rc2;
 600                 CDEBUG(D_QUOTA,
 601                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
 602                        rc == QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
 603         }
 604
 605         RETURN(0);
 606 }
 607
 608 static const char prefix[] = "OBJECTS/";
 609
 610 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 611 {
 612         struct mds_obd *mds = &obd->u.mds;
 613         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 614         struct obd_device_target *obt = &obd->u.obt;
 615         int rc = 0, i, rc1 = 0;
 616         char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 617         char name[64];
 618         struct lvfs_run_ctxt saved;
 619         ENTRY;
 620
 621         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 622
 623         if (oqctl->qc_type != USRQUOTA &&
 624             oqctl->qc_type != GRPQUOTA &&
 625             oqctl->qc_type != UGQUOTA)
 626                 RETURN(-EINVAL);
 627
 628         cfs_down(&obt->obt_quotachecking);
 629         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 630         cfs_down_write(&mds->mds_qonoff_sem);
 631
 632         for (i = 0; i < MAXQUOTAS; i++) {
 633                 struct file *fp;
 634
 635                 if (!Q_TYPESET(oqctl, i))
 636                         continue;
 637
 638                 /* quota file has been opened ? */
 639                 if (qinfo->qi_files[i]) {
 640                         CWARN("quota[%d] is on yet\n", i);
 641                         rc1 = -EBUSY;
 642                         continue;
 643                 }
 644
 645                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 646                 sprintf(name, "%s%s", prefix, quotafile[i]);
 647
 648                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 649                 if (IS_ERR(fp)) {
 650                         rc = PTR_ERR(fp);
 651                         CERROR("%s: error invalidating admin quotafile %s (rc:%d)\n",
 652                                obd->obd_name, name, rc);
 653                 }
 654                 else
 655                         filp_close(fp, 0);
 656         }
 657
 658         cfs_up_write(&mds->mds_qonoff_sem);
 659         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 660         cfs_up(&obt->obt_quotachecking);
 661         RETURN(rc ? : rc1);
 662 }
 663
 664 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 665 {
 666         struct mds_obd *mds = &obd->u.mds;
 667         struct obd_device_target *obt = &obd->u.obt;
 668         int rc;
 669         struct lvfs_run_ctxt saved;
 670         ENTRY;
 671
 672         if (oqctl->qc_type != USRQUOTA &&
 673             oqctl->qc_type != GRPQUOTA &&
 674             oqctl->qc_type != UGQUOTA)
 675                 RETURN(-EINVAL);
 676
 677         cfs_down(&obt->obt_quotachecking);
 678         if (obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))
 679                 GOTO(out, rc = -EBUSY);
 680         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 681         cfs_down_write(&mds->mds_qonoff_sem);
 682
 683         oqctl->qc_cmd = Q_FINVALIDATE;
 684         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 685         if (!rc)
 686                 rc = obd_quotactl(mds->mds_lov_exp, oqctl);
 687
 688         cfs_up_write(&mds->mds_qonoff_sem);
 689         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 690 out:
 691         cfs_up(&obt->obt_quotachecking);
 692         RETURN(rc);
 693 }
 694
 695 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 696 {
 697         struct mds_obd *mds = &obd->u.mds;
 698         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 699         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 700         struct lvfs_run_ctxt saved;
 701         char name[64];
 702         int i, rc = 0;
 703         ENTRY;
 704
 705         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 706
 707         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 708         cfs_down_write(&mds->mds_qonoff_sem);
 709
 710         for (i = 0; i < MAXQUOTAS && !rc; i++) {
 711                 struct file *fp;
 712
 713                 if (!Q_TYPESET(oqctl, i))
 714                         continue;
 715
 716                 /* quota file has been opened ? */
 717                 if (qinfo->qi_files[i]) {
 718                         CWARN("init %s admin quotafile while quota on.\n",
 719                               i == USRQUOTA ? "user" : "group");
 720                         continue;
 721                 }
 722
 723                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 724                 sprintf(name, "%s%s", prefix, quotafile[i]);
 725
 726                 /* check if quota file exists and is correct */
 727                 fp = filp_open(name, O_RDONLY, 0);
 728                 if (!IS_ERR(fp)) {
 729                         /* irregular file is not the right place for quota */
 730                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 731                                 CERROR("admin quota file %s is not "
 732                                        "regular!", name);
 733                                 filp_close(fp, 0);
 734                                 rc = -EINVAL;
 735                                 break;
 736                         }
 737                         qinfo->qi_files[i] = fp;
 738                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 739                         qinfo->qi_files[i] = NULL;
 740                         filp_close(fp, 0);
 741                 }
 742                 else
 743                         rc = PTR_ERR(fp);
 744
 745                 if (!rc)
 746                         continue;
 747
 748                 /* -EINVAL may be returned by quotainfo for bad quota file */
 749                 if (rc != -ENOENT && rc != -EINVAL) {
 750                         CERROR("%s: error opening old quota file %s (%d)\n",
 751                                obd->obd_name, name, rc);
 752                         break;
 753                 }
 754
 755                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
 756                        rc == -ENOENT ? "creating" : "overwriting");
 757
 758                 /* create quota file overwriting old if needed */
 759                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 760                 if (IS_ERR(fp)) {
 761                         rc = PTR_ERR(fp);
 762                         CERROR("%s: error creating admin quotafile %s (rc:%d)\n",
 763                                obd->obd_name, name, rc);
 764                         break;
 765                 }
 766
 767                 qinfo->qi_files[i] = fp;
 768
 769                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 770                 if (rc)
 771                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 772                                i == USRQUOTA ? "user" : "group", rc);
 773
 774                 filp_close(fp, 0);
 775                 qinfo->qi_files[i] = NULL;
 776         }
 777
 778         cfs_up_write(&mds->mds_qonoff_sem);
 779         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 780         RETURN(rc);
 781 }
 782
 783 static int close_quota_files(struct obd_quotactl *oqctl,
 784                              struct lustre_quota_info *qinfo)
 785 {
 786         int i, rc = 0;
 787         ENTRY;
 788
 789         for (i = 0; i < MAXQUOTAS; i++) {
 790                 if (!Q_TYPESET(oqctl, i))
 791                         continue;
 792                 if (qinfo->qi_files[i] == NULL) {
 793                         CDEBUG(D_QUOTA, "quota[%d] is off already\n", i);
 794                         rc = -EALREADY;
 795                         continue;
 796                 }
 797                 filp_close(qinfo->qi_files[i], 0);
 798                 qinfo->qi_files[i] = NULL;
 799         }
 800         RETURN(rc);
 801 }
 802
 803 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 804 {
 805         struct mds_obd *mds = &obd->u.mds;
 806         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 807         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 808         char name[64];
 809         int i, rc = 0, rc1 = 0;
 810         ENTRY;
 811
 812         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 813
 814         /* open admin quota files and read quotafile info */
 815         for (i = 0; i < MAXQUOTAS; i++) {
 816                 struct file *fp;
 817
 818                 if (!Q_TYPESET(oqctl, i) || qinfo->qi_files[i] != NULL)
 819                         continue;
 820
 821                 LASSERT(strlen(quotafile[i])
 822                         + sizeof(prefix) <= sizeof(name));
 823                 sprintf(name, "%s%s", prefix, quotafile[i]);
 824                 fp = filp_open(name, O_RDWR, 0);
 825                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 826                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
 827                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
 828                         break;
 829                 }
 830                 qinfo->qi_files[i] = fp;
 831
 832                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 833                 if (rc) {
 834                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
 835                         break;
 836                 }
 837
 838                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 839                 if (rc) {
 840                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
 841                                rc);
 842                         break;
 843                 }
 844         }
 845
 846         if (rc && rc1 != -EALREADY)
 847                 close_quota_files(oqctl, qinfo);
 848
 849         RETURN(rc ? : rc1);
 850 }
 851
 852 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 853 {
 854         int rc;
 855         ENTRY;
 856
 857         if (oqctl->qc_type != USRQUOTA &&
 858             oqctl->qc_type != GRPQUOTA &&
 859             oqctl->qc_type != UGQUOTA)
 860                 RETURN(-EINVAL);
 861
 862         rc = generic_quota_on(obd, oqctl, 1);
 863
 864         RETURN(rc);
 865 }
 866
 867
 868 int mds_admin_quota_off(struct obd_device *obd,
 869                         struct obd_quotactl *oqctl)
 870 {
 871         struct mds_obd *mds = &obd->u.mds;
 872         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 873         int rc;
 874         ENTRY;
 875
 876         /* close admin quota files */
 877         rc = close_quota_files(oqctl, qinfo);
 878         RETURN(rc);
 879 }
 880
 881
 882 /* with obt->obt_quotachecking held */
 883 int do_mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 884 {
 885         struct mds_obd *mds = &obd->u.mds;
 886         struct obd_device_target *obt = &obd->u.obt;
 887         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 888         struct lvfs_run_ctxt saved;
 889         int rc = 0, rc1 = 0, rc2 = 0;
 890         ENTRY;
 891
 892         LASSERT_SEM_LOCKED(&obt->obt_quotachecking);
 893
 894         if (oqctl->qc_type != USRQUOTA &&
 895             oqctl->qc_type != GRPQUOTA &&
 896             oqctl->qc_type != UGQUOTA)
 897                 RETURN(-EINVAL);
 898
 899         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 900         cfs_down_write(&mds->mds_qonoff_sem);
 901         /* close admin quota files */
 902         rc2 = mds_admin_quota_off(obd, oqctl);
 903         if (rc2 && rc2 != -EALREADY) {
 904                 CWARN("mds quota[%d] is failed to be off for %d\n", oqctl->qc_type, rc2);
 905                 GOTO(out, rc2);
 906         }
 907
 908         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 909         if (!rc1) {
 910                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
 911         } else if (quota_is_off(qctxt, oqctl)) {
 912                 CWARN("mds local quota[%d] is off already\n", oqctl->qc_type);
 913                 rc1 = -EALREADY;
 914         } else {
 915                 if (rc2 != -EALREADY) {
 916                         CWARN("mds local quota[%d] is failed to be off for %d\n",
 917                               oqctl->qc_type, rc1);
 918                         oqctl->qc_cmd = Q_QUOTAON;
 919                         mds_admin_quota_on(obd, oqctl);
 920                         oqctl->qc_cmd = Q_QUOTAOFF;
 921                 }
 922                 GOTO(out, rc1);
 923         }
 924
 925         rc = obd_quotactl(mds->mds_lov_exp, oqctl);
 926         if (rc && rc != -EALREADY) {
 927                 CWARN("mds remote quota[%d] is failed to be off for %d\n",
 928                       oqctl->qc_type, rc);
 929                 oqctl->qc_cmd = Q_QUOTAON;
 930                 if (rc2 != -EALREADY)
 931                         mds_admin_quota_on(obd, oqctl);
 932                 if (rc1 != -EALREADY) {
 933                         fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 934                         qctxt->lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
 935                 }
 936                 oqctl->qc_cmd = Q_QUOTAOFF;
 937         }
 938         EXIT;
 939
 940 out:
 941         CDEBUG(D_QUOTA, "%s: quotaoff type:flags:rc %u:%lu:%d\n",
 942                obd->obd_name, oqctl->qc_type, qctxt->lqc_flags, rc);
 943         cfs_up_write(&mds->mds_qonoff_sem);
 944         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 945         return rc ? : (rc1 ? : rc2);
 946 }
 947
 948 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 949 {
 950         struct obd_device_target *obt = &obd->u.obt;
 951         int rc;
 952         ENTRY;
 953
 954         cfs_down(&obt->obt_quotachecking);
 955         rc = do_mds_quota_off(obd, oqctl);
 956         cfs_up(&obt->obt_quotachecking);
 957         RETURN(rc);
 958 }
 959
 960 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 961 {
 962         struct mds_obd *mds = &obd->u.mds;
 963         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 964         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 965         int rc;
 966         ENTRY;
 967
 968         if (oqctl->qc_type != USRQUOTA &&
 969             oqctl->qc_type != GRPQUOTA)
 970                 RETURN(-EINVAL);
 971
 972         cfs_down_write(&mds->mds_qonoff_sem);
 973         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 974                 CWARN("quota[%u] is off\n", oqctl->qc_type);
 975                 GOTO(out, rc = -ESRCH);
 976         }
 977
 978         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 979         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 980         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 981
 982         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 983         EXIT;
 984
 985 out:
 986         cfs_up_write(&mds->mds_qonoff_sem);
 987         return rc;
 988 }
 989
 990 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 991 {
 992         struct mds_obd *mds = &obd->u.mds;
 993         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 994         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 995         int rc = 0;
 996         ENTRY;
 997
 998         if (oqctl->qc_type != USRQUOTA &&
 999             oqctl->qc_type != GRPQUOTA)
1000                 RETURN(-EINVAL);
1001
1002         cfs_down_read(&mds->mds_qonoff_sem);
1003         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1004                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1005                 GOTO(out, rc = -ESRCH);
1006         }
1007
1008         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
1009         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
1010         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
1011         EXIT;
1012
1013 out:
1014         cfs_up_read(&mds->mds_qonoff_sem);
1015         return rc;
1016 }
1017
1018 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
1019                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
1020                       int type, struct quota_adjust_qunit *oqaq)
1021 {
1022         __u64 bunit_curr_o, iunit_curr_o;
1023         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
1024         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
1025         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
1026                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
1027         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1028                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1029         int rc = 0;
1030         ENTRY;
1031
1032         if (!dquot || !oqaq)
1033                 RETURN(-EINVAL);
1034         LASSERT_MUTEX_LOCKED(&dquot->dq_mutex);
1035         LASSERT(oqaq->qaq_iunit_sz);
1036         LASSERT(oqaq->qaq_bunit_sz);
1037
1038         /* don't change qunit size */
1039         if (!qctxt->lqc_switch_qs)
1040                 RETURN(rc);
1041
1042         bunit_curr_o = oqaq->qaq_bunit_sz;
1043         iunit_curr_o = oqaq->qaq_iunit_sz;
1044
1045         if (dquot->dq_type == GRPQUOTA)
1046                 QAQ_SET_GRP(oqaq);
1047
1048         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1049                 __u64 b_limitation =
1050                         oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1051                 /* enlarge block qunit size */
1052                 while (blimit >
1053                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1054                         oqaq->qaq_bunit_sz =
1055                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1056                                 << QUOTABLOCK_BITS;
1057                         if (oqaq->qaq_bunit_sz >= qctxt->lqc_bunit_sz)
1058                                 break;
1059                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1060                                 shrink_qunit_limit;
1061                 }
1062
1063                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1064                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1065
1066                 /* shrink block qunit size */
1067                 while (blimit <
1068                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1069                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1070                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1071                                 QUOTABLOCK_BITS;
1072                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1073                                 shrink_qunit_limit;
1074                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1075                                 break;
1076                 }
1077
1078                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1079                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1080
1081                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1082                         QAQ_SET_ADJBLK(oqaq);
1083
1084         }
1085
1086         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1087                 __u64 i_limitation =
1088                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1089                 /* enlarge file qunit size */
1090                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1091                        + 2 * i_limitation) {
1092                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1093                         if (oqaq->qaq_iunit_sz >= qctxt->lqc_iunit_sz)
1094                                 break;
1095                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1096                                 shrink_qunit_limit;
1097                 }
1098
1099                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1100                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1101
1102                 /* shrink file qunit size */
1103                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1104                        + i_limitation) {
1105                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1106                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1107                                        shrink_qunit_limit;
1108                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1109                                 break;
1110                 }
1111
1112                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1113                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1114
1115                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1116                         QAQ_SET_ADJINO(oqaq);
1117
1118         }
1119
1120         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1121
1122         RETURN(rc);
1123 }
1124
1125 static int mds_init_slave_ilimits(struct obd_device *obd,
1126                                   struct obd_quotactl *oqctl, int set)
1127 {
1128         /* XXX: for file limits only adjust local now */
1129         struct obd_device_target *obt = &obd->u.obt;
1130         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1131         unsigned int id[MAXQUOTAS] = { 0, 0 };
1132         struct obd_quotactl *ioqc = NULL;
1133         struct lustre_qunit_size *lqs;
1134         int flag;
1135         int rc;
1136         ENTRY;
1137
1138         /* if we are going to set zero limit, needn't init slaves */
1139         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1140             !set)
1141                 RETURN(0);
1142
1143         OBD_ALLOC_PTR(ioqc);
1144         if (!ioqc)
1145                 RETURN(-ENOMEM);
1146
1147         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1148                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1149         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1150         ioqc->qc_id = oqctl->qc_id;
1151         ioqc->qc_type = oqctl->qc_type;
1152         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1153         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1154
1155         /* build lqs for mds */
1156         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1157                                qctxt, flag ? 1 : 0);
1158         if (lqs && !IS_ERR(lqs)) {
1159                 if (flag)
1160                         lqs->lqs_flags |= QI_SET;
1161                 else
1162                         lqs->lqs_flags &= ~QI_SET;
1163                 lqs_putref(lqs);
1164         } else {
1165                 CERROR("fail to %s lqs for inode(%s id: %u)!\n",
1166                        flag ? "create" : "search",
1167                        oqctl->qc_type ? "group" : "user",
1168                        oqctl->qc_id);
1169                 GOTO(out, rc = PTR_ERR(lqs));
1170         }
1171
1172         /* set local limit to MIN_QLIMIT */
1173         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1174         if (rc)
1175                 GOTO(out, rc);
1176
1177         /* trigger local qunit pre-acquire */
1178         if (oqctl->qc_type == USRQUOTA)
1179                 id[USRQUOTA] = oqctl->qc_id;
1180         else
1181                 id[GRPQUOTA] = oqctl->qc_id;
1182
1183         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1184         if (rc == -EDQUOT || rc == -EBUSY) {
1185                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1186                 rc = 0;
1187         }
1188         if (rc) {
1189                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1190                        rc);
1191                 GOTO(out, rc);
1192         }
1193         /* FIXME initialize all slaves in CMD */
1194         EXIT;
1195 out:
1196         if (ioqc)
1197                 OBD_FREE_PTR(ioqc);
1198         return rc;
1199 }
1200
1201 static int mds_init_slave_blimits(struct obd_device *obd,
1202                                   struct obd_quotactl *oqctl, int set)
1203 {
1204         struct obd_device_target *obt = &obd->u.obt;
1205         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1206         struct mds_obd *mds = &obd->u.mds;
1207         struct obd_quotactl *ioqc;
1208         struct lustre_qunit_size *lqs;
1209         unsigned int id[MAXQUOTAS] = { 0, 0 };
1210         int rc;
1211         int flag;
1212         ENTRY;
1213
1214         /* if we are going to set zero limit, needn't init slaves */
1215         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1216             !set)
1217                 RETURN(0);
1218
1219         OBD_ALLOC_PTR(ioqc);
1220         if (!ioqc)
1221                 RETURN(-ENOMEM);
1222
1223         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1224                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1225         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1226         ioqc->qc_id = oqctl->qc_id;
1227         ioqc->qc_type = oqctl->qc_type;
1228         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1229         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1230
1231         /* build lqs for mds */
1232         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1233                                qctxt, flag ? 1 : 0);
1234         if (lqs && !IS_ERR(lqs)) {
1235                 if (flag)
1236                         lqs->lqs_flags |= QB_SET;
1237                 else
1238                         lqs->lqs_flags &= ~QB_SET;
1239                 lqs_putref(lqs);
1240         } else {
1241                 CERROR("fail to %s lqs for block(%s id: %u)!\n",
1242                        flag ? "create" : "search",
1243                        oqctl->qc_type ? "group" : "user",
1244                        oqctl->qc_id);
1245                 GOTO(out, rc = PTR_ERR(lqs));
1246         }
1247
1248         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1249         if (rc)
1250                 GOTO(out, rc);
1251
1252         /* trigger local qunit pre-acquire */
1253         if (oqctl->qc_type == USRQUOTA)
1254                 id[USRQUOTA] = oqctl->qc_id;
1255         else
1256                 id[GRPQUOTA] = oqctl->qc_id;
1257
1258         /* initialize all slave's limit */
1259         rc = obd_quotactl(mds->mds_lov_exp, ioqc);
1260
1261         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1262         if (rc == -EDQUOT || rc == -EBUSY) {
1263                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1264                 rc = 0;
1265         }
1266         if (rc) {
1267                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1268                 GOTO(out, rc);
1269         }
1270
1271         EXIT;
1272 out:
1273         OBD_FREE_PTR(ioqc);
1274         return rc;
1275 }
1276
1277 static void adjust_lqs(struct obd_device *obd, struct quota_adjust_qunit *qaq)
1278 {
1279         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
1280         int rc = 0;
1281
1282         QAQ_SET_CREATE_LQS(qaq);
1283         /* adjust local lqs */
1284         rc = quota_adjust_slave_lqs(qaq, qctxt);
1285         if (rc < 0)
1286                 CERROR("adjust master's qunit size failed!(rc=%d)\n", rc);
1287
1288         /* adjust remote lqs */
1289         if (QAQ_IS_ADJBLK(qaq)) {
1290                 rc = obd_quota_adjust_qunit(obd->u.mds.mds_lov_exp, qaq, qctxt, NULL);
1291                 if (rc < 0)
1292                         CERROR("adjust slaves' qunit size failed!(rc=%d)\n", rc);
1293
1294         }
1295 }
1296
1297 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1298 {
1299         struct mds_obd *mds = &obd->u.mds;
1300         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1301         struct obd_device *lov_obd = class_exp2obd(mds->mds_lov_exp);
1302         struct lov_obd *lov = &lov_obd->u.lov;
1303         struct quota_adjust_qunit *oqaq = NULL;
1304         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1305         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1306         time_t btime, itime;
1307         struct lustre_dquot *dquot;
1308         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1309         /* orig_set means if quota was set before; now_set means we are
1310          * setting/cancelling quota */
1311         int orig_set, now_set;
1312         struct lustre_qunit_size *lqs;
1313         int rc = 0, rc2 = 0, flag = 0;
1314         ENTRY;
1315
1316         if (oqctl->qc_type != USRQUOTA &&
1317             oqctl->qc_type != GRPQUOTA)
1318                 RETURN(-EINVAL);
1319
1320         OBD_ALLOC_PTR(oqaq);
1321         if (!oqaq)
1322                 RETURN(-ENOMEM);
1323
1324         cfs_down_write(&mds->mds_qonoff_sem);
1325         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1326
1327         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1328                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1329                 GOTO(out_sem, rc = -ESRCH);
1330         }
1331
1332         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1333         if (IS_ERR(dquot))
1334                 GOTO(out_sem, rc = PTR_ERR(dquot));
1335         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1336         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1337
1338         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id), qctxt, 1);
1339         if (lqs == NULL)
1340                 rc = -ENOENT;
1341         if (IS_ERR(lqs))
1342                 rc = PTR_ERR(lqs);
1343         if (rc)
1344                 GOTO(out, rc);
1345
1346         cfs_mutex_lock(&dquot->dq_mutex);
1347         cfs_spin_lock(&lqs->lqs_lock);
1348         if (LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1349                 cfs_spin_unlock(&lqs->lqs_lock);
1350                 cfs_mutex_unlock(&dquot->dq_mutex);
1351                 GOTO(skip, rc = -EBUSY);
1352         }
1353         LQS_SET_SETQUOTA(lqs);
1354         cfs_spin_unlock(&lqs->lqs_lock);
1355
1356         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1357         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1358         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1359         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1360         btime = dquot->dq_dqb.dqb_btime;
1361         itime = dquot->dq_dqb.dqb_itime;
1362
1363         if (dqblk->dqb_valid & QIF_BTIME)
1364                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1365         if (dqblk->dqb_valid & QIF_ITIME)
1366                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1367
1368         if (dqblk->dqb_valid & QIF_BLIMITS) {
1369                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1370                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1371                 /* clear usage (limit pool) */
1372                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1373                     !dquot->dq_dqb.dqb_bsoftlimit)
1374                         dquot->dq_dqb.dqb_curspace = 0;
1375
1376                 /* clear grace time */
1377                 if (!dqblk->dqb_bsoftlimit ||
1378                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1379                         dquot->dq_dqb.dqb_btime = 0;
1380                 /* set grace only if user hasn't provided his own */
1381                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1382                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1383                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1384
1385                 flag |= LQUOTA_FLAGS_ADJBLK;
1386         }
1387
1388         if (dqblk->dqb_valid & QIF_ILIMITS) {
1389                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1390                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1391                 /* clear usage (limit pool) */
1392                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1393                     !dquot->dq_dqb.dqb_isoftlimit)
1394                         dquot->dq_dqb.dqb_curinodes = 0;
1395
1396                 if (!dqblk->dqb_isoftlimit ||
1397                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1398                         dquot->dq_dqb.dqb_itime = 0;
1399                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1400                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1401                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1402
1403                 flag |= LQUOTA_FLAGS_ADJINO;
1404         }
1405         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1406         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1407                                flag, oqaq);
1408         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1409         if (rc < 0)
1410                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1411
1412
1413         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1414
1415         cfs_mutex_unlock(&dquot->dq_mutex);
1416
1417         if (rc) {
1418                 CERROR("set limit failed! (rc:%d)\n", rc);
1419                 GOTO(update_fail, rc);
1420         }
1421
1422         cfs_up_write(&mds->mds_qonoff_sem);
1423         adjust_lqs(obd, oqaq);
1424
1425         orig_set = ihardlimit || isoftlimit;
1426         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1427         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1428                 cfs_mutex_lock(&dquot->dq_mutex);
1429                 dquot->dq_dqb.dqb_curinodes = 0;
1430                 cfs_mutex_unlock(&dquot->dq_mutex);
1431                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set);
1432                 if (rc) {
1433                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1434                         goto revoke_out;
1435                 }
1436         }
1437
1438         orig_set = bhardlimit || bsoftlimit;
1439         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1440         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1441                 cfs_mutex_lock(&dquot->dq_mutex);
1442                 dquot->dq_dqb.dqb_curspace = 0;
1443                 cfs_mutex_unlock(&dquot->dq_mutex);
1444                 rc = mds_init_slave_blimits(obd, oqctl, orig_set);
1445                 if (rc) {
1446                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1447                         goto revoke_out;
1448                 }
1449         }
1450
1451 revoke_out:
1452         cfs_down_write(&mds->mds_qonoff_sem);
1453         cfs_mutex_lock(&dquot->dq_mutex);
1454         if (rc) {
1455                 /* cancel previous setting */
1456                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1457                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1458                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1459                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1460                 dquot->dq_dqb.dqb_btime = btime;
1461                 dquot->dq_dqb.dqb_itime = itime;
1462         }
1463         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1464         cfs_mutex_unlock(&dquot->dq_mutex);
1465 update_fail:
1466         cfs_spin_lock(&lqs->lqs_lock);
1467         LQS_CLEAR_SETQUOTA(lqs);
1468         cfs_spin_unlock(&lqs->lqs_lock);
1469 skip:
1470         lqs_putref(lqs);
1471 out:
1472         lustre_dqput(dquot);
1473         EXIT;
1474 out_sem:
1475         cfs_up_write(&mds->mds_qonoff_sem);
1476
1477         if (oqaq)
1478                 OBD_FREE_PTR(oqaq);
1479
1480         return rc ? rc : rc2;
1481 }
1482
1483 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1484 {
1485         struct obd_quotactl *soqc;
1486         struct lvfs_run_ctxt saved;
1487         int rc, rc1;
1488         ENTRY;
1489
1490         OBD_ALLOC_PTR(soqc);
1491         if (!soqc)
1492                 RETURN(-ENOMEM);
1493
1494         soqc->qc_cmd = Q_GETOQUOTA;
1495         soqc->qc_id = oqctl->qc_id;
1496         soqc->qc_type = oqctl->qc_type;
1497
1498         /* get block usage from OSS */
1499         soqc->qc_dqblk.dqb_curspace = 0;
1500         rc = obd_quotactl(obd->u.mds.mds_lov_exp, soqc);
1501         if (!rc || rc == -EREMOTEIO) {
1502                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1503                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1504         }
1505
1506         /* get block/inode usage from MDS */
1507         soqc->qc_dqblk.dqb_curspace = 0;
1508         soqc->qc_dqblk.dqb_curinodes = 0;
1509         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1510         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1511         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1512         if (!rc1) {
1513                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1514                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1515                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1516         }
1517
1518         OBD_FREE_PTR(soqc);
1519
1520         RETURN(rc ? : rc1);
1521 }
1522
1523 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1524 {
1525         struct mds_obd *mds = &obd->u.mds;
1526         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1527         struct lustre_dquot *dquot;
1528         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1529         int rc;
1530         ENTRY;
1531
1532         if (oqctl->qc_type != USRQUOTA &&
1533             oqctl->qc_type != GRPQUOTA)
1534                 RETURN(-EINVAL);
1535
1536         cfs_down_read(&mds->mds_qonoff_sem);
1537         dqblk->dqb_valid = 0;
1538         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1539                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1540                 GOTO(out, rc = -ESRCH);
1541         }
1542
1543         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1544         if (IS_ERR(dquot))
1545                 GOTO(out, rc = PTR_ERR(dquot));
1546
1547         cfs_mutex_lock(&dquot->dq_mutex);
1548         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1549         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1550         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1551         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1552         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1553         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1554         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1555         cfs_mutex_unlock(&dquot->dq_mutex);
1556
1557         lustre_dqput(dquot);
1558         cfs_up_read(&mds->mds_qonoff_sem);
1559
1560         /* the usages in admin quota file is inaccurate */
1561         dqblk->dqb_curinodes = 0;
1562         dqblk->dqb_curspace = 0;
1563         rc = mds_get_space(obd, oqctl);
1564
1565         RETURN(rc);
1566
1567 out:
1568         cfs_up_read(&mds->mds_qonoff_sem);
1569         return rc;
1570 }
1571
1572 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1573 {
1574         struct lvfs_run_ctxt saved;
1575         int rc;
1576         ENTRY;
1577
1578         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1579         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1580         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1581
1582         RETURN(rc);
1583 }
1584
1585
1586 /* FIXME we only recovery block limit by now, need recovery inode
1587  * limits also after CMD involved in */
1588 static int
1589 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1590 {
1591         struct mds_obd *mds = &obd->u.mds;
1592         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1593         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1594         struct lustre_qunit_size *lqs;
1595         struct lustre_dquot *dquot;
1596         struct obd_quotactl *qctl;
1597         __u64 total_limits = 0;
1598         int rc = 0;
1599         ENTRY;
1600
1601         OBD_ALLOC_PTR(qctl);
1602         if (qctl == NULL)
1603                 RETURN(-ENOMEM);
1604
1605         dquot = lustre_dqget(obd, qinfo, id, type, 0);
1606         if (IS_ERR(dquot)) {
1607                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1608                 OBD_FREE_PTR(qctl);
1609                 RETURN(PTR_ERR(dquot));
1610         }
1611
1612         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 1);
1613         if (lqs == NULL)
1614                 rc = -ENOENT;
1615         if (IS_ERR(lqs))
1616                 rc = PTR_ERR(lqs);
1617         if (rc)
1618                 GOTO(skip, rc);
1619
1620         cfs_mutex_lock(&dquot->dq_mutex);
1621
1622         /* don't recover the dquot without limits or quota is setting or
1623          * another recovery is already going on */
1624         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1625             LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1626                 cfs_mutex_unlock(&dquot->dq_mutex);
1627                 GOTO(skip1, rc = 0);
1628         }
1629
1630         cfs_spin_lock(&lqs->lqs_lock);
1631         LQS_SET_RECOVERY(lqs);
1632         cfs_spin_unlock(&lqs->lqs_lock);
1633         cfs_mutex_unlock(&dquot->dq_mutex);
1634
1635         /* release mds_qonoff_sem during obd_quotactl ops here */
1636         cfs_up_write(&mds->mds_qonoff_sem);
1637
1638         /* get real bhardlimit from all slaves. */
1639         qctl->qc_cmd = Q_GETOQUOTA;
1640         qctl->qc_type = type;
1641         qctl->qc_id = id;
1642         qctl->qc_stat = QUOTA_RECOVERING;
1643         rc = obd_quotactl(mds->mds_lov_exp, qctl);
1644         cfs_down_write(&mds->mds_qonoff_sem);
1645         if (rc)
1646                 GOTO(out, rc);
1647         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1648
1649         /* get real bhardlimit from master */
1650         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1651         if (rc)
1652                 GOTO(out, rc);
1653         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1654
1655         /* amend the usage of the administrative quotafile */
1656         cfs_mutex_lock(&dquot->dq_mutex);
1657
1658         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1659
1660         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1661         if (rc)
1662                 CERROR("write dquot failed! (rc:%d)\n", rc);
1663
1664         cfs_mutex_unlock(&dquot->dq_mutex);
1665         EXIT;
1666 out:
1667         cfs_spin_lock(&lqs->lqs_lock);
1668         LQS_CLEAR_RECOVERY(lqs);
1669         cfs_spin_unlock(&lqs->lqs_lock);
1670 skip1:
1671         lqs_putref(lqs);
1672 skip:
1673         lustre_dqput(dquot);
1674         OBD_FREE_PTR(qctl);
1675         return rc;
1676 }
1677
1678 struct qmaster_recov_thread_data {
1679         struct obd_device *obd;
1680         cfs_completion_t comp;
1681 };
1682
1683 static int qmaster_recovery_main(void *arg)
1684 {
1685         struct qmaster_recov_thread_data *data = arg;
1686         struct obd_device *obd = data->obd;
1687         struct mds_obd *mds = &obd->u.mds;
1688         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1689         int rc = 0;
1690         unsigned short type;
1691         ENTRY;
1692
1693         cfs_daemonize_ctxt("qmaster_recovd");
1694
1695         /* for mds */
1696         class_incref(obd, "qmaster_recovd_mds", obd);
1697         /* for lov */
1698         class_incref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1699
1700         cfs_complete(&data->comp);
1701
1702         cfs_down_write(&mds->mds_qonoff_sem);
1703         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1704                 cfs_list_t id_list;
1705                 struct dquot_id *dqid, *tmp;
1706
1707                 if (qinfo->qi_files[type] == NULL)
1708                         continue;
1709
1710                 CFS_INIT_LIST_HEAD(&id_list);
1711                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1712                                  &id_list);
1713                 if (rc)
1714                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1715
1716                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1717                         cfs_list_del_init(&dqid->di_link);
1718                         if (rc)
1719                                 goto free;
1720
1721                         rc = dquot_recovery(obd, dqid->di_id, type);
1722                         if (rc)
1723                                 CERROR("%s: qmaster recovery failed for %sid %d"
1724                                        " rc:%d)\n", obd->obd_name,
1725                                        type ? "g" : "u", dqid->di_id, rc);
1726 free:
1727                         OBD_FREE_PTR(dqid);
1728                 }
1729         }
1730         cfs_up_write(&mds->mds_qonoff_sem);
1731         class_decref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1732         class_decref(obd, "qmaster_recovd_mds", obd);
1733         RETURN(rc);
1734 }
1735
1736 int mds_quota_recovery(struct obd_device *obd)
1737 {
1738         struct mds_obd *mds = &obd->u.mds;
1739         struct qmaster_recov_thread_data data;
1740         int rc = 0;
1741         ENTRY;
1742
1743         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
1744                 RETURN(0);
1745
1746         if (unlikely(!mds->mds_quota || obd->obd_stopping))
1747                 RETURN(rc);
1748
1749         cfs_mutex_lock(&obd->obd_dev_mutex);
1750         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1751                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1752                       mds->mds_lov_desc.ld_active_tgt_count,
1753                       mds->mds_lov_objid_count);
1754                 cfs_mutex_unlock(&obd->obd_dev_mutex);
1755                 RETURN(rc);
1756         }
1757         cfs_mutex_unlock(&obd->obd_dev_mutex);
1758
1759         data.obd = obd;
1760         cfs_init_completion(&data.comp);
1761
1762         rc = cfs_create_thread(qmaster_recovery_main, &data,
1763                                CFS_DAEMON_FLAGS);
1764         if (rc < 0)
1765                 CERROR("%s: cannot start quota recovery thread: rc %d\n",
1766                        obd->obd_name, rc);
1767
1768         cfs_wait_for_completion(&data.comp);
1769         RETURN(rc);
1770 }