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 #define DEBUG_SUBSYSTEM S_LQUOTA
  44
  45 #include <linux/version.h>
  46 #include <linux/fs.h>
  47 #include <asm/unistd.h>
  48 #include <linux/slab.h>
  49 #include <linux/quotaops.h>
  50 #include <linux/module.h>
  51 #include <linux/init.h>
  52 #include <linux/quota.h>
  53
  54 #include <obd_class.h>
  55 #include <lustre_quota.h>
  56 #include <lustre_fsfilt.h>
  57 #include <lustre_mds.h>
  58
  59 #include "quota_internal.h"
  60
  61 /* lock ordering: mds->mds_qonoff_sem > dquot->dq_mutex > lqs->lqs_lock */
  62 static cfs_list_t lustre_dquot_hash[NR_DQHASH];
  63 static DEFINE_RWLOCK(dquot_hash_lock);
  64
  65 cfs_mem_cache_t *lustre_dquot_cachep;
  66
  67 int lustre_dquot_init(void)
  68 {
  69         int i;
  70         ENTRY;
  71
  72         LASSERT(lustre_dquot_cachep == NULL);
  73         lustre_dquot_cachep = cfs_mem_cache_create("lustre_dquot_cache",
  74                                                    sizeof(struct lustre_dquot),
  75                                                    0, 0);
  76         if (!lustre_dquot_cachep)
  77                 return (-ENOMEM);
  78
  79         for (i = 0; i < NR_DQHASH; i++) {
  80                 CFS_INIT_LIST_HEAD(lustre_dquot_hash + i);
  81         }
  82         RETURN(0);
  83 }
  84
  85 void lustre_dquot_exit(void)
  86 {
  87         int i;
  88         ENTRY;
  89         /* FIXME cleanup work ?? */
  90
  91         for (i = 0; i < NR_DQHASH; i++) {
  92                 LASSERT(cfs_list_empty(lustre_dquot_hash + i));
  93         }
  94         if (lustre_dquot_cachep) {
  95                 int rc;
  96                 rc = cfs_mem_cache_destroy(lustre_dquot_cachep);
  97                 LASSERTF(rc == 0,"couldn't destroy lustre_dquot_cachep slab\n");
  98                 lustre_dquot_cachep = NULL;
  99         }
 100         EXIT;
 101 }
 102
 103 static inline int
 104 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 105              __attribute__((__const__));
 106
 107 static inline int
 108 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 109 {
 110         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
 111         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 112         return tmp;
 113 }
 114
 115 /* caller must hold dquot_hash_lock */
 116 static struct lustre_dquot *find_dquot(int hashent,
 117                                        struct lustre_quota_info *lqi, qid_t id,
 118                                        int type)
 119 {
 120         struct lustre_dquot *dquot;
 121         ENTRY;
 122
 123         cfs_list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
 124                 if (dquot->dq_info == lqi &&
 125                     dquot->dq_id == id && dquot->dq_type == type)
 126                         RETURN(dquot);
 127         }
 128         RETURN(NULL);
 129 }
 130
 131 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 132                                         qid_t id, int type)
 133 {
 134         struct lustre_dquot *dquot = NULL;
 135         ENTRY;
 136
 137         OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
 138         if (dquot == NULL)
 139                 RETURN(NULL);
 140
 141         CFS_INIT_LIST_HEAD(&dquot->dq_hash);
 142         cfs_mutex_init(&dquot->dq_mutex);
 143         cfs_mutex_lock(&dquot->dq_mutex);
 144         cfs_atomic_set(&dquot->dq_refcnt, 1);
 145         dquot->dq_info = lqi;
 146         dquot->dq_id = id;
 147         dquot->dq_type = type;
 148
 149         RETURN(dquot);
 150 }
 151
 152 static void free_dquot(struct lustre_dquot *dquot)
 153 {
 154         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 155 }
 156
 157 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 158 {
 159         cfs_list_t *head = lustre_dquot_hash +
 160             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 161         LASSERT(cfs_list_empty(&dquot->dq_hash));
 162         cfs_list_add(&dquot->dq_hash, head);
 163 }
 164
 165 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 166 {
 167         LASSERT(!cfs_list_empty(&dquot->dq_hash));
 168         cfs_list_del_init(&dquot->dq_hash);
 169 }
 170
 171 static void lustre_dqput(struct lustre_dquot *dquot)
 172 {
 173         ENTRY;
 174         cfs_write_lock(&dquot_hash_lock);
 175         LASSERT(cfs_atomic_read(&dquot->dq_refcnt));
 176         cfs_atomic_dec(&dquot->dq_refcnt);
 177         if (cfs_atomic_read(&dquot->dq_refcnt) == 0) {
 178                 remove_dquot_nolock(dquot);
 179                 free_dquot(dquot);
 180         }
 181         cfs_write_unlock(&dquot_hash_lock);
 182         EXIT;
 183 }
 184
 185 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 186                                          struct lustre_quota_info *lqi,
 187                                          qid_t id, int type, int can_fake)
 188 {
 189         unsigned int hashent = dquot_hashfn(lqi, id, type);
 190         struct lustre_dquot *dquot, *empty;
 191         int free_dq = 0;
 192         ENTRY;
 193
 194         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 195                 RETURN(ERR_PTR(-ENOMEM));
 196
 197         cfs_read_lock(&dquot_hash_lock);
 198         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 199                 cfs_atomic_inc(&dquot->dq_refcnt);
 200                 cfs_read_unlock(&dquot_hash_lock);
 201                 cfs_mutex_unlock(&empty->dq_mutex);
 202                 free_dq = 1;
 203         } else {
 204                 int rc;
 205
 206                 cfs_read_unlock(&dquot_hash_lock);
 207
 208                 dquot = empty;
 209                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
 210                 cfs_mutex_unlock(&dquot->dq_mutex);
 211                 if (rc) {
 212                         CERROR("can't read dquot from admin quotafile! "
 213                                "(rc:%d)\n", rc);
 214                         free_dquot(dquot);
 215                         RETURN(ERR_PTR(rc));
 216                 } else {
 217                         cfs_write_lock(&dquot_hash_lock);
 218                         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 219                                 cfs_atomic_inc(&dquot->dq_refcnt);
 220                                 free_dq = 1;
 221                         } else {
 222                                 dquot = empty;
 223                                 insert_dquot_nolock(dquot);
 224                         }
 225                         cfs_write_unlock(&dquot_hash_lock);
 226                 }
 227
 228         }
 229
 230         LASSERT(dquot);
 231         if (!can_fake && cfs_test_bit(DQ_FAKE_B, &dquot->dq_flags)) {
 232                 DQUOT_DEBUG(dquot, "It is a fake dquot: unexpected!\n");
 233                 lustre_dqput(dquot);
 234                 dquot = ERR_PTR(-ENOENT);
 235         }
 236
 237         if (free_dq)
 238                 free_dquot(empty);
 239
 240
 241         RETURN(dquot);
 242 }
 243
 244 static void init_oqaq(struct quota_adjust_qunit *oqaq,
 245                       struct lustre_quota_ctxt *qctxt,
 246                       qid_t id, int type)
 247 {
 248         struct lustre_qunit_size *lqs = NULL;
 249
 250         oqaq->qaq_id = id;
 251         oqaq->qaq_flags = type;
 252         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 0);
 253         if (lqs && !IS_ERR(lqs)) {
 254                 cfs_spin_lock(&lqs->lqs_lock);
 255                 oqaq->qaq_bunit_sz = lqs->lqs_bunit_sz;
 256                 oqaq->qaq_iunit_sz = lqs->lqs_iunit_sz;
 257                 oqaq->qaq_flags    = lqs->lqs_flags;
 258                 cfs_spin_unlock(&lqs->lqs_lock);
 259                 lqs_putref(lqs);
 260         } else {
 261                 CDEBUG(D_QUOTA, "Can't find the lustre qunit size!\n");
 262                 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
 263                 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
 264         }
 265 }
 266
 267 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
 268                           __u32 is_blk)
 269 {
 270         struct mds_obd *mds = &obd->u.mds;
 271         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 272         __u32 ost_num = mds->mds_lov_objid_count, mdt_num = 1;
 273         struct quota_adjust_qunit *oqaq = NULL;
 274         unsigned int qid[MAXQUOTAS] = { 0, 0 };
 275         struct lustre_quota_info *info = &mds->mds_quota_info;
 276         struct lustre_dquot *dquot = NULL;
 277         int adjust_res = 0;
 278         int rc = 0;
 279         ENTRY;
 280
 281         LASSERT(mds);
 282         cfs_down_read(&mds->mds_qonoff_sem);
 283         dquot = lustre_dqget(obd, info, id, type, 0);
 284         if (IS_ERR(dquot))
 285                 RETURN(PTR_ERR(dquot));
 286
 287         cfs_up_read(&mds->mds_qonoff_sem);
 288         OBD_ALLOC_PTR(oqaq);
 289         if (!oqaq)
 290                 GOTO(out, rc = -ENOMEM);
 291
 292         cfs_mutex_lock(&dquot->dq_mutex);
 293         init_oqaq(oqaq, qctxt, id, type);
 294
 295         rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
 296                                is_blk ? LQUOTA_FLAGS_ADJBLK :
 297                                LQUOTA_FLAGS_ADJINO, oqaq);
 298
 299         if (rc < 0) {
 300                 CERROR("create oqaq failed! (rc:%d)\n", rc);
 301                 GOTO(out_sem, rc);
 302         }
 303         QAQ_DEBUG(oqaq, "show oqaq.\n")
 304
 305         if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
 306                 GOTO(out_sem, rc);
 307
 308         /* adjust the mds slave qunit size */
 309         adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
 310         if (adjust_res <= 0) {
 311                 if (adjust_res < 0) {
 312                         rc = adjust_res;
 313                         CERROR("adjust mds slave's qunit size failed! "
 314                                "(rc:%d)\n", rc);
 315                 } else {
 316                         CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
 317                 }
 318                 GOTO(out_sem, rc);
 319         }
 320
 321         if (type)
 322                 qid[GRPQUOTA] = dquot->dq_id;
 323         else
 324                 qid[USRQUOTA] = dquot->dq_id;
 325
 326         cfs_mutex_unlock(&dquot->dq_mutex);
 327
 328         rc = qctxt_adjust_qunit(obd, qctxt, qid, is_blk, 0, NULL);
 329         if (rc == -EDQUOT || rc == -EBUSY) {
 330                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 331                 rc = 0;
 332         }
 333         if (rc) {
 334                 CERROR("%s: mds fail to adjust file quota! (rc:%d)\n",
 335                        obd->obd_name, rc);
 336                 GOTO(out, rc);
 337         }
 338
 339         /* only when block qunit is reduced, boardcast to osts */
 340         if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
 341                 rc = obd_quota_adjust_qunit(mds->mds_lov_exp, oqaq, qctxt, NULL);
 342
 343 out:
 344         lustre_dqput(dquot);
 345         if (oqaq)
 346                 OBD_FREE_PTR(oqaq);
 347
 348         RETURN(rc);
 349 out_sem:
 350         cfs_mutex_unlock(&dquot->dq_mutex);
 351         goto out;
 352 }
 353
 354 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
 355 {
 356         struct mds_obd *mds = &obd->u.mds;
 357         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 358         struct lustre_quota_info *info = &mds->mds_quota_info;
 359         struct lustre_dquot *dquot = NULL;
 360         __u64 *usage = NULL;
 361         __u64 hlimit = 0, slimit = 0;
 362         time_t *time = NULL;
 363         unsigned int grace = 0;
 364         struct lustre_qunit_size *lqs = NULL;
 365         int rc = 0;
 366         ENTRY;
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
 369                 RETURN(-EIO);
 370
 371         if (!ll_sb_has_quota_active(qctxt->lqc_sb,
 372                                     QDATA_IS_GRP(qdata) ? GRPQUOTA : USRQUOTA))
 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_mutex_lock(&dquot->dq_mutex);
 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)
 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_mutex_unlock(&dquot->dq_mutex);
 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                         CDEBUG(D_QUOTA, "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         CDEBUG(D_QUOTA, "%s: quotaoff type:flags:rc %u:%lu:%d\n",
 938                obd->obd_name, oqctl->qc_type, qctxt->lqc_flags, rc);
 939         cfs_up_write(&mds->mds_qonoff_sem);
 940         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 941         return rc ? : (rc1 ? : rc2);
 942 }
 943
 944 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 945 {
 946         struct obd_device_target *obt = &obd->u.obt;
 947         int rc;
 948         ENTRY;
 949
 950         cfs_down(&obt->obt_quotachecking);
 951         rc = do_mds_quota_off(obd, oqctl);
 952         cfs_up(&obt->obt_quotachecking);
 953         RETURN(rc);
 954 }
 955
 956 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 957 {
 958         struct mds_obd *mds = &obd->u.mds;
 959         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 960         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 961         int rc;
 962         ENTRY;
 963
 964         if (oqctl->qc_type != USRQUOTA &&
 965             oqctl->qc_type != GRPQUOTA)
 966                 RETURN(-EINVAL);
 967
 968         cfs_down_write(&mds->mds_qonoff_sem);
 969         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 970                 CWARN("quota[%u] is off\n", oqctl->qc_type);
 971                 GOTO(out, rc = -ESRCH);
 972         }
 973
 974         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 975         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 976         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 977
 978         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 979         EXIT;
 980
 981 out:
 982         cfs_up_write(&mds->mds_qonoff_sem);
 983         return rc;
 984 }
 985
 986 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 987 {
 988         struct mds_obd *mds = &obd->u.mds;
 989         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 990         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 991         int rc = 0;
 992         ENTRY;
 993
 994         if (oqctl->qc_type != USRQUOTA &&
 995             oqctl->qc_type != GRPQUOTA)
 996                 RETURN(-EINVAL);
 997
 998         cfs_down_read(&mds->mds_qonoff_sem);
 999         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1000                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1001                 GOTO(out, rc = -ESRCH);
1002         }
1003
1004         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
1005         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
1006         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
1007         EXIT;
1008
1009 out:
1010         cfs_up_read(&mds->mds_qonoff_sem);
1011         return rc;
1012 }
1013
1014 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
1015                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
1016                       int type, struct quota_adjust_qunit *oqaq)
1017 {
1018         __u64 bunit_curr_o, iunit_curr_o;
1019         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
1020         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
1021         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
1022                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
1023         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1024                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1025         int rc = 0;
1026         ENTRY;
1027
1028         if (!dquot || !oqaq)
1029                 RETURN(-EINVAL);
1030         LASSERT_MUTEX_LOCKED(&dquot->dq_mutex);
1031         LASSERT(oqaq->qaq_iunit_sz);
1032         LASSERT(oqaq->qaq_bunit_sz);
1033
1034         /* don't change qunit size */
1035         if (!qctxt->lqc_switch_qs)
1036                 RETURN(rc);
1037
1038         bunit_curr_o = oqaq->qaq_bunit_sz;
1039         iunit_curr_o = oqaq->qaq_iunit_sz;
1040
1041         if (dquot->dq_type == GRPQUOTA)
1042                 QAQ_SET_GRP(oqaq);
1043
1044         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1045                 __u64 b_limitation =
1046                         oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1047                 /* enlarge block qunit size */
1048                 while (blimit >
1049                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1050                         oqaq->qaq_bunit_sz =
1051                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1052                                 << QUOTABLOCK_BITS;
1053                         if (oqaq->qaq_bunit_sz >= qctxt->lqc_bunit_sz)
1054                                 break;
1055                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1056                                 shrink_qunit_limit;
1057                 }
1058
1059                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1060                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1061
1062                 /* shrink block qunit size */
1063                 while (blimit <
1064                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1065                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1066                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1067                                 QUOTABLOCK_BITS;
1068                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1069                                 shrink_qunit_limit;
1070                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1071                                 break;
1072                 }
1073
1074                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1075                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1076
1077                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1078                         QAQ_SET_ADJBLK(oqaq);
1079
1080         }
1081
1082         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1083                 __u64 i_limitation =
1084                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1085                 /* enlarge file qunit size */
1086                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1087                        + 2 * i_limitation) {
1088                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1089                         if (oqaq->qaq_iunit_sz >= qctxt->lqc_iunit_sz)
1090                                 break;
1091                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1092                                 shrink_qunit_limit;
1093                 }
1094
1095                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1096                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1097
1098                 /* shrink file qunit size */
1099                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1100                        + i_limitation) {
1101                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1102                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1103                                        shrink_qunit_limit;
1104                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1105                                 break;
1106                 }
1107
1108                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1109                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1110
1111                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1112                         QAQ_SET_ADJINO(oqaq);
1113
1114         }
1115
1116         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1117
1118         RETURN(rc);
1119 }
1120
1121 static int mds_init_slave_ilimits(struct obd_device *obd,
1122                                   struct obd_quotactl *oqctl, int set)
1123 {
1124         /* XXX: for file limits only adjust local now */
1125         struct obd_device_target *obt = &obd->u.obt;
1126         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1127         unsigned int id[MAXQUOTAS] = { 0, 0 };
1128         struct obd_quotactl *ioqc = NULL;
1129         struct lustre_qunit_size *lqs;
1130         int flag;
1131         int rc;
1132         ENTRY;
1133
1134         /* if we are going to set zero limit, needn't init slaves */
1135         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1136             !set)
1137                 RETURN(0);
1138
1139         OBD_ALLOC_PTR(ioqc);
1140         if (!ioqc)
1141                 RETURN(-ENOMEM);
1142
1143         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1144                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1145         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1146         ioqc->qc_id = oqctl->qc_id;
1147         ioqc->qc_type = oqctl->qc_type;
1148         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1149         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1150
1151         /* build lqs for mds */
1152         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1153                                qctxt, flag ? 1 : 0);
1154         if (lqs && !IS_ERR(lqs)) {
1155                 if (flag)
1156                         lqs->lqs_flags |= QI_SET;
1157                 else
1158                         lqs->lqs_flags &= ~QI_SET;
1159                 lqs_putref(lqs);
1160         } else {
1161                 CERROR("fail to %s lqs for inode(%s id: %u)!\n",
1162                        flag ? "create" : "search",
1163                        oqctl->qc_type ? "group" : "user",
1164                        oqctl->qc_id);
1165                 GOTO(out, rc = PTR_ERR(lqs));
1166         }
1167
1168         /* set local limit to MIN_QLIMIT */
1169         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1170         if (rc)
1171                 GOTO(out, rc);
1172
1173         /* trigger local qunit pre-acquire */
1174         if (oqctl->qc_type == USRQUOTA)
1175                 id[USRQUOTA] = oqctl->qc_id;
1176         else
1177                 id[GRPQUOTA] = oqctl->qc_id;
1178
1179         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1180         if (rc == -EDQUOT || rc == -EBUSY) {
1181                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1182                 rc = 0;
1183         }
1184         if (rc) {
1185                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1186                        rc);
1187                 GOTO(out, rc);
1188         }
1189         /* FIXME initialize all slaves in CMD */
1190         EXIT;
1191 out:
1192         if (ioqc)
1193                 OBD_FREE_PTR(ioqc);
1194         return rc;
1195 }
1196
1197 static int mds_init_slave_blimits(struct obd_device *obd,
1198                                   struct obd_quotactl *oqctl, int set)
1199 {
1200         struct obd_device_target *obt = &obd->u.obt;
1201         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1202         struct mds_obd *mds = &obd->u.mds;
1203         struct obd_quotactl *ioqc;
1204         struct lustre_qunit_size *lqs;
1205         unsigned int id[MAXQUOTAS] = { 0, 0 };
1206         int rc;
1207         int flag;
1208         ENTRY;
1209
1210         /* if we are going to set zero limit, needn't init slaves */
1211         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1212             !set)
1213                 RETURN(0);
1214
1215         OBD_ALLOC_PTR(ioqc);
1216         if (!ioqc)
1217                 RETURN(-ENOMEM);
1218
1219         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1220                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1221         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1222         ioqc->qc_id = oqctl->qc_id;
1223         ioqc->qc_type = oqctl->qc_type;
1224         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1225         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1226
1227         /* build lqs for mds */
1228         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1229                                qctxt, flag ? 1 : 0);
1230         if (lqs && !IS_ERR(lqs)) {
1231                 if (flag)
1232                         lqs->lqs_flags |= QB_SET;
1233                 else
1234                         lqs->lqs_flags &= ~QB_SET;
1235                 lqs_putref(lqs);
1236         } else {
1237                 CERROR("fail to %s lqs for block(%s id: %u)!\n",
1238                        flag ? "create" : "search",
1239                        oqctl->qc_type ? "group" : "user",
1240                        oqctl->qc_id);
1241                 GOTO(out, rc = PTR_ERR(lqs));
1242         }
1243
1244         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1245         if (rc)
1246                 GOTO(out, rc);
1247
1248         /* trigger local qunit pre-acquire */
1249         if (oqctl->qc_type == USRQUOTA)
1250                 id[USRQUOTA] = oqctl->qc_id;
1251         else
1252                 id[GRPQUOTA] = oqctl->qc_id;
1253
1254         /* initialize all slave's limit */
1255         rc = obd_quotactl(mds->mds_lov_exp, ioqc);
1256
1257         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1258         if (rc == -EDQUOT || rc == -EBUSY) {
1259                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1260                 rc = 0;
1261         }
1262         if (rc) {
1263                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1264                 GOTO(out, rc);
1265         }
1266
1267         EXIT;
1268 out:
1269         OBD_FREE_PTR(ioqc);
1270         return rc;
1271 }
1272
1273 static void adjust_lqs(struct obd_device *obd, struct quota_adjust_qunit *qaq)
1274 {
1275         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
1276         int rc = 0;
1277
1278         QAQ_SET_CREATE_LQS(qaq);
1279         /* adjust local lqs */
1280         rc = quota_adjust_slave_lqs(qaq, qctxt);
1281         if (rc < 0)
1282                 CERROR("adjust master's qunit size failed!(rc=%d)\n", rc);
1283
1284         /* adjust remote lqs */
1285         if (QAQ_IS_ADJBLK(qaq)) {
1286                 rc = obd_quota_adjust_qunit(obd->u.mds.mds_lov_exp, qaq, qctxt, NULL);
1287                 if (rc < 0)
1288                         CERROR("adjust slaves' qunit size failed!(rc=%d)\n", rc);
1289
1290         }
1291 }
1292
1293 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1294 {
1295         struct mds_obd *mds = &obd->u.mds;
1296         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1297         struct obd_device *lov_obd = class_exp2obd(mds->mds_lov_exp);
1298         struct lov_obd *lov = &lov_obd->u.lov;
1299         struct quota_adjust_qunit *oqaq = NULL;
1300         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1301         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1302         time_t btime, itime;
1303         struct lustre_dquot *dquot;
1304         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1305         /* orig_set means if quota was set before; now_set means we are
1306          * setting/cancelling quota */
1307         int orig_set, now_set;
1308         struct lustre_qunit_size *lqs;
1309         int rc = 0, rc2 = 0, flag = 0;
1310         ENTRY;
1311
1312         if (oqctl->qc_type != USRQUOTA &&
1313             oqctl->qc_type != GRPQUOTA)
1314                 RETURN(-EINVAL);
1315
1316         OBD_ALLOC_PTR(oqaq);
1317         if (!oqaq)
1318                 RETURN(-ENOMEM);
1319
1320         cfs_down_write(&mds->mds_qonoff_sem);
1321         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1322
1323         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1324                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1325                 GOTO(out_sem, rc = -ESRCH);
1326         }
1327
1328         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1329         if (IS_ERR(dquot))
1330                 GOTO(out_sem, rc = PTR_ERR(dquot));
1331         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1332         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1333
1334         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id), qctxt, 1);
1335         if (lqs == NULL)
1336                 rc = -ENOENT;
1337         if (IS_ERR(lqs))
1338                 rc = PTR_ERR(lqs);
1339         if (rc)
1340                 GOTO(out, rc);
1341
1342         cfs_mutex_lock(&dquot->dq_mutex);
1343         cfs_spin_lock(&lqs->lqs_lock);
1344         if (LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1345                 cfs_spin_unlock(&lqs->lqs_lock);
1346                 cfs_mutex_unlock(&dquot->dq_mutex);
1347                 GOTO(skip, rc = -EBUSY);
1348         }
1349         LQS_SET_SETQUOTA(lqs);
1350         cfs_spin_unlock(&lqs->lqs_lock);
1351
1352         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1353         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1354         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1355         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1356         btime = dquot->dq_dqb.dqb_btime;
1357         itime = dquot->dq_dqb.dqb_itime;
1358
1359         if (dqblk->dqb_valid & QIF_BTIME)
1360                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1361         if (dqblk->dqb_valid & QIF_ITIME)
1362                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1363
1364         if (dqblk->dqb_valid & QIF_BLIMITS) {
1365                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1366                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1367                 /* clear usage (limit pool) */
1368                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1369                     !dquot->dq_dqb.dqb_bsoftlimit)
1370                         dquot->dq_dqb.dqb_curspace = 0;
1371
1372                 /* clear grace time */
1373                 if (!dqblk->dqb_bsoftlimit ||
1374                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1375                         dquot->dq_dqb.dqb_btime = 0;
1376                 /* set grace only if user hasn't provided his own */
1377                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1378                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1379                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1380
1381                 flag |= LQUOTA_FLAGS_ADJBLK;
1382         }
1383
1384         if (dqblk->dqb_valid & QIF_ILIMITS) {
1385                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1386                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1387                 /* clear usage (limit pool) */
1388                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1389                     !dquot->dq_dqb.dqb_isoftlimit)
1390                         dquot->dq_dqb.dqb_curinodes = 0;
1391
1392                 if (!dqblk->dqb_isoftlimit ||
1393                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1394                         dquot->dq_dqb.dqb_itime = 0;
1395                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1396                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1397                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1398
1399                 flag |= LQUOTA_FLAGS_ADJINO;
1400         }
1401         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1402         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1403                                flag, oqaq);
1404         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1405         if (rc < 0)
1406                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1407
1408
1409         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1410
1411         cfs_mutex_unlock(&dquot->dq_mutex);
1412
1413         if (rc) {
1414                 CERROR("set limit failed! (rc:%d)\n", rc);
1415                 GOTO(update_fail, rc);
1416         }
1417
1418         cfs_up_write(&mds->mds_qonoff_sem);
1419         adjust_lqs(obd, oqaq);
1420
1421         orig_set = ihardlimit || isoftlimit;
1422         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1423         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1424                 cfs_mutex_lock(&dquot->dq_mutex);
1425                 dquot->dq_dqb.dqb_curinodes = 0;
1426                 cfs_mutex_unlock(&dquot->dq_mutex);
1427                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set);
1428                 if (rc) {
1429                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1430                         goto revoke_out;
1431                 }
1432         }
1433
1434         orig_set = bhardlimit || bsoftlimit;
1435         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1436         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1437                 cfs_mutex_lock(&dquot->dq_mutex);
1438                 dquot->dq_dqb.dqb_curspace = 0;
1439                 cfs_mutex_unlock(&dquot->dq_mutex);
1440                 rc = mds_init_slave_blimits(obd, oqctl, orig_set);
1441                 if (rc) {
1442                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1443                         goto revoke_out;
1444                 }
1445         }
1446
1447 revoke_out:
1448         cfs_down_write(&mds->mds_qonoff_sem);
1449         cfs_mutex_lock(&dquot->dq_mutex);
1450         if (rc) {
1451                 /* cancel previous setting */
1452                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1453                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1454                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1455                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1456                 dquot->dq_dqb.dqb_btime = btime;
1457                 dquot->dq_dqb.dqb_itime = itime;
1458         }
1459         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1460         cfs_mutex_unlock(&dquot->dq_mutex);
1461 update_fail:
1462         cfs_spin_lock(&lqs->lqs_lock);
1463         LQS_CLEAR_SETQUOTA(lqs);
1464         cfs_spin_unlock(&lqs->lqs_lock);
1465 skip:
1466         lqs_putref(lqs);
1467 out:
1468         lustre_dqput(dquot);
1469         EXIT;
1470 out_sem:
1471         cfs_up_write(&mds->mds_qonoff_sem);
1472
1473         if (oqaq)
1474                 OBD_FREE_PTR(oqaq);
1475
1476         return rc ? rc : rc2;
1477 }
1478
1479 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1480 {
1481         struct obd_quotactl *soqc;
1482         struct lvfs_run_ctxt saved;
1483         int rc, rc1;
1484         ENTRY;
1485
1486         OBD_ALLOC_PTR(soqc);
1487         if (!soqc)
1488                 RETURN(-ENOMEM);
1489
1490         soqc->qc_cmd = Q_GETOQUOTA;
1491         soqc->qc_id = oqctl->qc_id;
1492         soqc->qc_type = oqctl->qc_type;
1493
1494         /* get block usage from OSS */
1495         soqc->qc_dqblk.dqb_curspace = 0;
1496         rc = obd_quotactl(obd->u.mds.mds_lov_exp, soqc);
1497         if (!rc || rc == -EREMOTEIO) {
1498                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1499                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1500         }
1501
1502         /* get block/inode usage from MDS */
1503         soqc->qc_dqblk.dqb_curspace = 0;
1504         soqc->qc_dqblk.dqb_curinodes = 0;
1505         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1506         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1507         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1508         if (!rc1) {
1509                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1510                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1511                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1512         }
1513
1514         OBD_FREE_PTR(soqc);
1515
1516         RETURN(rc ? : rc1);
1517 }
1518
1519 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1520 {
1521         struct mds_obd *mds = &obd->u.mds;
1522         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1523         struct lustre_dquot *dquot;
1524         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1525         int rc;
1526         ENTRY;
1527
1528         if (oqctl->qc_type != USRQUOTA &&
1529             oqctl->qc_type != GRPQUOTA)
1530                 RETURN(-EINVAL);
1531
1532         cfs_down_read(&mds->mds_qonoff_sem);
1533         dqblk->dqb_valid = 0;
1534         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1535                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1536                 GOTO(out, rc = -ESRCH);
1537         }
1538
1539         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1540         if (IS_ERR(dquot))
1541                 GOTO(out, rc = PTR_ERR(dquot));
1542
1543         cfs_mutex_lock(&dquot->dq_mutex);
1544         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1545         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1546         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1547         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1548         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1549         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1550         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1551         cfs_mutex_unlock(&dquot->dq_mutex);
1552
1553         lustre_dqput(dquot);
1554         cfs_up_read(&mds->mds_qonoff_sem);
1555
1556         /* the usages in admin quota file is inaccurate */
1557         dqblk->dqb_curinodes = 0;
1558         dqblk->dqb_curspace = 0;
1559         rc = mds_get_space(obd, oqctl);
1560
1561         RETURN(rc);
1562
1563 out:
1564         cfs_up_read(&mds->mds_qonoff_sem);
1565         return rc;
1566 }
1567
1568 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1569 {
1570         struct lvfs_run_ctxt saved;
1571         int rc;
1572         ENTRY;
1573
1574         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1575         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1576         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1577
1578         RETURN(rc);
1579 }
1580
1581
1582 /* FIXME we only recovery block limit by now, need recovery inode
1583  * limits also after CMD involved in */
1584 static int
1585 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1586 {
1587         struct mds_obd *mds = &obd->u.mds;
1588         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1589         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1590         struct lustre_qunit_size *lqs;
1591         struct lustre_dquot *dquot;
1592         struct obd_quotactl *qctl;
1593         __u64 total_limits = 0;
1594         int rc = 0;
1595         ENTRY;
1596
1597         OBD_ALLOC_PTR(qctl);
1598         if (qctl == NULL)
1599                 RETURN(-ENOMEM);
1600
1601         dquot = lustre_dqget(obd, qinfo, id, type, 0);
1602         if (IS_ERR(dquot)) {
1603                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1604                 OBD_FREE_PTR(qctl);
1605                 RETURN(PTR_ERR(dquot));
1606         }
1607
1608         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 1);
1609         if (lqs == NULL)
1610                 rc = -ENOENT;
1611         if (IS_ERR(lqs))
1612                 rc = PTR_ERR(lqs);
1613         if (rc)
1614                 GOTO(skip, rc);
1615
1616         cfs_mutex_lock(&dquot->dq_mutex);
1617
1618         /* don't recover the dquot without limits or quota is setting or
1619          * another recovery is already going on */
1620         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1621             LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1622                 cfs_mutex_unlock(&dquot->dq_mutex);
1623                 GOTO(skip1, rc = 0);
1624         }
1625
1626         cfs_spin_lock(&lqs->lqs_lock);
1627         LQS_SET_RECOVERY(lqs);
1628         cfs_spin_unlock(&lqs->lqs_lock);
1629         cfs_mutex_unlock(&dquot->dq_mutex);
1630
1631         /* release mds_qonoff_sem during obd_quotactl ops here */
1632         cfs_up_write(&mds->mds_qonoff_sem);
1633
1634         /* get real bhardlimit from all slaves. */
1635         qctl->qc_cmd = Q_GETOQUOTA;
1636         qctl->qc_type = type;
1637         qctl->qc_id = id;
1638         qctl->qc_stat = QUOTA_RECOVERING;
1639         rc = obd_quotactl(mds->mds_lov_exp, qctl);
1640         cfs_down_write(&mds->mds_qonoff_sem);
1641         if (rc)
1642                 GOTO(out, rc);
1643         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1644
1645         /* get real bhardlimit from master */
1646         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1647         if (rc)
1648                 GOTO(out, rc);
1649         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1650
1651         /* amend the usage of the administrative quotafile */
1652         cfs_mutex_lock(&dquot->dq_mutex);
1653
1654         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1655
1656         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1657         if (rc)
1658                 CERROR("write dquot failed! (rc:%d)\n", rc);
1659
1660         cfs_mutex_unlock(&dquot->dq_mutex);
1661         EXIT;
1662 out:
1663         cfs_spin_lock(&lqs->lqs_lock);
1664         LQS_CLEAR_RECOVERY(lqs);
1665         cfs_spin_unlock(&lqs->lqs_lock);
1666 skip1:
1667         lqs_putref(lqs);
1668 skip:
1669         lustre_dqput(dquot);
1670         OBD_FREE_PTR(qctl);
1671         return rc;
1672 }
1673
1674 struct qmaster_recov_thread_data {
1675         struct obd_device *obd;
1676         cfs_completion_t comp;
1677 };
1678
1679 static int qmaster_recovery_main(void *arg)
1680 {
1681         struct qmaster_recov_thread_data *data = arg;
1682         struct obd_device *obd = data->obd;
1683         struct mds_obd *mds = &obd->u.mds;
1684         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1685         int rc = 0;
1686         unsigned short type;
1687         ENTRY;
1688
1689         cfs_daemonize_ctxt("qmaster_recovd");
1690
1691         /* for mds */
1692         class_incref(obd, "qmaster_recovd_mds", obd);
1693         /* for lov */
1694         class_incref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1695
1696         cfs_complete(&data->comp);
1697
1698         cfs_down_write(&mds->mds_qonoff_sem);
1699         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1700                 cfs_list_t id_list;
1701                 struct dquot_id *dqid, *tmp;
1702
1703                 if (qinfo->qi_files[type] == NULL)
1704                         continue;
1705
1706                 CFS_INIT_LIST_HEAD(&id_list);
1707                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1708                                  &id_list);
1709                 if (rc)
1710                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1711
1712                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1713                         cfs_list_del_init(&dqid->di_link);
1714                         if (rc)
1715                                 goto free;
1716
1717                         rc = dquot_recovery(obd, dqid->di_id, type);
1718                         if (rc)
1719                                 CERROR("%s: qmaster recovery failed for %sid %d"
1720                                        " rc:%d)\n", obd->obd_name,
1721                                        type ? "g" : "u", dqid->di_id, rc);
1722 free:
1723                         OBD_FREE_PTR(dqid);
1724                 }
1725         }
1726         cfs_up_write(&mds->mds_qonoff_sem);
1727         class_decref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1728         class_decref(obd, "qmaster_recovd_mds", obd);
1729         RETURN(rc);
1730 }
1731
1732 int mds_quota_recovery(struct obd_device *obd)
1733 {
1734         struct mds_obd *mds = &obd->u.mds;
1735         struct qmaster_recov_thread_data data;
1736         int rc = 0;
1737         ENTRY;
1738
1739         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
1740                 RETURN(0);
1741
1742         if (unlikely(!mds->mds_quota || obd->obd_stopping))
1743                 RETURN(rc);
1744
1745         cfs_mutex_lock(&obd->obd_dev_mutex);
1746         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1747                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1748                       mds->mds_lov_desc.ld_active_tgt_count,
1749                       mds->mds_lov_objid_count);
1750                 cfs_mutex_unlock(&obd->obd_dev_mutex);
1751                 RETURN(rc);
1752         }
1753         cfs_mutex_unlock(&obd->obd_dev_mutex);
1754
1755         data.obd = obd;
1756         cfs_init_completion(&data.comp);
1757
1758         rc = cfs_create_thread(qmaster_recovery_main, &data,
1759                                CFS_DAEMON_FLAGS);
1760         if (rc < 0)
1761                 CERROR("%s: cannot start quota recovery thread: rc %d\n",
1762                        obd->obd_name, rc);
1763
1764         cfs_wait_for_completion(&data.comp);
1765         RETURN(rc);
1766 }