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                 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(dquot, lustre_dquot_cachep, CFS_ALLOC_IO, sizeof(*dquot));
 145         if (dquot == NULL)
 146                 RETURN(NULL);
 147
 148         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         OBD_FAIL_RETURN(OBD_FAIL_OBD_DQACQ, -EIO);
 355
 356         dquot = lustre_dqget(obd, info, qdata->qd_id, QDATA_IS_GRP(qdata));
 357         if (IS_ERR(dquot))
 358                 RETURN(PTR_ERR(dquot));
 359
 360         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 361         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 362
 363         down(&mds->mds_qonoff_sem);
 364         down(&dquot->dq_sem);
 365
 366         if (dquot->dq_status & DQ_STATUS_RECOVERY) {
 367                 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
 368                 GOTO(out, rc = -EBUSY);
 369         }
 370
 371         if (QDATA_IS_BLK(qdata)) {
 372                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_bgrace;
 373                 usage = &dquot->dq_dqb.dqb_curspace;
 374                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 375                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 376                 time = &dquot->dq_dqb.dqb_btime;
 377         } else {
 378                 grace = info->qi_info[QDATA_IS_GRP(qdata)].dqi_igrace;
 379                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 380                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 381                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 382                 time = &dquot->dq_dqb.dqb_itime;
 383         }
 384
 385         /* if the quota limit in admin quotafile is zero, we just inform
 386          * slave to clear quota limit with zero qd_count */
 387         if (hlimit == 0 && slimit == 0) {
 388                 qdata->qd_count = 0;
 389                 GOTO(out, rc);
 390         }
 391
 392         switch (opc) {
 393         case QUOTA_DQACQ:
 394                 if (hlimit &&
 395                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > hlimit)
 396                 {
 397                         if (QDATA_IS_CHANGE_QS(qdata) &&
 398                             QUSG(*usage, QDATA_IS_BLK(qdata)) < hlimit)
 399                                 qdata->qd_count = (hlimit -
 400                                         QUSG(*usage, QDATA_IS_BLK(qdata)))
 401                                         * (QDATA_IS_BLK(qdata) ?
 402                                            QUOTABLOCK_SIZE : 1);
 403                         else
 404                                 GOTO(out, rc = -EDQUOT);
 405                 }
 406
 407                 if (slimit &&
 408                     QUSG(*usage + qdata->qd_count, QDATA_IS_BLK(qdata)) > slimit) {
 409                         if (*time && cfs_time_current_sec() >= *time)
 410                                 GOTO(out, rc = -EDQUOT);
 411                         else if (!*time)
 412                                 *time = cfs_time_current_sec() + grace;
 413                 }
 414
 415                 *usage += qdata->qd_count;
 416                 break;
 417         case QUOTA_DQREL:
 418                 /* The usage in administrative file might be incorrect before
 419                  * recovery done */
 420                 if (*usage - qdata->qd_count < 0)
 421                         *usage = 0;
 422                 else
 423                         *usage -= qdata->qd_count;
 424
 425                 /* (usage <= soft limit) but not (usage < soft limit) */
 426                 if (!slimit || QUSG(*usage, QDATA_IS_BLK(qdata)) <= slimit)
 427                         *time = 0;
 428                 break;
 429         default:
 430                 LBUG();
 431         }
 432
 433         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 434         EXIT;
 435 out:
 436         up(&dquot->dq_sem);
 437         up(&mds->mds_qonoff_sem);
 438         lustre_dqput(dquot);
 439         if (rc != -EDQUOT)
 440                 dqacq_adjust_qunit_sz(obd, qdata->qd_id, QDATA_IS_GRP(qdata),
 441                                       QDATA_IS_BLK(qdata));
 442
 443         quota_search_lqs(qdata, NULL, qctxt, &lqs);
 444         if (QDATA_IS_BLK(qdata)) {
 445                 if (!lqs) {
 446                         CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
 447                         qdata->qd_qunit  = qctxt->lqc_bunit_sz;
 448                 } else {
 449                         spin_lock(&lqs->lqs_lock);
 450                         qdata->qd_qunit  = lqs->lqs_bunit_sz;
 451                         spin_unlock(&lqs->lqs_lock);
 452                 }
 453                 QDATA_SET_ADJBLK(qdata);
 454         } else {
 455                 if (!lqs) {
 456                         CDEBUG(D_INFO, "Can't find the lustre qunit size!\n");
 457                         qdata->qd_qunit  = qctxt->lqc_iunit_sz;
 458                 } else {
 459                         spin_lock(&lqs->lqs_lock);
 460                         qdata->qd_qunit  = lqs->lqs_iunit_sz;
 461                         spin_unlock(&lqs->lqs_lock);
 462                 }
 463                 QDATA_SET_ADJINO(qdata);
 464         }
 465
 466         QDATA_DEBUG(qdata, "alloc/release qunit in dqacq_handler\n");
 467         if (lqs)
 468                 lqs_putref(lqs);
 469
 470         return rc;
 471 }
 472
 473 int mds_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 474                      unsigned int qpids[], int rc, int opc)
 475 {
 476         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 477         int rc2 = 0;
 478         ENTRY;
 479
 480         if (rc && rc != -EDQUOT && rc != ENOLCK)
 481                 RETURN(0);
 482
 483         switch (opc) {
 484         case FSFILT_OP_RENAME:
 485                 /* acquire/release block quota on owner of original parent */
 486                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[2], qpids[3], 1, 0,
 487                                          NULL);
 488                 /* fall-through */
 489         case FSFILT_OP_SETATTR:
 490                 /* acquire/release file quota on original owner */
 491                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 0, 0,
 492                                           NULL);
 493                 /* fall-through */
 494         case FSFILT_OP_CREATE:
 495         case FSFILT_OP_UNLINK:
 496                 /* acquire/release file/block quota on owner of child
 497                  * (or current owner) */
 498                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0,
 499                                           NULL);
 500                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 501                                           NULL);
 502                 /* acquire/release block quota on owner of parent
 503                  * (or original owner) */
 504                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 505                                           NULL);
 506                 break;
 507         default:
 508                 LBUG();
 509                 break;
 510         }
 511
 512         if (rc2)
 513                 CDEBUG(rc2 == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
 514                        "mds adjust qunit %ssuccessfully! (opc:%d rc:%d)\n",
 515                        rc2 == QUOTA_REQ_RETURNED ? "" : "un", opc, rc2);
 516         RETURN(0);
 517 }
 518
 519 int filter_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 520                         unsigned int qpids[], int rc, int opc)
 521 {
 522         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 523         int rc2 = 0;
 524         ENTRY;
 525
 526         if (rc && rc != -EDQUOT)
 527                 RETURN(0);
 528
 529         switch (opc) {
 530         case FSFILT_OP_SETATTR:
 531                 /* acquire/release block quota on original & current owner */
 532                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 533                                         NULL);
 534                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0,
 535                                          NULL);
 536                 break;
 537         case FSFILT_OP_UNLINK:
 538                 /* release block quota on this owner */
 539         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 540                 /* acquire block quota on this owner */
 541                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0,
 542                                         NULL);
 543                 break;
 544         default:
 545                 LBUG();
 546                 break;
 547         }
 548
 549         if (rc || rc2) {
 550                 if (!rc)
 551                         rc = rc2;
 552                 CDEBUG(rc == QUOTA_REQ_RETURNED ? D_QUOTA: D_ERROR,
 553                        "filter adjust qunit %ssuccessfully! (opc:%d rc%d)\n",
 554                        QUOTA_REQ_RETURNED ? "" : "un", opc, rc);
 555         }
 556
 557         RETURN(0);
 558 }
 559
 560 static const char prefix[] = "OBJECTS/";
 561
 562 int mds_quota_get_version(struct obd_device *obd,
 563                           lustre_quota_version_t *version)
 564 {
 565         struct mds_obd *mds = &obd->u.mds;
 566         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 567
 568         *version = qinfo->qi_version;
 569
 570         return 0;
 571 }
 572
 573 int mds_quota_set_version(struct obd_device *obd, lustre_quota_version_t version)
 574 {
 575         struct mds_obd *mds = &obd->u.mds;
 576         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 577         int rc = 0, i;
 578
 579         if (version != LUSTRE_QUOTA_V1 && version != LUSTRE_QUOTA_V2)
 580                 return -EINVAL;
 581
 582         down(&mds->mds_qonoff_sem);
 583
 584         /* no need to change version? nothing to do then */
 585         if (qinfo->qi_version == version)
 586                 goto out;
 587
 588         for (i = 0; i < MAXQUOTAS; i++) {
 589                 /* quota file has been opened ? */
 590                 if (qinfo->qi_files[i]) {
 591                         rc = -EBUSY;
 592                         goto out;
 593                 }
 594         }
 595
 596         CDEBUG(D_INFO, "changing quota version %d -> %d\n", qinfo->qi_version,
 597                version);
 598
 599         qinfo->qi_version = version;
 600
 601 out:
 602         up(&mds->mds_qonoff_sem);
 603
 604         return rc;
 605 }
 606
 607 int mds_quota_invalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 608 {
 609         struct mds_obd *mds = &obd->u.mds;
 610         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 611         int rc = 0, i;
 612         char *quotafiles_v1[] = LUSTRE_ADMIN_QUOTAFILES_V1;
 613         char *quotafiles_v2[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 614         char name[64];
 615         struct lvfs_run_ctxt saved;
 616
 617         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 618
 619         down(&mds->mds_qonoff_sem);
 620
 621         for (i = 0; i < MAXQUOTAS; i++) {
 622                 struct file *fp;
 623                 char* quotafile = (qinfo->qi_version == LUSTRE_QUOTA_V1)?
 624                                    quotafiles_v1[i]:quotafiles_v2[i];
 625
 626                 if (!Q_TYPESET(oqctl, i))
 627                         continue;
 628
 629                 /* quota file has been opened ? */
 630                 if (qinfo->qi_files[i]) {
 631                         rc = -EBUSY;
 632                         goto out;
 633                 }
 634
 635                 LASSERT(strlen(quotafile) + sizeof(prefix) <= sizeof(name));
 636                 sprintf(name, "%s%s", prefix, quotafile);
 637
 638                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 639                 if (IS_ERR(fp)) {
 640                         rc = PTR_ERR(fp);
 641                         CERROR("error invalidating admin quotafile %s (rc:%d)\n",
 642                                name, rc);
 643                 }
 644                 else
 645                         filp_close(fp, 0);
 646         }
 647
 648 out:
 649         up(&mds->mds_qonoff_sem);
 650
 651         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 652
 653         return rc;
 654 }
 655
 656 int mds_quota_finvalidate(struct obd_device *obd, struct obd_quotactl *oqctl)
 657 {
 658         struct mds_obd *mds = &obd->u.mds;
 659         int rc;
 660         struct lvfs_run_ctxt saved;
 661
 662         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 663         down(&mds->mds_qonoff_sem);
 664
 665         oqctl->qc_cmd = Q_FINVALIDATE;
 666         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 667         if (!rc)
 668                 rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 669
 670         up(&mds->mds_qonoff_sem);
 671         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 672
 673         return rc;
 674 }
 675
 676 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 677 {
 678         struct mds_obd *mds = &obd->u.mds;
 679         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 680         char *quotafiles_v1[] = LUSTRE_ADMIN_QUOTAFILES_V1;
 681         char *quotafiles_v2[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 682         struct lvfs_run_ctxt saved;
 683         char name[64];
 684         int i, rc = 0;
 685         ENTRY;
 686
 687         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 688
 689         down(&mds->mds_qonoff_sem);
 690
 691         for (i = 0; i < MAXQUOTAS && !rc; i++) {
 692                 struct file *fp;
 693                 char* quotafile = (qinfo->qi_version == LUSTRE_QUOTA_V1)?
 694                                         quotafiles_v1[i]:quotafiles_v2[i];
 695
 696                 if (!Q_TYPESET(oqctl, i))
 697                         continue;
 698
 699                 /* quota file has been opened ? */
 700                 if (qinfo->qi_files[i]) {
 701                         CWARN("init %s admin quotafile while quota on.\n",
 702                               i == USRQUOTA ? "user" : "group");
 703                         continue;
 704                 }
 705
 706                 LASSERT(strlen(quotafile) + sizeof(prefix) <= sizeof(name));
 707                 sprintf(name, "%s%s", prefix, quotafile);
 708
 709                 /* check if quota file exists and is correct */
 710                 fp = filp_open(name, O_RDONLY, 0);
 711                 if (!IS_ERR(fp)) {
 712                         /* irregular file is not the right place for quota */
 713                         if (!S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 714                                 CERROR("admin quota file %s is not "
 715                                        "regular!", quotafile);
 716                                 filp_close(fp, 0);
 717                                 rc = -EINVAL;
 718                                 break;
 719                         }
 720                         qinfo->qi_files[i] = fp;
 721                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 722                         qinfo->qi_files[i] = 0;
 723                         filp_close(fp, 0);
 724                 }
 725                 else
 726                         rc = PTR_ERR(fp);
 727
 728                 if (!rc)
 729                         continue;
 730
 731                 /* -EINVAL may be returned by quotainfo for bad quota file */
 732                 if (rc != -ENOENT && rc != -EINVAL) {
 733                         CERROR("error opening old quota file %s (%d)\n",
 734                                name, rc);
 735                         break;
 736                 }
 737
 738                 CDEBUG(D_INFO, "%s new quota file %s\n", name,
 739                        rc == -ENOENT ? "creating" : "overwriting");
 740
 741                 /* create quota file overwriting old if needed */
 742                 fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 743                 if (IS_ERR(fp)) {
 744                         rc = PTR_ERR(fp);
 745                         CERROR("error creating admin quotafile %s (rc:%d)\n",
 746                                name, rc);
 747                         break;
 748                 }
 749
 750                 qinfo->qi_files[i] = fp;
 751
 752                 switch (qinfo->qi_version) {
 753                 case LUSTRE_QUOTA_V1:
 754                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 755                         if (rc)
 756                                 CERROR("error init %s admin quotafile! (rc:%d)\n",
 757                                        i == USRQUOTA ? "user" : "group", rc);
 758                         break;
 759                 case LUSTRE_QUOTA_V2:
 760                         rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CONVERT);
 761                         if (rc)
 762                                 CERROR("error convert %s admin quotafile! (rc:%d)\n",
 763                                        i == USRQUOTA ? "user" : "group", rc);
 764                         break;
 765                 default:
 766                         LBUG();
 767                 }
 768
 769                 filp_close(fp, 0);
 770                 qinfo->qi_files[i] = NULL;
 771         }
 772         up(&mds->mds_qonoff_sem);
 773
 774         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 775         RETURN(rc);
 776 }
 777
 778 static int close_quota_files(struct obd_quotactl *oqctl,
 779                              struct lustre_quota_info *qinfo)
 780 {
 781         int i, rc = 0;
 782         ENTRY;
 783
 784         for (i = 0; i < MAXQUOTAS; i++) {
 785                 if (!Q_TYPESET(oqctl, i))
 786                         continue;
 787                 if (qinfo->qi_files[i] == NULL) {
 788                         rc = -ESRCH;
 789                         continue;
 790                 }
 791                 filp_close(qinfo->qi_files[i], 0);
 792                 qinfo->qi_files[i] = NULL;
 793         }
 794         RETURN(rc);
 795 }
 796
 797 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 798 {
 799         struct mds_obd *mds = &obd->u.mds;
 800         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 801         const char *quotafiles_v1[] = LUSTRE_ADMIN_QUOTAFILES_V1;
 802         const char *quotafiles_v2[] = LUSTRE_ADMIN_QUOTAFILES_V2;
 803         char name[64];
 804         int i, rc = 0;
 805         ENTRY;
 806
 807         /* open admin quota files and read quotafile info */
 808         for (i = 0; i < MAXQUOTAS; i++) {
 809                 struct file *fp;
 810                 const char* quotafile = qinfo->qi_version == LUSTRE_QUOTA_V1?
 811                                         quotafiles_v1[i] : quotafiles_v2[i];
 812
 813                 if (!Q_TYPESET(oqctl, i))
 814                         continue;
 815
 816                 LASSERT(strlen(quotafile)
 817                         + sizeof(prefix) <= sizeof(name));
 818                 sprintf(name, "%s%s", prefix, quotafile);
 819
 820                 if (qinfo->qi_files[i] != NULL) {
 821                         rc = -EBUSY;
 822                         break;
 823                 }
 824
 825                 fp = filp_open(name, O_RDWR, 0);
 826                 /* handle transparent migration to 64 bit quota file */
 827                 if (IS_ERR(fp) && PTR_ERR(fp) == -ENOENT &&
 828                     qinfo->qi_version == LUSTRE_QUOTA_V2) {
 829                         CDEBUG(D_INFO, "attempting to convert V1 quota file to"
 830                                        " V2 format\n");
 831                         fp = filp_open(name, O_CREAT | O_TRUNC | O_RDWR, 0644);
 832                         if (!IS_ERR(fp)) {
 833                                 qinfo->qi_files[i] = fp;
 834                                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CONVERT);
 835                                 if (rc) {
 836                                         CERROR("error convert %s admin "
 837                                                "quotafile! (rc:%d)\n",
 838                                                i == USRQUOTA ? "user" : "group",
 839                                                rc);
 840                                         break;
 841                                 }
 842                         }
 843                 }
 844
 845                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 846                         rc = IS_ERR(fp) ? PTR_ERR(fp) : -EINVAL;
 847                         CERROR("error open/create %s! (rc:%d)\n", name, rc);
 848                         break;
 849                 }
 850                 qinfo->qi_files[i] = fp;
 851
 852                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_CHK);
 853                 if (rc) {
 854                         CERROR("invalid quota file %s! (rc:%d)\n", name, rc);
 855                         break;
 856                 }
 857
 858                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 859                 if (rc) {
 860                         CERROR("error read quotainfo of %s! (rc:%d)\n", name,
 861                                rc);
 862                         break;
 863                 }
 864         }
 865
 866         if (rc && rc != -EBUSY)
 867                 close_quota_files(oqctl, qinfo);
 868
 869         RETURN(rc);
 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 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 886 {
 887         struct mds_obd *mds = &obd->u.mds;
 888         struct obd_device_target *obt = &obd->u.obt;
 889         struct lvfs_run_ctxt saved;
 890         int rc;
 891         ENTRY;
 892
 893         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 894                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 895                 atomic_inc(&obt->obt_quotachecking);
 896                 RETURN(-EBUSY);
 897         }
 898
 899         down(&mds->mds_qonoff_sem);
 900         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 901         rc = mds_admin_quota_on(obd, oqctl);
 902         if (rc)
 903                 GOTO(out, rc);
 904
 905         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 906         if (!rc)
 907                 obt->obt_qctxt.lqc_flags |= UGQUOTA2LQC(oqctl->qc_type);
 908         else
 909                 GOTO(out, rc);
 910
 911         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 912
 913 out:
 914         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 915         up(&mds->mds_qonoff_sem);
 916         atomic_inc(&obt->obt_quotachecking);
 917         RETURN(rc);
 918 }
 919
 920 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 921 {
 922         struct mds_obd *mds = &obd->u.mds;
 923         struct obd_device_target *obt = &obd->u.obt;
 924         struct lvfs_run_ctxt saved;
 925         int rc, rc2;
 926         ENTRY;
 927
 928         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 929                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 930                 atomic_inc(&obt->obt_quotachecking);
 931                 RETURN(-EBUSY);
 932         }
 933
 934         down(&mds->mds_qonoff_sem);
 935         /* close admin quota files */
 936         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 937         mds_admin_quota_off(obd, oqctl);
 938
 939         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 940         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 941         if (!rc2)
 942                 obt->obt_qctxt.lqc_flags &= ~UGQUOTA2LQC(oqctl->qc_type);
 943
 944         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 945         up(&mds->mds_qonoff_sem);
 946         atomic_inc(&obt->obt_quotachecking);
 947
 948         RETURN(rc ?: rc2);
 949 }
 950
 951 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 952 {
 953         struct mds_obd *mds = &obd->u.mds;
 954         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 955         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 956         int rc;
 957         ENTRY;
 958
 959         down(&mds->mds_qonoff_sem);
 960         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 961                 rc = -ESRCH;
 962                 goto out;
 963         }
 964
 965         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 966         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 967         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 968
 969         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 970
 971 out:
 972         up(&mds->mds_qonoff_sem);
 973         RETURN(rc);
 974 }
 975
 976 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 977 {
 978         struct mds_obd *mds = &obd->u.mds;
 979         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 980         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 981         int rc = 0;
 982         ENTRY;
 983
 984         down(&mds->mds_qonoff_sem);
 985         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 986                 rc = -ESRCH;
 987                 goto out;
 988         }
 989
 990         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
 991         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
 992         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
 993
 994 out:
 995         up(&mds->mds_qonoff_sem);
 996         RETURN(rc);
 997 }
 998
 999 int dquot_create_oqaq(struct lustre_quota_ctxt *qctxt,
1000                       struct lustre_dquot *dquot, __u32 ost_num, __u32 mdt_num,
1001                       int type, struct quota_adjust_qunit *oqaq)
1002 {
1003         __u64 bunit_curr_o, iunit_curr_o;
1004         unsigned long shrink_qunit_limit = qctxt->lqc_cqs_boundary_factor;
1005         unsigned long cqs_factor = qctxt->lqc_cqs_qs_factor;
1006         __u64 blimit = dquot->dq_dqb.dqb_bhardlimit ?
1007                 dquot->dq_dqb.dqb_bhardlimit : dquot->dq_dqb.dqb_bsoftlimit;
1008         __u64 ilimit = dquot->dq_dqb.dqb_ihardlimit ?
1009                 dquot->dq_dqb.dqb_ihardlimit : dquot->dq_dqb.dqb_isoftlimit;
1010         int rc = 0;
1011         ENTRY;
1012
1013         if (!dquot || !oqaq)
1014                 RETURN(-EINVAL);
1015         LASSERT_SEM_LOCKED(&dquot->dq_sem);
1016         LASSERT(oqaq->qaq_iunit_sz);
1017         LASSERT(oqaq->qaq_bunit_sz);
1018
1019         /* don't change qunit size */
1020         if (!qctxt->lqc_switch_qs)
1021                 RETURN(rc);
1022
1023         bunit_curr_o = oqaq->qaq_bunit_sz;
1024         iunit_curr_o = oqaq->qaq_iunit_sz;
1025
1026         if (dquot->dq_type == GRPQUOTA)
1027                 QAQ_SET_GRP(oqaq);
1028
1029         if ((type & LQUOTA_FLAGS_ADJBLK) && blimit) {
1030                 __u64 b_limitation =
1031                         oqaq->qaq_bunit_sz * ost_num * shrink_qunit_limit;
1032                 /* enlarge block qunit size */
1033                 while (blimit >
1034                        QUSG(dquot->dq_dqb.dqb_curspace + 2 * b_limitation, 1)) {
1035                         oqaq->qaq_bunit_sz =
1036                                 QUSG(oqaq->qaq_bunit_sz * cqs_factor, 1)
1037                                 << QUOTABLOCK_BITS;
1038                         b_limitation = oqaq->qaq_bunit_sz * ost_num *
1039                                 shrink_qunit_limit;
1040                 }
1041
1042                 if (oqaq->qaq_bunit_sz > qctxt->lqc_bunit_sz)
1043                         oqaq->qaq_bunit_sz = qctxt->lqc_bunit_sz;
1044
1045                 /* shrink block qunit size */
1046                 while (blimit <
1047                        QUSG(dquot->dq_dqb.dqb_curspace + b_limitation, 1)) {
1048                         do_div(oqaq->qaq_bunit_sz , cqs_factor);
1049                         oqaq->qaq_bunit_sz = QUSG(oqaq->qaq_bunit_sz, 1) <<
1050                                 QUOTABLOCK_BITS;
1051                         b_limitation = oqaq->qaq_bunit_sz * ost_num *
1052                                 shrink_qunit_limit;
1053                         if (oqaq->qaq_bunit_sz <  qctxt->lqc_cqs_least_bunit)
1054                                 break;
1055                 }
1056
1057                 if (oqaq->qaq_bunit_sz < qctxt->lqc_cqs_least_bunit)
1058                         oqaq->qaq_bunit_sz = qctxt->lqc_cqs_least_bunit;
1059
1060                 if (bunit_curr_o != oqaq->qaq_bunit_sz)
1061                         QAQ_SET_ADJBLK(oqaq);
1062
1063         }
1064
1065         if ((type & LQUOTA_FLAGS_ADJINO) && ilimit) {
1066                 __u64 i_limitation =
1067                         oqaq->qaq_iunit_sz * mdt_num * shrink_qunit_limit;
1068                 /* enlarge file qunit size */
1069                 while (ilimit > dquot->dq_dqb.dqb_curinodes
1070                        + 2 * i_limitation) {
1071                         oqaq->qaq_iunit_sz = oqaq->qaq_iunit_sz * cqs_factor;
1072                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1073                                 shrink_qunit_limit;
1074                 }
1075
1076                 if (oqaq->qaq_iunit_sz > qctxt->lqc_iunit_sz)
1077                         oqaq->qaq_iunit_sz = qctxt->lqc_iunit_sz;
1078
1079                 /* shrink file qunit size */
1080                 while (ilimit < dquot->dq_dqb.dqb_curinodes
1081                        + i_limitation) {
1082                         do_div(oqaq->qaq_iunit_sz, cqs_factor);
1083                         i_limitation = oqaq->qaq_iunit_sz * mdt_num *
1084                                        shrink_qunit_limit;
1085                         if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1086                                 break;
1087                 }
1088
1089                 if (oqaq->qaq_iunit_sz < qctxt->lqc_cqs_least_iunit)
1090                         oqaq->qaq_iunit_sz = qctxt->lqc_cqs_least_iunit;
1091
1092                 if (iunit_curr_o != oqaq->qaq_iunit_sz)
1093                         QAQ_SET_ADJINO(oqaq);
1094
1095         }
1096
1097         if (!dquot->dq_dqb.dqb_bhardlimit && !dquot->dq_dqb.dqb_bsoftlimit &&
1098             !dquot->dq_dqb.dqb_ihardlimit && !dquot->dq_dqb.dqb_isoftlimit) {
1099                 oqaq->qaq_bunit_sz = 0;
1100                 oqaq->qaq_iunit_sz = 0;
1101                 QAQ_SET_ADJBLK(oqaq);
1102                 QAQ_SET_ADJINO(oqaq);
1103         }
1104
1105         QAQ_DEBUG(oqaq, "the oqaq computed\n");
1106
1107         RETURN(rc);
1108 }
1109
1110 static int mds_init_slave_ilimits(struct obd_device *obd,
1111                                   struct obd_quotactl *oqctl, int set,
1112                                   struct quota_adjust_qunit *oqaq)
1113 {
1114         /* XXX: for file limits only adjust local now */
1115         struct obd_device_target *obt = &obd->u.obt;
1116         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1117         unsigned int uid = 0, gid = 0;
1118         struct obd_quotactl *ioqc = NULL;
1119         int flag;
1120         int rc;
1121         ENTRY;
1122
1123         /* if we are going to set zero limit, needn't init slaves */
1124         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
1125             !set)
1126                 RETURN(0);
1127
1128         OBD_ALLOC_PTR(ioqc);
1129         if (!ioqc)
1130                 RETURN(-ENOMEM);
1131
1132         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
1133                oqctl->qc_dqblk.dqb_isoftlimit || !set;
1134         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1135         ioqc->qc_id = oqctl->qc_id;
1136         ioqc->qc_type = oqctl->qc_type;
1137         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
1138         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
1139
1140         if (QAQ_IS_ADJINO(oqaq)) {
1141                 /* adjust the mds slave's inode qunit size */
1142                 rc = quota_adjust_slave_lqs(oqaq, qctxt);
1143                 if (rc < 0)
1144                         CDEBUG(D_ERROR, "adjust mds slave's inode qunit size \
1145                                failed! (rc:%d)\n", rc);
1146         }
1147
1148         /* set local limit to MIN_QLIMIT */
1149         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1150         if (rc)
1151                 GOTO(out, rc);
1152
1153         /* trigger local qunit pre-acquire */
1154         if (oqctl->qc_type == USRQUOTA)
1155                 uid = oqctl->qc_id;
1156         else
1157                 gid = oqctl->qc_id;
1158
1159         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 0, 0,
1160                                 NULL);
1161         if (rc == -EDQUOT || rc == -EBUSY) {
1162                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1163                 rc = 0;
1164         }
1165         if (rc) {
1166                 CDEBUG(D_QUOTA,"error mds adjust local file quota! (rc:%d)\n",
1167                        rc);
1168                 GOTO(out, rc);
1169         }
1170         /* FIXME initialize all slaves in CMD */
1171         EXIT;
1172 out:
1173         if (ioqc)
1174                 OBD_FREE_PTR(ioqc);
1175         return rc;
1176 }
1177
1178 static int mds_init_slave_blimits(struct obd_device *obd,
1179                                   struct obd_quotactl *oqctl, int set,
1180                                   struct quota_adjust_qunit *oqaq)
1181 {
1182         struct obd_device_target *obt = &obd->u.obt;
1183         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
1184         struct mds_obd *mds = &obd->u.mds;
1185         struct obd_quotactl *ioqc;
1186         unsigned int uid = 0, gid = 0;
1187         int rc, rc1 = 0;
1188         int flag;
1189         ENTRY;
1190
1191         /* if we are going to set zero limit, needn't init slaves */
1192         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
1193             !set)
1194                 RETURN(0);
1195
1196         OBD_ALLOC_PTR(ioqc);
1197         if (!ioqc)
1198                 RETURN(-ENOMEM);
1199
1200         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
1201                oqctl->qc_dqblk.dqb_bsoftlimit || !set;
1202         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
1203         ioqc->qc_id = oqctl->qc_id;
1204         ioqc->qc_type = oqctl->qc_type;
1205         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
1206         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
1207         if (QAQ_IS_ADJBLK(oqaq)) {
1208                 /* adjust the mds slave's block qunit size */
1209                 rc1 = quota_adjust_slave_lqs(oqaq, qctxt);
1210                 if (rc1 < 0)
1211                         CERROR("adjust mds slave's block qunit size failed!"
1212                                "(rc:%d)\n", rc1);
1213         }
1214
1215         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
1216         if (rc)
1217                 GOTO(out, rc);
1218
1219         /* trigger local qunit pre-acquire */
1220         if (oqctl->qc_type == USRQUOTA)
1221                 uid = oqctl->qc_id;
1222         else
1223                 gid = oqctl->qc_id;
1224
1225         /* initialize all slave's limit */
1226         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
1227
1228         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 1, 0,
1229                                 NULL);
1230         if (rc == -EDQUOT || rc == -EBUSY) {
1231                 CDEBUG(D_QUOTA, "rc: %d.\n", rc);
1232                 rc = 0;
1233         }
1234         if (rc) {
1235                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
1236                 GOTO(out, rc);
1237         }
1238
1239         /* adjust all slave's qunit size when setting quota
1240          * this is will create a lqs for every ost, which will present
1241          * certain uid/gid is set quota or not */
1242         QAQ_SET_ADJBLK(oqaq);
1243         rc = obd_quota_adjust_qunit(mds->mds_osc_exp, oqaq, qctxt);
1244
1245         EXIT;
1246 out:
1247         OBD_FREE_PTR(ioqc);
1248         return rc;
1249 }
1250
1251 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1252 {
1253         struct mds_obd *mds = &obd->u.mds;
1254         struct lustre_quota_ctxt *qctxt = &mds->mds_obt.obt_qctxt;
1255         struct obd_device *lov_obd = class_exp2obd(mds->mds_osc_exp);
1256         struct lov_obd *lov = &lov_obd->u.lov;
1257         struct quota_adjust_qunit *oqaq = NULL;
1258         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1259         __u64 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
1260         time_t btime, itime;
1261         struct lustre_dquot *dquot;
1262         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1263         /* orig_set means if quota was set before; now_set means we are
1264          * setting/cancelling quota */
1265         int orig_set, now_set;
1266         int rc, rc2 = 0, flag = 0;
1267         ENTRY;
1268
1269         OBD_ALLOC_PTR(oqaq);
1270         if (!oqaq)
1271                 RETURN(-ENOMEM);
1272         down(&mds->mds_qonoff_sem);
1273         init_oqaq(oqaq, qctxt, oqctl->qc_id, oqctl->qc_type);
1274
1275         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1276                 GOTO(out_sem, rc = -ESRCH);
1277
1278         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1279         if (IS_ERR(dquot))
1280                 GOTO(out_sem, rc = PTR_ERR(dquot));
1281         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
1282         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
1283
1284         down(&dquot->dq_sem);
1285
1286         if (dquot->dq_status) {
1287                 up(&dquot->dq_sem);
1288                 lustre_dqput(dquot);
1289                 GOTO(out_sem, rc = -EBUSY);
1290         }
1291         dquot->dq_status |= DQ_STATUS_SET;
1292
1293         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1294         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1295         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1296         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1297         btime = dquot->dq_dqb.dqb_btime;
1298         itime = dquot->dq_dqb.dqb_itime;
1299
1300         if (dqblk->dqb_valid & QIF_BTIME)
1301                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
1302         if (dqblk->dqb_valid & QIF_ITIME)
1303                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
1304
1305         if (dqblk->dqb_valid & QIF_BLIMITS) {
1306                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
1307                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
1308                 /* clear usage (limit pool) */
1309                 if (!dquot->dq_dqb.dqb_bhardlimit &&
1310                     !dquot->dq_dqb.dqb_bsoftlimit)
1311                         dquot->dq_dqb.dqb_curspace = 0;
1312
1313                 /* clear grace time */
1314                 if (!dqblk->dqb_bsoftlimit ||
1315                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
1316                         dquot->dq_dqb.dqb_btime = 0;
1317                 /* set grace only if user hasn't provided his own */
1318                 else if (!(dqblk->dqb_valid & QIF_BTIME))
1319                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
1320                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
1321
1322                 flag |= LQUOTA_FLAGS_ADJBLK;
1323         }
1324
1325         if (dqblk->dqb_valid & QIF_ILIMITS) {
1326                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
1327                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
1328                 /* clear usage (limit pool) */
1329                 if (!dquot->dq_dqb.dqb_ihardlimit &&
1330                     !dquot->dq_dqb.dqb_isoftlimit)
1331                         dquot->dq_dqb.dqb_curinodes = 0;
1332
1333                 if (!dqblk->dqb_isoftlimit ||
1334                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
1335                         dquot->dq_dqb.dqb_itime = 0;
1336                 else if (!(dqblk->dqb_valid & QIF_ITIME))
1337                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
1338                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
1339
1340                 flag |= LQUOTA_FLAGS_ADJINO;
1341         }
1342         QAQ_DEBUG(oqaq, "before dquot_create_oqaq\n");
1343         rc = dquot_create_oqaq(qctxt, dquot, lov->desc.ld_tgt_count, 1,
1344                                flag, oqaq);
1345         QAQ_DEBUG(oqaq, "after dquot_create_oqaq\n");
1346         if (rc < 0)
1347                 CDEBUG(D_QUOTA, "adjust qunit size failed! (rc:%d)\n", rc);
1348
1349
1350         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1351
1352         up(&dquot->dq_sem);
1353
1354         if (rc) {
1355                 CERROR("set limit failed! (rc:%d)\n", rc);
1356                 goto out;
1357         }
1358
1359         up(&mds->mds_qonoff_sem);
1360         orig_set = ihardlimit || isoftlimit;
1361         now_set  = dqblk->dqb_ihardlimit || dqblk->dqb_isoftlimit;
1362         if (dqblk->dqb_valid & QIF_ILIMITS && orig_set != now_set) {
1363                 down(&dquot->dq_sem);
1364                 dquot->dq_dqb.dqb_curinodes = 0;
1365                 up(&dquot->dq_sem);
1366                 rc = mds_init_slave_ilimits(obd, oqctl, orig_set, oqaq);
1367                 if (rc) {
1368                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
1369                         goto revoke_out;
1370                 }
1371         }
1372
1373         orig_set = bhardlimit || bsoftlimit;
1374         now_set  = dqblk->dqb_bhardlimit || dqblk->dqb_bsoftlimit;
1375         if (dqblk->dqb_valid & QIF_BLIMITS && orig_set != now_set) {
1376                 down(&dquot->dq_sem);
1377                 dquot->dq_dqb.dqb_curspace = 0;
1378                 up(&dquot->dq_sem);
1379                 rc = mds_init_slave_blimits(obd, oqctl, orig_set, oqaq);
1380                 if (rc) {
1381                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
1382                         goto revoke_out;
1383                 }
1384         }
1385
1386 revoke_out:
1387         down(&mds->mds_qonoff_sem);
1388         down(&dquot->dq_sem);
1389         if (rc) {
1390                 /* cancel previous setting */
1391                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
1392                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
1393                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
1394                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
1395                 dquot->dq_dqb.dqb_btime = btime;
1396                 dquot->dq_dqb.dqb_itime = itime;
1397         }
1398         rc2 = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1399         up(&dquot->dq_sem);
1400
1401 out:
1402         down(&dquot->dq_sem);
1403         dquot->dq_status &= ~DQ_STATUS_SET;
1404         up(&dquot->dq_sem);
1405         lustre_dqput(dquot);
1406         EXIT;
1407 out_sem:
1408         up(&mds->mds_qonoff_sem);
1409
1410         if (oqaq)
1411                 OBD_FREE_PTR(oqaq);
1412
1413         return rc ? rc : rc2;
1414 }
1415
1416 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
1417 {
1418         struct obd_quotactl *soqc;
1419         struct lvfs_run_ctxt saved;
1420         int rc, rc1;
1421         ENTRY;
1422
1423         OBD_ALLOC_PTR(soqc);
1424         if (!soqc)
1425                 RETURN(-ENOMEM);
1426
1427         soqc->qc_cmd = Q_GETOQUOTA;
1428         soqc->qc_id = oqctl->qc_id;
1429         soqc->qc_type = oqctl->qc_type;
1430
1431         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
1432
1433         oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
1434
1435         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1436         soqc->qc_dqblk.dqb_curspace = 0;
1437         rc1 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
1438         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1439
1440         oqctl->qc_dqblk.dqb_curinodes += soqc->qc_dqblk.dqb_curinodes;
1441         if (!rc1)
1442                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1443         oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
1444         if (!rc && !rc1)
1445                 oqctl->qc_dqblk.dqb_valid |= QIF_USAGE;
1446
1447         OBD_FREE_PTR(soqc);
1448
1449         if (!rc)
1450                 rc = rc1;
1451         RETURN(rc);
1452 }
1453
1454 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
1455 {
1456         struct mds_obd *mds = &obd->u.mds;
1457         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1458         struct lustre_dquot *dquot;
1459         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
1460         int rc;
1461         ENTRY;
1462
1463         down(&mds->mds_qonoff_sem);
1464         dqblk->dqb_valid = 0;
1465         if (qinfo->qi_files[oqctl->qc_type] == NULL)
1466                 GOTO(out, rc = -ESRCH);
1467
1468         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
1469         if (IS_ERR(dquot))
1470                 GOTO(out, rc = PTR_ERR(dquot));
1471
1472         down(&dquot->dq_sem);
1473         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
1474         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
1475         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
1476         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
1477         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
1478         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
1479         dqblk->dqb_valid |= QIF_LIMITS | QIF_TIMES;
1480         up(&dquot->dq_sem);
1481
1482         lustre_dqput(dquot);
1483
1484         /* the usages in admin quota file is inaccurate */
1485         dqblk->dqb_curinodes = 0;
1486         dqblk->dqb_curspace = 0;
1487         rc = mds_get_space(obd, oqctl);
1488         EXIT;
1489 out:
1490         up(&mds->mds_qonoff_sem);
1491         return rc;
1492 }
1493
1494 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
1495 {
1496         struct lvfs_run_ctxt saved;
1497         int rc;
1498         ENTRY;
1499
1500         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1501         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
1502         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1503
1504         RETURN(rc);
1505 }
1506
1507
1508 /* FIXME we only recovery block limit by now, need recovery inode
1509  * limits also after CMD involved in */
1510 static int
1511 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
1512 {
1513         struct mds_obd *mds = &obd->u.mds;
1514         struct lustre_quota_info *qinfo= &obd->u.mds.mds_quota_info;
1515         struct lustre_dquot *dquot;
1516         struct obd_quotactl *qctl;
1517         __u64 total_limits = 0;
1518         int rc;
1519         ENTRY;
1520
1521         OBD_ALLOC_PTR(qctl);
1522         if (qctl == NULL)
1523                 RETURN(-ENOMEM);
1524
1525         dquot = lustre_dqget(obd, qinfo, id, type);
1526         if (IS_ERR(dquot)) {
1527                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
1528                 OBD_FREE_PTR(qctl);
1529                 RETURN(PTR_ERR(dquot));
1530         }
1531
1532         down(&dquot->dq_sem);
1533
1534         /* don't recovery the dquot without limits or under setting */
1535         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
1536             dquot->dq_status)
1537                 GOTO(skip, rc = 0);
1538         dquot->dq_status |= DQ_STATUS_RECOVERY;
1539
1540         up(&dquot->dq_sem);
1541
1542         /* get real bhardlimit from all slaves. */
1543         qctl->qc_cmd = Q_GETOQUOTA;
1544         qctl->qc_type = type;
1545         qctl->qc_id = id;
1546         qctl->qc_stat = QUOTA_RECOVERING;
1547         rc = obd_quotactl(obd->u.mds.mds_osc_exp, qctl);
1548         if (rc)
1549                 GOTO(out, rc);
1550         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1551
1552         /* get real bhardlimit from master */
1553         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1554         if (rc)
1555                 GOTO(out, rc);
1556         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1557
1558         /* amend the usage of the administrative quotafile */
1559         down(&mds->mds_qonoff_sem);
1560         down(&dquot->dq_sem);
1561
1562         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1563
1564         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1565         if (rc)
1566                 CERROR("write dquot failed! (rc:%d)\n", rc);
1567
1568         up(&dquot->dq_sem);
1569         up(&mds->mds_qonoff_sem);
1570         EXIT;
1571 out:
1572         down(&dquot->dq_sem);
1573         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1574 skip:
1575         up(&dquot->dq_sem);
1576
1577         lustre_dqput(dquot);
1578         OBD_FREE_PTR(qctl);
1579         return rc;
1580 }
1581
1582 struct qmaster_recov_thread_data {
1583         struct obd_device *obd;
1584         struct completion comp;
1585 };
1586
1587 static int qmaster_recovery_main(void *arg)
1588 {
1589         struct qmaster_recov_thread_data *data = arg;
1590         struct obd_device *obd = data->obd;
1591         int rc = 0;
1592         unsigned short type;
1593         ENTRY;
1594
1595         ptlrpc_daemonize("qmaster_recovd");
1596
1597         complete(&data->comp);
1598
1599         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1600                 struct mds_obd *mds = &obd->u.mds;
1601                 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1602                 struct list_head id_list;
1603                 struct dquot_id *dqid, *tmp;
1604
1605                 down(&mds->mds_qonoff_sem);
1606                 if (qinfo->qi_files[type] == NULL) {
1607                         up(&mds->mds_qonoff_sem);
1608                         continue;
1609                 }
1610                 INIT_LIST_HEAD(&id_list);
1611                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1612                                  &id_list);
1613                 up(&mds->mds_qonoff_sem);
1614
1615                 if (rc)
1616                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1617
1618                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1619                         list_del_init(&dqid->di_link);
1620                         if (rc)
1621                                 goto free;
1622
1623                         rc = dquot_recovery(obd, dqid->di_id, type);
1624                         if (rc)
1625                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1626                                        " rc:%d)\n", dqid->di_id, type, rc);
1627 free:
1628                         kfree(dqid);
1629                 }
1630         }
1631         RETURN(rc);
1632 }
1633
1634 int mds_quota_recovery(struct obd_device *obd)
1635 {
1636         struct mds_obd *mds = &obd->u.mds;
1637         struct qmaster_recov_thread_data data;
1638         int rc = 0;
1639         ENTRY;
1640
1641         mutex_down(&obd->obd_dev_sem);
1642         if (mds->mds_lov_desc.ld_active_tgt_count != mds->mds_lov_objid_count) {
1643                 CWARN("Only %u/%u OSTs are active, abort quota recovery\n",
1644                       mds->mds_lov_desc.ld_active_tgt_count,
1645                       mds->mds_lov_objid_count);
1646                 mutex_up(&obd->obd_dev_sem);
1647                 RETURN(rc);
1648         }
1649         mutex_up(&obd->obd_dev_sem);
1650
1651         data.obd = obd;
1652         init_completion(&data.comp);
1653
1654         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1655         if (rc < 0)
1656                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1657
1658         wait_for_completion(&data.comp);
1659         RETURN(rc);
1660 }
1661
1662 #endif /* HAVE_QUOTA_SUPPORT */