lustre/quota/quota_master.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/quota/quota_master.c
  39  *
  40  * Lustre Quota Master request handler
  41  *
  42  * Author: Niu YaWei <niu@clusterfs.com>
  43  */
  44
  45 #ifndef EXPORT_SYMTAB
  46 # define EXPORT_SYMTAB
  47 #endif
  48
  49 #define DEBUG_SUBSYSTEM S_LQUOTA
  50
  51 #include <linux/version.h>
  52 #include <linux/fs.h>
  53 #include <asm/unistd.h>
  54 #include <linux/slab.h>
  55 #include <linux/quotaops.h>
  56 #include <linux/module.h>
  57 #include <linux/init.h>
  58 #include <linux/quota.h>
  59
  60 #include <obd_class.h>
  61 #include <lustre_quota.h>
  62 #include <lustre_fsfilt.h>
  63 #include <lustre_mds.h>
  64
  65 #include "quota_internal.h"
  66
  67 #ifdef HAVE_QUOTA_SUPPORT
  68
  69 /* lock ordering: mds->mds_qonoff_sem > dquot->dq_mutex > lqs->lqs_lock */
  70 static cfs_list_t lustre_dquot_hash[NR_DQHASH];
  71 static cfs_rwlock_t dquot_hash_lock = CFS_RW_LOCK_UNLOCKED;
  72
  73 cfs_mem_cache_t *lustre_dquot_cachep;
  74
  75 int lustre_dquot_init(void)
  76 {
  77         int i;
  78         ENTRY;
  79
  80         LASSERT(lustre_dquot_cachep == NULL);
  81         lustre_dquot_cachep = cfs_mem_cache_create("lustre_dquot_cache",
  82                                                    sizeof(struct lustre_dquot),
  83                                                    0, 0);
  84         if (!lustre_dquot_cachep)
  85                 return (-ENOMEM);
  86
  87         for (i = 0; i < NR_DQHASH; i++) {
  88                 CFS_INIT_LIST_HEAD(lustre_dquot_hash + i);
  89         }
  90         RETURN(0);
  91 }
  92
  93 void lustre_dquot_exit(void)
  94 {
  95         int i;
  96         ENTRY;
  97         /* FIXME cleanup work ?? */
  98
  99         for (i = 0; i < NR_DQHASH; i++) {
 100                 LASSERT(cfs_list_empty(lustre_dquot_hash + i));
 101         }
 102         if (lustre_dquot_cachep) {
 103                 int rc;
 104                 rc = cfs_mem_cache_destroy(lustre_dquot_cachep);
 105                 LASSERTF(rc == 0,"couldn't destroy lustre_dquot_cachep slab\n");
 106                 lustre_dquot_cachep = NULL;
 107         }
 108         EXIT;
 109 }
 110
 111 static inline int
 112 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 113              __attribute__((__const__));
 114
 115 static inline int
 116 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 117 {
 118         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
 119         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 120         return tmp;
 121 }
 122
 123 /* caller must hold dquot_hash_lock */
 124 static struct lustre_dquot *find_dquot(int hashent,
 125                                        struct lustre_quota_info *lqi, qid_t id,
 126                                        int type)
 127 {
 128         struct lustre_dquot *dquot;
 129         ENTRY;
 130
 131         cfs_list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
 132                 if (dquot->dq_info == lqi &&
 133                     dquot->dq_id == id && dquot->dq_type == type)
 134                         RETURN(dquot);
 135         }
 136         RETURN(NULL);
 137 }
 138
 139 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 140                                         qid_t id, int type)
 141 {
 142         struct lustre_dquot *dquot = NULL;
 143         ENTRY;
 144
 145         OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
 146         if (dquot == NULL)
 147                 RETURN(NULL);
 148
 149         CFS_INIT_LIST_HEAD(&dquot->dq_hash);
 150         cfs_mutex_init(&dquot->dq_mutex);
 151         cfs_mutex_lock(&dquot->dq_mutex);
 152         cfs_atomic_set(&dquot->dq_refcnt, 1);
 153         dquot->dq_info = lqi;
 154         dquot->dq_id = id;
 155         dquot->dq_type = type;
 156
 157         RETURN(dquot);
 158 }
 159
 160 static void free_dquot(struct lustre_dquot *dquot)
 161 {
 162         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 163 }
 164
 165 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 166 {
 167         cfs_list_t *head = lustre_dquot_hash +
 168             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 169         LASSERT(cfs_list_empty(&dquot->dq_hash));
 170         cfs_list_add(&dquot->dq_hash, head);
 171 }
 172
 173 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 174 {
 175         LASSERT(!cfs_list_empty(&dquot->dq_hash));
 176         cfs_list_del_init(&dquot->dq_hash);
 177 }
 178
 179 static void lustre_dqput(struct lustre_dquot *dquot)
 180 {
 181         ENTRY;
 182         cfs_write_lock(&dquot_hash_lock);
 183         LASSERT(cfs_atomic_read(&dquot->dq_refcnt));
 184         cfs_atomic_dec(&dquot->dq_refcnt);
 185         if (cfs_atomic_read(&dquot->dq_refcnt) == 0) {
 186                 remove_dquot_nolock(dquot);
 187                 free_dquot(dquot);
 188         }
 189         cfs_write_unlock(&dquot_hash_lock);
 190         EXIT;
 191 }
 192
 193 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 194                                          struct lustre_quota_info *lqi,
 195                                          qid_t id, int type, int can_fake)
 196 {
 197         unsigned int hashent = dquot_hashfn(lqi, id, type);
 198         struct lustre_dquot *dquot, *empty;
 199         int free_dq = 0;
 200         ENTRY;
 201
 202         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 203                 RETURN(ERR_PTR(-ENOMEM));
 204
 205         cfs_read_lock(&dquot_hash_lock);
 206         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 207                 cfs_atomic_inc(&dquot->dq_refcnt);
 208                 cfs_read_unlock(&dquot_hash_lock);
 209                 cfs_mutex_unlock(&empty->dq_mutex);
 210                 free_dq = 1;
 211         } else {
 212                 int rc;
 213
 214                 cfs_read_unlock(&dquot_hash_lock);
 215
 216                 dquot = empty;
 217                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
 218                 cfs_mutex_unlock(&dquot->dq_mutex);
 219                 if (rc) {
 220                         CERROR("can't read dquot from admin quotafile! "
 221                                "(rc:%d)\n", rc);
 222                         free_dquot(dquot);
 223                         RETURN(ERR_PTR(rc));
 224                 } else {
 225                         cfs_write_lock(&dquot_hash_lock);
 226                         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 227                                 cfs_atomic_inc(&dquot->dq_refcnt);
 228                                 free_dq = 1;
 229                         } else {
 230                                 dquot = empty;
 231                                 insert_dquot_nolock(dquot);
 232                         }
 233                         cfs_write_unlock(&dquot_hash_lock);
 234                 }
 235
 236         }
 237
 238         LASSERT(dquot);
 239         if (!can_fake && cfs_test_bit(DQ_FAKE_B, &dquot->dq_flags)) {
 240                 DQUOT_DEBUG(dquot, "It is a fake dquot: unexpected!\n");
 241                 lustre_dqput(dquot);
 242                 dquot = ERR_PTR(-ENOENT);
 243         }
 244
 245         if (free_dq)
 246                 free_dquot(empty);
 247
 248
 249         RETURN(dquot);
 250 }
 251
 252 static void init_oqaq(struct quota_adjust_qunit *oqaq,
 253                       struct lustre_quota_ctxt *qctxt,
 254                       qid_t id, int type)
 255 {
 256         struct lustre_qunit_size *lqs = NULL;
 257
 258         oqaq->qaq_id = id;
 259         oqaq->qaq_flags = type;
 260         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 0);
 261         if (lqs && !IS_ERR(lqs)) {
 262                 cfs_spin_lock(&lqs->lqs_lock);
 263                 oqaq->qaq_bunit_sz = lqs->lqs_bunit_sz;
 264                 oqaq->qaq_iunit_sz = lqs->lqs_iunit_sz;
 265                 oqaq->qaq_flags    = lqs->lqs_flags;
 266                 cfs_spin_unlock(&lqs->lqs_lock);
 267                 lqs_putref(lqs);
 268         } else {
 269                 CDEBUG(D_QUOTA, "Can't find the lustre qunit size!\n");
 270                 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
 271                 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
 272         }
 273 }
 274
 275 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
 276                           __u32 is_blk)
 277 {
 278         struct mds_obd *mds = &obd->u.mds;
 279         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 280         __u32 ost_num = mds->mds_lov_objid_count, mdt_num = 1;
 281         struct quota_adjust_qunit *oqaq = NULL;
 282         unsigned int qid[MAXQUOTAS] = { 0, 0 };
 283         struct lustre_quota_info *info = &mds->mds_quota_info;
 284         struct lustre_dquot *dquot = NULL;
 285         int adjust_res = 0;
 286         int rc = 0;
 287         ENTRY;
 288
 289         LASSERT(mds);
 290         cfs_down_read(&mds->mds_qonoff_sem);
 291         dquot = lustre_dqget(obd, info, id, type, 0);
 292         if (IS_ERR(dquot))
 293                 RETURN(PTR_ERR(dquot));
 294
 295         cfs_up_read(&mds->mds_qonoff_sem);
 296         OBD_ALLOC_PTR(oqaq);
 297         if (!oqaq)
 298                 GOTO(out, rc = -ENOMEM);
 299
 300         cfs_mutex_lock(&dquot->dq_mutex);
 301         init_oqaq(oqaq, qctxt, id, type);
 302
 303         rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
 304                                is_blk ? LQUOTA_FLAGS_ADJBLK :
 305                                LQUOTA_FLAGS_ADJINO, oqaq);
 306
 307         if (rc < 0) {
 308                 CERROR("create oqaq failed! (rc:%d)\n", rc);
 309                 GOTO(out_sem, rc);
 310         }
 311         QAQ_DEBUG(oqaq, "show oqaq.\n")
 312
 313         if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
 314                 GOTO(out_sem, rc);
 315
 316         /* adjust the mds slave qunit size */
 317         adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
 318         if (adjust_res <= 0) {
 319                 if (adjust_res < 0) {
 320                         rc = adjust_res;
 321                         CERROR("adjust mds slave's qunit size failed! "
 322                                "(rc:%d)\n", rc);
 323                 } else {
 324                         CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
 325                 }
 326                 GOTO(out_sem, rc);
 327         }
 328
 329         if (type)
 330                 qid[GRPQUOTA] = dquot->dq_id;
 331         else
 332                 qid[USRQUOTA] = dquot->dq_id;
 333
 334         cfs_mutex_unlock(&dquot->dq_mutex);
 335
 336         rc = qctxt_adjust_qunit(obd, qctxt, qid, is_blk, 0, NULL);
 337         if (rc == -EDQUOT || rc == -EBUSY) {
 338                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 339                 rc = 0;
 340         }
 341         if (rc) {
 342                 CERROR("%s: mds fail to adjust file quota! (rc:%d)\n",
 343                        obd->obd_name, rc);
 344                 GOTO(out, rc);
 345         }
 346
 347         /* only when block qunit is reduced, boardcast to osts */
 348         if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
 349                 rc = obd_quota_adjust_qunit(mds->mds_lov_exp, oqaq, qctxt, NULL);
 350
 351 out:
 352         lustre_dqput(dquot);
 353         if (oqaq)
 354                 OBD_FREE_PTR(oqaq);
 355
 356         RETURN(rc);
 357 out_sem:
 358         cfs_mutex_unlock(&dquot->dq_mutex);
 359         goto out;
 360 }
 361
 362 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
 363 {
 364         struct mds_obd *mds = &obd->u.mds;
 365         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 366         struct lustre_quota_info *info = &mds->mds_quota_info;
 367         struct lustre_dquot *dquot = NULL;
 368         __u64 *usage = NULL;
 369         __u64 hlimit = 0, slimit = 0;
 370         time_t *time = NULL;
 371         unsigned int grace = 0;
 372         struct lustre_qunit_size *lqs = NULL;
 373         int rc = 0;
 374         ENTRY;
 375
 376         if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
 377                 RETURN(-EIO);
 378
 379         if (!ll_sb_has_quota_active(qctxt->lqc_sb,
 380                                     QDATA_IS_GRP(qdata) ? GRPQUOTA : USRQUOTA))
 381                 RETURN(-EIO);
 382
 383         lqs = quota_search_lqs(LQS_KEY(QDATA_IS_GRP(qdata), qdata->qd_id),
 384                                qctxt, 0);
 385         if (lqs == NULL)
 386                 rc = -ENOENT;
 387         if (IS_ERR(lqs))
 388                 rc = PTR_ERR(lqs);
 389         if (rc)
 390                 RETURN(rc);
 391
 392         cfs_spin_lock(&lqs->lqs_lock);
 393         if (LQS_IS_RECOVERY(lqs)) {
 394                 cfs_spin_unlock(&lqs->lqs_lock);
 395                 LQS_DEBUG(lqs, "this lqs is under recovery\n");
 396                 GOTO(skip, rc = -EBUSY);
 397         }
 398         cfs_spin_unlock(&lqs->lqs_lock);
 399
 400         cfs_down_write(&mds->mds_qonoff_sem);
 401         dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata), 0);
 402         if (IS_ERR(dquot)) {
 403                 cfs_up_write(&mds->mds_qonoff_sem);
 404                 GOTO(skip, rc = PTR_ERR(dquot));
 405         }
 406
 407         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 408         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 409
 410         cfs_mutex_lock(&dquot->dq_mutex);
 411
 412         if (QDATA_IS_BLK(qdata)) {
 413                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
 414                 usage = &dquot->dq_dqb.dqb_curspace;
 415                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 416                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 417                 time = &dquot->dq_dqb.dqb_btime;
 418         } else {
 419                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
 420                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 421                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 422                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 423                 time = &dquot->dq_dqb.dqb_itime;
 424         }
 425
 426         /* if the quota limit in admin quotafile is zero, we just inform
 427          * slave to clear quota limit with zero qd_count */
 428         if (hlimit == 0 && slimit == 0) {
 429                 qdata->qd_count = 0;
 430                 GOTO(out, rc);
 431         }
 432
 433         switch (opc) {
 434         case QUOTA_DQACQ:
 435                 if (hlimit &&
 436                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
 437                 {
 438                         if (QDATA_IS_CHANGE_QS(qdata) &&
 439                             QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
 440                                 qdata->qd_count = (hlimit -
 441                                         QUSG(*usage, QDATA_IS_BLK(qdata)))
 442                                         * (QDATA_IS_BLK(qdata) ?
 443                                            QUOTABLOCK_SIZE : 1);
 444                         else
 445                                 GOTO(out, rc = -EDQUOT);
 446                 }
 447
 448                 if (slimit &&
 449                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
 450                         if (*time && cfs_time_current_sec() >= *time)
 451                                 GOTO(out, rc = -EDQUOT);
 452                         else if (!*time)
 453                                 *time = cfs_time_current_sec() + grace;
 454                 }
 455
 456                 *usage += qdata->qd_count;
 457                 break;
 458         case QUOTA_DQREL:
 459                 /* The usage in administrative file might be incorrect before
 460                  * recovery done */
 461                 if (*usage < qdata->qd_count)
 462                         *usage = 0;
 463                 else
 464                         *usage -= qdata->qd_count;
 465
 466                 /* (usage <= soft limit) but not (usage < soft limit) */
 467                 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
 468                         *time = 0;
 469                 break;
 470         default:
 471                 LBUG();
 472         }
 473
 474         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 475         EXIT;
 476 out:
 477         cfs_mutex_unlock(&dquot->dq_mutex);
 478         cfs_up_write(&mds->mds_qonoff_sem);
 479         lustre_dqput(dquot);
 480         if (rc != -EDQUOT)
 481                 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
 482                                       QDATA_IS_BLK(qdata));
 483
 484         cfs_spin_lock(&lqs->lqs_lock);
 485         qdata->qd_qunit  = QDATA_IS_BLK(qdata) ? lqs->lqs_bunit_sz :
 486                 lqs->lqs_iunit_sz;
 487         cfs_spin_unlock(&lqs->lqs_lock);
 488
 489         if (QDATA_IS_BLK(qdata))
 490                 QDATA_SET_ADJBLK(qdata);
 491         else
 492                 QDATA_SET_ADJINO(qdata);
 493
 494         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
 495 skip:
 496         lqs_putref(lqs);
 497
 498         return rc;
 499 }
 500
 501 int mds_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
 502                      const unsigned int qpids[], int rc, int opc)
 503 {
 504         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 505         int rc2 = 0;
 506         ENTRY;
 507
 508         if (rc && rc != -EDQUOT && rc != ENOLCK)
 509                 RETURN(0);
 510
 511         switch (opc) {
 512         case FSFILT_OP_SETATTR:
 513                 /* release file quota on original owner */
 514                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 0, 0, NULL);
 515                 /* release block quota on original owner */
 516                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 517                 /* acquire file quota on current owner */
 518                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 519                 /* acquire block quota on current owner */
 520                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 521                 break;
 522         case FSFILT_OP_UNLINK_PARTIAL_CHILD:
 523                 /* release file quota on child */
 524                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 525                 /* rlease block quota on child */
 526                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 527                 break;
 528         case FSFILT_OP_CREATE_PARTIAL_CHILD:
 529                 /* acquire file quota on child */
 530                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 531                 /* acquire block quota on child */
 532                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 533                 break;
 534         case FSFILT_OP_LINK:
 535                 /* acquire block quota on parent */
 536                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 537                 break;
 538         case FSFILT_OP_UNLINK:
 539                 /* release block quota on parent */
 540                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 541                 /* release file quota on child */
 542                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 543                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 544                         /* release block quota on child */
 545                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
 546                                                   NULL);
 547                 break;
 548         case FSFILT_OP_UNLINK_PARTIAL_PARENT:
 549                 /* release block quota on parent */
 550                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 551                 break;
 552         case FSFILT_OP_CREATE:
 553                 /* acquire block quota on parent */
 554                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 555                 /* acquire file quota on child */
 556                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 0, 0, NULL);
 557                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 558                         /* acquire block quota on child */
 559                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0,
 560                                                   NULL);
 561                 break;
 562         default:
 563                 LBUG();
 564                 break;
 565         }
 566
 567         if (rc2)
 568                 CDEBUG(D_QUOTA,
 569                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
 570                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
 571         RETURN(0);
 572 }
 573
 574 int filter_quota_adjust(struct obd_device *obd, const unsigned int qcids[],
 575                         const unsigned int qpids[], int rc, int opc)
 576 {
 577         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 578         int rc2 = 0;
 579         ENTRY;
 580
 581         if (rc && rc != -EDQUOT)
 582                 RETURN(0);
 583
 584         switch (opc) {
 585         case FSFILT_OP_SETATTR:
 586                 /* acquire/release block quota on original & current owner */
 587                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 588                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids, 1, 0, NULL);
 589                 break;
 590         case FSFILT_OP_UNLINK:
 591                 /* release block quota on this owner */
 592         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 593                 /* acquire block quota on this owner */
 594                 rc = qctxt_adjust_qunit(obd, qctxt, qcids, 1, 0, NULL);
 595                 break;
 596         default:
 597                 LBUG();
 598                 break;
 599         }
 600
 601         if (rc || rc2) {
 602                 if (!rc)
 603                         rc = rc2;
 604                 CDEBUG(D_QUOTA,
 605                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
 606                        rc == QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
 607         }
 608
 609         RETURN(0);
 610 }
 611
 612 static const char prefix[] = "OBJECTS/";
 613
 614 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 615 {
 616         struct mds_obd *mds = &obd->u.mds;
 617         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 618         struct obd_device_target *obt = &obd->u.obt;
 619         int rc = 0, i, rc1 = 0;
 620         char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 621         char name[64];
 622         struct lvfs_run_ctxt saved;
 623         ENTRY;
 624
 625         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 626
 627         if (oqctl->qc_type != USRQUOTA &&
 628             oqctl->qc_type != GRPQUOTA &&
 629             oqctl->qc_type != UGQUOTA)
 630                 RETURN(-EINVAL);
 631
 632         cfs_down(&obt->obt_quotachecking);
 633         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 634         cfs_down_write(&mds->mds_qonoff_sem);
 635
 636         for (i = 0; i < MAXQUOTAS; i++) {
 637                 struct file *fp;
 638
 639                 if (!Q_TYPESET(oqctl, i))
 640                         continue;
 641
 642                 /* quota file has been opened ? */
 643                 if (qinfo->qi_files[i]) {
 644                         CWARN("quota[%d] is on yet\n", i);
 645                         rc1 = -EBUSY;
 646                         continue;
 647                 }
 648
 649                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 650                 sprintf(name, "%s%s", prefix, quotafile[i]);
 651
 652                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 653                 if (IS_ERR(fp)) {
 654                         rc = PTR_ERR(fp);
 655                         CERROR("%s: error invalidating admin quotafile %s (rc:%d)\n",
 656                                obd->obd_name, name, rc);
 657                 }
 658                 else
 659                         filp_close(fp, 0);
 660         }
 661
 662         cfs_up_write(&mds->mds_qonoff_sem);
 663         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 664         cfs_up(&obt->obt_quotachecking);
 665         RETURN(rc ? : rc1);
 666 }
 667
 668 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 669 {
 670         struct mds_obd *mds = &obd->u.mds;
 671         struct obd_device_target *obt = &obd->u.obt;
 672         int rc;
 673         struct lvfs_run_ctxt saved;
 674         ENTRY;
 675
 676         if (oqctl->qc_type != USRQUOTA &&
 677             oqctl->qc_type != GRPQUOTA &&
 678             oqctl->qc_type != UGQUOTA)
 679                 RETURN(-EINVAL);
 680
 681         cfs_down(&obt->obt_quotachecking);
 682         if (obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))
 683                 GOTO(out, rc = -EBUSY);
 684         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 685         cfs_down_write(&mds->mds_qonoff_sem);
 686
 687         oqctl->qc_cmd = Q_FINVALIDATE;
 688         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 689         if (!rc)
 690                 rc = obd_quotactl(mds->mds_lov_exp, oqctl);
 691
 692         cfs_up_write(&mds->mds_qonoff_sem);
 693         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 694 out:
 695         cfs_up(&obt->obt_quotachecking);
 696         RETURN(rc);
 697 }
 698
 699 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 700 {
 701         struct mds_obd *mds = &obd->u.mds;
 702         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 703         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 704         struct lvfs_run_ctxt saved;
 705         char name[64];
 706         int i, rc = 0;
 707         ENTRY;
 708
 709         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 710
 711         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 712         cfs_down_write(&mds->mds_qonoff_sem);
 713
 714         for (i = 0; i < MAXQUOTAS && !rc; i++) {
 715                 struct file *fp;
 716
 717                 if (!Q_TYPESET(oqctl, i))
 718                         continue;
 719
 720                 /* quota file has been opened ? */
 721                 if (qinfo->qi_files[i]) {
 722                         CWARN("init %s admin quotafile while quota on.\n",
 723                               i == USRQUOTA ? "user" : "group");
 724                         continue;
 725                 }
 726
 727                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 728                 sprintf(name, "%s%s", prefix, quotafile[i]);
 729
 730                 /* check if quota file exists and is correct */
 731                 fp = filp_open(name, O_RDONLY, 0);
 732                 if (!IS_ERR(fp)) {
 733                         /* irregular file is not the right place for quota */
 734                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 735                                 CERROR("admin quota file %s is not "
 736                                        "regular!", name);
 737                                 filp_close(fp, 0);
 738                                 rc = -EINVAL;
 739                                 break;
 740                         }
 741                         qinfo->qi_files[i] = fp;
 742                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 743                         qinfo->qi_files[i] = NULL;
 744                         filp_close(fp, 0);
 745                 }
 746                 else
 747                         rc = PTR_ERR(fp);
 748
 749                 if (!rc)
 750                         continue;
 751
 752                 /* -EINVAL may be returned by quotainfo for bad quota file */
 753                 if (rc != -ENOENT && rc != -EINVAL) {
 754                         CERROR("%s: error opening old quota file %s (%d)\n",
 755                                obd->obd_name, name, rc);
 756                         break;
 757                 }
 758
 759                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
 760                        rc == -ENOENT ? "creating" : "overwriting");
 761
 762                 /* create quota file overwriting old if needed */
 763                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 764                 if (IS_ERR(fp)) {
 765                         rc = PTR_ERR(fp);
 766                         CERROR("%s: error creating admin quotafile %s (rc:%d)\n",
 767                                obd->obd_name, name, rc);
 768                         break;
 769                 }
 770
 771                 qinfo->qi_files[i] = fp;
 772
 773                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 774                 if (rc)
 775                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 776                                i == USRQUOTA ? "user" : "group", rc);
 777
 778                 filp_close(fp, 0);
 779                 qinfo->qi_files[i] = NULL;
 780         }
 781
 782         cfs_up_write(&mds->mds_qonoff_sem);
 783         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 784         RETURN(rc);
 785 }
 786
 787 static int close_quota_files(struct obd_quotactl *oqctl,
 788                              struct lustre_quota_info *qinfo)
 789 {
 790         int i, rc = 0;
 791         ENTRY;
 792
 793         for (i = 0; i < MAXQUOTAS; i++) {
 794                 if (!Q_TYPESET(oqctl, i))
 795                         continue;
 796                 if (qinfo->qi_files[i] == NULL) {
 797                         CDEBUG(D_QUOTA, "quota[%d] is off already\n", i);
 798                         rc = -EALREADY;
 799                         continue;
 800                 }
 801                 filp_close(qinfo->qi_files[i], 0);
 802                 qinfo->qi_files[i] = NULL;
 803         }
 804         RETURN(rc);
 805 }
 806
 807 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 808 {
 809         struct mds_obd *mds = &obd->u.mds;
 810         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 811         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 812         char name[64];
 813         int i, rc = 0, rc1 = 0;
 814         ENTRY;
 815
 816         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 817
 818         /* open admin quota files and read quotafile info */
 819         for (i = 0; i < MAXQUOTAS; i++) {
 820                 struct file *fp;
 821
 822                 if (!Q_TYPESET(oqctl, i) || qinfo->qi_files[i] != NULL)
 823                         continue;
 824
 825                 LASSERT(strlen(quotafile[i])
 826                         + sizeof(prefix) <= sizeof(name));
 827                 sprintf(name, "%s%s", prefix, quotafile[i]);
 828                 fp = filp_open(name, O_RDWR, 0);
 829                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 830                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
 831                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
 832                         break;
 833                 }
 834                 qinfo->qi_files[i] = fp;
 835
 836                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 837                 if (rc) {
 838                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
 839                         break;
 840                 }
 841
 842                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 843                 if (rc) {
 844                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
 845                                rc);
 846                         break;
 847                 }
 848         }
 849
 850         if (rc && rc1 != -EALREADY)
 851                 close_quota_files(oqctl, qinfo);
 852
 853         RETURN(rc ? : rc1);
 854 }
 855
 856 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 857 {
 858         int rc;
 859         ENTRY;
 860
 861         if (oqctl->qc_type != USRQUOTA &&
 862             oqctl->qc_type != GRPQUOTA &&
 863             oqctl->qc_type != UGQUOTA)
 864                 RETURN(-EINVAL);
 865
 866         rc = generic_quota_on(obd, oqctl, 1);
 867
 868         RETURN(rc);
 869 }
 870
 871
 872 int mds_admin_quota_off(struct obd_device *obd,
 873                         struct obd_quotactl *oqctl)
 874 {
 875         struct mds_obd *mds = &obd->u.mds;
 876         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 877         int rc;
 878         ENTRY;
 879
 880         /* close admin quota files */
 881         rc = close_quota_files(oqctl, qinfo);
 882         RETURN(rc);
 883 }
 884
 885
 886 /* with obt->obt_quotachecking held */
 887 int do_mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 888 {
 889         struct mds_obd *mds = &obd->u.mds;
 890         struct obd_device_target *obt = &obd->u.obt;
 891         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 892         struct lvfs_run_ctxt saved;
 893         int rc = 0, rc1 = 0, rc2 = 0;
 894         ENTRY;
 895
 896         LASSERT_SEM_LOCKED(&obt->obt_quotachecking);
 897
 898         if (oqctl->qc_type != USRQUOTA &&
 899             oqctl->qc_type != GRPQUOTA &&
 900             oqctl->qc_type != UGQUOTA)
 901                 RETURN(-EINVAL);
 902
 903         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 904         cfs_down_write(&mds->mds_qonoff_sem);
 905         /* close admin quota files */
 906         rc2 = mds_admin_quota_off(obd, oqctl);
 907         if (rc2 && rc2 != -EALREADY) {
 908                 CWARN("mds quota[%d] is failed to be off for %d\n", oqctl->qc_type, rc2);
 909                 GOTO(out, rc2);
 910         }
 911
 912         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 913         if (!rc1) {
 914                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
 915         } else if (quota_is_off(qctxt, oqctl)) {
 916                 CWARN("mds local quota[%d] is off already\n", oqctl->qc_type);
 917                 rc1 = -EALREADY;
 918         } else {
 919                 if (rc2 != -EALREADY) {
 920                         CWARN("mds local quota[%d] is failed to be off for %d\n",
 921                               oqctl->qc_type, rc1);
 922                         oqctl->qc_cmd = Q_QUOTAON;
 923                         mds_admin_quota_on(obd, oqctl);
 924                         oqctl->qc_cmd = Q_QUOTAOFF;
 925                 }
 926                 GOTO(out, rc1);
 927         }
 928
 929         rc = obd_quotactl(mds->mds_lov_exp, oqctl);
 930         if (rc && rc != -EALREADY) {
 931                 CWARN("mds remote quota[%d] is failed to be off for %d\n",
 932                       oqctl->qc_type, rc);
 933                 oqctl->qc_cmd = Q_QUOTAON;
 934                 if (rc2 != -EALREADY)
 935                         mds_admin_quota_on(obd, oqctl);
 936                 if (rc1 != -EALREADY) {
 937                         fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 938                         qctxt->lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
 939                 }
 940                 oqctl->qc_cmd = Q_QUOTAOFF;
 941         }
 942         EXIT;
 943
 944 out:
 945         CDEBUG(D_QUOTA, "%s: quotaoff type:flags:rc %u:%lu:%d\n",
 946                obd->obd_name, oqctl->qc_type, qctxt->lqc_flags, rc);
 947         cfs_up_write(&mds->mds_qonoff_sem);
 948         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 949         return rc ? : (rc1 ? : rc2);
 950 }
 951
 952 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 953 {
 954         struct obd_device_target *obt = &obd->u.obt;
 955         int rc;
 956         ENTRY;
 957
 958         cfs_down(&obt->obt_quotachecking);
 959         rc = do_mds_quota_off(obd, oqctl);
 960         cfs_up(&obt->obt_quotachecking);
 961         RETURN(rc);
 962 }
 963
 964 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 965 {
 966         struct mds_obd *mds = &obd->u.mds;
 967         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 968         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 969         int rc;
 970         ENTRY;
 971
 972         if (oqctl->qc_type != USRQUOTA &&
 973             oqctl->qc_type != GRPQUOTA)
 974                 RETURN(-EINVAL);
 975
 976         cfs_down_write(&mds->mds_qonoff_sem);
 977         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 978                 CWARN("quota[%u] is off\n", oqctl->qc_type);
 979                 GOTO(out, rc = -ESRCH);
 980         }
 981
 982         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 983         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 984         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 985
 986         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 987         EXIT;
 988
 989 out:
 990         cfs_up_write(&mds->mds_qonoff_sem);
 991         return rc;
 992 }
 993
 994 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 995 {
 996         struct mds_obd *mds = &obd->u.mds;
 997         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 998         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 999         int rc = 0;
1000         ENTRY;
1001
1002         if (oqctl->qc_type != USRQUOTA &&
1003             oqctl->qc_type != GRPQUOTA)
1004                 RETURN(-EINVAL);
1005
1006         cfs_down_read(&mds->mds_qonoff_sem);
1007         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1008                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1009                 GOTO(out, rc = -ESRCH);
1010         }
1011
1012         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
1013         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
1014         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
1015         EXIT;
1016
1017 out:
1018         cfs_up_read(&mds->mds_qonoff_sem);
1019         return rc;
1020 }
1021
1022 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
1023                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
1024                       int type, struct quota_adjust_qunit *oqaq)
1025 {
1026         __u64 bunit_curr_o, iunit_curr_o;
1027         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
1028         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
1029         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
1030                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
1031         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1032                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1033         int rc = 0;
1034         ENTRY;
1035
1036         if (!dquot || !oqaq)
1037                 RETURN(-EINVAL);
1038         LASSERT_MUTEX_LOCKED(&dquot->dq_mutex);
1039         LASSERT(oqaq->qaq_iunit_sz);
1040         LASSERT(oqaq->qaq_bunit_sz);
1041
1042         /* don't change qunit size */
1043         if (!qctxt->lqc_switch_qs)
1044                 RETURN(rc);
1045
1046         bunit_curr_o = oqaq->qaq_bunit_sz;
1047         iunit_curr_o = oqaq->qaq_iunit_sz;
1048
1049         if (dquot->dq_type == GRPQUOTA)
1050                 QAQ_SET_GRP(oqaq);
1051
1052         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1053                 __u64 b_limitation =
1054                         oqaq->qaq_bunit_sz * (ost_num + 1) * shrink_qunit_limit;
1055                 /* enlarge block qunit size */
1056                 while (blimit >
1057                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1058                         oqaq->qaq_bunit_sz =
1059                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1060                                 << QUOTABLOCK_BITS;
1061                         if (oqaq->qaq_bunit_sz >= qctxt->lqc_bunit_sz)
1062                                 break;
1063                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1064                                 shrink_qunit_limit;
1065                 }
1066
1067                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1068                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1069
1070                 /* shrink block qunit size */
1071                 while (blimit <
1072                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1073                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1074                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1075                                 QUOTABLOCK_BITS;
1076                         b_limitation = oqaq->qaq_bunit_sz * (ost_num + 1) *
1077                                 shrink_qunit_limit;
1078                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1079                                 break;
1080                 }
1081
1082                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1083                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1084
1085                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1086                         QAQ_SET_ADJBLK(oqaq);
1087
1088         }
1089
1090         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1091                 __u64 i_limitation =
1092                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1093                 /* enlarge file qunit size */
1094                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1095                        + 2 * i_limitation) {
1096                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1097                         if (oqaq->qaq_iunit_sz >= qctxt->lqc_iunit_sz)
1098                                 break;
1099                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1100                                 shrink_qunit_limit;
1101                 }
1102
1103                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1104                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1105
1106                 /* shrink file qunit size */
1107                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1108                        + i_limitation) {
1109                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1110                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1111                                        shrink_qunit_limit;
1112                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1113                                 break;
1114                 }
1115
1116                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1117                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1118
1119                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1120                         QAQ_SET_ADJINO(oqaq);
1121
1122         }
1123
1124         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1125
1126         RETURN(rc);
1127 }
1128
1129 static int mds_init_slave_ilimits(struct obd_device *obd,
1130                                   struct obd_quotactl *oqctl, int set)
1131 {
1132         /* XXX: for file limits only adjust local now */
1133         struct obd_device_target *obt = &obd->u.obt;
1134         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1135         unsigned int id[MAXQUOTAS] = { 0, 0 };
1136         struct obd_quotactl *ioqc = NULL;
1137         struct lustre_qunit_size *lqs;
1138         int flag;
1139         int rc;
1140         ENTRY;
1141
1142         /* if we are going to set zero limit, needn't init slaves */
1143         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1144             !set)
1145                 RETURN(0);
1146
1147         OBD_ALLOC_PTR(ioqc);
1148         if (!ioqc)
1149                 RETURN(-ENOMEM);
1150
1151         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1152                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1153         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1154         ioqc->qc_id = oqctl->qc_id;
1155         ioqc->qc_type = oqctl->qc_type;
1156         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1157         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1158
1159         /* build lqs for mds */
1160         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1161                                qctxt, flag ? 1 : 0);
1162         if (lqs && !IS_ERR(lqs)) {
1163                 if (flag)
1164                         lqs->lqs_flags |= QI_SET;
1165                 else
1166                         lqs->lqs_flags &= ~QI_SET;
1167                 lqs_putref(lqs);
1168         } else {
1169                 CERROR("fail to %s lqs for inode(%s id: %u)!\n",
1170                        flag ? "create" : "search",
1171                        oqctl->qc_type ? "group" : "user",
1172                        oqctl->qc_id);
1173                 GOTO(out, rc = PTR_ERR(lqs));
1174         }
1175
1176         /* set local limit to MIN_QLIMIT */
1177         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1178         if (rc)
1179                 GOTO(out, rc);
1180
1181         /* trigger local qunit pre-acquire */
1182         if (oqctl->qc_type == USRQUOTA)
1183                 id[USRQUOTA] = oqctl->qc_id;
1184         else
1185                 id[GRPQUOTA] = oqctl->qc_id;
1186
1187         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 0, 0, NULL);
1188         if (rc == -EDQUOT || rc == -EBUSY) {
1189                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1190                 rc = 0;
1191         }
1192         if (rc) {
1193                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1194                        rc);
1195                 GOTO(out, rc);
1196         }
1197         /* FIXME initialize all slaves in CMD */
1198         EXIT;
1199 out:
1200         if (ioqc)
1201                 OBD_FREE_PTR(ioqc);
1202         return rc;
1203 }
1204
1205 static int mds_init_slave_blimits(struct obd_device *obd,
1206                                   struct obd_quotactl *oqctl, int set)
1207 {
1208         struct obd_device_target *obt = &obd->u.obt;
1209         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1210         struct mds_obd *mds = &obd->u.mds;
1211         struct obd_quotactl *ioqc;
1212         struct lustre_qunit_size *lqs;
1213         unsigned int id[MAXQUOTAS] = { 0, 0 };
1214         int rc;
1215         int flag;
1216         ENTRY;
1217
1218         /* if we are going to set zero limit, needn't init slaves */
1219         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1220             !set)
1221                 RETURN(0);
1222
1223         OBD_ALLOC_PTR(ioqc);
1224         if (!ioqc)
1225                 RETURN(-ENOMEM);
1226
1227         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1228                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1229         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1230         ioqc->qc_id = oqctl->qc_id;
1231         ioqc->qc_type = oqctl->qc_type;
1232         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1233         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1234
1235         /* build lqs for mds */
1236         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
1237                                qctxt, flag ? 1 : 0);
1238         if (lqs && !IS_ERR(lqs)) {
1239                 if (flag)
1240                         lqs->lqs_flags |= QB_SET;
1241                 else
1242                         lqs->lqs_flags &= ~QB_SET;
1243                 lqs_putref(lqs);
1244         } else {
1245                 CERROR("fail to %s lqs for block(%s id: %u)!\n",
1246                        flag ? "create" : "search",
1247                        oqctl->qc_type ? "group" : "user",
1248                        oqctl->qc_id);
1249                 GOTO(out, rc = PTR_ERR(lqs));
1250         }
1251
1252         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1253         if (rc)
1254                 GOTO(out, rc);
1255
1256         /* trigger local qunit pre-acquire */
1257         if (oqctl->qc_type == USRQUOTA)
1258                 id[USRQUOTA] = oqctl->qc_id;
1259         else
1260                 id[GRPQUOTA] = oqctl->qc_id;
1261
1262         /* initialize all slave's limit */
1263         rc = obd_quotactl(mds->mds_lov_exp, ioqc);
1264
1265         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, id, 1, 0, NULL);
1266         if (rc == -EDQUOT || rc == -EBUSY) {
1267                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1268                 rc = 0;
1269         }
1270         if (rc) {
1271                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1272                 GOTO(out, rc);
1273         }
1274
1275         EXIT;
1276 out:
1277         OBD_FREE_PTR(ioqc);
1278         return rc;
1279 }
1280
1281 static void adjust_lqs(struct obd_device *obd, struct quota_adjust_qunit *qaq)
1282 {
1283         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
1284         int rc = 0;
1285
1286         QAQ_SET_CREATE_LQS(qaq);
1287         /* adjust local lqs */
1288         rc = quota_adjust_slave_lqs(qaq, qctxt);
1289         if (rc < 0)
1290                 CERROR("adjust master's qunit size failed!(rc=%d)\n", rc);
1291
1292         /* adjust remote lqs */
1293         if (QAQ_IS_ADJBLK(qaq)) {
1294                 rc = obd_quota_adjust_qunit(obd->u.mds.mds_lov_exp, qaq, qctxt, NULL);
1295                 if (rc < 0)
1296                         CERROR("adjust slaves' qunit size failed!(rc=%d)\n", rc);
1297
1298         }
1299 }
1300
1301 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1302 {
1303         struct mds_obd *mds = &obd->u.mds;
1304         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1305         struct obd_device *lov_obd = class_exp2obd(mds->mds_lov_exp);
1306         struct lov_obd *lov = &lov_obd->u.lov;
1307         struct quota_adjust_qunit *oqaq = NULL;
1308         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1309         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1310         time_t btime, itime;
1311         struct lustre_dquot *dquot;
1312         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1313         /* orig_set means if quota was set before; now_set means we are
1314          * setting/cancelling quota */
1315         int orig_set, now_set;
1316         struct lustre_qunit_size *lqs;
1317         int rc = 0, rc2 = 0, flag = 0;
1318         ENTRY;
1319
1320         if (oqctl->qc_type != USRQUOTA &&
1321             oqctl->qc_type != GRPQUOTA)
1322                 RETURN(-EINVAL);
1323
1324         OBD_ALLOC_PTR(oqaq);
1325         if (!oqaq)
1326                 RETURN(-ENOMEM);
1327
1328         cfs_down_write(&mds->mds_qonoff_sem);
1329         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1330
1331         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1332                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1333                 GOTO(out_sem, rc = -ESRCH);
1334         }
1335
1336         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1337         if (IS_ERR(dquot))
1338                 GOTO(out_sem, rc = PTR_ERR(dquot));
1339         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1340         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1341
1342         lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id), qctxt, 1);
1343         if (lqs == NULL)
1344                 rc = -ENOENT;
1345         if (IS_ERR(lqs))
1346                 rc = PTR_ERR(lqs);
1347         if (rc)
1348                 GOTO(out, rc);
1349
1350         cfs_mutex_lock(&dquot->dq_mutex);
1351         cfs_spin_lock(&lqs->lqs_lock);
1352         if (LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1353                 cfs_spin_unlock(&lqs->lqs_lock);
1354                 cfs_mutex_unlock(&dquot->dq_mutex);
1355                 GOTO(skip, rc = -EBUSY);
1356         }
1357         LQS_SET_SETQUOTA(lqs);
1358         cfs_spin_unlock(&lqs->lqs_lock);
1359
1360         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1361         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1362         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1363         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1364         btime = dquot->dq_dqb.dqb_btime;
1365         itime = dquot->dq_dqb.dqb_itime;
1366
1367         if (dqblk->dqb_valid & QIF_BTIME)
1368                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1369         if (dqblk->dqb_valid & QIF_ITIME)
1370                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1371
1372         if (dqblk->dqb_valid & QIF_BLIMITS) {
1373                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1374                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1375                 /* clear usage (limit pool) */
1376                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1377                     !dquot->dq_dqb.dqb_bsoftlimit)
1378                         dquot->dq_dqb.dqb_curspace = 0;
1379
1380                 /* clear grace time */
1381                 if (!dqblk->dqb_bsoftlimit ||
1382                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1383                         dquot->dq_dqb.dqb_btime = 0;
1384                 /* set grace only if user hasn't provided his own */
1385                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1386                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1387                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1388
1389                 flag |= LQUOTA_FLAGS_ADJBLK;
1390         }
1391
1392         if (dqblk->dqb_valid & QIF_ILIMITS) {
1393                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1394                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1395                 /* clear usage (limit pool) */
1396                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1397                     !dquot->dq_dqb.dqb_isoftlimit)
1398                         dquot->dq_dqb.dqb_curinodes = 0;
1399
1400                 if (!dqblk->dqb_isoftlimit ||
1401                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1402                         dquot->dq_dqb.dqb_itime = 0;
1403                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1404                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1405                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1406
1407                 flag |= LQUOTA_FLAGS_ADJINO;
1408         }
1409         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1410         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1411                                flag, oqaq);
1412         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1413         if (rc < 0)
1414                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1415
1416
1417         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1418
1419         cfs_mutex_unlock(&dquot->dq_mutex);
1420
1421         if (rc) {
1422                 CERROR("set limit failed! (rc:%d)\n", rc);
1423                 GOTO(update_fail, rc);
1424         }
1425
1426         cfs_up_write(&mds->mds_qonoff_sem);
1427         adjust_lqs(obd, oqaq);
1428
1429         orig_set = ihardlimit || isoftlimit;
1430         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1431         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1432                 cfs_mutex_lock(&dquot->dq_mutex);
1433                 dquot->dq_dqb.dqb_curinodes = 0;
1434                 cfs_mutex_unlock(&dquot->dq_mutex);
1435                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set);
1436                 if (rc) {
1437                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1438                         goto revoke_out;
1439                 }
1440         }
1441
1442         orig_set = bhardlimit || bsoftlimit;
1443         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1444         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1445                 cfs_mutex_lock(&dquot->dq_mutex);
1446                 dquot->dq_dqb.dqb_curspace = 0;
1447                 cfs_mutex_unlock(&dquot->dq_mutex);
1448                 rc = mds_init_slave_blimits(obd, oqctl, orig_set);
1449                 if (rc) {
1450                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1451                         goto revoke_out;
1452                 }
1453         }
1454
1455 revoke_out:
1456         cfs_down_write(&mds->mds_qonoff_sem);
1457         cfs_mutex_lock(&dquot->dq_mutex);
1458         if (rc) {
1459                 /* cancel previous setting */
1460                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1461                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1462                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1463                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1464                 dquot->dq_dqb.dqb_btime = btime;
1465                 dquot->dq_dqb.dqb_itime = itime;
1466         }
1467         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1468         cfs_mutex_unlock(&dquot->dq_mutex);
1469 update_fail:
1470         cfs_spin_lock(&lqs->lqs_lock);
1471         LQS_CLEAR_SETQUOTA(lqs);
1472         cfs_spin_unlock(&lqs->lqs_lock);
1473 skip:
1474         lqs_putref(lqs);
1475 out:
1476         lustre_dqput(dquot);
1477         EXIT;
1478 out_sem:
1479         cfs_up_write(&mds->mds_qonoff_sem);
1480
1481         if (oqaq)
1482                 OBD_FREE_PTR(oqaq);
1483
1484         return rc ? rc : rc2;
1485 }
1486
1487 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1488 {
1489         struct obd_quotactl *soqc;
1490         struct lvfs_run_ctxt saved;
1491         int rc, rc1;
1492         ENTRY;
1493
1494         OBD_ALLOC_PTR(soqc);
1495         if (!soqc)
1496                 RETURN(-ENOMEM);
1497
1498         soqc->qc_cmd = Q_GETOQUOTA;
1499         soqc->qc_id = oqctl->qc_id;
1500         soqc->qc_type = oqctl->qc_type;
1501
1502         /* get block usage from OSS */
1503         soqc->qc_dqblk.dqb_curspace = 0;
1504         rc = obd_quotactl(obd->u.mds.mds_lov_exp, soqc);
1505         if (!rc || rc == -EREMOTEIO) {
1506                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1507                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1508         }
1509
1510         /* get block/inode usage from MDS */
1511         soqc->qc_dqblk.dqb_curspace = 0;
1512         soqc->qc_dqblk.dqb_curinodes = 0;
1513         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1514         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1515         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1516         if (!rc1) {
1517                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1518                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1519                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1520         }
1521
1522         OBD_FREE_PTR(soqc);
1523
1524         RETURN(rc ? : rc1);
1525 }
1526
1527 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1528 {
1529         struct mds_obd *mds = &obd->u.mds;
1530         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1531         struct lustre_dquot *dquot;
1532         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1533         int rc;
1534         ENTRY;
1535
1536         if (oqctl->qc_type != USRQUOTA &&
1537             oqctl->qc_type != GRPQUOTA)
1538                 RETURN(-EINVAL);
1539
1540         cfs_down_read(&mds->mds_qonoff_sem);
1541         dqblk->dqb_valid = 0;
1542         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
1543                 CWARN("quota[%u] is off\n", oqctl->qc_type);
1544                 GOTO(out, rc = -ESRCH);
1545         }
1546
1547         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type, 1);
1548         if (IS_ERR(dquot))
1549                 GOTO(out, rc = PTR_ERR(dquot));
1550
1551         cfs_mutex_lock(&dquot->dq_mutex);
1552         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1553         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1554         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1555         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1556         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1557         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1558         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1559         cfs_mutex_unlock(&dquot->dq_mutex);
1560
1561         lustre_dqput(dquot);
1562         cfs_up_read(&mds->mds_qonoff_sem);
1563
1564         /* the usages in admin quota file is inaccurate */
1565         dqblk->dqb_curinodes = 0;
1566         dqblk->dqb_curspace = 0;
1567         rc = mds_get_space(obd, oqctl);
1568
1569         RETURN(rc);
1570
1571 out:
1572         cfs_up_read(&mds->mds_qonoff_sem);
1573         return rc;
1574 }
1575
1576 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1577 {
1578         struct lvfs_run_ctxt saved;
1579         int rc;
1580         ENTRY;
1581
1582         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1583         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1584         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1585
1586         RETURN(rc);
1587 }
1588
1589
1590 /* FIXME we only recovery block limit by now, need recovery inode
1591  * limits also after CMD involved in */
1592 static int
1593 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1594 {
1595         struct mds_obd *mds = &obd->u.mds;
1596         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1597         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1598         struct lustre_qunit_size *lqs;
1599         struct lustre_dquot *dquot;
1600         struct obd_quotactl *qctl;
1601         __u64 total_limits = 0;
1602         int rc = 0;
1603         ENTRY;
1604
1605         OBD_ALLOC_PTR(qctl);
1606         if (qctl == NULL)
1607                 RETURN(-ENOMEM);
1608
1609         dquot = lustre_dqget(obd, qinfo, id, type, 0);
1610         if (IS_ERR(dquot)) {
1611                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1612                 OBD_FREE_PTR(qctl);
1613                 RETURN(PTR_ERR(dquot));
1614         }
1615
1616         lqs = quota_search_lqs(LQS_KEY(type, id), qctxt, 1);
1617         if (lqs == NULL)
1618                 rc = -ENOENT;
1619         if (IS_ERR(lqs))
1620                 rc = PTR_ERR(lqs);
1621         if (rc)
1622                 GOTO(skip, rc);
1623
1624         cfs_mutex_lock(&dquot->dq_mutex);
1625
1626         /* don't recover the dquot without limits or quota is setting or
1627          * another recovery is already going on */
1628         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1629             LQS_IS_SETQUOTA(lqs) || LQS_IS_RECOVERY(lqs)) {
1630                 cfs_mutex_unlock(&dquot->dq_mutex);
1631                 GOTO(skip1, rc = 0);
1632         }
1633
1634         cfs_spin_lock(&lqs->lqs_lock);
1635         LQS_SET_RECOVERY(lqs);
1636         cfs_spin_unlock(&lqs->lqs_lock);
1637         cfs_mutex_unlock(&dquot->dq_mutex);
1638
1639         /* release mds_qonoff_sem during obd_quotactl ops here */
1640         cfs_up_write(&mds->mds_qonoff_sem);
1641
1642         /* get real bhardlimit from all slaves. */
1643         qctl->qc_cmd = Q_GETOQUOTA;
1644         qctl->qc_type = type;
1645         qctl->qc_id = id;
1646         qctl->qc_stat = QUOTA_RECOVERING;
1647         rc = obd_quotactl(mds->mds_lov_exp, qctl);
1648         cfs_down_write(&mds->mds_qonoff_sem);
1649         if (rc)
1650                 GOTO(out, rc);
1651         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1652
1653         /* get real bhardlimit from master */
1654         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1655         if (rc)
1656                 GOTO(out, rc);
1657         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1658
1659         /* amend the usage of the administrative quotafile */
1660         cfs_mutex_lock(&dquot->dq_mutex);
1661
1662         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1663
1664         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1665         if (rc)
1666                 CERROR("write dquot failed! (rc:%d)\n", rc);
1667
1668         cfs_mutex_unlock(&dquot->dq_mutex);
1669         EXIT;
1670 out:
1671         cfs_spin_lock(&lqs->lqs_lock);
1672         LQS_CLEAR_RECOVERY(lqs);
1673         cfs_spin_unlock(&lqs->lqs_lock);
1674 skip1:
1675         lqs_putref(lqs);
1676 skip:
1677         lustre_dqput(dquot);
1678         OBD_FREE_PTR(qctl);
1679         return rc;
1680 }
1681
1682 struct qmaster_recov_thread_data {
1683         struct obd_device *obd;
1684         cfs_completion_t comp;
1685 };
1686
1687 static int qmaster_recovery_main(void *arg)
1688 {
1689         struct qmaster_recov_thread_data *data = arg;
1690         struct obd_device *obd = data->obd;
1691         struct mds_obd *mds = &obd->u.mds;
1692         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1693         int rc = 0;
1694         unsigned short type;
1695         ENTRY;
1696
1697         cfs_daemonize_ctxt("qmaster_recovd");
1698
1699         /* for mds */
1700         class_incref(obd, "qmaster_recovd_mds", obd);
1701         /* for lov */
1702         class_incref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1703
1704         cfs_complete(&data->comp);
1705
1706         cfs_down_write(&mds->mds_qonoff_sem);
1707         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1708                 cfs_list_t id_list;
1709                 struct dquot_id *dqid, *tmp;
1710
1711                 if (qinfo->qi_files[type] == NULL)
1712                         continue;
1713
1714                 CFS_INIT_LIST_HEAD(&id_list);
1715                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1716                                  &id_list);
1717                 if (rc)
1718                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1719
1720                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1721                         cfs_list_del_init(&dqid->di_link);
1722                         if (rc)
1723                                 goto free;
1724
1725                         rc = dquot_recovery(obd, dqid->di_id, type);
1726                         if (rc)
1727                                 CERROR("%s: qmaster recovery failed for %sid %d"
1728                                        " rc:%d)\n", obd->obd_name,
1729                                        type ? "g" : "u", dqid->di_id, rc);
1730 free:
1731                         OBD_FREE_PTR(dqid);
1732                 }
1733         }
1734         cfs_up_write(&mds->mds_qonoff_sem);
1735         class_decref(mds->mds_lov_obd, "qmaster_recovd_lov", mds->mds_lov_obd);
1736         class_decref(obd, "qmaster_recovd_mds", obd);
1737         RETURN(rc);
1738 }
1739
1740 int mds_quota_recovery(struct obd_device *obd)
1741 {
1742         struct mds_obd *mds = &obd->u.mds;
1743         struct qmaster_recov_thread_data data;
1744         int rc = 0;
1745         ENTRY;
1746
1747         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
1748                 RETURN(0);
1749
1750         if (unlikely(!mds->mds_quota || obd->obd_stopping))
1751                 RETURN(rc);
1752
1753         cfs_mutex_lock(&obd->obd_dev_mutex);
1754         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1755                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1756                       mds->mds_lov_desc.ld_active_tgt_count,
1757                       mds->mds_lov_objid_count);
1758                 cfs_mutex_unlock(&obd->obd_dev_mutex);
1759                 RETURN(rc);
1760         }
1761         cfs_mutex_unlock(&obd->obd_dev_mutex);
1762
1763         data.obd = obd;
1764         cfs_init_completion(&data.comp);
1765
1766         rc = cfs_create_thread(qmaster_recovery_main, &data,
1767                                CFS_DAEMON_FLAGS);
1768         if (rc < 0)
1769                 CERROR("%s: cannot start quota recovery thread: rc %d\n",
1770                        obd->obd_name, rc);
1771
1772         cfs_wait_for_completion(&data.comp);
1773         RETURN(rc);
1774 }
1775
1776 #endif /* HAVE_QUOTA_SUPPORT */