lustre/quota/quota_master.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/quota/quota_master.c
  37  *
  38  * Lustre Quota Master request handler
  39  *
  40  * Author: Niu YaWei <niu@clusterfs.com>
  41  */
  42
  43 #ifndef EXPORT_SYMTAB
  44 # define EXPORT_SYMTAB
  45 #endif
  46
  47 #define DEBUG_SUBSYSTEM S_LQUOTA
  48
  49 #include <linux/version.h>
  50 #include <linux/fs.h>
  51 #include <asm/unistd.h>
  52 #include <linux/slab.h>
  53 #include <linux/quotaops.h>
  54 #include <linux/module.h>
  55 #include <linux/init.h>
  56 #include <linux/quota.h>
  57
  58 #include <obd_class.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_fsfilt.h>
  61 #include <lustre_mds.h>
  62
  63 #include "quota_internal.h"
  64
  65 #ifdef HAVE_QUOTA_SUPPORT
  66
  67 /* lock ordering: mds->mds_qonoff_sem > dquot->dq_sem */
  68 static struct list_head lustre_dquot_hash[NR_DQHASH];
  69 static spinlock_t dquot_hash_lock = SPIN_LOCK_UNLOCKED;
  70
  71 cfs_mem_cache_t *lustre_dquot_cachep;
  72
  73 int lustre_dquot_init(void)
  74 {
  75         int i;
  76         ENTRY;
  77
  78         LASSERT(lustre_dquot_cachep == NULL);
  79         lustre_dquot_cachep = cfs_mem_cache_create("lustre_dquot_cache",
  80                                                    sizeof(struct lustre_dquot),
  81                                                    0, 0);
  82         if (!lustre_dquot_cachep)
  83                 return (-ENOMEM);
  84
  85         for (i = 0; i < NR_DQHASH; i++) {
  86                 CFS_INIT_LIST_HEAD(lustre_dquot_hash + i);
  87         }
  88         RETURN(0);
  89 }
  90
  91 void lustre_dquot_exit(void)
  92 {
  93         int i;
  94         ENTRY;
  95         /* FIXME cleanup work ?? */
  96
  97         for (i = 0; i < NR_DQHASH; i++) {
  98                 LASSERT(list_empty(lustre_dquot_hash + i));
  99         }
 100         if (lustre_dquot_cachep) {
 101                 int rc;
 102                 rc = cfs_mem_cache_destroy(lustre_dquot_cachep);
 103                 LASSERTF(rc == 0,"couldn't destroy lustre_dquot_cachep slab\n");
 104                 lustre_dquot_cachep = NULL;
 105         }
 106         EXIT;
 107 }
 108
 109 static inline int
 110 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 111              __attribute__((__const__));
 112
 113 static inline int
 114 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
 115 {
 116         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
 117         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 118         return tmp;
 119 }
 120
 121 /* caller must hold dquot_hash_lock */
 122 static struct lustre_dquot *find_dquot(int hashent,
 123                                        struct lustre_quota_info *lqi, qid_t id,
 124                                        int type)
 125 {
 126         struct lustre_dquot *dquot;
 127         ENTRY;
 128
 129         LASSERT_SPIN_LOCKED(&dquot_hash_lock);
 130         list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
 131                 if (dquot->dq_info == lqi &&
 132                     dquot->dq_id == id && dquot->dq_type == type)
 133                         RETURN(dquot);
 134         }
 135         RETURN(NULL);
 136 }
 137
 138 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 139                                         qid_t id, int type)
 140 {
 141         struct lustre_dquot *dquot = NULL;
 142         ENTRY;
 143
 144         OBD_SLAB_ALLOC_PTR_GFP(dquot, lustre_dquot_cachep, CFS_ALLOC_IO);
 145         if (dquot == NULL)
 146                 RETURN(NULL);
 147
 148         CFS_INIT_LIST_HEAD(&dquot->dq_hash);
 149         init_mutex_locked(&dquot->dq_sem);
 150         dquot->dq_refcnt = 1;
 151         dquot->dq_info = lqi;
 152         dquot->dq_id = id;
 153         dquot->dq_type = type;
 154         dquot->dq_status = DQ_STATUS_AVAIL;
 155
 156         RETURN(dquot);
 157 }
 158
 159 static void free_dquot(struct lustre_dquot *dquot)
 160 {
 161         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 162 }
 163
 164 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 165 {
 166         struct list_head *head = lustre_dquot_hash +
 167             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 168         LASSERT(list_empty(&dquot->dq_hash));
 169         list_add(&dquot->dq_hash, head);
 170 }
 171
 172 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 173 {
 174         LASSERT(!list_empty(&dquot->dq_hash));
 175         list_del_init(&dquot->dq_hash);
 176 }
 177
 178 static void lustre_dqput(struct lustre_dquot *dquot)
 179 {
 180         ENTRY;
 181         spin_lock(&dquot_hash_lock);
 182         LASSERT(dquot->dq_refcnt);
 183         dquot->dq_refcnt--;
 184         if (!dquot->dq_refcnt) {
 185                 remove_dquot_nolock(dquot);
 186                 free_dquot(dquot);
 187         }
 188         spin_unlock(&dquot_hash_lock);
 189         EXIT;
 190 }
 191
 192 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 193                                          struct lustre_quota_info *lqi,
 194                                          qid_t id, int type)
 195 {
 196         unsigned int hashent = dquot_hashfn(lqi, id, type);
 197         struct lustre_dquot *dquot, *empty;
 198         ENTRY;
 199
 200         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 201                 RETURN(ERR_PTR(-ENOMEM));
 202
 203         spin_lock(&dquot_hash_lock);
 204         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 205                 dquot->dq_refcnt++;
 206                 spin_unlock(&dquot_hash_lock);
 207                 free_dquot(empty);
 208         } else {
 209                 int rc;
 210
 211                 dquot = empty;
 212                 insert_dquot_nolock(dquot);
 213                 spin_unlock(&dquot_hash_lock);
 214
 215                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
 216                 up(&dquot->dq_sem);
 217                 if (rc) {
 218                         CERROR("can't read dquot from admin quotafile! "
 219                                "(rc:%d)\n", rc);
 220                         lustre_dqput(dquot);
 221                         RETURN(ERR_PTR(rc));
 222                 }
 223
 224         }
 225
 226         LASSERT(dquot);
 227         RETURN(dquot);
 228 }
 229
 230 static void init_oqaq(struct quota_adjust_qunit *oqaq,
 231                       struct lustre_quota_ctxt *qctxt,
 232                       qid_t id, int type)
 233 {
 234         struct lustre_qunit_size *lqs = NULL;
 235
 236         oqaq->qaq_id = id;
 237         oqaq->qaq_flags = type;
 238         quota_search_lqs(NULL, oqaq, qctxt, &lqs);
 239         if (lqs) {
 240                 spin_lock(&lqs->lqs_lock);
 241                 oqaq->qaq_bunit_sz = lqs->lqs_bunit_sz;
 242                 oqaq->qaq_iunit_sz = lqs->lqs_iunit_sz;
 243                 oqaq->qaq_flags    = lqs->lqs_flags;
 244                 spin_unlock(&lqs->lqs_lock);
 245                 lqs_putref(lqs);
 246         } else {
 247                 CDEBUG(D_QUOTA, "Can't find the lustre qunit size!\n");
 248                 oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
 249                 oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
 250         }
 251 }
 252
 253 int dqacq_adjust_qunit_sz(struct obd_device *obd, qid_t id, int type,
 254                           __u32 is_blk)
 255 {
 256         struct mds_obd *mds = &obd->u.mds;
 257         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 258         struct obd_device *lov_mds_obd = class_exp2obd(mds->mds_osc_exp);
 259         struct lov_obd *lov = &lov_mds_obd->u.lov;
 260         __u32 ost_num = lov->desc.ld_tgt_count, mdt_num = 1;
 261         struct quota_adjust_qunit *oqaq = NULL;
 262         unsigned int uid = 0, gid = 0;
 263         struct lustre_quota_info *info = &mds->mds_quota_info;
 264         struct lustre_dquot *dquot = NULL;
 265         int adjust_res = 0;
 266         int rc = 0;
 267         ENTRY;
 268
 269         LASSERT(mds);
 270         dquot = lustre_dqget(obd, info, id, type);
 271         if (IS_ERR(dquot))
 272                 RETURN(PTR_ERR(dquot));
 273
 274         OBD_ALLOC_PTR(oqaq);
 275         if (!oqaq)
 276                 GOTO(out, rc = -ENOMEM);
 277
 278         down(&dquot->dq_sem);
 279         init_oqaq(oqaq, qctxt, id, type);
 280
 281         rc = dquot_create_oqaq(qctxt, dquot, ost_num, mdt_num,
 282                                is_blk ? LQUOTA_FLAGS_ADJBLK :
 283                                LQUOTA_FLAGS_ADJINO, oqaq);
 284
 285         if (rc < 0) {
 286                 CDEBUG(D_ERROR, "create oqaq failed! (rc:%d)\n", rc);
 287                 GOTO(out_sem, rc);
 288         }
 289         QAQ_DEBUG(oqaq, "show oqaq.\n")
 290
 291         if (!QAQ_IS_ADJBLK(oqaq) && !QAQ_IS_ADJINO(oqaq))
 292                 GOTO(out_sem, rc);
 293
 294         /* adjust the mds slave qunit size */
 295         adjust_res = quota_adjust_slave_lqs(oqaq, qctxt);
 296         if (adjust_res <= 0) {
 297                 if (adjust_res < 0) {
 298                         rc = adjust_res;
 299                         CDEBUG(D_ERROR, "adjust mds slave's qunit size failed! \
 300                                (rc:%d)\n", rc);
 301                 } else {
 302                         CDEBUG(D_QUOTA, "qunit doesn't need to be adjusted.\n");
 303                 }
 304                 GOTO(out_sem, rc);
 305         }
 306
 307         if (type)
 308                 gid = dquot->dq_id;
 309         else
 310                 uid = dquot->dq_id;
 311
 312         up(&dquot->dq_sem);
 313
 314         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, is_blk, 0, NULL);
 315         if (rc == -EDQUOT || rc == -EBUSY) {
 316                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 317                 rc = 0;
 318         }
 319         if (rc) {
 320                 CDEBUG(D_ERROR, "mds fail to adjust file quota! \
 321                                (rc:%d)\n", rc);
 322                 GOTO(out, rc);
 323         }
 324
 325         /* only when block qunit is reduced, boardcast to osts */
 326         if ((adjust_res & LQS_BLK_DECREASE) && QAQ_IS_ADJBLK(oqaq))
 327                 rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
 328
 329 out:
 330         lustre_dqput(dquot);
 331         if (oqaq)
 332                 OBD_FREE_PTR(oqaq);
 333
 334         RETURN(rc);
 335 out_sem:
 336         up(&dquot->dq_sem);
 337         goto out;
 338 }
 339
 340 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
 341 {
 342         struct mds_obd *mds = &obd->u.mds;
 343         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
 344         struct lustre_quota_info *info = &mds->mds_quota_info;
 345         struct lustre_dquot *dquot = NULL;
 346         __u64 *usage = NULL;
 347         __u64 hlimit = 0, slimit = 0;
 348         time_t *time = NULL;
 349         unsigned int grace = 0;
 350         struct lustre_qunit_size *lqs = NULL;
 351         int rc = 0;
 352         ENTRY;
 353
 354         if (OBD_FAIL_CHECK(OBD_FAIL_OBD_DQACQ))
 355                 RETURN(-EIO);
 356
 357         dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata));
 358         if (IS_ERR(dquot))
 359                 RETURN(PTR_ERR(dquot));
 360
 361         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 362         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 363
 364         down(&mds->mds_qonoff_sem);
 365         down(&dquot->dq_sem);
 366
 367         if (dquot->dq_status & DQ_STATUS_RECOVERY) {
 368                 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
 369                 GOTO(out, rc = -EBUSY);
 370         }
 371
 372         if (QDATA_IS_BLK(qdata)) {
 373                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
 374                 usage = &dquot->dq_dqb.dqb_curspace;
 375                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 376                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 377                 time = &dquot->dq_dqb.dqb_btime;
 378         } else {
 379                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
 380                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 381                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 382                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 383                 time = &dquot->dq_dqb.dqb_itime;
 384         }
 385
 386         /* if the quota limit in admin quotafile is zero, we just inform
 387          * slave to clear quota limit with zero qd_count */
 388         if (hlimit == 0 && slimit == 0) {
 389                 qdata->qd_count = 0;
 390                 GOTO(out, rc);
 391         }
 392
 393         switch (opc) {
 394         case QUOTA_DQACQ:
 395                 if (hlimit &&
 396                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
 397                 {
 398                         if (QDATA_IS_CHANGE_QS(qdata) &&
 399                             QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
 400                                 qdata->qd_count = (hlimit -
 401                                         QUSG(*usage, QDATA_IS_BLK(qdata)))
 402                                         * (QDATA_IS_BLK(qdata) ?
 403                                            QUOTABLOCK_SIZE : 1);
 404                         else
 405                                 GOTO(out, rc = -EDQUOT);
 406                 }
 407
 408                 if (slimit &&
 409                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
 410                         if (*time && cfs_time_current_sec() >= *time)
 411                                 GOTO(out, rc = -EDQUOT);
 412                         else if (!*time)
 413                                 *time = cfs_time_current_sec() + grace;
 414                 }
 415
 416                 *usage += qdata->qd_count;
 417                 break;
 418         case QUOTA_DQREL:
 419                 /* The usage in administrative file might be incorrect before
 420                  * recovery done */
 421                 if (*usage - qdata->qd_count < 0)
 422                         *usage = 0;
 423                 else
 424                         *usage -= qdata->qd_count;
 425
 426                 /* (usage <= soft limit) but not (usage < soft limit) */
 427                 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
 428                         *time = 0;
 429                 break;
 430         default:
 431                 LBUG();
 432         }
 433
 434         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 435         EXIT;
 436 out:
 437         up(&dquot->dq_sem);
 438         up(&mds->mds_qonoff_sem);
 439         lustre_dqput(dquot);
 440         if (rc != -EDQUOT)
 441                 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
 442                                       QDATA_IS_BLK(qdata));
 443
 444         quota_search_lqs(qdata, NULL, qctxt, &lqs);
 445         if (QDATA_IS_BLK(qdata)) {
 446                 if (!lqs) {
 447                         CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
 448                         qdata->qd_qunit  = qctxt->lqc_bunit_sz;
 449                 } else {
 450                         spin_lock(&lqs->lqs_lock);
 451                         qdata->qd_qunit  = lqs->lqs_bunit_sz;
 452                         spin_unlock(&lqs->lqs_lock);
 453                 }
 454                 QDATA_SET_ADJBLK(qdata);
 455         } else {
 456                 if (!lqs) {
 457                         CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
 458                         qdata->qd_qunit  = qctxt->lqc_iunit_sz;
 459                 } else {
 460                         spin_lock(&lqs->lqs_lock);
 461                         qdata->qd_qunit  = lqs->lqs_iunit_sz;
 462                         spin_unlock(&lqs->lqs_lock);
 463                 }
 464                 QDATA_SET_ADJINO(qdata);
 465         }
 466
 467         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
 468         if (lqs)
 469                 lqs_putref(lqs);
 470
 471         return rc;
 472 }
 473
 474 int mds_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 475                      unsigned int qpids[], int rc, int opc)
 476 {
 477         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 478         int rc2 = 0;
 479         ENTRY;
 480
 481         if (rc && rc != -EDQUOT && rc != ENOLCK)
 482                 RETURN(0);
 483
 484         switch (opc) {
 485         case FSFILT_OP_SETATTR:
 486                 /* release file quota on original owner */
 487                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 0, 0,
 488                                           NULL);
 489                 /* release block quota on original owner */
 490                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 491                                           NULL);
 492                 /* acquire file quota on current owner */
 493                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
 494                                           NULL);
 495                 /* acquire block quota on current owner */
 496                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 497                                           NULL);
 498                 break;
 499         case FSFILT_OP_UNLINK_PARTIAL_CHILD:
 500                 /* release file quota on child */
 501                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
 502                                           NULL);
 503                 /* rlease block quota on child */
 504                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 505                                           NULL);
 506                 break;
 507         case FSFILT_OP_CREATE_PARTIAL_CHILD:
 508                 /* acquire file quota on child */
 509                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
 510                                           NULL);
 511                 /* acquire block quota on child */
 512                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 513                                           NULL);
 514                 break;
 515         case FSFILT_OP_LINK:
 516                 /* acquire block quota on parent */
 517                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 518                                           NULL);
 519                 break;
 520         case FSFILT_OP_UNLINK:
 521                 /* release block quota on parent */
 522                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 523                                           NULL);
 524                 /* release file quota on child */
 525                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
 526                                           NULL);
 527                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 528                         /* release block quota on child */
 529                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0],
 530                                                   qcids[1], 1, 0, NULL);
 531                 break;
 532         case FSFILT_OP_UNLINK_PARTIAL_PARENT:
 533                 /* release block quota on parent */
 534                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 535                                           NULL);
 536                 break;
 537         case FSFILT_OP_CREATE:
 538                 /* acquire block quota on parent */
 539                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 540                                           NULL);
 541                 /* acquire file quota on child */
 542                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
 543                                           NULL);
 544                 if (qpids[0] != qcids[0] || qpids[1] != qcids[1])
 545                         /* acquire block quota on child */
 546                         rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0],
 547                                                   qcids[1], 1, 0, NULL);
 548                 break;
 549         default:
 550                 LBUG();
 551                 break;
 552         }
 553
 554         if (rc2)
 555                 CDEBUG(rc2 == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
 556                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
 557                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
 558         RETURN(0);
 559 }
 560
 561 int filter_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 562                         unsigned int qpids[], int rc, int opc)
 563 {
 564         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 565         int rc2 = 0;
 566         ENTRY;
 567
 568         if (rc && rc != -EDQUOT)
 569                 RETURN(0);
 570
 571         switch (opc) {
 572         case FSFILT_OP_SETATTR:
 573                 /* acquire/release block quota on original & current owner */
 574                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 575                                         NULL);
 576                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 577                                          NULL);
 578                 break;
 579         case FSFILT_OP_UNLINK:
 580                 /* release block quota on this owner */
 581         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 582                 /* acquire block quota on this owner */
 583                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 584                                         NULL);
 585                 break;
 586         default:
 587                 LBUG();
 588                 break;
 589         }
 590
 591         if (rc || rc2) {
 592                 if (!rc)
 593                         rc = rc2;
 594                 CDEBUG(rc == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
 595                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
 596                        QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
 597         }
 598
 599         RETURN(0);
 600 }
 601
 602 static const char prefix[] = "OBJECTS/";
 603
 604 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 605 {
 606         struct mds_obd *mds = &obd->u.mds;
 607         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 608         int rc = 0, i;
 609         char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 610         char name[64];
 611         struct lvfs_run_ctxt saved;
 612
 613         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 614
 615         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 616
 617         down(&mds->mds_qonoff_sem);
 618
 619         for (i = 0; i < MAXQUOTAS; i++) {
 620                 struct file *fp;
 621
 622                 if (!Q_TYPESET(oqctl, i))
 623                         continue;
 624
 625                 /* quota file has been opened ? */
 626                 if (qinfo->qi_files[i]) {
 627                         rc = -EBUSY;
 628                         goto out;
 629                 }
 630
 631                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 632                 sprintf(name, "%s%s", prefix, quotafile[i]);
 633
 634                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 635                 if (IS_ERR(fp)) {
 636                         rc = PTR_ERR(fp);
 637                         CERROR("error invalidating admin quotafile %s (rc:%d)\n",
 638                                name, rc);
 639                 }
 640                 else
 641                         filp_close(fp, 0);
 642         }
 643
 644 out:
 645         up(&mds->mds_qonoff_sem);
 646
 647         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 648
 649         return rc;
 650 }
 651
 652 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 653 {
 654         struct mds_obd *mds = &obd->u.mds;
 655         int rc;
 656         struct lvfs_run_ctxt saved;
 657
 658         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 659         down(&mds->mds_qonoff_sem);
 660
 661         oqctl->qc_cmd = Q_FINVALIDATE;
 662         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 663         if (!rc)
 664                 rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 665
 666         up(&mds->mds_qonoff_sem);
 667         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 668
 669         return rc;
 670 }
 671
 672 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 673 {
 674         struct mds_obd *mds = &obd->u.mds;
 675         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 676         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 677         struct lvfs_run_ctxt saved;
 678         char name[64];
 679         int i, rc = 0;
 680         ENTRY;
 681
 682         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 683
 684         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 685
 686         down(&mds->mds_qonoff_sem);
 687
 688         for (i = 0; i < MAXQUOTAS && !rc; i++) {
 689                 struct file *fp;
 690
 691                 if (!Q_TYPESET(oqctl, i))
 692                         continue;
 693
 694                 /* quota file has been opened ? */
 695                 if (qinfo->qi_files[i]) {
 696                         CWARN("init %s admin quotafile while quota on.\n",
 697                               i == USRQUOTA ? "user" : "group");
 698                         continue;
 699                 }
 700
 701                 LASSERT(strlen(quotafile[i]) + sizeof(prefix) <= sizeof(name));
 702                 sprintf(name, "%s%s", prefix, quotafile[i]);
 703
 704                 /* check if quota file exists and is correct */
 705                 fp = filp_open(name, O_RDONLY, 0);
 706                 if (!IS_ERR(fp)) {
 707                         /* irregular file is not the right place for quota */
 708                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 709                                 CERROR("admin quota file %s is not "
 710                                        "regular!", name);
 711                                 filp_close(fp, 0);
 712                                 rc = -EINVAL;
 713                                 break;
 714                         }
 715                         qinfo->qi_files[i] = fp;
 716                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 717                         qinfo->qi_files[i] = 0;
 718                         filp_close(fp, 0);
 719                 }
 720                 else
 721                         rc = PTR_ERR(fp);
 722
 723                 if (!rc)
 724                         continue;
 725
 726                 /* -EINVAL may be returned by quotainfo for bad quota file */
 727                 if (rc != -ENOENT && rc != -EINVAL) {
 728                         CERROR("error opening old quota file %s (%d)\n",
 729                                name, rc);
 730                         break;
 731                 }
 732
 733                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
 734                        rc == -ENOENT ? "creating" : "overwriting");
 735
 736                 /* create quota file overwriting old if needed */
 737                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 738                 if (IS_ERR(fp)) {
 739                         rc = PTR_ERR(fp);
 740                         CERROR("error creating admin quotafile %s (rc:%d)\n",
 741                                name, rc);
 742                         break;
 743                 }
 744
 745                 qinfo->qi_files[i] = fp;
 746
 747                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 748                 if (rc)
 749                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 750                                i == USRQUOTA ? "user" : "group", rc);
 751
 752                 filp_close(fp, 0);
 753                 qinfo->qi_files[i] = NULL;
 754         }
 755         up(&mds->mds_qonoff_sem);
 756
 757         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 758         RETURN(rc);
 759 }
 760
 761 static int close_quota_files(struct obd_quotactl *oqctl,
 762                              struct lustre_quota_info *qinfo)
 763 {
 764         int i, rc = 0;
 765         ENTRY;
 766
 767         for (i = 0; i < MAXQUOTAS; i++) {
 768                 if (!Q_TYPESET(oqctl, i))
 769                         continue;
 770                 if (qinfo->qi_files[i] == NULL) {
 771                         rc = -ESRCH;
 772                         continue;
 773                 }
 774                 filp_close(qinfo->qi_files[i], 0);
 775                 qinfo->qi_files[i] = NULL;
 776         }
 777         RETURN(rc);
 778 }
 779
 780 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 781 {
 782         struct mds_obd *mds = &obd->u.mds;
 783         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 784         const char *quotafile[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 785         char name[64];
 786         int i, rc = 0;
 787         ENTRY;
 788
 789         LASSERT(qinfo->qi_version == LUSTRE_QUOTA_V2);
 790
 791         /* open admin quota files and read quotafile info */
 792         for (i = 0; i < MAXQUOTAS; i++) {
 793                 struct file *fp;
 794
 795                 if (!Q_TYPESET(oqctl, i))
 796                         continue;
 797
 798                 LASSERT(strlen(quotafile[i])
 799                         + sizeof(prefix) <= sizeof(name));
 800                 sprintf(name, "%s%s", prefix, quotafile[i]);
 801
 802                 if (qinfo->qi_files[i] != NULL) {
 803                         rc = -EBUSY;
 804                         break;
 805                 }
 806
 807                 fp = filp_open(name, O_RDWR, 0);
 808                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 809                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
 810                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
 811                         break;
 812                 }
 813                 qinfo->qi_files[i] = fp;
 814
 815                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 816                 if (rc) {
 817                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
 818                         break;
 819                 }
 820
 821                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 822                 if (rc) {
 823                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
 824                                rc);
 825                         break;
 826                 }
 827         }
 828
 829         if (rc && rc != -EBUSY)
 830                 close_quota_files(oqctl, qinfo);
 831
 832         RETURN(rc);
 833 }
 834
 835 int mds_admin_quota_off(struct obd_device *obd,
 836                         struct obd_quotactl *oqctl)
 837 {
 838         struct mds_obd *mds = &obd->u.mds;
 839         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 840         int rc;
 841         ENTRY;
 842
 843         /* close admin quota files */
 844         rc = close_quota_files(oqctl, qinfo);
 845         RETURN(rc);
 846 }
 847
 848 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 849 {
 850         struct mds_obd *mds = &obd->u.mds;
 851         struct obd_device_target *obt = &obd->u.obt;
 852         struct lvfs_run_ctxt saved;
 853         int rc;
 854         ENTRY;
 855
 856         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 857                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 858                 atomic_inc(&obt->obt_quotachecking);
 859                 RETURN(-EBUSY);
 860         }
 861
 862         down(&mds->mds_qonoff_sem);
 863         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 864         rc = mds_admin_quota_on(obd, oqctl);
 865         if (rc)
 866                 GOTO(out, rc);
 867
 868         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 869         if (!rc)
 870                 obt->obt_qctxt.lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
 871         else
 872                 GOTO(out, rc);
 873
 874         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 875
 876 out:
 877         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 878         up(&mds->mds_qonoff_sem);
 879         atomic_inc(&obt->obt_quotachecking);
 880         RETURN(rc);
 881 }
 882
 883 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 884 {
 885         struct mds_obd *mds = &obd->u.mds;
 886         struct obd_device_target *obt = &obd->u.obt;
 887         struct lvfs_run_ctxt saved;
 888         int rc, rc2;
 889         ENTRY;
 890
 891         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 892                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 893                 atomic_inc(&obt->obt_quotachecking);
 894                 RETURN(-EBUSY);
 895         }
 896
 897         down(&mds->mds_qonoff_sem);
 898         /* close admin quota files */
 899         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 900         mds_admin_quota_off(obd, oqctl);
 901
 902         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 903         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 904         if (!rc2)
 905                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
 906
 907         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 908         up(&mds->mds_qonoff_sem);
 909         atomic_inc(&obt->obt_quotachecking);
 910
 911         RETURN(rc ?: rc2);
 912 }
 913
 914 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 915 {
 916         struct mds_obd *mds = &obd->u.mds;
 917         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 918         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 919         int rc;
 920         ENTRY;
 921
 922         down(&mds->mds_qonoff_sem);
 923         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 924                 rc = -ESRCH;
 925                 goto out;
 926         }
 927
 928         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 929         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 930         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 931
 932         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 933
 934 out:
 935         up(&mds->mds_qonoff_sem);
 936         RETURN(rc);
 937 }
 938
 939 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 940 {
 941         struct mds_obd *mds = &obd->u.mds;
 942         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 943         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 944         int rc = 0;
 945         ENTRY;
 946
 947         down(&mds->mds_qonoff_sem);
 948         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 949                 rc = -ESRCH;
 950                 goto out;
 951         }
 952
 953         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
 954         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
 955         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
 956
 957 out:
 958         up(&mds->mds_qonoff_sem);
 959         RETURN(rc);
 960 }
 961
 962 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
 963                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
 964                       int type, struct quota_adjust_qunit *oqaq)
 965 {
 966         __u64 bunit_curr_o, iunit_curr_o;
 967         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
 968         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
 969         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
 970                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
 971         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
 972                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
 973         int rc = 0;
 974         ENTRY;
 975
 976         if (!dquot || !oqaq)
 977                 RETURN(-EINVAL);
 978         LASSERT_SEM_LOCKED(&dquot->dq_sem);
 979         LASSERT(oqaq->qaq_iunit_sz);
 980         LASSERT(oqaq->qaq_bunit_sz);
 981
 982         /* don't change qunit size */
 983         if (!qctxt->lqc_switch_qs)
 984                 RETURN(rc);
 985
 986         bunit_curr_o = oqaq->qaq_bunit_sz;
 987         iunit_curr_o = oqaq->qaq_iunit_sz;
 988
 989         if (dquot->dq_type == GRPQUOTA)
 990                 QAQ_SET_GRP(oqaq);
 991
 992         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
 993                 __u64 b_limitation =
 994                         oqaq->qaq_bunit_sz * ost_num * shrink_qunit_limit;
 995                 /* enlarge block qunit size */
 996                 while (blimit >
 997                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
 998                         oqaq->qaq_bunit_sz =
 999                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1000                                 << QUOTABLOCK_BITS;
1001                         b_limitation = oqaq->qaq_bunit_sz * ost_num *
1002                                 shrink_qunit_limit;
1003                 }
1004
1005                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1006                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1007
1008                 /* shrink block qunit size */
1009                 while (blimit <
1010                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1011                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1012                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1013                                 QUOTABLOCK_BITS;
1014                         b_limitation = oqaq->qaq_bunit_sz * ost_num *
1015                                 shrink_qunit_limit;
1016                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1017                                 break;
1018                 }
1019
1020                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1021                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1022
1023                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1024                         QAQ_SET_ADJBLK(oqaq);
1025
1026         }
1027
1028         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1029                 __u64 i_limitation =
1030                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1031                 /* enlarge file qunit size */
1032                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1033                        + 2 * i_limitation) {
1034                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1035                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1036                                 shrink_qunit_limit;
1037                 }
1038
1039                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1040                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1041
1042                 /* shrink file qunit size */
1043                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1044                        + i_limitation) {
1045                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1046                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1047                                        shrink_qunit_limit;
1048                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1049                                 break;
1050                 }
1051
1052                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1053                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1054
1055                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1056                         QAQ_SET_ADJINO(oqaq);
1057
1058         }
1059
1060         if (!dquot->dq_dqb.dqb_bhardlimit && !dquot->dq_dqb.dqb_bsoftlimit &&
1061             !dquot->dq_dqb.dqb_ihardlimit && !dquot->dq_dqb.dqb_isoftlimit) {
1062                 oqaq->qaq_bunit_sz = 0;
1063                 oqaq->qaq_iunit_sz = 0;
1064                 QAQ_SET_ADJBLK(oqaq);
1065                 QAQ_SET_ADJINO(oqaq);
1066         }
1067
1068         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1069
1070         RETURN(rc);
1071 }
1072
1073 static int mds_init_slave_ilimits(struct obd_device *obd,
1074                                   struct obd_quotactl *oqctl, int set,
1075                                   struct quota_adjust_qunit *oqaq)
1076 {
1077         /* XXX: for file limits only adjust local now */
1078         struct obd_device_target *obt = &obd->u.obt;
1079         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1080         unsigned int uid = 0, gid = 0;
1081         struct obd_quotactl *ioqc = NULL;
1082         int flag;
1083         int rc;
1084         ENTRY;
1085
1086         /* if we are going to set zero limit, needn't init slaves */
1087         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1088             !set)
1089                 RETURN(0);
1090
1091         OBD_ALLOC_PTR(ioqc);
1092         if (!ioqc)
1093                 RETURN(-ENOMEM);
1094
1095         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1096                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1097         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1098         ioqc->qc_id = oqctl->qc_id;
1099         ioqc->qc_type = oqctl->qc_type;
1100         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1101         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1102
1103         if (QAQ_IS_ADJINO(oqaq)) {
1104                 /* adjust the mds slave's inode qunit size */
1105                 rc = quota_adjust_slave_lqs(oqaq, qctxt);
1106                 if (rc < 0)
1107                         CDEBUG(D_ERROR, "adjust mds slave's inode qunit size \
1108                                failed! (rc:%d)\n", rc);
1109         }
1110
1111         /* set local limit to MIN_QLIMIT */
1112         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1113         if (rc)
1114                 GOTO(out, rc);
1115
1116         /* trigger local qunit pre-acquire */
1117         if (oqctl->qc_type == USRQUOTA)
1118                 uid = oqctl->qc_id;
1119         else
1120                 gid = oqctl->qc_id;
1121
1122         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 0, 0,
1123                                 NULL);
1124         if (rc == -EDQUOT || rc == -EBUSY) {
1125                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1126                 rc = 0;
1127         }
1128         if (rc) {
1129                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1130                        rc);
1131                 GOTO(out, rc);
1132         }
1133         /* FIXME initialize all slaves in CMD */
1134         EXIT;
1135 out:
1136         if (ioqc)
1137                 OBD_FREE_PTR(ioqc);
1138         return rc;
1139 }
1140
1141 static int mds_init_slave_blimits(struct obd_device *obd,
1142                                   struct obd_quotactl *oqctl, int set,
1143                                   struct quota_adjust_qunit *oqaq)
1144 {
1145         struct obd_device_target *obt = &obd->u.obt;
1146         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1147         struct mds_obd *mds = &obd->u.mds;
1148         struct obd_quotactl *ioqc;
1149         unsigned int uid = 0, gid = 0;
1150         int rc, rc1 = 0;
1151         int flag;
1152         ENTRY;
1153
1154         /* if we are going to set zero limit, needn't init slaves */
1155         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1156             !set)
1157                 RETURN(0);
1158
1159         OBD_ALLOC_PTR(ioqc);
1160         if (!ioqc)
1161                 RETURN(-ENOMEM);
1162
1163         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1164                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1165         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1166         ioqc->qc_id = oqctl->qc_id;
1167         ioqc->qc_type = oqctl->qc_type;
1168         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1169         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1170         if (QAQ_IS_ADJBLK(oqaq)) {
1171                 /* adjust the mds slave's block qunit size */
1172                 rc1 = quota_adjust_slave_lqs(oqaq, qctxt);
1173                 if (rc1 < 0)
1174                         CERROR("adjust mds slave's block qunit size failed!"
1175                                "(rc:%d)\n", rc1);
1176         }
1177
1178         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1179         if (rc)
1180                 GOTO(out, rc);
1181
1182         /* trigger local qunit pre-acquire */
1183         if (oqctl->qc_type == USRQUOTA)
1184                 uid = oqctl->qc_id;
1185         else
1186                 gid = oqctl->qc_id;
1187
1188         /* initialize all slave's limit */
1189         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
1190
1191         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 1, 0,
1192                                 NULL);
1193         if (rc == -EDQUOT || rc == -EBUSY) {
1194                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1195                 rc = 0;
1196         }
1197         if (rc) {
1198                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1199                 GOTO(out, rc);
1200         }
1201
1202         /* adjust all slave's qunit size when setting quota
1203          * this is will create a lqs for every ost, which will present
1204          * certain uid/gid is set quota or not */
1205         QAQ_SET_ADJBLK(oqaq);
1206         rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
1207
1208         EXIT;
1209 out:
1210         OBD_FREE_PTR(ioqc);
1211         return rc;
1212 }
1213
1214 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1215 {
1216         struct mds_obd *mds = &obd->u.mds;
1217         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1218         struct obd_device *lov_obd = class_exp2obd(mds->mds_osc_exp);
1219         struct lov_obd *lov = &lov_obd->u.lov;
1220         struct quota_adjust_qunit *oqaq = NULL;
1221         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1222         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1223         time_t btime, itime;
1224         struct lustre_dquot *dquot;
1225         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1226         /* orig_set means if quota was set before; now_set means we are
1227          * setting/cancelling quota */
1228         int orig_set, now_set;
1229         int rc, rc2 = 0, flag = 0;
1230         ENTRY;
1231
1232         OBD_ALLOC_PTR(oqaq);
1233         if (!oqaq)
1234                 RETURN(-ENOMEM);
1235         down(&mds->mds_qonoff_sem);
1236         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1237
1238         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1239                 GOTO(out_sem, rc = -ESRCH);
1240
1241         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1242         if (IS_ERR(dquot))
1243                 GOTO(out_sem, rc = PTR_ERR(dquot));
1244         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1245         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1246
1247         down(&dquot->dq_sem);
1248
1249         if (dquot->dq_status) {
1250                 up(&dquot->dq_sem);
1251                 lustre_dqput(dquot);
1252                 GOTO(out_sem, rc = -EBUSY);
1253         }
1254         dquot->dq_status |= DQ_STATUS_SET;
1255
1256         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1257         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1258         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1259         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1260         btime = dquot->dq_dqb.dqb_btime;
1261         itime = dquot->dq_dqb.dqb_itime;
1262
1263         if (dqblk->dqb_valid & QIF_BTIME)
1264                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1265         if (dqblk->dqb_valid & QIF_ITIME)
1266                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1267
1268         if (dqblk->dqb_valid & QIF_BLIMITS) {
1269                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1270                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1271                 /* clear usage (limit pool) */
1272                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1273                     !dquot->dq_dqb.dqb_bsoftlimit)
1274                         dquot->dq_dqb.dqb_curspace = 0;
1275
1276                 /* clear grace time */
1277                 if (!dqblk->dqb_bsoftlimit ||
1278                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1279                         dquot->dq_dqb.dqb_btime = 0;
1280                 /* set grace only if user hasn't provided his own */
1281                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1282                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1283                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1284
1285                 flag |= LQUOTA_FLAGS_ADJBLK;
1286         }
1287
1288         if (dqblk->dqb_valid & QIF_ILIMITS) {
1289                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1290                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1291                 /* clear usage (limit pool) */
1292                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1293                     !dquot->dq_dqb.dqb_isoftlimit)
1294                         dquot->dq_dqb.dqb_curinodes = 0;
1295
1296                 if (!dqblk->dqb_isoftlimit ||
1297                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1298                         dquot->dq_dqb.dqb_itime = 0;
1299                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1300                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1301                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1302
1303                 flag |= LQUOTA_FLAGS_ADJINO;
1304         }
1305         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1306         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1307                                flag, oqaq);
1308         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1309         if (rc < 0)
1310                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1311
1312
1313         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1314
1315         up(&dquot->dq_sem);
1316
1317         if (rc) {
1318                 CERROR("set limit failed! (rc:%d)\n", rc);
1319                 goto out;
1320         }
1321
1322         up(&mds->mds_qonoff_sem);
1323         orig_set = ihardlimit || isoftlimit;
1324         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1325         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1326                 down(&dquot->dq_sem);
1327                 dquot->dq_dqb.dqb_curinodes = 0;
1328                 up(&dquot->dq_sem);
1329                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set, oqaq);
1330                 if (rc) {
1331                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1332                         goto revoke_out;
1333                 }
1334         }
1335
1336         orig_set = bhardlimit || bsoftlimit;
1337         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1338         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1339                 down(&dquot->dq_sem);
1340                 dquot->dq_dqb.dqb_curspace = 0;
1341                 up(&dquot->dq_sem);
1342                 rc = mds_init_slave_blimits(obd, oqctl, orig_set, oqaq);
1343                 if (rc) {
1344                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1345                         goto revoke_out;
1346                 }
1347         }
1348
1349 revoke_out:
1350         down(&mds->mds_qonoff_sem);
1351         down(&dquot->dq_sem);
1352         if (rc) {
1353                 /* cancel previous setting */
1354                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1355                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1356                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1357                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1358                 dquot->dq_dqb.dqb_btime = btime;
1359                 dquot->dq_dqb.dqb_itime = itime;
1360         }
1361         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1362         up(&dquot->dq_sem);
1363
1364 out:
1365         down(&dquot->dq_sem);
1366         dquot->dq_status &= ~DQ_STATUS_SET;
1367         up(&dquot->dq_sem);
1368         lustre_dqput(dquot);
1369         EXIT;
1370 out_sem:
1371         up(&mds->mds_qonoff_sem);
1372
1373         if (oqaq)
1374                 OBD_FREE_PTR(oqaq);
1375
1376         return rc ? rc : rc2;
1377 }
1378
1379 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1380 {
1381         struct obd_quotactl *soqc;
1382         struct lvfs_run_ctxt saved;
1383         int rc, rc1;
1384         ENTRY;
1385
1386         OBD_ALLOC_PTR(soqc);
1387         if (!soqc)
1388                 RETURN(-ENOMEM);
1389
1390         soqc->qc_cmd = Q_GETOQUOTA;
1391         soqc->qc_id = oqctl->qc_id;
1392         soqc->qc_type = oqctl->qc_type;
1393
1394         /* get block usage from OSS */
1395         soqc->qc_dqblk.dqb_curspace = 0;
1396         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
1397         if (!rc) {
1398                 oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1399                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1400         }
1401
1402         /* get block/inode usage from MDS */
1403         soqc->qc_dqblk.dqb_curspace = 0;
1404         soqc->qc_dqblk.dqb_curinodes = 0;
1405         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1406         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1407         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1408         if (!rc1) {
1409                 oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1410                 oqctl->qc_dqblk.dqb_curinodes = soqc->qc_dqblk.dqb_curinodes;
1411                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1412         }
1413
1414         OBD_FREE_PTR(soqc);
1415
1416         RETURN(rc ? : rc1);
1417 }
1418
1419 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1420 {
1421         struct mds_obd *mds = &obd->u.mds;
1422         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1423         struct lustre_dquot *dquot;
1424         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1425         int rc;
1426         ENTRY;
1427
1428         down(&mds->mds_qonoff_sem);
1429         dqblk->dqb_valid = 0;
1430         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1431                 GOTO(out, rc = -ESRCH);
1432
1433         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1434         if (IS_ERR(dquot))
1435                 GOTO(out, rc = PTR_ERR(dquot));
1436
1437         down(&dquot->dq_sem);
1438         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1439         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1440         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1441         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1442         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1443         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1444         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1445         up(&dquot->dq_sem);
1446
1447         lustre_dqput(dquot);
1448
1449         /* the usages in admin quota file is inaccurate */
1450         dqblk->dqb_curinodes = 0;
1451         dqblk->dqb_curspace = 0;
1452         rc = mds_get_space(obd, oqctl);
1453         EXIT;
1454 out:
1455         up(&mds->mds_qonoff_sem);
1456         return rc;
1457 }
1458
1459 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1460 {
1461         struct lvfs_run_ctxt saved;
1462         int rc;
1463         ENTRY;
1464
1465         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1466         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1467         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1468
1469         RETURN(rc);
1470 }
1471
1472
1473 /* FIXME we only recovery block limit by now, need recovery inode
1474  * limits also after CMD involved in */
1475 static int
1476 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1477 {
1478         struct mds_obd *mds = &obd->u.mds;
1479         struct lustre_quota_info *qinfo= &mds->mds_quota_info;
1480         struct lustre_dquot *dquot;
1481         struct obd_quotactl *qctl;
1482         __u64 total_limits = 0;
1483         int rc;
1484         ENTRY;
1485
1486         OBD_ALLOC_PTR(qctl);
1487         if (qctl == NULL)
1488                 RETURN(-ENOMEM);
1489
1490         dquot = lustre_dqget(obd, qinfo, id, type);
1491         if (IS_ERR(dquot)) {
1492                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1493                 OBD_FREE_PTR(qctl);
1494                 RETURN(PTR_ERR(dquot));
1495         }
1496
1497         down(&dquot->dq_sem);
1498
1499         /* don't recovery the dquot without limits or under setting */
1500         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1501             dquot->dq_status)
1502                 GOTO(skip, rc = 0);
1503         dquot->dq_status |= DQ_STATUS_RECOVERY;
1504
1505         up(&dquot->dq_sem);
1506
1507         /* get real bhardlimit from all slaves. */
1508         qctl->qc_cmd = Q_GETOQUOTA;
1509         qctl->qc_type = type;
1510         qctl->qc_id = id;
1511         qctl->qc_stat = QUOTA_RECOVERING;
1512         rc = obd_quotactl(mds->mds_osc_exp, qctl);
1513         if (rc)
1514                 GOTO(out, rc);
1515         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1516
1517         /* get real bhardlimit from master */
1518         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1519         if (rc)
1520                 GOTO(out, rc);
1521         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1522
1523         /* amend the usage of the administrative quotafile */
1524         down(&mds->mds_qonoff_sem);
1525         down(&dquot->dq_sem);
1526
1527         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1528
1529         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1530         if (rc)
1531                 CERROR("write dquot failed! (rc:%d)\n", rc);
1532
1533         up(&dquot->dq_sem);
1534         up(&mds->mds_qonoff_sem);
1535         EXIT;
1536 out:
1537         down(&dquot->dq_sem);
1538         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1539 skip:
1540         up(&dquot->dq_sem);
1541
1542         lustre_dqput(dquot);
1543         OBD_FREE_PTR(qctl);
1544         return rc;
1545 }
1546
1547 struct qmaster_recov_thread_data {
1548         struct obd_device *obd;
1549         struct completion comp;
1550 };
1551
1552 static int qmaster_recovery_main(void *arg)
1553 {
1554         struct qmaster_recov_thread_data *data = arg;
1555         struct obd_device *obd = data->obd;
1556         int rc = 0;
1557         unsigned short type;
1558         ENTRY;
1559
1560         ptlrpc_daemonize("qmaster_recovd");
1561
1562         complete(&data->comp);
1563
1564         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1565                 struct mds_obd *mds = &obd->u.mds;
1566                 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1567                 struct list_head id_list;
1568                 struct dquot_id *dqid, *tmp;
1569
1570                 down(&mds->mds_qonoff_sem);
1571                 if (qinfo->qi_files[type] == NULL) {
1572                         up(&mds->mds_qonoff_sem);
1573                         continue;
1574                 }
1575                 CFS_INIT_LIST_HEAD(&id_list);
1576                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1577                                  &id_list);
1578                 up(&mds->mds_qonoff_sem);
1579
1580                 if (rc)
1581                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1582
1583                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1584                         list_del_init(&dqid->di_link);
1585                         if (rc)
1586                                 goto free;
1587
1588                         rc = dquot_recovery(obd, dqid->di_id, type);
1589                         if (rc)
1590                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1591                                        " rc:%d)\n", dqid->di_id, type, rc);
1592 free:
1593                         kfree(dqid);
1594                 }
1595         }
1596         RETURN(rc);
1597 }
1598
1599 int mds_quota_recovery(struct obd_device *obd)
1600 {
1601         struct mds_obd *mds = &obd->u.mds;
1602         struct qmaster_recov_thread_data data;
1603         int rc = 0;
1604         ENTRY;
1605
1606         if (unlikely(!mds->mds_quota))
1607                 RETURN(rc);
1608
1609         mutex_down(&obd->obd_dev_sem);
1610         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1611                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1612                       mds->mds_lov_desc.ld_active_tgt_count,
1613                       mds->mds_lov_objid_count);
1614                 mutex_up(&obd->obd_dev_sem);
1615                 RETURN(rc);
1616         }
1617         mutex_up(&obd->obd_dev_sem);
1618
1619         data.obd = obd;
1620         init_completion(&data.comp);
1621
1622         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1623         if (rc < 0)
1624                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1625
1626         wait_for_completion(&data.comp);
1627         RETURN(rc);
1628 }
1629
1630 #endif /* HAVE_QUOTA_SUPPORT */