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         /* slaves never acquires qunit for user root */
 218         LASSERT(qdata->qd_id || qdata_type);
 219
 220         dquot = lustre_dqget(obd, info, qdata->qd_id, qdata_type);
 221         if (IS_ERR(dquot))
 222                 RETURN(PTR_ERR(dquot));
 223
 224         DQUOT_DEBUG(dquot, "get dquot in dqacq_handler\n");
 225         QINFO_DEBUG(dquot->dq_info, "get dquot in dqadq_handler\n");
 226
 227         down(&mds->mds_qonoff_sem);
 228         down(&dquot->dq_sem);
 229
 230         if (dquot->dq_status & DQ_STATUS_RECOVERY) {
 231                 DQUOT_DEBUG(dquot, "this dquot is under recovering.\n");
 232                 GOTO(out, rc = -EBUSY);
 233         }
 234
 235         if (is_blk) {
 236                 grace = info->qi_info[qdata_type].dqi_bgrace;
 237                 usage = &dquot->dq_dqb.dqb_curspace;
 238                 hlimit = dquot->dq_dqb.dqb_bhardlimit;
 239                 slimit = dquot->dq_dqb.dqb_bsoftlimit;
 240                 time = &dquot->dq_dqb.dqb_btime;
 241         } else {
 242                 grace = info->qi_info[qdata_type].dqi_igrace;
 243                 usage = (__u64 *) & dquot->dq_dqb.dqb_curinodes;
 244                 hlimit = dquot->dq_dqb.dqb_ihardlimit;
 245                 slimit = dquot->dq_dqb.dqb_isoftlimit;
 246                 time = &dquot->dq_dqb.dqb_itime;
 247         }
 248
 249         /* if the quota limit in admin quotafile is zero, we just inform
 250          * slave to clear quota limit with zero qd_count */
 251         if (hlimit == 0 && slimit == 0) {
 252                 qdata->qd_count = 0;
 253                 GOTO(out, rc);
 254         }
 255
 256         switch (opc) {
 257         case QUOTA_DQACQ:
 258                 if (hlimit &&
 259                     QUSG(*usage + qdata->qd_count, is_blk) > hlimit)
 260                         GOTO(out, rc = -EDQUOT);
 261
 262                 if (slimit &&
 263                     QUSG(*usage + qdata->qd_count, is_blk) > slimit) {
 264                         if (*time && cfs_time_current_sec() >= *time)
 265                                 GOTO(out, rc = -EDQUOT);
 266                         else if (!*time)
 267                                 *time = cfs_time_current_sec() + grace;
 268                 }
 269
 270                 *usage += qdata->qd_count;
 271                 break;
 272         case QUOTA_DQREL:
 273                 /* The usage in administrative file might be incorrect before
 274                  * recovery done */
 275                 if (*usage - qdata->qd_count < 0)
 276                         *usage = 0;
 277                 else
 278                         *usage -= qdata->qd_count;
 279
 280                 /* (usage <= soft limit) but not (usage < soft limit) */
 281                 if (!slimit || QUSG(*usage, is_blk) <= slimit)
 282                         *time = 0;
 283                 break;
 284         default:
 285                 LBUG();
 286         }
 287
 288         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 289         EXIT;
 290 out:
 291         up(&dquot->dq_sem);
 292         up(&mds->mds_qonoff_sem);
 293         lustre_dqput(dquot);
 294         return rc;
 295 }
 296
 297 int mds_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 298                      unsigned int qpids[], int rc, int opc)
 299 {
 300         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 301         int rc2 = 0;
 302         ENTRY;
 303
 304         if (rc && rc != -EDQUOT)
 305                 RETURN(0);
 306
 307         switch (opc) {
 308         case FSFILT_OP_RENAME:
 309                 /* acquire/release block quota on owner of original parent */
 310                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[2], qpids[3], 1, 0);
 311                 /* fall-through */
 312         case FSFILT_OP_SETATTR:
 313                 /* acquire/release file quota on original owner */
 314                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 0, 0);
 315                 /* fall-through */
 316         case FSFILT_OP_CREATE:
 317         case FSFILT_OP_UNLINK:
 318                 /* acquire/release file/block quota on owner of child (or current owner) */
 319                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 0, 0);
 320                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 321                 /* acquire/release block quota on owner of parent (or original owner) */
 322                 rc2 |= qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0);
 323                 break;
 324         default:
 325                 LBUG();
 326                 break;
 327         }
 328
 329         if (rc2)
 330                 CERROR("mds adjust qunit failed! (opc:%d rc:%d)\n", opc, rc2);
 331         RETURN(0);
 332 }
 333
 334 int filter_quota_adjust(struct obd_device *obd, unsigned int qcids[],
 335                         unsigned int qpids[], int rc, int opc)
 336 {
 337         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 338         int rc2 = 0;
 339         ENTRY;
 340
 341         if (rc && rc != -EDQUOT && rc != ENOLCK)
 342                 RETURN(0);
 343
 344         switch (opc) {
 345         case FSFILT_OP_SETATTR:
 346                 /* acquire/release block quota on original & current owner */
 347                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 348                 rc2 = qctxt_adjust_qunit(obd, qctxt, qpids[0], qpids[1], 1, 0);
 349                 break;
 350         case FSFILT_OP_UNLINK:
 351                 /* release block quota on this owner */
 352         case FSFILT_OP_CREATE: /* XXX for write operation on obdfilter */
 353                 /* acquire block quota on this owner */
 354                 rc = qctxt_adjust_qunit(obd, qctxt, qcids[0], qcids[1], 1, 0);
 355                 break;
 356         default:
 357                 LBUG();
 358                 break;
 359         }
 360
 361         if (rc || rc2)
 362                 CERROR("filter adjust qunit failed! (opc:%d rc%d)\n",
 363                        opc, rc ?: rc2);
 364         RETURN(0);
 365 }
 366
 367 #define LUSTRE_ADMIN_QUOTAFILES {\
 368         "admin_quotafile.usr",  /* user admin quotafile */\
 369         "admin_quotafile.grp"   /* group admin quotafile */\
 370 }
 371 static const char prefix[] = "OBJECTS/";
 372
 373 int init_admin_quotafiles(struct obd_device *obd, struct obd_quotactl *oqctl)
 374 {
 375         struct mds_obd *mds = &obd->u.mds;
 376         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 377         const char *quotafiles[] = LUSTRE_ADMIN_QUOTAFILES;
 378         struct lvfs_run_ctxt saved;
 379         char name[64];
 380         int i, rc = 0;
 381         struct dentry *dparent = mds->mds_objects_dir;
 382         struct inode *iparent = dparent->d_inode;
 383         ENTRY;
 384
 385         LASSERT(iparent);
 386         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 387
 388         down(&mds->mds_qonoff_sem);
 389         for (i = 0; i < MAXQUOTAS; i++) {
 390                 struct dentry *de;
 391                 struct file *fp;
 392
 393                 if (!Q_TYPESET(oqctl, i))
 394                         continue;
 395
 396                 /* quota file has been opened ? */
 397                 if (qinfo->qi_files[i]) {
 398                         CWARN("init %s admin quotafile while quota on.\n",
 399                               i == USRQUOTA ? "user" : "group");
 400                         continue;
 401                 }
 402
 403                 /* lookup quota file */
 404                 rc = 0;
 405                 LOCK_INODE_MUTEX(iparent);
 406                 de = lookup_one_len(quotafiles[i], dparent,
 407                                     strlen(quotafiles[i]));
 408                 UNLOCK_INODE_MUTEX(iparent);
 409                 if (IS_ERR(de) || de->d_inode == NULL ||
 410                     !S_ISREG(de->d_inode->i_mode))
 411                         rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
 412                 if (!IS_ERR(de))
 413                         dput(de);
 414
 415                 if (rc && rc != -ENOENT) {
 416                         CERROR("error lookup quotafile %s! (rc:%d)\n",
 417                                name, rc);
 418                         break;
 419                 } else if (!rc) {
 420                         continue;
 421                 }
 422
 423                 LASSERT(strlen(quotafiles[i]) + sizeof(prefix) <= sizeof(name));
 424                 sprintf(name, "%s%s", prefix, quotafiles[i]);
 425
 426                 LASSERT(rc == -ENOENT);
 427                 /* create quota file */
 428                 fp = filp_open(name, O_CREAT | O_EXCL, 0644);
 429                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 430                         rc = PTR_ERR(fp);
 431                         CERROR("error creating admin quotafile %s (rc:%d)\n",
 432                                name, rc);
 433                         break;
 434                 }
 435
 436                 qinfo->qi_files[i] = fp;
 437                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_INIT_INFO);
 438                 filp_close(fp, 0);
 439                 qinfo->qi_files[i] = NULL;
 440
 441                 if (rc) {
 442                         CERROR("error init %s admin quotafile! (rc:%d)\n",
 443                                i == USRQUOTA ? "user" : "group", rc);
 444                         break;
 445                 }
 446         }
 447         up(&mds->mds_qonoff_sem);
 448
 449         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 450         RETURN(rc);
 451 }
 452
 453 static int close_quota_files(struct obd_quotactl *oqctl,
 454                              struct lustre_quota_info *qinfo)
 455 {
 456         int i, rc = 0;
 457         ENTRY;
 458
 459         for (i = 0; i < MAXQUOTAS; i++) {
 460                 if (!Q_TYPESET(oqctl, i))
 461                         continue;
 462                 if (qinfo->qi_files[i] == NULL) {
 463                         rc = -ESRCH;
 464                         continue;
 465                 }
 466                 filp_close(qinfo->qi_files[i], 0);
 467                 qinfo->qi_files[i] = NULL;
 468         }
 469         RETURN(rc);
 470 }
 471
 472 int mds_admin_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 473 {
 474         struct mds_obd *mds = &obd->u.mds;
 475         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 476         const char *quotafiles[] = LUSTRE_ADMIN_QUOTAFILES;
 477         char name[64];
 478         int i, rc = 0;
 479         struct inode *iparent = mds->mds_objects_dir->d_inode;
 480         ENTRY;
 481
 482         LASSERT(iparent);
 483
 484         /* open admin quota files and read quotafile info */
 485         for (i = 0; i < MAXQUOTAS; i++) {
 486                 struct file *fp;
 487
 488                 if (!Q_TYPESET(oqctl, i))
 489                         continue;
 490
 491                 LASSERT(strlen(quotafiles[i]) + sizeof(prefix) <= sizeof(name));
 492                 sprintf(name, "%s%s", prefix, quotafiles[i]);
 493
 494                 if (qinfo->qi_files[i] != NULL) {
 495                         rc = -EBUSY;
 496                         break;
 497                 }
 498
 499                 fp = filp_open(name, O_RDWR | O_EXCL, 0644);
 500                 if (IS_ERR(fp) || !S_ISREG(fp->f_dentry->d_inode->i_mode)) {
 501                         rc = PTR_ERR(fp);
 502                         CDEBUG(rc == -ENOENT ? D_QUOTA : D_ERROR,
 503                                "open %s failed! (rc:%d)\n", name, rc);
 504                         break;
 505                 }
 506                 qinfo->qi_files[i] = fp;
 507
 508                 rc = fsfilt_quotainfo(obd, qinfo, i, QFILE_RD_INFO);
 509                 if (rc) {
 510                         CERROR("error read quotainfo of %s! (rc:%d)\n",
 511                                name, rc);
 512                         break;
 513                 }
 514         }
 515
 516         if (rc && rc != -EBUSY)
 517                 close_quota_files(oqctl, qinfo);
 518
 519         RETURN(rc);
 520 }
 521
 522 static int mds_admin_quota_off(struct obd_device *obd,
 523                                struct obd_quotactl *oqctl)
 524 {
 525         struct mds_obd *mds = &obd->u.mds;
 526         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 527         int rc;
 528         ENTRY;
 529
 530         /* close admin quota files */
 531         rc = close_quota_files(oqctl, qinfo);
 532         RETURN(rc);
 533 }
 534
 535 int mds_quota_on(struct obd_device *obd, struct obd_quotactl *oqctl)
 536 {
 537         struct mds_obd *mds = &obd->u.mds;
 538         struct obd_device_target *obt = &obd->u.obt;
 539         struct lvfs_run_ctxt saved;
 540         int rc;
 541         ENTRY;
 542
 543         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 544                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 545                 atomic_inc(&obt->obt_quotachecking);
 546                 RETURN(-EBUSY);
 547         }
 548
 549         down(&mds->mds_qonoff_sem);
 550         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 551         rc = mds_admin_quota_on(obd, oqctl);
 552         if (rc)
 553                 goto out;
 554
 555         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 556         if (rc)
 557                 goto out;
 558
 559         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 560         if (!rc)
 561                 obt->obt_qctxt.lqc_status = 1;
 562 out:
 563         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 564         up(&mds->mds_qonoff_sem);
 565         atomic_inc(&obt->obt_quotachecking);
 566         RETURN(rc);
 567 }
 568
 569 int mds_quota_off(struct obd_device *obd, struct obd_quotactl *oqctl)
 570 {
 571         struct mds_obd *mds = &obd->u.mds;
 572         struct obd_device_target *obt = &obd->u.obt;
 573         struct lvfs_run_ctxt saved;
 574         int rc, rc2;
 575         ENTRY;
 576
 577         if (!atomic_dec_and_test(&obt->obt_quotachecking)) {
 578                 CDEBUG(D_INFO, "other people are doing quotacheck\n");
 579                 atomic_inc(&obt->obt_quotachecking);
 580                 RETURN(-EBUSY);
 581         }
 582
 583         down(&mds->mds_qonoff_sem);
 584         /* close admin quota files */
 585         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 586         mds_admin_quota_off(obd, oqctl);
 587
 588         rc = obd_quotactl(mds->mds_osc_exp, oqctl);
 589         rc2 = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 590         if (!rc2)
 591                 obt->obt_qctxt.lqc_status = 0;
 592
 593         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 594         up(&mds->mds_qonoff_sem);
 595         atomic_inc(&obt->obt_quotachecking);
 596
 597         RETURN(rc ?: rc2);
 598 }
 599
 600 int mds_set_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 601 {
 602         struct mds_obd *mds = &obd->u.mds;
 603         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 604         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 605         int rc;
 606         ENTRY;
 607
 608         down(&mds->mds_qonoff_sem);
 609         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 610                 rc = -ESRCH;
 611                 goto out;
 612         }
 613
 614         qinfo->qi_info[oqctl->qc_type].dqi_bgrace = dqinfo->dqi_bgrace;
 615         qinfo->qi_info[oqctl->qc_type].dqi_igrace = dqinfo->dqi_igrace;
 616         qinfo->qi_info[oqctl->qc_type].dqi_flags = dqinfo->dqi_flags;
 617
 618         rc = fsfilt_quotainfo(obd, qinfo, oqctl->qc_type, QFILE_WR_INFO);
 619
 620 out:
 621         up(&mds->mds_qonoff_sem);
 622         RETURN(rc);
 623 }
 624
 625 int mds_get_dqinfo(struct obd_device *obd, struct obd_quotactl *oqctl)
 626 {
 627         struct mds_obd *mds = &obd->u.mds;
 628         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 629         struct obd_dqinfo *dqinfo = &oqctl->qc_dqinfo;
 630         int rc = 0;
 631         ENTRY;
 632
 633         down(&mds->mds_qonoff_sem);
 634         if (qinfo->qi_files[oqctl->qc_type] == NULL) {
 635                 rc = -ESRCH;
 636                 goto out;
 637         }
 638
 639         dqinfo->dqi_bgrace = qinfo->qi_info[oqctl->qc_type].dqi_bgrace;
 640         dqinfo->dqi_igrace = qinfo->qi_info[oqctl->qc_type].dqi_igrace;
 641         dqinfo->dqi_flags = qinfo->qi_info[oqctl->qc_type].dqi_flags;
 642
 643 out:
 644         up(&mds->mds_qonoff_sem);
 645         RETURN(rc);
 646 }
 647
 648 static int mds_init_slave_ilimits(struct obd_device *obd,
 649                                   struct obd_quotactl *oqctl, int set)
 650 {
 651         /* XXX: for file limits only adjust local now */
 652         unsigned int uid = 0, gid = 0;
 653         struct obd_quotactl *ioqc = NULL;
 654         int rc;
 655         ENTRY;
 656
 657         /* if we are going to set zero limit, needn't init slaves */
 658         if (!oqctl->qc_dqblk.dqb_ihardlimit && !oqctl->qc_dqblk.dqb_isoftlimit)
 659                 RETURN(0);
 660
 661         if (!set)
 662                 goto acquire;
 663
 664         OBD_ALLOC_PTR(ioqc);
 665         if (!ioqc)
 666                 RETURN(-ENOMEM);
 667
 668         ioqc->qc_cmd = Q_INITQUOTA;
 669         ioqc->qc_id = oqctl->qc_id;
 670         ioqc->qc_type = oqctl->qc_type;
 671         ioqc->qc_dqblk.dqb_valid = QIF_ILIMITS;
 672         ioqc->qc_dqblk.dqb_ihardlimit = MIN_QLIMIT;
 673
 674         /* set local limit to MIN_QLIMIT */
 675         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
 676         if (rc)
 677                 GOTO(out, rc);
 678 acquire:
 679         /* trigger local qunit pre-acquire */
 680         if (oqctl->qc_type == USRQUOTA)
 681                 uid = oqctl->qc_id;
 682         else
 683                 gid = oqctl->qc_id;
 684
 685         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 0, 0);
 686         if (rc) {
 687                 CERROR("error mds adjust local file quota! (rc:%d)\n", rc);
 688                 GOTO(out, rc);
 689         }
 690         /* FIXME initialize all slaves in CMD */
 691         EXIT;
 692 out:
 693         if (ioqc)
 694                 OBD_FREE_PTR(ioqc);
 695         return rc;
 696 }
 697
 698 static int mds_init_slave_blimits(struct obd_device *obd,
 699                                   struct obd_quotactl *oqctl, int set)
 700 {
 701         struct mds_obd *mds = &obd->u.mds;
 702         struct obd_quotactl *ioqc;
 703         unsigned int uid = 0, gid = 0;
 704         int rc;
 705         ENTRY;
 706
 707         /* if we are going to set zero limit, needn't init slaves */
 708         if (!oqctl->qc_dqblk.dqb_bhardlimit && !oqctl->qc_dqblk.dqb_bsoftlimit)
 709                 RETURN(0);
 710
 711         OBD_ALLOC_PTR(ioqc);
 712         if (!ioqc)
 713                 RETURN(-ENOMEM);
 714
 715         ioqc->qc_cmd = Q_INITQUOTA;
 716         ioqc->qc_id = oqctl->qc_id;
 717         ioqc->qc_type = oqctl->qc_type;
 718         ioqc->qc_dqblk.dqb_valid = QIF_BLIMITS;
 719         ioqc->qc_dqblk.dqb_bhardlimit = set ? MIN_QLIMIT : 0;
 720
 721         /* set local limit to MIN_QLIMIT */
 722         if (set) {
 723                 rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, ioqc);
 724                 if (rc)
 725                         GOTO(out, rc);
 726         }
 727
 728         /* trigger local qunit pre-acquire */
 729         if (oqctl->qc_type == USRQUOTA)
 730                 uid = oqctl->qc_id;
 731         else
 732                 gid = oqctl->qc_id;
 733
 734         rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt, uid, gid, 1, 0);
 735         if (rc) {
 736                 CERROR("error mds adjust local block quota! (rc:%d)\n", rc);
 737                 GOTO(out, rc);
 738         }
 739
 740         /* initialize all slave's limit */
 741         rc = obd_quotactl(mds->mds_osc_exp, ioqc);
 742         EXIT;
 743 out:
 744         OBD_FREE_PTR(ioqc);
 745         return rc;
 746 }
 747
 748 int mds_set_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
 749 {
 750         struct mds_obd *mds = &obd->u.mds;
 751         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 752         __u32 ihardlimit, isoftlimit, bhardlimit, bsoftlimit;
 753         time_t btime, itime;
 754         struct lustre_dquot *dquot;
 755         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
 756         int set, rc;
 757         ENTRY;
 758
 759         down(&mds->mds_qonoff_sem);
 760         if (qinfo->qi_files[oqctl->qc_type] == NULL)
 761                 GOTO(out_sem, rc = -ESRCH);
 762
 763         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
 764         if (IS_ERR(dquot))
 765                 GOTO(out_sem, rc = PTR_ERR(dquot));
 766         DQUOT_DEBUG(dquot, "get dquot in mds_set_blk\n");
 767         QINFO_DEBUG(dquot->dq_info, "get dquot in mds_set_blk\n");
 768
 769         down(&dquot->dq_sem);
 770
 771         if (dquot->dq_status) {
 772                 up(&dquot->dq_sem);
 773                 lustre_dqput(dquot);
 774                 GOTO(out_sem, rc = -EBUSY);
 775         }
 776         dquot->dq_status |= DQ_STATUS_SET;
 777
 778         ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
 779         isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
 780         bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
 781         bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
 782         btime = dquot->dq_dqb.dqb_btime;
 783         itime = dquot->dq_dqb.dqb_itime;
 784
 785         if (dqblk->dqb_valid & QIF_BTIME)
 786                 dquot->dq_dqb.dqb_btime = dqblk->dqb_btime;
 787         if (dqblk->dqb_valid & QIF_ITIME)
 788                 dquot->dq_dqb.dqb_itime = dqblk->dqb_itime;
 789
 790         if (dqblk->dqb_valid & QIF_BLIMITS) {
 791                 dquot->dq_dqb.dqb_bhardlimit = dqblk->dqb_bhardlimit;
 792                 dquot->dq_dqb.dqb_bsoftlimit = dqblk->dqb_bsoftlimit;
 793                 /* clear usage (limit pool) */
 794                 if (!dquot->dq_dqb.dqb_bhardlimit &&
 795                     !dquot->dq_dqb.dqb_bsoftlimit)
 796                         dquot->dq_dqb.dqb_curspace = 0;
 797
 798                 /* clear grace time */
 799                 if (!dqblk->dqb_bsoftlimit ||
 800                     toqb(dquot->dq_dqb.dqb_curspace) <= dqblk->dqb_bsoftlimit)
 801                         dquot->dq_dqb.dqb_btime = 0;
 802                 /* set grace only if user hasn't provided his own */
 803                 else if (!(dqblk->dqb_valid & QIF_BTIME))
 804                         dquot->dq_dqb.dqb_btime = cfs_time_current_sec() +
 805                                 qinfo->qi_info[dquot->dq_type].dqi_bgrace;
 806         }
 807
 808         if (dqblk->dqb_valid & QIF_ILIMITS) {
 809                 dquot->dq_dqb.dqb_ihardlimit = dqblk->dqb_ihardlimit;
 810                 dquot->dq_dqb.dqb_isoftlimit = dqblk->dqb_isoftlimit;
 811                 /* clear usage (limit pool) */
 812                 if (!dquot->dq_dqb.dqb_ihardlimit &&
 813                     !dquot->dq_dqb.dqb_isoftlimit)
 814                         dquot->dq_dqb.dqb_curinodes = 0;
 815
 816                 if (!dqblk->dqb_isoftlimit ||
 817                     dquot->dq_dqb.dqb_curinodes <= dqblk->dqb_isoftlimit)
 818                         dquot->dq_dqb.dqb_itime = 0;
 819                 else if (!(dqblk->dqb_valid & QIF_ITIME))
 820                         dquot->dq_dqb.dqb_itime = cfs_time_current_sec() +
 821                                 qinfo->qi_info[dquot->dq_type].dqi_igrace;
 822         }
 823
 824         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 825
 826         up(&dquot->dq_sem);
 827
 828         if (rc) {
 829                 CERROR("set limit failed! (rc:%d)\n", rc);
 830                 goto out;
 831         }
 832
 833         up(&mds->mds_qonoff_sem);
 834         if (dqblk->dqb_valid & QIF_ILIMITS) {
 835                 set = !(ihardlimit || isoftlimit);
 836                 rc = mds_init_slave_ilimits(obd, oqctl, set);
 837                 if (rc) {
 838                         CERROR("init slave ilimits failed! (rc:%d)\n", rc);
 839                         goto revoke_out;
 840                 }
 841         }
 842
 843         if (dqblk->dqb_valid & QIF_BLIMITS) {
 844                 set = !(bhardlimit || bsoftlimit);
 845                 rc = mds_init_slave_blimits(obd, oqctl, set);
 846                 if (rc) {
 847                         CERROR("init slave blimits failed! (rc:%d)\n", rc);
 848                         goto revoke_out;
 849                 }
 850         }
 851         down(&mds->mds_qonoff_sem);
 852
 853 revoke_out:
 854         if (rc) {
 855                 /* cancel previous setting */
 856                 down(&dquot->dq_sem);
 857                 dquot->dq_dqb.dqb_ihardlimit = ihardlimit;
 858                 dquot->dq_dqb.dqb_isoftlimit = isoftlimit;
 859                 dquot->dq_dqb.dqb_bhardlimit = bhardlimit;
 860                 dquot->dq_dqb.dqb_bsoftlimit = bsoftlimit;
 861                 dquot->dq_dqb.dqb_btime = btime;
 862                 dquot->dq_dqb.dqb_itime = itime;
 863                 fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
 864                 up(&dquot->dq_sem);
 865         }
 866 out:
 867         down(&dquot->dq_sem);
 868         dquot->dq_status &= ~DQ_STATUS_SET;
 869         up(&dquot->dq_sem);
 870         lustre_dqput(dquot);
 871         EXIT;
 872 out_sem:
 873         up(&mds->mds_qonoff_sem);
 874         return rc;
 875 }
 876
 877 static int mds_get_space(struct obd_device *obd, struct obd_quotactl *oqctl)
 878 {
 879         struct obd_quotactl *soqc;
 880         struct lvfs_run_ctxt saved;
 881         int rc;
 882         ENTRY;
 883
 884         OBD_ALLOC_PTR(soqc);
 885         if (!soqc)
 886                 RETURN(-ENOMEM);
 887
 888         soqc->qc_cmd = Q_GETOQUOTA;
 889         soqc->qc_id = oqctl->qc_id;
 890         soqc->qc_type = oqctl->qc_type;
 891
 892         rc = obd_quotactl(obd->u.mds.mds_osc_exp, soqc);
 893         if (rc)
 894                GOTO(out, rc);
 895
 896         oqctl->qc_dqblk.dqb_curspace = soqc->qc_dqblk.dqb_curspace;
 897
 898         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 899         soqc->qc_dqblk.dqb_curspace = 0;
 900         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, soqc);
 901         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 902
 903         if (rc)
 904                 GOTO(out, rc);
 905
 906         oqctl->qc_dqblk.dqb_curinodes += soqc->qc_dqblk.dqb_curinodes;
 907         oqctl->qc_dqblk.dqb_curspace += soqc->qc_dqblk.dqb_curspace;
 908         EXIT;
 909 out:
 910         OBD_FREE_PTR(soqc);
 911         return rc;
 912 }
 913
 914 int mds_get_dqblk(struct obd_device *obd, struct obd_quotactl *oqctl)
 915 {
 916         struct mds_obd *mds = &obd->u.mds;
 917         struct lustre_quota_info *qinfo = &mds->mds_quota_info;
 918         struct lustre_dquot *dquot;
 919         struct obd_dqblk *dqblk = &oqctl->qc_dqblk;
 920         int rc;
 921         ENTRY;
 922
 923         down(&mds->mds_qonoff_sem);
 924         if (qinfo->qi_files[oqctl->qc_type] == NULL)
 925                 GOTO(out, rc = -ESRCH);
 926
 927         dquot = lustre_dqget(obd, qinfo, oqctl->qc_id, oqctl->qc_type);
 928         if (IS_ERR(dquot))
 929                 GOTO(out, rc = PTR_ERR(dquot));
 930
 931         down(&dquot->dq_sem);
 932         dqblk->dqb_ihardlimit = dquot->dq_dqb.dqb_ihardlimit;
 933         dqblk->dqb_isoftlimit = dquot->dq_dqb.dqb_isoftlimit;
 934         dqblk->dqb_bhardlimit = dquot->dq_dqb.dqb_bhardlimit;
 935         dqblk->dqb_bsoftlimit = dquot->dq_dqb.dqb_bsoftlimit;
 936         dqblk->dqb_btime = dquot->dq_dqb.dqb_btime;
 937         dqblk->dqb_itime = dquot->dq_dqb.dqb_itime;
 938         up(&dquot->dq_sem);
 939
 940         lustre_dqput(dquot);
 941
 942         /* the usages in admin quota file is inaccurate */
 943         dqblk->dqb_curinodes = 0;
 944         dqblk->dqb_curspace = 0;
 945         rc = mds_get_space(obd, oqctl);
 946         EXIT;
 947 out:
 948         up(&mds->mds_qonoff_sem);
 949         return rc;
 950 }
 951
 952 int mds_get_obd_quota(struct obd_device *obd, struct obd_quotactl *oqctl)
 953 {
 954         struct lvfs_run_ctxt saved;
 955         int rc;
 956         ENTRY;
 957
 958         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 959         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
 960         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 961
 962         RETURN(rc);
 963 }
 964
 965
 966 /* FIXME we only recovery block limit by now, need recovery inode
 967  * limits also after CMD involved in */
 968 static int
 969 dquot_recovery(struct obd_device *obd, unsigned int id, unsigned short type)
 970 {
 971         struct mds_obd *mds = &obd->u.mds;
 972         struct lustre_quota_info *qinfo= &obd->u.mds.mds_quota_info;
 973         struct lustre_dquot *dquot;
 974         struct obd_quotactl *qctl;
 975         __u64 total_limits = 0;
 976         int rc;
 977         ENTRY;
 978
 979         OBD_ALLOC_PTR(qctl);
 980         if (qctl == NULL)
 981                 RETURN(-ENOMEM);
 982
 983         dquot = lustre_dqget(obd, qinfo, id, type);
 984         if (IS_ERR(dquot)) {
 985                 CERROR("Get dquot failed. (rc:%ld)\n", PTR_ERR(dquot));
 986                 OBD_FREE_PTR(qctl);
 987                 RETURN(PTR_ERR(dquot));
 988         }
 989
 990         down(&dquot->dq_sem);
 991
 992         /* don't recovery the dquot without limits or under setting */
 993         if (!(dquot->dq_dqb.dqb_bhardlimit || dquot->dq_dqb.dqb_bsoftlimit) ||
 994             dquot->dq_status)
 995                 GOTO(skip, rc = 0);
 996         dquot->dq_status |= DQ_STATUS_RECOVERY;
 997
 998         up(&dquot->dq_sem);
 999
1000         /* get real bhardlimit from all slaves. */
1001         qctl->qc_cmd = Q_GETOQUOTA;
1002         qctl->qc_type = type;
1003         qctl->qc_id = id;
1004         qctl->qc_stat = QUOTA_RECOVERING;
1005         rc = obd_quotactl(obd->u.mds.mds_osc_exp, qctl);
1006         if (rc)
1007                 GOTO(out, rc);
1008         total_limits = qctl->qc_dqblk.dqb_bhardlimit;
1009
1010         /* get real bhardlimit from master */
1011         rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, qctl);
1012         if (rc)
1013                 GOTO(out, rc);
1014         total_limits += qctl->qc_dqblk.dqb_bhardlimit;
1015
1016         /* amend the usage of the administrative quotafile */
1017         down(&mds->mds_qonoff_sem);
1018         down(&dquot->dq_sem);
1019
1020         dquot->dq_dqb.dqb_curspace = total_limits << QUOTABLOCK_BITS;
1021
1022         rc = fsfilt_dquot(obd, dquot, QFILE_WR_DQUOT);
1023         if (rc)
1024                 CERROR("write dquot failed! (rc:%d)\n", rc);
1025
1026         up(&dquot->dq_sem);
1027         up(&mds->mds_qonoff_sem);
1028         EXIT;
1029 out:
1030         down(&dquot->dq_sem);
1031         dquot->dq_status &= ~DQ_STATUS_RECOVERY;
1032 skip:
1033         up(&dquot->dq_sem);
1034
1035         lustre_dqput(dquot);
1036         OBD_FREE_PTR(qctl);
1037         return rc;
1038 }
1039
1040 struct qmaster_recov_thread_data {
1041         struct obd_device *obd;
1042         struct completion comp;
1043 };
1044
1045 static int qmaster_recovery_main(void *arg)
1046 {
1047         struct qmaster_recov_thread_data *data = arg;
1048         struct obd_device *obd = data->obd;
1049         int rc = 0;
1050         unsigned short type;
1051         ENTRY;
1052
1053         ptlrpc_daemonize("qmaster_recovd");
1054
1055         complete(&data->comp);
1056
1057         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
1058                 struct mds_obd *mds = &obd->u.mds;
1059                 struct lustre_quota_info *qinfo = &mds->mds_quota_info;
1060                 struct list_head id_list;
1061                 struct dquot_id *dqid, *tmp;
1062
1063                 down(&mds->mds_qonoff_sem);
1064                 if (qinfo->qi_files[type] == NULL) {
1065                         up(&mds->mds_qonoff_sem);
1066                         continue;
1067                 }
1068                 INIT_LIST_HEAD(&id_list);
1069                 rc = fsfilt_qids(obd, qinfo->qi_files[type], NULL, type,
1070                                  &id_list);
1071                 up(&mds->mds_qonoff_sem);
1072
1073                 if (rc)
1074                         CERROR("error get ids from admin quotafile.(%d)\n", rc);
1075
1076                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1077                         list_del_init(&dqid->di_link);
1078                         if (rc)
1079                                 goto free;
1080
1081                         rc = dquot_recovery(obd, dqid->di_id, type);
1082                         if (rc)
1083                                 CERROR("qmaster recovery failed! (id:%d type:%d"
1084                                        " rc:%d)\n", dqid->di_id, type, rc);
1085 free:
1086                         kfree(dqid);
1087                 }
1088         }
1089         RETURN(rc);
1090 }
1091
1092 int mds_quota_recovery(struct obd_device *obd)
1093 {
1094         struct lov_obd *lov = &obd->u.mds.mds_osc_obd->u.lov;
1095         struct qmaster_recov_thread_data data;
1096         int rc = 0;
1097         ENTRY;
1098
1099         mutex_down(&lov->lov_lock);
1100         if (lov->desc.ld_tgt_count != lov->desc.ld_active_tgt_count) {
1101                 CWARN("Not all osts are active, abort quota recovery\n");
1102                 mutex_up(&lov->lov_lock);
1103                 RETURN(rc);
1104         }
1105         mutex_up(&lov->lov_lock);
1106
1107         data.obd = obd;
1108         init_completion(&data.comp);
1109
1110         rc = kernel_thread(qmaster_recovery_main, &data, CLONE_VM|CLONE_FILES);
1111         if (rc < 0)
1112                 CERROR("Cannot start quota recovery thread: rc %d\n", rc);
1113
1114         wait_for_completion(&data.comp);
1115         RETURN(rc);
1116 }