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 <obd_class.h>
  31 #include <lustre_quota.h>
  32 #include <lustre_fsfilt.h>
  33 #include <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 cfs_mem_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 = cfs_mem_cache_create("lustre_dquot_cache",
  51                                                    sizeof(struct lustre_dquot),
  52                                                    0, 0);
  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 = cfs_mem_cache_destroy(lustre_dquot_cachep);
  74                 LASSERTF(rc == 0,"couldn't destroy lustre_dquot_cachep slab\n");
  75                 lustre_dquot_cachep = NULL;
  76         }
  77         EXIT;
  78 }
  79
  80 static inline int
  81 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
  82              __attribute__((__const__));
  83
  84 static inline int
  85 dquot_hashfn(struct lustre_quota_info *info, unsigned int id, int type)
  86 {
  87         unsigned long tmp = ((unsigned long)info >> L1_CACHE_SHIFT) ^ id;
  88         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
  89         return tmp;
  90 }
  91
  92 /* caller must hold dquot_hash_lock */
  93 static struct lustre_dquot *find_dquot(int hashent,
  94                                        struct lustre_quota_info *lqi, qid_t id,
  95                                        int type)
  96 {
  97         struct lustre_dquot *dquot;
  98         ENTRY;
  99
 100         LASSERT_SPIN_LOCKED(&dquot_hash_lock);
 101         list_for_each_entry(dquot, &lustre_dquot_hash[hashent], dq_hash) {
 102                 if (dquot->dq_info == lqi &&
 103                     dquot->dq_id == id && dquot->dq_type == type)
 104                         RETURN(dquot);
 105         }
 106         RETURN(NULL);
 107 }
 108
 109 static struct lustre_dquot *alloc_dquot(struct lustre_quota_info *lqi,
 110                                         qid_t id, int type)
 111 {
 112         struct lustre_dquot *dquot = NULL;
 113         ENTRY;
 114
 115         OBD_SLAB_ALLOC(dquot, lustre_dquot_cachep, CFS_ALLOC_IO, sizeof(*dquot));
 116         if (dquot == NULL)
 117                 RETURN(NULL);
 118
 119         INIT_LIST_HEAD(&dquot->dq_hash);
 120         init_mutex_locked(&dquot->dq_sem);
 121         dquot->dq_refcnt = 1;
 122         dquot->dq_info = lqi;
 123         dquot->dq_id = id;
 124         dquot->dq_type = type;
 125         dquot->dq_status = DQ_STATUS_AVAIL;
 126
 127         RETURN(dquot);
 128 }
 129
 130 static void free_dquot(struct lustre_dquot *dquot)
 131 {
 132         OBD_SLAB_FREE(dquot, lustre_dquot_cachep, sizeof(*dquot));
 133 }
 134
 135 static void insert_dquot_nolock(struct lustre_dquot *dquot)
 136 {
 137         struct list_head *head = lustre_dquot_hash +
 138             dquot_hashfn(dquot->dq_info, dquot->dq_id, dquot->dq_type);
 139         LASSERT(list_empty(&dquot->dq_hash));
 140         list_add(&dquot->dq_hash, head);
 141 }
 142
 143 static void remove_dquot_nolock(struct lustre_dquot *dquot)
 144 {
 145         LASSERT(!list_empty(&dquot->dq_hash));
 146         list_del_init(&dquot->dq_hash);
 147 }
 148
 149 static void lustre_dqput(struct lustre_dquot *dquot)
 150 {
 151         ENTRY;
 152         spin_lock(&dquot_hash_lock);
 153         LASSERT(dquot->dq_refcnt);
 154         dquot->dq_refcnt--;
 155         if (!dquot->dq_refcnt) {
 156                 remove_dquot_nolock(dquot);
 157                 free_dquot(dquot);
 158         }
 159         spin_unlock(&dquot_hash_lock);
 160         EXIT;
 161 }
 162
 163 static struct lustre_dquot *lustre_dqget(struct obd_device *obd,
 164                                          struct lustre_quota_info *lqi,
 165                                          qid_t id, int type)
 166 {
 167         unsigned int hashent = dquot_hashfn(lqi, id, type);
 168         struct lustre_dquot *dquot, *empty;
 169         ENTRY;
 170
 171         if ((empty = alloc_dquot(lqi, id, type)) == NULL)
 172                 RETURN(ERR_PTR(-ENOMEM));
 173
 174         spin_lock(&dquot_hash_lock);
 175         if ((dquot = find_dquot(hashent, lqi, id, type)) != NULL) {
 176                 dquot->dq_refcnt++;
 177                 spin_unlock(&dquot_hash_lock);
 178                 free_dquot(empty);
 179         } else {
 180                 int rc;
 181
 182                 dquot = empty;
 183                 insert_dquot_nolock(dquot);
 184                 spin_unlock(&dquot_hash_lock);
 185
 186                 rc = fsfilt_dquot(obd, dquot, QFILE_RD_DQUOT);
 187                 up(&dquot->dq_sem);
 188                 if (rc) {
 189                         CERROR("can't read dquot from admin quotafile! "
 190                                "(rc:%d)\n", rc);
 191                         lustre_dqput(dquot);
 192                         RETURN(ERR_PTR(rc));
 193                 }
 194
 195         }
 196
 197         LASSERT(dquot);
 198         RETURN(dquot);
 199 }
 200
 201 int dqacq_handler(struct obd_device *obd, struct qunit_data *qdata, int opc)
 202 {
 203         struct mds_obd *mds = &obd->u.mds;
 204         struct lustre_quota_info *info = &mds->mds_quota_info;
 205         struct lustre_dquot *dquot = NULL;
 206         __u64 *usage = NULL;
 207         __u32 hlimit = 0, slimit = 0;
 208         __u32 qdata_type = qdata->qd_flags & QUOTA_IS_GRP;
 209         __u32 is_blk = (qdata->qd_flags & QUOTA_IS_BLOCK) >> 1;
 210         time_t *time = NULL;
 211         unsigned int grace = 0;
 212         int rc = 0;
 213         ENTRY;
 214
 215         OBD_FAIL_RETURN(OBD_FAIL_OBD_DQACQ, -EIO);
 216
 217         dquot = lustre_dqget(obd, info, qdata->qd_id, qdata_type);
 218         if (IS_ERR(dquot))
 219                 RETURN(PTR_ERR(dquot));
 220
 221         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 222         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 223
 224         down(&mds->mds_qonoff_sem);
 225         down(&dquot->dq_sem);
 226
 227         if (dquot->dq_status & DQ_STATUS_RECOVERY) {
 228                 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
 229                 GOTO(out, rc = -EBUSY);
 230         }
 231
 232         if (is_blk) {
 233                 grace = info->qi_info[qdata_type].dqi_bgrace;
 234                 usage = &dquot->dq_dqb.dqb_curspace;
 235                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 236                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 237                 time = &dquot->dq_dqb.dqb_btime;
 238         } else {
 239                 grace = info->qi_info[qdata_type].dqi_igrace;
 240                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 241                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 242                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 243                 time = &dquot->dq_dqb.dqb_itime;
 244         }
 245
 246         /* if the quota limit in admin quotafile is zero, we just inform
 247          * slave to clear quota limit with zero qd_count */
 248         if (hlimit == 0 && slimit == 0) {
 249                 qdata->qd_count = 0;
 250                 GOTO(out, rc);
 251         }
 252
 253         switch (opc) {
 254         case QUOTA_DQACQ:
 255                 if (hlimit &&
 256                     QUSG(*usage + qdata->qd_count, is_blk) > hlimit)
 257                         GOTO(out, rc = -EDQUOT);
 258
 259                 if (slimit &&
 260                     QUSG(*usage + qdata->qd_count, is_blk) > slimit) {
 261                         if (*time && cfs_time_current_sec() >= *time)
 262                                 GOTO(out, rc = -EDQUOT);
 263                         else if (!*time)
 264                                 *time = cfs_time_current_sec() + grace;
 265                 }
 266
 267                 *usage += qdata->qd_count;
 268                 break;
 269         case QUOTA_DQREL:
 270                 /* The usage in administrative file might be incorrect before
 271                  * recovery done */
 272                 if (*usage - qdata->qd_count < 0)
 273                         *usage = 0;
 274                 else
 275                         *usage -= qdata->qd_count;
 276
 277                 /* (usage <= soft limit) but not (usage < soft limit) */
 278                 if (!slimit || QUSG(*usage, is_blk) <= slimit)
 279                         *time = 0;
 280                 break;
 281         default:
 282                 LBUG();
 283         }
 284
 285         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 286         EXIT;
 287 out:
 288         up(&dquot->dq_sem);
 289         up(&mds->mds_qonoff_sem);
 290         lustre_dqput(dquot);
 291         return rc;
 292 }
 293
 294 int mds_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 295                      unsigned int qpids[], int rc, int opc)
 296 {
 297         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 298         int rc2 = 0;
 299         ENTRY;
 300
 301         if (rc && rc != -EDQUOT)
 302                 RETURN(0);
 303
 304         switch (opc) {
 305         case FSFILT_OP_RENAME:
 306                 /* acquire/release block quota on owner of original parent */
 307                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[2], qpids[3], 1, 0);
 308                 /* fall-through */
 309         case FSFILT_OP_SETATTR:
 310                 /* acquire/release file quota on original owner */
 311                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 0, 0);
 312                 /* fall-through */
 313         case FSFILT_OP_CREATE:
 314         case FSFILT_OP_UNLINK:
 315                 /* acquire/release file/block quota on owner of child (or current owner) */
 316                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0);
 317                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 318                 /* acquire/release block quota on owner of parent (or original owner) */
 319                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0);
 320                 break;
 321         default:
 322                 LBUG();
 323                 break;
 324         }
 325
 326         if (rc2)
 327                 CERROR("mds adjust qunit failed! (opc:%d rc:%d)\n", opc, rc2);
 328         RETURN(0);
 329 }
 330
 331 int filter_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 332                         unsigned int qpids[], int rc, int opc)
 333 {
 334         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 335         int rc2 = 0;
 336         ENTRY;
 337
 338         if (rc && rc != -EDQUOT)
 339                 RETURN(0);
 340
 341         switch (opc) {
 342         case FSFILT_OP_SETATTR:
 343                 /* acquire/release block quota on original & current owner */
 344                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 345                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0);
 346                 break;
 347         case FSFILT_OP_UNLINK:
 348                 /* release block quota on this owner */
 349         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 350                 /* acquire block quota on this owner */
 351                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 352                 break;
 353         default:
 354                 LBUG();
 355                 break;
 356         }
 357
 358         if (rc || rc2)
 359                 CERROR("filter adjust qunit failed! (opc:%d rc%d)\n",
 360                        opc, rc ?: rc2);
 361         RETURN(0);
 362 }
 363
 364 #define LUSTRE_ADMIN_QUOTAFILES {\
 365         "admin_quotafile.usr",  /* user admin quotafile */\
 366         "admin_quotafile.grp"   /* group admin quotafile */\
 367 }
 368 static const char prefix[] = "OBJECTS/";
 369
 370 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 371 {
 372         struct mds_obd *mds = &obd->u.mds;
 373         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 374         const char *quotafiles[] = LUSTRE_ADMIN_QUOTAFILES;
 375         struct lvfs_run_ctxt saved;
 376         char name[64];
 377         int i, rc = 0;
 378         struct dentry *dparent = mds->mds_objects_dir;
 379         struct inode *iparent = dparent->d_inode;
 380         ENTRY;
 381
 382         LASSERT(iparent);
 383         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 384
 385         down(&mds->mds_qonoff_sem);
 386         for (i = 0; i < MAXQUOTAS; i++) {
 387                 struct dentry *de;
 388                 struct file *fp;
 389
 390                 if (!Q_TYPESET(oqctl, i))
 391                         continue;
 392
 393                 /* quota file has been opened ? */
 394                 if (qinfo->qi_files[i]) {
 395                         CWARN("init %s admin quotafile while quota on.\n",
 396                               i == USRQUOTA ? "user" : "group");
 397                         continue;
 398                 }
 399
 400                 /* lookup quota file */
 401                 rc = 0;
 402                 LOCK_INODE_MUTEX(iparent);
 403                 de = lookup_one_len(quotafiles[i], dparent,
 404                                     strlen(quotafiles[i]));
 405                 UNLOCK_INODE_MUTEX(iparent);
 406                 if (IS_ERR(de) || de->d_inode == NULL ||
 407                     !S_ISREG(de->d_inode->i_mode))
 408                         rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
 409                 if (!IS_ERR(de))
 410                         dput(de);
 411
 412                 if (rc && rc != -ENOENT) {
 413                         CERROR("error lookup quotafile %s! (rc:%d)\n",
 414                                name, rc);
 415                         break;
 416                 } else if (!rc) {
 417                         continue;
 418                 }
 419
 420                 LASSERT(strlen(quotafiles[i]) + sizeof(prefix) <= sizeof(name));
 421                 sprintf(name, "%s%s", prefix, quotafiles[i]);
 422
 423                 LASSERT(rc == -ENOENT);
 424                 /* create quota file */
 425                 fp = filp_open(name, O_CREAT | O_EXCL, 0644);
 426                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 427                         rc = PTR_ERR(fp);
 428                         CERROR("error creating admin quotafile %s (rc:%d)\n",
 429                                name, rc);
 430                         break;
 431                 }
 432
 433                 qinfo->qi_files[i] = fp;
 434                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 435                 filp_close(fp, 0);
 436                 qinfo->qi_files[i] = NULL;
 437
 438                 if (rc) {
 439                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 440                                i == USRQUOTA ? "user" : "group", rc);
 441                         break;
 442                 }
 443         }
 444         up(&mds->mds_qonoff_sem);
 445
 446         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 447         RETURN(rc);
 448 }
 449
 450 static int close_quota_files(struct obd_quotactl *oqctl,
 451                              struct lustre_quota_info *qinfo)
 452 {
 453         int i, rc = 0;
 454         ENTRY;
 455
 456         for (i = 0; i < MAXQUOTAS; i++) {
 457                 if (!Q_TYPESET(oqctl, i))
 458                         continue;
 459                 if (qinfo->qi_files[i] == NULL) {
 460                         rc = -ESRCH;
 461                         continue;
 462                 }
 463                 filp_close(qinfo->qi_files[i], 0);
 464                 qinfo->qi_files[i] = NULL;
 465         }
 466         RETURN(rc);
 467 }
 468
 469 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 470 {
 471         struct mds_obd *mds = &obd->u.mds;
 472         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 473         const char *quotafiles[] = LUSTRE_ADMIN_QUOTAFILES;
 474         char name[64];
 475         int i, rc = 0;
 476         struct inode *iparent = mds->mds_objects_dir->d_inode;
 477         ENTRY;
 478
 479         LASSERT(iparent);
 480
 481         /* open admin quota files and read quotafile info */
 482         for (i = 0; i < MAXQUOTAS; i++) {
 483                 struct file *fp;
 484
 485                 if (!Q_TYPESET(oqctl, i))
 486                         continue;
 487
 488                 LASSERT(strlen(quotafiles[i]) + sizeof(prefix) <= sizeof(name));
 489                 sprintf(name, "%s%s", prefix, quotafiles[i]);
 490
 491                 if (qinfo->qi_files[i] != NULL) {
 492                         rc = -EBUSY;
 493                         break;
 494                 }
 495
 496                 fp = filp_open(name, O_RDWR | O_EXCL, 0644);
 497                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 498                         rc = PTR_ERR(fp);
 499                         CDEBUG(rc == -ENOENT ? D_QUOTA : D_ERROR,
 500                                "open %s failed! (rc:%d)\n", name, rc);
 501                         break;
 502                 }
 503                 qinfo->qi_files[i] = fp;
 504
 505                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 506                 if (rc) {
 507                         CERROR("error read quotainfo of %s! (rc:%d)\n",
 508                                name, rc);
 509                         break;
 510                 }
 511         }
 512
 513         if (rc && rc != -EBUSY)
 514                 close_quota_files(oqctl, qinfo);
 515
 516         RETURN(rc);
 517 }
 518
 519 static int mds_admin_quota_off(struct obd_device *obd,
 520                                struct obd_quotactl *oqctl)
 521 {
 522         struct mds_obd *mds = &obd->u.mds;
 523         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 524         int rc;
 525         ENTRY;
 526
 527         /* close admin quota files */
 528         rc = close_quota_files(oqctl, qinfo);
 529         RETURN(rc);
 530 }
 531
 532 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 533 {
 534         struct mds_obd *mds = &obd->u.mds;
 535         struct obd_device_target *obt = &obd->u.obt;
 536         struct lvfs_run_ctxt saved;
 537         int rc;
 538         ENTRY;
 539
 540         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 541                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 542                 atomic_inc(&obt->obt_quotachecking);
 543                 RETURN(-EBUSY);
 544         }
 545
 546         down(&mds->mds_qonoff_sem);
 547         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 548         rc = mds_admin_quota_on(obd, oqctl);
 549         if (rc)
 550                 goto out;
 551
 552         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 553         if (rc)
 554                 goto out;
 555
 556         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 557         if (!rc)
 558                 obt->obt_qctxt.lqc_status = 1;
 559 out:
 560         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 561         up(&mds->mds_qonoff_sem);
 562         atomic_inc(&obt->obt_quotachecking);
 563         RETURN(rc);
 564 }
 565
 566 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 567 {
 568         struct mds_obd *mds = &obd->u.mds;
 569         struct obd_device_target *obt = &obd->u.obt;
 570         struct lvfs_run_ctxt saved;
 571         int rc, rc2;
 572         ENTRY;
 573
 574         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 575                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 576                 atomic_inc(&obt->obt_quotachecking);
 577                 RETURN(-EBUSY);
 578         }
 579
 580         down(&mds->mds_qonoff_sem);
 581         /* close admin quota files */
 582         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 583         mds_admin_quota_off(obd, oqctl);
 584
 585         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 586         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 587         if (!rc2)
 588                 obt->obt_qctxt.lqc_status = 0;
 589
 590         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 591         up(&mds->mds_qonoff_sem);
 592         atomic_inc(&obt->obt_quotachecking);
 593
 594         RETURN(rc ?: rc2);
 595 }
 596
 597 int mds_set_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;
 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         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 612         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 613         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 614
 615         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 616
 617 out:
 618         up(&mds->mds_qonoff_sem);
 619         RETURN(rc);
 620 }
 621
 622 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 623 {
 624         struct mds_obd *mds = &obd->u.mds;
 625         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 626         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 627         int rc = 0;
 628         ENTRY;
 629
 630         down(&mds->mds_qonoff_sem);
 631         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 632                 rc = -ESRCH;
 633                 goto out;
 634         }
 635
 636         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
 637         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
 638         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
 639
 640 out:
 641         up(&mds->mds_qonoff_sem);
 642         RETURN(rc);
 643 }
 644
 645 static int mds_init_slave_ilimits(struct obd_device *obd,
 646                                   struct obd_quotactl *oqctl, int set)
 647 {
 648         /* XXX: for file limits only adjust local now */
 649         unsigned int uid = 0, gid = 0;
 650         struct obd_quotactl *ioqc = NULL;
 651         int flag;
 652         int rc;
 653         ENTRY;
 654
 655         /* if we are going to set zero limit, needn't init slaves */
 656         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit &&
 657             set)
 658                 RETURN(0);
 659
 660         OBD_ALLOC_PTR(ioqc);
 661         if (!ioqc)
 662                 RETURN(-ENOMEM);
 663
 664         flag = oqctl->qc_dqblk.dqb_ihardlimit ||
 665                oqctl->qc_dqblk.dqb_isoftlimit || set;
 666         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
 667         ioqc->qc_id = oqctl->qc_id;
 668         ioqc->qc_type = oqctl->qc_type;
 669         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
 670         ioqc->qc_dqblk.dqb_ihardlimit = flag ? MIN_QLIMIT : 0;
 671
 672         /* set local limit to MIN_QLIMIT */
 673         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
 674         if (rc)
 675                 GOTO(out, rc);
 676
 677         /* trigger local qunit pre-acquire */
 678         if (oqctl->qc_type == USRQUOTA)
 679                 uid = oqctl->qc_id;
 680         else
 681                 gid = oqctl->qc_id;
 682
 683         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 0, 0);
 684         if (rc) {
 685                 CERROR("error mds adjust local file quota! (rc:%d)\n", rc);
 686                 GOTO(out, rc);
 687         }
 688         /* FIXME initialize all slaves in CMD */
 689         EXIT;
 690 out:
 691         if (ioqc)
 692                 OBD_FREE_PTR(ioqc);
 693         return rc;
 694 }
 695
 696 static int mds_init_slave_blimits(struct obd_device *obd,
 697                                   struct obd_quotactl *oqctl, int set)
 698 {
 699         struct mds_obd *mds = &obd->u.mds;
 700         struct obd_quotactl *ioqc;
 701         unsigned int uid = 0, gid = 0;
 702         int flag;
 703         int rc;
 704         ENTRY;
 705
 706         /* if we are going to set zero limit, needn't init slaves */
 707         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit &&
 708             set)
 709                 RETURN(0);
 710
 711         OBD_ALLOC_PTR(ioqc);
 712         if (!ioqc)
 713                 RETURN(-ENOMEM);
 714
 715         flag = oqctl->qc_dqblk.dqb_bhardlimit ||
 716                oqctl->qc_dqblk.dqb_bsoftlimit || set;
 717         ioqc->qc_cmd = flag ? Q_INITQUOTA : Q_SETQUOTA;
 718         ioqc->qc_id = oqctl->qc_id;
 719         ioqc->qc_type = oqctl->qc_type;
 720         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
 721         ioqc->qc_dqblk.dqb_bhardlimit = flag ? MIN_QLIMIT : 0;
 722
 723         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
 724         if (rc)
 725                 GOTO(out, rc);
 726
 727         /* trigger local qunit pre-acquire */
 728         if (oqctl->qc_type == USRQUOTA)
 729                 uid = oqctl->qc_id;
 730         else
 731                 gid = oqctl->qc_id;
 732
 733         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 1, 0);
 734         if (rc) {
 735                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
 736                 GOTO(out, rc);
 737         }
 738
 739         /* initialize all slave's limit */
 740         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
 741         EXIT;
 742 out:
 743         OBD_FREE_PTR(ioqc);
 744         return rc;
 745 }
 746
 747 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
 748 {
 749         struct mds_obd *mds = &obd->u.mds;
 750         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 751         __u32 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
 752         time_t btime, itime;
 753         struct lustre_dquot *dquot;
 754         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
 755         int set, rc;
 756         ENTRY;
 757
 758         down(&mds->mds_qonoff_sem);
 759         if (qinfo->qi_files[oqctl->qc_type] == NULL)
 760                 GOTO(out_sem, rc = -ESRCH);
 761
 762         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
 763         if (IS_ERR(dquot))
 764                 GOTO(out_sem, rc = PTR_ERR(dquot));
 765         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
 766         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
 767
 768         down(&dquot->dq_sem);
 769
 770         if (dquot->dq_status) {
 771                 up(&dquot->dq_sem);
 772                 lustre_dqput(dquot);
 773                 GOTO(out_sem, rc = -EBUSY);
 774         }
 775         dquot->dq_status |= DQ_STATUS_SET;
 776
 777         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
 778         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
 779         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
 780         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
 781         btime = dquot->dq_dqb.dqb_btime;
 782         itime = dquot->dq_dqb.dqb_itime;
 783
 784         if (dqblk->dqb_valid & QIF_BTIME)
 785                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
 786         if (dqblk->dqb_valid & QIF_ITIME)
 787                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
 788
 789         if (dqblk->dqb_valid & QIF_BLIMITS) {
 790                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
 791                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
 792                 /* clear usage (limit pool) */
 793                 if (!dquot->dq_dqb.dqb_bhardlimit &&
 794                     !dquot->dq_dqb.dqb_bsoftlimit)
 795                         dquot->dq_dqb.dqb_curspace = 0;
 796
 797                 /* clear grace time */
 798                 if (!dqblk->dqb_bsoftlimit ||
 799                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
 800                         dquot->dq_dqb.dqb_btime = 0;
 801                 /* set grace only if user hasn't provided his own */
 802                 else if (!(dqblk->dqb_valid & QIF_BTIME))
 803                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
 804                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
 805         }
 806
 807         if (dqblk->dqb_valid & QIF_ILIMITS) {
 808                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
 809                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
 810                 /* clear usage (limit pool) */
 811                 if (!dquot->dq_dqb.dqb_ihardlimit &&
 812                     !dquot->dq_dqb.dqb_isoftlimit)
 813                         dquot->dq_dqb.dqb_curinodes = 0;
 814
 815                 if (!dqblk->dqb_isoftlimit ||
 816                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
 817                         dquot->dq_dqb.dqb_itime = 0;
 818                 else if (!(dqblk->dqb_valid & QIF_ITIME))
 819                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
 820                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
 821         }
 822
 823         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 824
 825         up(&dquot->dq_sem);
 826
 827         if (rc) {
 828                 CERROR("set limit failed! (rc:%d)\n", rc);
 829                 goto out;
 830         }
 831
 832         up(&mds->mds_qonoff_sem);
 833         if (dqblk->dqb_valid & QIF_ILIMITS) {
 834                 set = !(ihardlimit || isoftlimit);
 835                 rc = mds_init_slave_ilimits(obd, oqctl, set);
 836                 if (rc) {
 837                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
 838                         goto revoke_out;
 839                 }
 840         }
 841
 842         if (dqblk->dqb_valid & QIF_BLIMITS) {
 843                 set = !(bhardlimit || bsoftlimit);
 844                 rc = mds_init_slave_blimits(obd, oqctl, set);
 845                 if (rc) {
 846                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
 847                         goto revoke_out;
 848                 }
 849         }
 850         down(&mds->mds_qonoff_sem);
 851
 852 revoke_out:
 853         if (rc) {
 854                 /* cancel previous setting */
 855                 down(&dquot->dq_sem);
 856                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
 857                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
 858                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
 859                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
 860                 dquot->dq_dqb.dqb_btime = btime;
 861                 dquot->dq_dqb.dqb_itime = itime;
 862                 fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 863                 up(&dquot->dq_sem);
 864         }
 865 out:
 866         down(&dquot->dq_sem);
 867         dquot->dq_status &= ~DQ_STATUS_SET;
 868         up(&dquot->dq_sem);
 869         lustre_dqput(dquot);
 870         EXIT;
 871 out_sem:
 872         up(&mds->mds_qonoff_sem);
 873         return rc;
 874 }
 875
 876 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
 877 {
 878         struct obd_quotactl *soqc;
 879         struct lvfs_run_ctxt saved;
 880         int rc;
 881         ENTRY;
 882
 883         OBD_ALLOC_PTR(soqc);
 884         if (!soqc)
 885                 RETURN(-ENOMEM);
 886
 887         soqc->qc_cmd = Q_GETOQUOTA;
 888         soqc->qc_id = oqctl->qc_id;
 889         soqc->qc_type = oqctl->qc_type;
 890
 891         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
 892         if (rc)
 893                GOTO(out, rc);
 894
 895         oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
 896
 897         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 898         soqc->qc_dqblk.dqb_curspace = 0;
 899         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
 900         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 901
 902         if (rc)
 903                 GOTO(out, rc);
 904
 905         oqctl->qc_dqblk.dqb_curinodes += soqc->qc_dqblk.dqb_curinodes;
 906         oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
 907         EXIT;
 908 out:
 909         OBD_FREE_PTR(soqc);
 910         return rc;
 911 }
 912
 913 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
 914 {
 915         struct mds_obd *mds = &obd->u.mds;
 916         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 917         struct lustre_dquot *dquot;
 918         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
 919         int rc;
 920         ENTRY;
 921
 922         down(&mds->mds_qonoff_sem);
 923         if (qinfo->qi_files[oqctl->qc_type] == NULL)
 924                 GOTO(out, rc = -ESRCH);
 925
 926         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
 927         if (IS_ERR(dquot))
 928                 GOTO(out, rc = PTR_ERR(dquot));
 929
 930         down(&dquot->dq_sem);
 931         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
 932         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
 933         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
 934         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
 935         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
 936         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
 937         up(&dquot->dq_sem);
 938
 939         lustre_dqput(dquot);
 940
 941         /* the usages in admin quota file is inaccurate */
 942         dqblk->dqb_curinodes = 0;
 943         dqblk->dqb_curspace = 0;
 944         rc = mds_get_space(obd, oqctl);
 945         EXIT;
 946 out:
 947         up(&mds->mds_qonoff_sem);
 948         return rc;
 949 }
 950
 951 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
 952 {
 953         struct lvfs_run_ctxt saved;
 954         int rc;
 955         ENTRY;
 956
 957         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 958         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 959         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 960
 961         RETURN(rc);
 962 }
 963
 964
 965 /* FIXME we only recovery block limit by now, need recovery inode
 966  * limits also after CMD involved in */
 967 static int
 968 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
 969 {
 970         struct mds_obd *mds = &obd->u.mds;
 971         struct lustre_quota_info *qinfo= &obd->u.mds.mds_quota_info;
 972         struct lustre_dquot *dquot;
 973         struct obd_quotactl *qctl;
 974         __u64 total_limits = 0;
 975         int rc;
 976         ENTRY;
 977
 978         OBD_ALLOC_PTR(qctl);
 979         if (qctl == NULL)
 980                 RETURN(-ENOMEM);
 981
 982         dquot = lustre_dqget(obd, qinfo, id, type);
 983         if (IS_ERR(dquot)) {
 984                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
 985                 OBD_FREE_PTR(qctl);
 986                 RETURN(PTR_ERR(dquot));
 987         }
 988
 989         down(&dquot->dq_sem);
 990
 991         /* don't recovery the dquot without limits or under setting */
 992         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
 993             dquot->dq_status)
 994                 GOTO(skip, rc = 0);
 995         dquot->dq_status |= DQ_STATUS_RECOVERY;
 996
 997         up(&dquot->dq_sem);
 998
 999         /* get real bhardlimit from all slaves. */
1000         qctl->qc_cmd = Q_GETOQUOTA;
1001         qctl->qc_type = type;
1002         qctl->qc_id = id;
1003         qctl->qc_stat = QUOTA_RECOVERING;
1004         rc = obd_quotactl(obd->u.mds.mds_osc_exp, qctl);
1005         if (rc)
1006                 GOTO(out, rc);
1007         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1008
1009         /* get real bhardlimit from master */
1010         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1011         if (rc)
1012                 GOTO(out, rc);
1013         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1014
1015         /* amend the usage of the administrative quotafile */
1016         down(&mds->mds_qonoff_sem);
1017         down(&dquot->dq_sem);
1018
1019         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1020
1021         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1022         if (rc)
1023                 CERROR("write dquot failed! (rc:%d)\n", rc);
1024
1025         up(&dquot->dq_sem);
1026         up(&mds->mds_qonoff_sem);
1027         EXIT;
1028 out:
1029         down(&dquot->dq_sem);
1030         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1031 skip:
1032         up(&dquot->dq_sem);
1033
1034         lustre_dqput(dquot);
1035         OBD_FREE_PTR(qctl);
1036         return rc;
1037 }
1038
1039 struct qmaster_recov_thread_data {
1040         struct obd_device *obd;
1041         struct completion comp;
1042 };
1043
1044 static int qmaster_recovery_main(void *arg)
1045 {
1046         struct qmaster_recov_thread_data *data = arg;
1047         struct obd_device *obd = data->obd;
1048         int rc = 0;
1049         unsigned short type;
1050         ENTRY;
1051
1052         ptlrpc_daemonize("qmaster_recovd");
1053
1054         complete(&data->comp);
1055
1056         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1057                 struct mds_obd *mds = &obd->u.mds;
1058                 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1059                 struct list_head id_list;
1060                 struct dquot_id *dqid, *tmp;
1061
1062                 down(&mds->mds_qonoff_sem);
1063                 if (qinfo->qi_files[type] == NULL) {
1064                         up(&mds->mds_qonoff_sem);
1065                         continue;
1066                 }
1067                 INIT_LIST_HEAD(&id_list);
1068                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1069                                  &id_list);
1070                 up(&mds->mds_qonoff_sem);
1071
1072                 if (rc)
1073                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1074
1075                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1076                         list_del_init(&dqid->di_link);
1077                         if (rc)
1078                                 goto free;
1079
1080                         rc = dquot_recovery(obd, dqid->di_id, type);
1081                         if (rc)
1082                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1083                                        " rc:%d)\n", dqid->di_id, type, rc);
1084 free:
1085                         kfree(dqid);
1086                 }
1087         }
1088         RETURN(rc);
1089 }
1090
1091 int mds_quota_recovery(struct obd_device *obd)
1092 {
1093         struct lov_obd *lov = &obd->u.mds.mds_osc_obd->u.lov;
1094         struct qmaster_recov_thread_data data;
1095         int rc = 0;
1096         ENTRY;
1097
1098         mutex_down(&lov->lov_lock);
1099         if (lov->desc.ld_tgt_count != lov->desc.ld_active_tgt_count) {
1100                 CWARN("Not all osts are active, abort quota recovery\n");
1101                 mutex_up(&lov->lov_lock);
1102                 RETURN(rc);
1103         }
1104         mutex_up(&lov->lov_lock);
1105
1106         data.obd = obd;
1107         init_completion(&data.comp);
1108
1109         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1110         if (rc < 0)
1111                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1112
1113         wait_for_completion(&data.comp);
1114         RETURN(rc);
1115 }