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  *  lustre/quota/quota_master.c
   5  *  Lustre Quota Master request handler
   6  *
   7  *  Copyright (c) 2001-2005 Cluster File Systems, Inc.
   8  *   Author: Niu YaWei <niu@clusterfs.com>
   9  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   No redistribution or use is permitted outside of Cluster File Systems, Inc.
  13  *
  14  */
  15 #ifndef EXPORT_SYMTAB
  16 # define EXPORT_SYMTAB
  17 #endif
  18
  19 #define DEBUG_SUBSYSTEM S_MDS
  20
  21 #include <linux/version.h>
  22 #include <linux/fs.h>
  23 #include <asm/unistd.h>
  24 #include <linux/slab.h>
  25 #include <linux/quotaops.h>
  26 #include <linux/module.h>
  27 #include <linux/init.h>
  28 #include <linux/quota.h>
  29
  30 #include <linux/obd_class.h>
  31 #include <linux/lustre_quota.h>
  32 #include <linux/lustre_fsfilt.h>
  33 #include <linux/lustre_mds.h>
  34
  35 #include "quota_internal.h"
  36
  37 /* lock ordering:
  38  * mds->mds_qonoff_sem > dquot->dq_sem */
  39 static struct list_head lustre_dquot_hash[NR_DQHASH];
  40 static spinlock_t dquot_hash_lock = SPIN_LOCK_UNLOCKED;
  41
  42 kmem_cache_t *lustre_dquot_cachep;
  43
  44 int lustre_dquot_init(void)
  45 {
  46         int i;
  47         ENTRY;
  48
  49         LASSERT(lustre_dquot_cachep == NULL);
  50         lustre_dquot_cachep = kmem_cache_create("lustre_dquot_cache",
  51                                                 sizeof(struct lustre_dquot),
  52                                                 0, 0, NULL, NULL);
  53         if (!lustre_dquot_cachep)
  54                 return (-ENOMEM);
  55
  56         for (i = 0; i < NR_DQHASH; i++) {
  57                 INIT_LIST_HEAD(lustre_dquot_hash + i);
  58         }
  59         RETURN(0);
  60 }
  61
  62 void lustre_dquot_exit(void)
  63 {
  64         int i;
  65         ENTRY;
  66         /* FIXME cleanup work ?? */
  67
  68         for (i = 0; i < NR_DQHASH; i++) {
  69                 LASSERT(list_empty(lustre_dquot_hash + i));
  70         }
  71         if (lustre_dquot_cachep) {
  72                 int rc;
  73                 rc = kmem_cache_destroy(lustre_dquot_cachep);
  74                 LASSERT(rc == 0);
  75                 lustre_dquot_cachep = NULL;
  76         }
  77         EXIT;
  78 }
  79
  80 static inline int const dquot_hashfn(struct lustre_quota_info *info,
  81                                      unsigned int id, int type)
  82 {
  83         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
  84         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
  85         return tmp;
  86 }
  87
  88 /* caller must hold dquot_hash_lock */
  89 static struct lustre_dquot *find_dquot(int hashent,
  90                                        struct lustre_quota_info *lqi, qid_t id,
  91                                        int type)
  92 {
  93         struct lustre_dquot *dquot;
  94         ENTRY;
  95
  96         LASSERT_SPIN_LOCKED(&dquot_hash_lock);
  97         list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
  98                 if (dquot->dq_info == lqi &&
  99                     dquot->dq_id == id && dquot->dq_type == type)
 100                         RETURN(dquot);
 101         }
 102         RETURN(NULL);
 103 }
 104
 105 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 106                                         qid_t id, int type)
 107 {
 108         struct lustre_dquot *dquot = NULL;
 109         ENTRY;
 110
 111         OBD_SLAB_ALLOC(dquot, lustre_dquot_cachep, SLAB_NOFS, sizeof(*dquot));
 112         if (dquot == NULL)
 113                 RETURN(NULL);
 114
 115         INIT_LIST_HEAD(&dquot->dq_hash);
 116         init_mutex_locked(&dquot->dq_sem);
 117         dquot->dq_refcnt = 1;
 118         dquot->dq_info = lqi;
 119         dquot->dq_id = id;
 120         dquot->dq_type = type;
 121         dquot->dq_status = DQ_STATUS_AVAIL;
 122
 123         RETURN(dquot);
 124 }
 125
 126 static void free_dquot(struct lustre_dquot *dquot)
 127 {
 128         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 129 }
 130
 131 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 132 {
 133         struct list_head *head = lustre_dquot_hash +
 134             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 135         LASSERT(list_empty(&dquot->dq_hash));
 136         list_add(&dquot->dq_hash, head);
 137 }
 138
 139 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 140 {
 141         LASSERT(!list_empty(&dquot->dq_hash));
 142         list_del_init(&dquot->dq_hash);
 143 }
 144
 145 static void lustre_dqput(struct lustre_dquot *dquot)
 146 {
 147         ENTRY;
 148         spin_lock(&dquot_hash_lock);
 149         LASSERT(dquot->dq_refcnt);
 150         dquot->dq_refcnt--;
 151         if (!dquot->dq_refcnt) {
 152                 remove_dquot_nolock(dquot);
 153                 free_dquot(dquot);
 154         }
 155         spin_unlock(&dquot_hash_lock);
 156         EXIT;
 157 }
 158
 159 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 160                                          struct lustre_quota_info *lqi,
 161                                          qid_t id, int type)
 162 {
 163         unsigned int hashent = dquot_hashfn(lqi, id, type);
 164         struct lustre_dquot *dquot, *empty;
 165         ENTRY;
 166
 167         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 168                 RETURN(ERR_PTR(-ENOMEM));
 169
 170         spin_lock(&dquot_hash_lock);
 171         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 172                 dquot->dq_refcnt++;
 173                 spin_unlock(&dquot_hash_lock);
 174                 free_dquot(empty);
 175         } else {
 176                 int rc;
 177
 178                 dquot = empty;
 179                 insert_dquot_nolock(dquot);
 180                 spin_unlock(&dquot_hash_lock);
 181
 182                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
 183                 up(&dquot->dq_sem);
 184                 if (rc) {
 185                         CERROR("can't read dquot from admin quotafile! "
 186                                "(rc:%d)\n", rc);
 187                         lustre_dqput(dquot);
 188                         RETURN(ERR_PTR(rc));
 189                 }
 190
 191         }
 192
 193         LASSERT(dquot);
 194         RETURN(dquot);
 195 }
 196
 197 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
 198 {
 199         struct mds_obd *mds = &obd->u.mds;
 200         struct lustre_quota_info *info = &mds->mds_quota_info;
 201         struct lustre_dquot *dquot = NULL;
 202         __u64 *usage = NULL;
 203         __u32 hlimit = 0, slimit = 0;
 204         time_t *time = NULL;
 205         unsigned int grace = 0;
 206         int rc = 0;
 207         ENTRY;
 208
 209         OBD_FAIL_RETURN(OBD_FAIL_OBD_DQACQ, -EIO);
 210
 211         /* slaves never acquires qunit for user root */
 212         LASSERT(qdata->qd_id || qdata->qd_type == GRPQUOTA);
 213
 214         dquot = lustre_dqget(obd, info, qdata->qd_id, qdata->qd_type);
 215         if (IS_ERR(dquot))
 216                 RETURN(PTR_ERR(dquot));
 217
 218         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 219         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 220
 221         down(&mds->mds_qonoff_sem);
 222         down(&dquot->dq_sem);
 223
 224         if (dquot->dq_status & DQ_STATUS_RECOVERY) {
 225                 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
 226                 GOTO(out, rc = -EBUSY);
 227         }
 228
 229         if (qdata->qd_isblk) {
 230                 grace = info->qi_info[qdata->qd_type].dqi_bgrace;
 231                 usage = &dquot->dq_dqb.dqb_curspace;
 232                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 233                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 234                 time = &dquot->dq_dqb.dqb_btime;
 235         } else {
 236                 grace = info->qi_info[qdata->qd_type].dqi_igrace;
 237                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 238                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 239                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 240                 time = &dquot->dq_dqb.dqb_itime;
 241         }
 242
 243         /* if the quota limit in admin quotafile is zero, we just inform
 244          * slave to clear quota limit with zero qd_count */
 245         if (hlimit == 0 && slimit == 0) {
 246                 qdata->qd_count = 0;
 247                 GOTO(out, rc);
 248         }
 249
 250         switch (opc) {
 251         case QUOTA_DQACQ:
 252                 if (hlimit &&
 253                     QUSG(*usage + qdata->qd_count, qdata->qd_isblk) > hlimit)
 254                         GOTO(out, rc = -EDQUOT);
 255
 256                 if (slimit &&
 257                     QUSG(*usage + qdata->qd_count, qdata->qd_isblk) > slimit) {
 258                         if (*time && cfs_time_current_sec() >= *time)
 259                                 GOTO(out, rc = -EDQUOT);
 260                         else if (!*time)
 261                                 *time = cfs_time_current_sec() + grace;
 262                 }
 263
 264                 *usage += qdata->qd_count;
 265                 break;
 266         case QUOTA_DQREL:
 267                 /* The usage in administrative file might be incorrect before
 268                  * recovery done */
 269                 if (*usage - qdata->qd_count < 0)
 270                         *usage = 0;
 271                 else
 272                         *usage -= qdata->qd_count;
 273
 274                 /* (usage <= soft limit) but not (usage < soft limit) */
 275                 if (!slimit || QUSG(*usage, qdata->qd_isblk) <= slimit)
 276                         *time = 0;
 277                 break;
 278         default:
 279                 LBUG();
 280         }
 281
 282         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 283         EXIT;
 284 out:
 285         up(&dquot->dq_sem);
 286         up(&mds->mds_qonoff_sem);
 287         lustre_dqput(dquot);
 288         return rc;
 289 }
 290
 291 int mds_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 292                      unsigned int qpids[], int rc, int opc)
 293 {
 294         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 295         int rc2 = 0;
 296         ENTRY;
 297
 298         if (rc && rc != -EDQUOT)
 299                 RETURN(0);
 300
 301         switch (opc) {
 302         case FSFILT_OP_RENAME:
 303                 /* acquire/release block quota on owner of original parent */
 304                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[2], qpids[3], 1, 0);
 305                 /* fall-through */
 306         case FSFILT_OP_SETATTR:
 307                 /* acquire/release file quota on original owner */
 308                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 0, 0);
 309                 /* fall-through */
 310         case FSFILT_OP_CREATE:
 311         case FSFILT_OP_UNLINK:
 312                 /* acquire/release file/block quota on owner of child (or current owner) */
 313                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0);
 314                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 315                 /* acquire/release block quota on owner of parent (or original owner) */
 316                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0);
 317                 break;
 318         default:
 319                 LBUG();
 320                 break;
 321         }
 322
 323         if (rc2)
 324                 CERROR("mds adjust qunit failed! (opc:%d rc:%d)\n", opc, rc2);
 325         RETURN(0);
 326 }
 327
 328 int filter_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 329                         unsigned int qpids[], int rc, int opc)
 330 {
 331         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 332         int rc2 = 0;
 333         ENTRY;
 334
 335         if (rc && rc != -EDQUOT)
 336                 RETURN(0);
 337
 338         switch (opc) {
 339         case FSFILT_OP_SETATTR:
 340                 /* acquire/release block quota on original & current owner */
 341                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 342                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0);
 343                 break;
 344         case FSFILT_OP_UNLINK:
 345                 /* release block quota on this owner */
 346         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 347                 /* acquire block quota on this owner */
 348                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 349                 break;
 350         default:
 351                 LBUG();
 352                 break;
 353         }
 354
 355         if (rc || rc2)
 356                 CERROR("filter adjust qunit failed! (opc:%d rc%d)\n",
 357                        opc, rc ?: rc2);
 358         RETURN(0);
 359 }
 360
 361 #define LUSTRE_ADMIN_QUOTAFILES {\
 362         "admin_quotafile.usr",  /* user admin quotafile */\
 363         "admin_quotafile.grp"   /* group admin quotafile */\
 364 }
 365 static const char prefix[] = "OBJECTS/";
 366
 367 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 368 {
 369         struct mds_obd *mds = &obd->u.mds;
 370         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 371         const char *quotafiles[] = LUSTRE_ADMIN_QUOTAFILES;
 372         struct lvfs_run_ctxt saved;
 373         char name[64];
 374         int i, rc = 0;
 375         struct dentry *dparent = mds->mds_objects_dir;
 376         struct inode *iparent = dparent->d_inode;
 377         ENTRY;
 378
 379         LASSERT(iparent);
 380         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 381
 382         down(&mds->mds_qonoff_sem);
 383         for (i = 0; i < MAXQUOTAS; i++) {
 384                 struct dentry *de;
 385                 struct file *fp;
 386
 387                 if (!Q_TYPESET(oqctl, i))
 388                         continue;
 389
 390                 /* quota file has been opened ? */
 391                 if (qinfo->qi_files[i]) {
 392                         CWARN("init %s admin quotafile while quota on.\n",
 393                               i == USRQUOTA ? "user" : "group");
 394                         continue;
 395                 }
 396
 397                 /* lookup quota file */
 398                 rc = 0;
 399                 LOCK_INODE_MUTEX(iparent);
 400                 de = lookup_one_len(quotafiles[i], dparent,
 401                                     strlen(quotafiles[i]));
 402                 UNLOCK_INODE_MUTEX(iparent);
 403                 if (IS_ERR(de) || de->d_inode == NULL ||
 404                     !S_ISREG(de->d_inode->i_mode))
 405                         rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
 406                 if (!IS_ERR(de))
 407                         dput(de);
 408
 409                 if (rc && rc != -ENOENT) {
 410                         CERROR("error lookup quotafile %s! (rc:%d)\n",
 411                                name, rc);
 412                         break;
 413                 } else if (!rc) {
 414                         continue;
 415                 }
 416
 417                 LASSERT(strlen(quotafiles[i]) + sizeof(prefix) <= sizeof(name));
 418                 sprintf(name, "%s%s", prefix, quotafiles[i]);
 419
 420                 LASSERT(rc == -ENOENT);
 421                 /* create quota file */
 422                 fp = filp_open(name, O_CREAT | O_EXCL, 0644);
 423                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 424                         rc = PTR_ERR(fp);
 425                         CERROR("error creating admin quotafile %s (rc:%d)\n",
 426                                name, rc);
 427                         break;
 428                 }
 429
 430                 qinfo->qi_files[i] = fp;
 431                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 432                 filp_close(fp, 0);
 433                 qinfo->qi_files[i] = NULL;
 434
 435                 if (rc) {
 436                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 437                                i == USRQUOTA ? "user" : "group", rc);
 438                         break;
 439                 }
 440         }
 441         up(&mds->mds_qonoff_sem);
 442
 443         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 444         RETURN(rc);
 445 }
 446
 447 static int close_quota_files(struct obd_quotactl *oqctl,
 448                              struct lustre_quota_info *qinfo)
 449 {
 450         int i, rc = 0;
 451         ENTRY;
 452
 453         for (i = 0; i < MAXQUOTAS; i++) {
 454                 if (!Q_TYPESET(oqctl, i))
 455                         continue;
 456                 if (qinfo->qi_files[i] == NULL) {
 457                         rc = -ESRCH;
 458                         continue;
 459                 }
 460                 filp_close(qinfo->qi_files[i], 0);
 461                 qinfo->qi_files[i] = NULL;
 462         }
 463         RETURN(rc);
 464 }
 465
 466 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 467 {
 468         struct mds_obd *mds = &obd->u.mds;
 469         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 470         const char *quotafiles[] = LUSTRE_ADMIN_QUOTAFILES;
 471         char name[64];
 472         int i, rc = 0;
 473         struct inode *iparent = mds->mds_objects_dir->d_inode;
 474         ENTRY;
 475
 476         LASSERT(iparent);
 477
 478         /* open admin quota files and read quotafile info */
 479         for (i = 0; i < MAXQUOTAS; i++) {
 480                 struct file *fp;
 481
 482                 if (!Q_TYPESET(oqctl, i))
 483                         continue;
 484
 485                 LASSERT(strlen(quotafiles[i]) + sizeof(prefix) <= sizeof(name));
 486                 sprintf(name, "%s%s", prefix, quotafiles[i]);
 487
 488                 if (qinfo->qi_files[i] != NULL) {
 489                         rc = -EBUSY;
 490                         break;
 491                 }
 492
 493                 fp = filp_open(name, O_RDWR | O_EXCL, 0644);
 494                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 495                         rc = PTR_ERR(fp);
 496                         CERROR("error open %s! (rc:%d)\n", name, rc);
 497                         break;
 498                 }
 499                 qinfo->qi_files[i] = fp;
 500
 501                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 502                 if (rc) {
 503                         CERROR("error read quotainfo of %s! (rc:%d)\n",
 504                                name, rc);
 505                         break;
 506                 }
 507         }
 508
 509         if (rc && rc != -EBUSY)
 510                 close_quota_files(oqctl, qinfo);
 511
 512         RETURN(rc);
 513 }
 514
 515 static int mds_admin_quota_off(struct obd_device *obd,
 516                                struct obd_quotactl *oqctl)
 517 {
 518         struct mds_obd *mds = &obd->u.mds;
 519         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 520         int rc;
 521         ENTRY;
 522
 523         /* close admin quota files */
 524         rc = close_quota_files(oqctl, qinfo);
 525         RETURN(rc);
 526 }
 527
 528 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 529 {
 530         struct mds_obd *mds = &obd->u.mds;
 531         struct lvfs_run_ctxt saved;
 532         int rc;
 533         ENTRY;
 534
 535         down(&mds->mds_qonoff_sem);
 536         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 537         rc = mds_admin_quota_on(obd, oqctl);
 538         if (rc)
 539                 goto out;
 540
 541         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 542         if (rc)
 543                 goto out;
 544
 545         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 546 out:
 547         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 548         up(&mds->mds_qonoff_sem);
 549         RETURN(rc);
 550 }
 551
 552 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 553 {
 554         struct mds_obd *mds = &obd->u.mds;
 555         struct lvfs_run_ctxt saved;
 556         int rc, rc2;
 557         ENTRY;
 558
 559         down(&mds->mds_qonoff_sem);
 560         /* close admin quota files */
 561         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 562         mds_admin_quota_off(obd, oqctl);
 563
 564         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 565         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 566
 567         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 568         up(&mds->mds_qonoff_sem);
 569         RETURN(rc ?: rc2);
 570 }
 571
 572 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 573 {
 574         struct mds_obd *mds = &obd->u.mds;
 575         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 576         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 577         int rc;
 578         ENTRY;
 579
 580         down(&mds->mds_qonoff_sem);
 581         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 582                 rc = -ESRCH;
 583                 goto out;
 584         }
 585
 586         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 587         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 588         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 589
 590         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 591
 592 out:
 593         up(&mds->mds_qonoff_sem);
 594         RETURN(rc);
 595 }
 596
 597 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 598 {
 599         struct mds_obd *mds = &obd->u.mds;
 600         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 601         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 602         int rc = 0;
 603         ENTRY;
 604
 605         down(&mds->mds_qonoff_sem);
 606         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 607                 rc = -ESRCH;
 608                 goto out;
 609         }
 610
 611         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
 612         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
 613         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
 614
 615 out:
 616         up(&mds->mds_qonoff_sem);
 617         RETURN(rc);
 618 }
 619
 620 static int mds_init_slave_ilimits(struct obd_device *obd,
 621                                   struct obd_quotactl *oqctl, int set)
 622 {
 623         /* XXX: for file limits only adjust local now */
 624         unsigned int uid = 0, gid = 0;
 625         struct obd_quotactl *ioqc = NULL;
 626         int rc;
 627         ENTRY;
 628
 629         /* if we are going to set zero limit, needn't init slaves */
 630         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit)
 631                 RETURN(0);
 632
 633         if (!set)
 634                 goto acquire;
 635
 636         OBD_ALLOC_PTR(ioqc);
 637         if (!ioqc)
 638                 RETURN(-ENOMEM);
 639
 640         ioqc->qc_cmd = Q_INITQUOTA;
 641         ioqc->qc_id = oqctl->qc_id;
 642         ioqc->qc_type = oqctl->qc_type;
 643         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
 644         ioqc->qc_dqblk.dqb_ihardlimit = MIN_QLIMIT;
 645
 646         /* set local limit to MIN_QLIMIT */
 647         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
 648         if (rc)
 649                 GOTO(out, rc);
 650 acquire:
 651         /* trigger local qunit pre-acquire */
 652         if (oqctl->qc_type == USRQUOTA)
 653                 uid = oqctl->qc_id;
 654         else
 655                 gid = oqctl->qc_id;
 656
 657         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 0, 0);
 658         if (rc) {
 659                 CERROR("error mds adjust local file quota! (rc:%d)\n", rc);
 660                 GOTO(out, rc);
 661         }
 662         /* FIXME initialize all slaves in CMD */
 663         EXIT;
 664 out:
 665         if (ioqc)
 666                 OBD_FREE_PTR(ioqc);
 667         return rc;
 668 }
 669
 670 static int mds_init_slave_blimits(struct obd_device *obd,
 671                                   struct obd_quotactl *oqctl, int set)
 672 {
 673         struct mds_obd *mds = &obd->u.mds;
 674         struct obd_quotactl *ioqc;
 675         unsigned int uid = 0, gid = 0;
 676         int rc;
 677         ENTRY;
 678
 679         /* if we are going to set zero limit, needn't init slaves */
 680         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit)
 681                 RETURN(0);
 682
 683         OBD_ALLOC_PTR(ioqc);
 684         if (!ioqc)
 685                 RETURN(-ENOMEM);
 686
 687         ioqc->qc_cmd = Q_INITQUOTA;
 688         ioqc->qc_id = oqctl->qc_id;
 689         ioqc->qc_type = oqctl->qc_type;
 690         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
 691         ioqc->qc_dqblk.dqb_bhardlimit = set ? MIN_QLIMIT : 0;
 692
 693         /* set local limit to MIN_QLIMIT */
 694         if (set) {
 695                 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
 696                 if (rc)
 697                         GOTO(out, rc);
 698         }
 699
 700         /* trigger local qunit pre-acquire */
 701         if (oqctl->qc_type == USRQUOTA)
 702                 uid = oqctl->qc_id;
 703         else
 704                 gid = oqctl->qc_id;
 705
 706         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 1, 0);
 707         if (rc) {
 708                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
 709                 GOTO(out, rc);
 710         }
 711
 712         /* initialize all slave's limit */
 713         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
 714         EXIT;
 715 out:
 716         OBD_FREE_PTR(ioqc);
 717         return rc;
 718 }
 719
 720 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
 721 {
 722         struct mds_obd *mds = &obd->u.mds;
 723         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 724         __u32 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
 725         time_t btime, itime;
 726         struct lustre_dquot *dquot;
 727         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
 728         int set, rc;
 729         ENTRY;
 730
 731         down(&mds->mds_qonoff_sem);
 732         if (qinfo->qi_files[oqctl->qc_type] == NULL)
 733                 GOTO(out_sem, rc = -ESRCH);
 734
 735         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
 736         if (IS_ERR(dquot))
 737                 GOTO(out_sem, rc = PTR_ERR(dquot));
 738         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
 739         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
 740
 741         down(&dquot->dq_sem);
 742
 743         if (dquot->dq_status) {
 744                 up(&dquot->dq_sem);
 745                 lustre_dqput(dquot);
 746                 GOTO(out_sem, rc = -EBUSY);
 747         }
 748         dquot->dq_status |= DQ_STATUS_SET;
 749
 750         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
 751         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
 752         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
 753         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
 754         btime = dquot->dq_dqb.dqb_btime;
 755         itime = dquot->dq_dqb.dqb_itime;
 756
 757         if (dqblk->dqb_valid & QIF_BTIME)
 758                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
 759         if (dqblk->dqb_valid & QIF_ITIME)
 760                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
 761
 762         if (dqblk->dqb_valid & QIF_BLIMITS) {
 763                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
 764                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
 765                 /* clear usage (limit pool) */
 766                 if (!dquot->dq_dqb.dqb_bhardlimit &&
 767                     !dquot->dq_dqb.dqb_bsoftlimit)
 768                         dquot->dq_dqb.dqb_curspace = 0;
 769
 770                 /* clear grace time */
 771                 if (!dqblk->dqb_bsoftlimit ||
 772                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
 773                         dquot->dq_dqb.dqb_btime = 0;
 774                 /* set grace only if user hasn't provided his own */
 775                 else if (!(dqblk->dqb_valid & QIF_BTIME))
 776                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
 777                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
 778         }
 779
 780         if (dqblk->dqb_valid & QIF_ILIMITS) {
 781                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
 782                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
 783                 /* clear usage (limit pool) */
 784                 if (!dquot->dq_dqb.dqb_ihardlimit &&
 785                     !dquot->dq_dqb.dqb_isoftlimit)
 786                         dquot->dq_dqb.dqb_curinodes = 0;
 787
 788                 if (!dqblk->dqb_isoftlimit ||
 789                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
 790                         dquot->dq_dqb.dqb_itime = 0;
 791                 else if (!(dqblk->dqb_valid & QIF_ITIME))
 792                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
 793                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
 794         }
 795
 796         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 797
 798         up(&dquot->dq_sem);
 799
 800         if (rc) {
 801                 CERROR("set limit failed! (rc:%d)\n", rc);
 802                 goto out;
 803         }
 804
 805         up(&mds->mds_qonoff_sem);
 806         if (dqblk->dqb_valid & QIF_ILIMITS) {
 807                 set = !(ihardlimit || isoftlimit);
 808                 rc = mds_init_slave_ilimits(obd, oqctl, set);
 809                 if (rc) {
 810                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
 811                         goto revoke_out;
 812                 }
 813         }
 814
 815         if (dqblk->dqb_valid & QIF_BLIMITS) {
 816                 set = !(bhardlimit || bsoftlimit);
 817                 rc = mds_init_slave_blimits(obd, oqctl, set);
 818                 if (rc) {
 819                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
 820                         goto revoke_out;
 821                 }
 822         }
 823         down(&mds->mds_qonoff_sem);
 824
 825 revoke_out:
 826         if (rc) {
 827                 /* cancel previous setting */
 828                 down(&dquot->dq_sem);
 829                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
 830                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
 831                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
 832                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
 833                 dquot->dq_dqb.dqb_btime = btime;
 834                 dquot->dq_dqb.dqb_itime = itime;
 835                 fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 836                 up(&dquot->dq_sem);
 837         }
 838 out:
 839         down(&dquot->dq_sem);
 840         dquot->dq_status &= ~DQ_STATUS_SET;
 841         up(&dquot->dq_sem);
 842         lustre_dqput(dquot);
 843         EXIT;
 844 out_sem:
 845         up(&mds->mds_qonoff_sem);
 846         return rc;
 847 }
 848
 849 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
 850 {
 851         struct obd_quotactl *soqc;
 852         struct lvfs_run_ctxt saved;
 853         int rc;
 854         ENTRY;
 855
 856         OBD_ALLOC_PTR(soqc);
 857         if (!soqc)
 858                 RETURN(-ENOMEM);
 859
 860         soqc->qc_cmd = Q_GETOQUOTA;
 861         soqc->qc_id = oqctl->qc_id;
 862         soqc->qc_type = oqctl->qc_type;
 863
 864         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
 865         if (rc)
 866                GOTO(out, rc);
 867
 868         oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
 869
 870         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 871         soqc->qc_dqblk.dqb_curspace = 0;
 872         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
 873         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 874
 875         if (rc)
 876                 GOTO(out, rc);
 877
 878         oqctl->qc_dqblk.dqb_curinodes += soqc->qc_dqblk.dqb_curinodes;
 879         oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
 880         EXIT;
 881 out:
 882         OBD_FREE_PTR(soqc);
 883         return rc;
 884 }
 885
 886 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
 887 {
 888         struct mds_obd *mds = &obd->u.mds;
 889         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 890         struct lustre_dquot *dquot;
 891         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
 892         int rc;
 893         ENTRY;
 894
 895         down(&mds->mds_qonoff_sem);
 896         if (qinfo->qi_files[oqctl->qc_type] == NULL)
 897                 GOTO(out, rc = -ESRCH);
 898
 899         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
 900         if (IS_ERR(dquot))
 901                 GOTO(out, rc = PTR_ERR(dquot));
 902
 903         down(&dquot->dq_sem);
 904         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
 905         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
 906         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
 907         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
 908         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
 909         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
 910         up(&dquot->dq_sem);
 911
 912         lustre_dqput(dquot);
 913
 914         /* the usages in admin quota file is inaccurate */
 915         dqblk->dqb_curinodes = 0;
 916         dqblk->dqb_curspace = 0;
 917         rc = mds_get_space(obd, oqctl);
 918         EXIT;
 919 out:
 920         up(&mds->mds_qonoff_sem);
 921         return rc;
 922 }
 923
 924 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
 925 {
 926         struct lvfs_run_ctxt saved;
 927         int rc;
 928         ENTRY;
 929
 930         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 931         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 932         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 933
 934         RETURN(rc);
 935 }
 936
 937
 938 /* FIXME we only recovery block limit by now, need recovery inode
 939  * limits also after CMD involved in */
 940 static int
 941 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
 942 {
 943         struct mds_obd *mds = &obd->u.mds;
 944         struct lustre_quota_info *qinfo= &obd->u.mds.mds_quota_info;
 945         struct lustre_dquot *dquot;
 946         struct obd_quotactl *qctl;
 947         __u64 total_limits = 0;
 948         int rc;
 949         ENTRY;
 950
 951         OBD_ALLOC_PTR(qctl);
 952         if (qctl == NULL)
 953                 RETURN(-ENOMEM);
 954
 955         dquot = lustre_dqget(obd, qinfo, id, type);
 956         if (IS_ERR(dquot)) {
 957                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
 958                 OBD_FREE_PTR(qctl);
 959                 RETURN(PTR_ERR(dquot));
 960         }
 961
 962         down(&dquot->dq_sem);
 963
 964         /* don't recovery the dquot without limits or under setting */
 965         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
 966             dquot->dq_status)
 967                 GOTO(skip, rc = 0);
 968         dquot->dq_status |= DQ_STATUS_RECOVERY;
 969
 970         up(&dquot->dq_sem);
 971
 972         /* get real bhardlimit from all slaves. */
 973         qctl->qc_cmd = Q_GETOQUOTA;
 974         qctl->qc_type = type;
 975         qctl->qc_id = id;
 976         qctl->qc_stat = QUOTA_RECOVERING;
 977         rc = obd_quotactl(obd->u.mds.mds_osc_exp, qctl);
 978         if (rc)
 979                 GOTO(out, rc);
 980         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
 981
 982         /* get real bhardlimit from master */
 983         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
 984         if (rc)
 985                 GOTO(out, rc);
 986         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
 987
 988         /* amend the usage of the administrative quotafile */
 989         down(&mds->mds_qonoff_sem);
 990         down(&dquot->dq_sem);
 991
 992         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
 993
 994         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 995         if (rc)
 996                 CERROR("write dquot failed! (rc:%d)\n", rc);
 997
 998         up(&dquot->dq_sem);
 999         up(&mds->mds_qonoff_sem);
1000         EXIT;
1001 out:
1002         down(&dquot->dq_sem);
1003         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1004 skip:
1005         up(&dquot->dq_sem);
1006
1007         lustre_dqput(dquot);
1008         OBD_FREE_PTR(qctl);
1009         return rc;
1010 }
1011
1012 struct qmaster_recov_thread_data {
1013         struct obd_device *obd;
1014         struct completion comp;
1015 };
1016
1017 static int qmaster_recovery_main(void *arg)
1018 {
1019         struct qmaster_recov_thread_data *data = arg;
1020         struct obd_device *obd = data->obd;
1021         int rc = 0;
1022         unsigned short type;
1023         ENTRY;
1024
1025         ptlrpc_daemonize("qmaster_recovd");
1026
1027         complete(&data->comp);
1028
1029         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1030                 struct mds_obd *mds = &obd->u.mds;
1031                 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1032                 struct list_head id_list;
1033                 struct dquot_id *dqid, *tmp;
1034
1035                 down(&mds->mds_qonoff_sem);
1036                 if (qinfo->qi_files[type] == NULL) {
1037                         up(&mds->mds_qonoff_sem);
1038                         continue;
1039                 }
1040                 INIT_LIST_HEAD(&id_list);
1041                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1042                                  &id_list);
1043                 up(&mds->mds_qonoff_sem);
1044
1045                 if (rc)
1046                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1047
1048                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1049                         list_del_init(&dqid->di_link);
1050                         if (rc)
1051                                 goto free;
1052
1053                         rc = dquot_recovery(obd, dqid->di_id, type);
1054                         if (rc)
1055                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1056                                        " rc:%d)\n", dqid->di_id, type, rc);
1057 free:
1058                         kfree(dqid);
1059                 }
1060         }
1061         RETURN(rc);
1062 }
1063
1064 int mds_quota_recovery(struct obd_device *obd)
1065 {
1066         struct lov_obd *lov = &obd->u.mds.mds_osc_obd->u.lov;
1067         struct qmaster_recov_thread_data data;
1068         int rc = 0;
1069         ENTRY;
1070
1071         down(&lov->lov_lock);
1072         if (lov->desc.ld_tgt_count != lov->desc.ld_active_tgt_count) {
1073                 CWARN("Not all osts are active, abort quota recovery\n");
1074                 up(&lov->lov_lock);
1075                 RETURN(rc);
1076         }
1077         up(&lov->lov_lock);
1078
1079         data.obd = obd;
1080         init_completion(&data.comp);
1081
1082         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1083         if (rc < 0)
1084                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1085
1086         wait_for_completion(&data.comp);
1087         RETURN(rc);
1088 }