lustre/quota/quota_interface.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/quota/quota_interface.c
   5  *
   6  *  Copyright (c) 2001-2005 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   No redistribution or use is permitted outside of Cluster File Systems, Inc.
  11  *
  12  */
  13 #ifndef EXPORT_SYMTAB
  14 # define EXPORT_SYMTAB
  15 #endif
  16 #define DEBUG_SUBSYSTEM S_MDS
  17
  18 #ifdef __KERNEL__
  19 # include <linux/version.h>
  20 # include <linux/module.h>
  21 # include <linux/init.h>
  22 # include <linux/fs.h>
  23 # include <linux/jbd.h>
  24 # include <linux/ext3_fs.h>
  25 # if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
  26 #  include <linux/smp_lock.h>
  27 #  include <linux/buffer_head.h>
  28 #  include <linux/workqueue.h>
  29 #  include <linux/mount.h>
  30 # else
  31 #  include <linux/locks.h>
  32 # endif
  33 #else /* __KERNEL__ */
  34 # include <liblustre.h>
  35 #endif
  36
  37 #include <obd_class.h>
  38 #include <lustre_mds.h>
  39 #include <lustre_dlm.h>
  40 #include <lustre_cfg.h>
  41 #include <obd_ost.h>
  42 #include <lustre_fsfilt.h>
  43 #include <lustre_quota.h>
  44 #include <lprocfs_status.h>
  45 #include "quota_internal.h"
  46
  47
  48 #ifdef __KERNEL__
  49
  50 /* quota proc file handling functions */
  51 #ifdef LPROCFS
  52
  53 int lprocfs_quota_rd_bunit(char *page, char **start, off_t off, int count,
  54                            int *eof, void *data)
  55 {
  56         struct obd_device *obd = (struct obd_device *)data;
  57         LASSERT(obd != NULL);
  58
  59         return snprintf(page, count, "%lu\n",
  60                         obd->u.obt.obt_qctxt.lqc_bunit_sz);
  61 }
  62 EXPORT_SYMBOL(lprocfs_quota_rd_bunit);
  63
  64 int lprocfs_quota_wr_bunit(struct file *file, const char *buffer,
  65                            unsigned long count, void *data)
  66 {
  67         struct obd_device *obd = (struct obd_device *)data;
  68         int val, rc;
  69         LASSERT(obd != NULL);
  70
  71         rc = lprocfs_write_helper(buffer, count, &val);
  72         if (rc)
  73                 return rc;
  74
  75         if (val % QUOTABLOCK_SIZE ||
  76             val <= obd->u.obt.obt_qctxt.lqc_btune_sz)
  77                 return -EINVAL;
  78
  79         obd->u.obt.obt_qctxt.lqc_bunit_sz = val;
  80         return count;
  81 }
  82 EXPORT_SYMBOL(lprocfs_quota_wr_bunit);
  83
  84 int lprocfs_quota_rd_btune(char *page, char **start, off_t off, int count,
  85                            int *eof, void *data)
  86 {
  87         struct obd_device *obd = (struct obd_device *)data;
  88         LASSERT(obd != NULL);
  89
  90         return snprintf(page, count, "%lu\n",
  91                         obd->u.obt.obt_qctxt.lqc_btune_sz);
  92 }
  93 EXPORT_SYMBOL(lprocfs_quota_rd_btune);
  94
  95 int lprocfs_quota_wr_btune(struct file *file, const char *buffer,
  96                            unsigned long count, void *data)
  97 {
  98         struct obd_device *obd = (struct obd_device *)data;
  99         int val, rc;
 100         LASSERT(obd != NULL);
 101
 102         rc = lprocfs_write_helper(buffer, count, &val);
 103         if (rc)
 104                 return rc;
 105
 106         if (val <= QUOTABLOCK_SIZE * MIN_QLIMIT || val % QUOTABLOCK_SIZE ||
 107             val >= obd->u.obt.obt_qctxt.lqc_bunit_sz)
 108                 return -EINVAL;
 109
 110         obd->u.obt.obt_qctxt.lqc_btune_sz = val;
 111         return count;
 112 }
 113 EXPORT_SYMBOL(lprocfs_quota_wr_btune);
 114
 115 int lprocfs_quota_rd_iunit(char *page, char **start, off_t off, int count,
 116                            int *eof, void *data)
 117 {
 118         struct obd_device *obd = (struct obd_device *)data;
 119         LASSERT(obd != NULL);
 120
 121         return snprintf(page, count, "%lu\n",
 122                         obd->u.obt.obt_qctxt.lqc_iunit_sz);
 123 }
 124 EXPORT_SYMBOL(lprocfs_quota_rd_iunit);
 125
 126 int lprocfs_quota_wr_iunit(struct file *file, const char *buffer,
 127                            unsigned long count, void *data)
 128 {
 129         struct obd_device *obd = (struct obd_device *)data;
 130         int val, rc;
 131         LASSERT(obd != NULL);
 132
 133         rc = lprocfs_write_helper(buffer, count, &val);
 134         if (rc)
 135                 return rc;
 136
 137         if (val <= obd->u.obt.obt_qctxt.lqc_itune_sz)
 138                 return -EINVAL;
 139
 140         obd->u.obt.obt_qctxt.lqc_iunit_sz = val;
 141         return count;
 142 }
 143 EXPORT_SYMBOL(lprocfs_quota_wr_iunit);
 144
 145 int lprocfs_quota_rd_itune(char *page, char **start, off_t off, int count,
 146                            int *eof, void *data)
 147 {
 148         struct obd_device *obd = (struct obd_device *)data;
 149         LASSERT(obd != NULL);
 150
 151         return snprintf(page, count, "%lu\n",
 152                         obd->u.obt.obt_qctxt.lqc_itune_sz);
 153 }
 154 EXPORT_SYMBOL(lprocfs_quota_rd_itune);
 155
 156 int lprocfs_quota_wr_itune(struct file *file, const char *buffer,
 157                            unsigned long count, void *data)
 158 {
 159         struct obd_device *obd = (struct obd_device *)data;
 160         int val, rc;
 161         LASSERT(obd != NULL);
 162
 163         rc = lprocfs_write_helper(buffer, count, &val);
 164         if (rc)
 165                 return rc;
 166
 167         if (val <= MIN_QLIMIT ||
 168             val >= obd->u.obt.obt_qctxt.lqc_iunit_sz)
 169                 return -EINVAL;
 170
 171         obd->u.obt.obt_qctxt.lqc_itune_sz = val;
 172         return count;
 173 }
 174 EXPORT_SYMBOL(lprocfs_quota_wr_itune);
 175
 176 int lprocfs_quota_rd_switch_seconds(char *page, char **start, off_t off,
 177                                     int count, int *eof, void *data)
 178 {
 179         struct obd_device *obd = (struct obd_device *)data;
 180         LASSERT(obd != NULL);
 181
 182         return snprintf(page, count, "%d\n",
 183                         obd->u.obt.obt_qctxt.lqc_switch_seconds);
 184 }
 185 EXPORT_SYMBOL(lprocfs_quota_rd_switch_seconds);
 186
 187 int lprocfs_quota_wr_switch_seconds(struct file *file, const char *buffer,
 188                                     unsigned long count, void *data)
 189 {
 190         struct obd_device *obd = (struct obd_device *)data;
 191         int val, rc;
 192         LASSERT(obd != NULL);
 193
 194         rc = lprocfs_write_helper(buffer, count, &val);
 195         if (rc)
 196                 return rc;
 197
 198         if (val <= 10)
 199                 return -EINVAL;
 200
 201         obd->u.obt.obt_qctxt.lqc_switch_seconds = val;
 202         return count;
 203 }
 204 EXPORT_SYMBOL(lprocfs_quota_wr_switch_seconds);
 205
 206 #define USER_QUOTA      1
 207 #define GROUP_QUOTA     2
 208
 209 #define MAX_STYPE_SIZE  5
 210 int lprocfs_quota_rd_type(char *page, char **start, off_t off, int count,
 211                           int *eof, void *data)
 212 {
 213         struct obd_device *obd = (struct obd_device *)data;
 214         char stype[MAX_STYPE_SIZE + 1] = "";
 215         int type = obd->u.obt.obt_qctxt.lqc_atype;
 216         LASSERT(obd != NULL);
 217
 218         if (type == 0) {
 219                 strcpy(stype, "off");
 220         } else {
 221                 if (type & USER_QUOTA)
 222                         strcat(stype, "u");
 223                 if (type & GROUP_QUOTA)
 224                         strcat(stype, "g");
 225         }
 226
 227         /* append with quota version on MDS */
 228         if (!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME)) {
 229                 int rc;
 230                 lustre_quota_version_t version;
 231
 232                 rc = mds_quota_get_version(obd, &version);
 233                 if (rc)
 234                         return rc;
 235
 236                 switch (version) {
 237                         case LUSTRE_QUOTA_V1:
 238                                 strcat(stype, "1");
 239                                 break;
 240                         case LUSTRE_QUOTA_V2:
 241                                 strcat(stype, "2");
 242                                 break;
 243                         default:
 244                                 return -ENOSYS;
 245                 }
 246         }
 247
 248         return snprintf(page, count, "%s\n", stype);
 249 }
 250 EXPORT_SYMBOL(lprocfs_quota_rd_type);
 251
 252 static int auto_quota_on(struct obd_device *obd, int type,
 253                          struct super_block *sb, int is_master)
 254 {
 255         struct obd_quotactl *oqctl;
 256         struct lvfs_run_ctxt saved;
 257         int rc;
 258         ENTRY;
 259
 260         LASSERT(type == USRQUOTA || type == GRPQUOTA || type == UGQUOTA);
 261
 262         /* quota already turned on */
 263         if (obd->u.obt.obt_qctxt.lqc_status)
 264                 RETURN(0);
 265
 266         OBD_ALLOC_PTR(oqctl);
 267         if (!oqctl)
 268                 RETURN(-ENOMEM);
 269
 270         oqctl->qc_type = type;
 271         oqctl->qc_cmd = Q_QUOTAON;
 272         oqctl->qc_id = QFMT_LDISKFS;
 273
 274         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 275
 276         if (!is_master)
 277                 goto local_quota;
 278
 279         /* turn on cluster wide quota */
 280         rc = mds_admin_quota_on(obd, oqctl);
 281         if (rc) {
 282                 CDEBUG(rc == -ENOENT ? D_QUOTA : D_ERROR,
 283                        "auto-enable admin quota failed. rc=%d\n", rc);
 284                 GOTO(out_pop, rc);
 285         }
 286 local_quota:
 287         /* turn on local quota */
 288         rc = fsfilt_quotactl(obd, sb, oqctl);
 289         if (rc) {
 290                 CDEBUG(rc == -ENOENT ? D_QUOTA : D_ERROR,
 291                        "auto-enable local quota failed. rc=%d\n", rc);
 292                 if (is_master)
 293                         mds_quota_off(obd, oqctl);
 294         } else {
 295                 obd->u.obt.obt_qctxt.lqc_status = 1;
 296         }
 297 out_pop:
 298         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 299
 300         OBD_FREE_PTR(oqctl);
 301         RETURN(rc);
 302 }
 303
 304 int lprocfs_quota_wr_type(struct file *file, const char *buffer,
 305                           unsigned long count, void *data)
 306 {
 307         struct obd_device *obd = (struct obd_device *)data;
 308         struct obd_device_target *obt = &obd->u.obt;
 309         int type = 0;
 310         unsigned long i;
 311         char stype[MAX_STYPE_SIZE + 1] = "";
 312         LASSERT(obd != NULL);
 313
 314         if (count > MAX_STYPE_SIZE)
 315                 return -EINVAL;
 316
 317         if (copy_from_user(stype, buffer, count))
 318                 return -EFAULT;
 319
 320         for (i = 0 ; i < count ; i++) {
 321                 int rc;
 322
 323                 switch (stype[i]) {
 324                 case 'u' :
 325                         type |= USER_QUOTA;
 326                         break;
 327                 case 'g' :
 328                         type |= GROUP_QUOTA;
 329                         break;
 330                 /* quota version specifiers */
 331                 case '1' :
 332                         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
 333                                 break;
 334
 335                         rc = mds_quota_set_version(obd, LUSTRE_QUOTA_V1);
 336                         if (rc) {
 337                                 CDEBUG(D_QUOTA, "failed to set quota v1! %d\n", rc);
 338                                 return rc;
 339                         }
 340                         break;
 341                 case '2' :
 342                         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
 343                                 break;
 344
 345                         rc = mds_quota_set_version(obd, LUSTRE_QUOTA_V2);
 346                         if (rc) {
 347                                 CDEBUG(D_QUOTA, "could not set quota v2! %d\n", rc);
 348                                 return rc;
 349                         }
 350                         break;
 351                 default  : /* just skip stray symbols like \n */
 352                         break;
 353                 }
 354         }
 355
 356         obt->obt_qctxt.lqc_atype = type;
 357
 358         if (type == 0)
 359                 return count;
 360
 361         if (!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
 362                 auto_quota_on(obd, type - 1, obt->obt_sb, 1);
 363         else if (!strcmp(obd->obd_type->typ_name, LUSTRE_OST_NAME))
 364                 auto_quota_on(obd, type - 1, obt->obt_sb, 0);
 365         else
 366                 return -EFAULT;
 367
 368         return count;
 369 }
 370 EXPORT_SYMBOL(lprocfs_quota_wr_type);
 371
 372
 373 #endif /* LPROCFS */
 374
 375 static int filter_quota_setup(struct obd_device *obd)
 376 {
 377         int rc = 0;
 378         struct obd_device_target *obt = &obd->u.obt;
 379         ENTRY;
 380
 381         atomic_set(&obt->obt_quotachecking, 1);
 382         rc = qctxt_init(&obt->obt_qctxt, obt->obt_sb, NULL);
 383         if (rc) {
 384                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 385                 RETURN(rc);
 386         }
 387         RETURN(rc);
 388 }
 389
 390 static int filter_quota_cleanup(struct obd_device *obd)
 391 {
 392         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 393         return 0;
 394 }
 395
 396 static int filter_quota_setinfo(struct obd_export *exp, struct obd_device *obd)
 397 {
 398         struct obd_import *imp;
 399
 400         /* setup the quota context import */
 401         spin_lock(&obd->u.obt.obt_qctxt.lqc_lock);
 402         obd->u.obt.obt_qctxt.lqc_import = exp->exp_imp_reverse;
 403         spin_unlock(&obd->u.obt.obt_qctxt.lqc_lock);
 404
 405         /* make imp's connect flags equal relative exp's connect flags
 406          * adding it to avoid the scan export list
 407          */
 408         imp = exp->exp_imp_reverse;
 409         if (imp)
 410                 imp->imp_connect_data.ocd_connect_flags |=
 411                         (exp->exp_connect_flags &
 412                          (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 413
 414         /* start quota slave recovery thread. (release high limits) */
 415         qslave_start_recovery(obd, &obd->u.obt.obt_qctxt);
 416         return 0;
 417 }
 418
 419 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 420 {
 421         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 422
 423         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 424          * should be invalid b=12374 */
 425         if (qctxt->lqc_import == exp->exp_imp_reverse) {
 426                 spin_lock(&qctxt->lqc_lock);
 427                 qctxt->lqc_import = NULL;
 428                 spin_unlock(&qctxt->lqc_lock);
 429         }
 430
 431         return 0;
 432 }
 433
 434 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 435 {
 436         ENTRY;
 437
 438         if (!sb_any_quota_enabled(obd->u.obt.obt_sb))
 439                 RETURN(0);
 440
 441         if (ignore)
 442                 cap_raise(current->cap_effective, CAP_SYS_RESOURCE);
 443         else
 444                 cap_lower(current->cap_effective, CAP_SYS_RESOURCE);
 445
 446         RETURN(0);
 447 }
 448
 449 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 450 {
 451         struct obd_device_target *obt = &obd->u.obt;
 452         int err, cnt, rc = 0;
 453         struct obd_quotactl *oqctl;
 454         ENTRY;
 455
 456         if (!sb_any_quota_enabled(obt->obt_sb))
 457                 RETURN(0);
 458
 459         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 460
 461         OBD_ALLOC_PTR(oqctl);
 462         if (!oqctl) {
 463                 CERROR("Not enough memory!");
 464                 RETURN(-ENOMEM);
 465         }
 466
 467         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 468                 memset(oqctl, 0, sizeof(*oqctl));
 469
 470                 oqctl->qc_cmd = Q_GETQUOTA;
 471                 oqctl->qc_type = cnt;
 472                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 473                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 474                 if (err) {
 475                         if (!rc)
 476                                 rc = err;
 477                         continue;
 478                 }
 479
 480                 /* set over quota flags for a uid/gid */
 481                 oa->o_valid |= (cnt == USRQUOTA) ?
 482                                OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA;
 483                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 484                    (toqb(oqctl->qc_dqblk.dqb_curspace) >
 485                     oqctl->qc_dqblk.dqb_bhardlimit))
 486                         oa->o_flags |= (cnt == USRQUOTA) ?
 487                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 488         }
 489         OBD_FREE_PTR(oqctl);
 490         RETURN(rc);
 491 }
 492
 493 static int filter_quota_acquire(struct obd_device *obd, unsigned int uid,
 494                                 unsigned int gid)
 495 {
 496         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 497         int rc;
 498         ENTRY;
 499
 500         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, LQUOTA_FLAGS_BLK, 1);
 501         RETURN(rc);
 502 }
 503
 504 /* check whether the left quota of certain uid and gid can satisfy a block_write
 505  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA */
 506 static int quota_check_common(struct obd_device *obd, unsigned int uid,
 507                               unsigned int gid, int count, int cycle, int isblk)
 508 {
 509         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 510         int i;
 511         __u32 id[MAXQUOTAS] = { uid, gid };
 512         struct qunit_data qdata[MAXQUOTAS];
 513         int rc = 0, rc2[2] = { 0, 0 };
 514         ENTRY;
 515
 516         CLASSERT(MAXQUOTAS < 4);
 517         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 518                 RETURN(rc);
 519
 520         for (i = 0; i < MAXQUOTAS; i++) {
 521                 struct lustre_qunit_size *lqs = NULL;
 522
 523                 qdata[i].qd_id = id[i];
 524                 qdata[i].qd_flags = i;
 525                 if (isblk)
 526                         QDATA_SET_BLK(&qdata[i]);
 527                 qdata[i].qd_count = 0;
 528
 529                 /* ignore root user */
 530                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 531                         continue;
 532
 533                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 534                 if (!lqs)
 535                         continue;
 536
 537                 qctxt_wait_pending_dqacq(qctxt, id[i], i, isblk);
 538                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 539                 spin_lock(&lqs->lqs_lock);
 540                 if (!cycle) {
 541                         rc = QUOTA_RET_INC_PENDING;
 542                         if (isblk)
 543                                 lqs->lqs_bwrite_pending += count;
 544                         else
 545                                 lqs->lqs_iwrite_pending += count;
 546                 }
 547
 548                 CDEBUG(D_QUOTA, "write pending: %lu, qd_count: "LPU64".\n",
 549                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 550                        qdata[i].qd_count);
 551                 if (rc2[i] == QUOTA_RET_OK) {
 552                         if (isblk && qdata[i].qd_count <
 553                             lqs->lqs_bwrite_pending * CFS_PAGE_SIZE)
 554                                 rc2[i] = QUOTA_RET_ACQUOTA;
 555                         if (!isblk && qdata[i].qd_count <
 556                             lqs->lqs_iwrite_pending)
 557                                 rc2[i] = QUOTA_RET_ACQUOTA;
 558                 }
 559
 560                 spin_unlock(&lqs->lqs_lock);
 561                 lqs_putref(lqs);
 562         }
 563
 564         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 565                 RETURN(rc | QUOTA_RET_ACQUOTA);
 566         else
 567                 RETURN(rc);
 568 }
 569
 570 static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid,
 571                                 unsigned int gid, int count, int *pending,
 572                                 int isblk, quota_acquire acquire)
 573 {
 574         int rc = 0, cycle = 0;
 575         ENTRY;
 576
 577         while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk)) &
 578                QUOTA_RET_ACQUOTA) {
 579
 580                 if (rc & QUOTA_RET_INC_PENDING)
 581                         *pending = 1;
 582
 583                 cycle++;
 584                 if (isblk)
 585                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 586                 rc = acquire(obd, uid, gid);
 587
 588                 /* please reference to dqacq_completion for the below */
 589                 /* a new request is finished, try again */
 590                 if (rc == -EAGAIN) {
 591                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 592                         continue;
 593                 }
 594
 595                 /* it is out of quota already */
 596                 if (rc == -EDQUOT) {
 597                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 598                         break;
 599                 }
 600
 601                 /* -EBUSY and others, try 10 times */
 602                 if (rc < 0 && cycle < 10) {
 603                         CDEBUG(D_QUOTA, "rc: %d, cycle: %d\n", rc, cycle);
 604                         cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE, HZ);
 605                         continue;
 606                 }
 607
 608                 CDEBUG(D_QUOTA, "exit with rc: %d\n", rc);
 609                 break;
 610         }
 611
 612         if (!cycle && rc & QUOTA_RET_INC_PENDING)
 613                 *pending = 1;
 614
 615         RETURN(rc);
 616 }
 617
 618
 619 static int filter_quota_check(struct obd_device *obd, unsigned int uid,
 620                               unsigned int gid, int npage, int *flag,
 621                               quota_acquire acquire)
 622 {
 623         return quota_chk_acq_common(obd, uid, gid, npage, flag, LQUOTA_FLAGS_BLK,
 624                                     acquire);
 625 }
 626
 627 /* when a block_write or inode_create rpc is finished, adjust the record for
 628  * pending blocks and inodes*/
 629 static int quota_pending_commit(struct obd_device *obd, unsigned int uid,
 630                                 unsigned int gid, int count, int isblk)
 631 {
 632         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 633         int i;
 634         __u32 id[MAXQUOTAS] = { uid, gid };
 635         struct qunit_data qdata[MAXQUOTAS];
 636         ENTRY;
 637
 638         CLASSERT(MAXQUOTAS < 4);
 639         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 640                 RETURN(0);
 641
 642         for (i = 0; i < MAXQUOTAS; i++) {
 643                 struct lustre_qunit_size *lqs = NULL;
 644
 645                 qdata[i].qd_id = id[i];
 646                 qdata[i].qd_flags = i;
 647                 if (isblk)
 648                         QDATA_SET_BLK(&qdata[i]);
 649                 qdata[i].qd_count = 0;
 650
 651                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 652                         continue;
 653
 654                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 655                 if (lqs) {
 656                         spin_lock(&lqs->lqs_lock);
 657                         CDEBUG(D_QUOTA, "pending: %lu, count: %d.\n",
 658                                isblk ? lqs->lqs_bwrite_pending :
 659                                lqs->lqs_iwrite_pending, count);
 660
 661                         if (isblk) {
 662                                 if (lqs->lqs_bwrite_pending >= count)
 663                                         lqs->lqs_bwrite_pending -= count;
 664                                 else
 665                                         CDEBUG(D_ERROR,
 666                                                "there are too many blocks!\n");
 667                         } else {
 668                                 if (lqs->lqs_iwrite_pending >= count)
 669                                         lqs->lqs_iwrite_pending -= count;
 670                                 else
 671                                         CDEBUG(D_ERROR,
 672                                                "there are too many files!\n");
 673                         }
 674
 675                         spin_unlock(&lqs->lqs_lock);
 676                         lqs_putref(lqs);
 677                 }
 678         }
 679
 680         RETURN(0);
 681 }
 682
 683 static int filter_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 684                                        unsigned int gid, int npage)
 685 {
 686         return quota_pending_commit(obd, uid, gid, npage, LQUOTA_FLAGS_BLK);
 687 }
 688
 689 static int mds_quota_init(void)
 690 {
 691         return lustre_dquot_init();
 692 }
 693
 694 static int mds_quota_exit(void)
 695 {
 696         lustre_dquot_exit();
 697         return 0;
 698 }
 699
 700 static int mds_quota_setup(struct obd_device *obd)
 701 {
 702         struct obd_device_target *obt = &obd->u.obt;
 703         struct mds_obd *mds = &obd->u.mds;
 704         int rc;
 705         ENTRY;
 706
 707         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 708         atomic_set(&obt->obt_quotachecking, 1);
 709         /* initialize quota master and quota context */
 710         sema_init(&mds->mds_qonoff_sem, 1);
 711         rc = qctxt_init(&obt->obt_qctxt, obt->obt_sb, dqacq_handler);
 712         if (rc) {
 713                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 714                 RETURN(rc);
 715         }
 716         RETURN(rc);
 717 }
 718
 719 static int mds_quota_cleanup(struct obd_device *obd)
 720 {
 721         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 722         RETURN(0);
 723 }
 724
 725 static int mds_quota_fs_cleanup(struct obd_device *obd)
 726 {
 727         struct mds_obd *mds = &obd->u.mds;
 728         int i;
 729         ENTRY;
 730
 731         /* close admin quota files */
 732         down(&mds->mds_qonoff_sem);
 733         for (i = 0; i < MAXQUOTAS; i++) {
 734                 if (mds->mds_quota_info.qi_files[i]) {
 735                         filp_close(mds->mds_quota_info.qi_files[i], 0);
 736                         mds->mds_quota_info.qi_files[i] = NULL;
 737                 }
 738         }
 739         up(&mds->mds_qonoff_sem);
 740         RETURN(0);
 741 }
 742
 743 static int mds_quota_check(struct obd_device *obd, unsigned int uid,
 744                            unsigned int gid, int inodes, int *flag,
 745                            quota_acquire acquire)
 746 {
 747         return quota_chk_acq_common(obd, uid, gid, inodes, flag, 0, acquire);
 748 }
 749
 750 static int mds_quota_acquire(struct obd_device *obd, unsigned int uid,
 751                              unsigned int gid)
 752 {
 753         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 754         int rc;
 755         ENTRY;
 756
 757         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, 0, 1);
 758         RETURN(rc);
 759 }
 760
 761 static int mds_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 762                                     unsigned int gid, int inodes)
 763 {
 764         return quota_pending_commit(obd, uid, gid, inodes, 0);
 765 }
 766 #endif /* __KERNEL__ */
 767
 768 struct osc_quota_info {
 769         struct list_head        oqi_hash;       /* hash list */
 770         struct client_obd      *oqi_cli;        /* osc obd */
 771         unsigned int            oqi_id;         /* uid/gid of a file */
 772         short                   oqi_type;       /* quota type */
 773 };
 774
 775 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 776
 777 static struct list_head qinfo_hash[NR_DQHASH];
 778 /* SLAB cache for client quota context */
 779 cfs_mem_cache_t *qinfo_cachep = NULL;
 780
 781 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 782                          __attribute__((__const__));
 783
 784 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 785 {
 786         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 787         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 788         return tmp;
 789 }
 790
 791 /* caller must hold qinfo_list_lock */
 792 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 793 {
 794         struct list_head *head = qinfo_hash +
 795                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 796
 797         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 798         list_add(&oqi->oqi_hash, head);
 799 }
 800
 801 /* caller must hold qinfo_list_lock */
 802 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 803 {
 804         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 805         list_del_init(&oqi->oqi_hash);
 806 }
 807
 808 /* caller must hold qinfo_list_lock */
 809 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 810                                                 unsigned int id, int type)
 811 {
 812         unsigned int hashent = hashfn(cli, id, type);
 813         struct osc_quota_info *oqi;
 814
 815         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 816         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 817                 if (oqi->oqi_cli == cli &&
 818                     oqi->oqi_id == id && oqi->oqi_type == type)
 819                         return oqi;
 820         }
 821         return NULL;
 822 }
 823
 824 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 825                                           unsigned int id, int type)
 826 {
 827         struct osc_quota_info *oqi;
 828         ENTRY;
 829
 830         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 831         if(!oqi)
 832                 RETURN(NULL);
 833
 834         INIT_LIST_HEAD(&oqi->oqi_hash);
 835         oqi->oqi_cli = cli;
 836         oqi->oqi_id = id;
 837         oqi->oqi_type = type;
 838
 839         RETURN(oqi);
 840 }
 841
 842 static void free_qinfo(struct osc_quota_info *oqi)
 843 {
 844         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 845 }
 846
 847 int osc_quota_chkdq(struct client_obd *cli,
 848                     unsigned int uid, unsigned int gid)
 849 {
 850         unsigned int id;
 851         int cnt, rc = QUOTA_OK;
 852         ENTRY;
 853
 854         spin_lock(&qinfo_list_lock);
 855         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 856                 struct osc_quota_info *oqi = NULL;
 857
 858                 id = (cnt == USRQUOTA) ? uid : gid;
 859                 oqi = find_qinfo(cli, id, cnt);
 860                 if (oqi) {
 861                         rc = NO_QUOTA;
 862                         break;
 863                 }
 864         }
 865         spin_unlock(&qinfo_list_lock);
 866
 867         RETURN(rc);
 868 }
 869
 870 int osc_quota_setdq(struct client_obd *cli,
 871                     unsigned int uid, unsigned int gid,
 872                     obd_flag valid, obd_flag flags)
 873 {
 874         unsigned int id;
 875         obd_flag noquota;
 876         int cnt, rc = 0;
 877         ENTRY;
 878
 879
 880         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 881                 struct osc_quota_info *oqi, *old;
 882
 883                 if (!(valid & ((cnt == USRQUOTA) ?
 884                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 885                         continue;
 886
 887                 id = (cnt == USRQUOTA) ? uid : gid;
 888                 noquota = (cnt == USRQUOTA) ?
 889                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 890
 891                 oqi = alloc_qinfo(cli, id, cnt);
 892                 if (oqi) {
 893                         spin_lock(&qinfo_list_lock);
 894
 895                         old = find_qinfo(cli, id, cnt);
 896                         if (old && !noquota)
 897                                 remove_qinfo_hash(old);
 898                         else if (!old && noquota)
 899                                 insert_qinfo_hash(oqi);
 900
 901                         spin_unlock(&qinfo_list_lock);
 902
 903                         if (old || !noquota)
 904                                 free_qinfo(oqi);
 905                         if (old && !noquota)
 906                                 free_qinfo(old);
 907                 } else {
 908                         CERROR("not enough mem!\n");
 909                         rc = -ENOMEM;
 910                         break;
 911                 }
 912         }
 913
 914         RETURN(rc);
 915 }
 916
 917 int osc_quota_cleanup(struct obd_device *obd)
 918 {
 919         struct client_obd *cli = &obd->u.cli;
 920         struct osc_quota_info *oqi, *n;
 921         int i;
 922         ENTRY;
 923
 924         spin_lock(&qinfo_list_lock);
 925         for (i = 0; i < NR_DQHASH; i++) {
 926                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 927                         if (oqi->oqi_cli != cli)
 928                                 continue;
 929                         remove_qinfo_hash(oqi);
 930                         free_qinfo(oqi);
 931                 }
 932         }
 933         spin_unlock(&qinfo_list_lock);
 934
 935         RETURN(0);
 936 }
 937
 938 int osc_quota_init(void)
 939 {
 940         int i;
 941         ENTRY;
 942
 943         LASSERT(qinfo_cachep == NULL);
 944         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 945                                             sizeof(struct osc_quota_info),
 946                                             0, 0);
 947         if (!qinfo_cachep)
 948                 RETURN(-ENOMEM);
 949
 950         for (i = 0; i < NR_DQHASH; i++)
 951                 INIT_LIST_HEAD(qinfo_hash + i);
 952
 953         RETURN(0);
 954 }
 955
 956 int osc_quota_exit(void)
 957 {
 958         struct osc_quota_info *oqi, *n;
 959         int i, rc;
 960         ENTRY;
 961
 962         spin_lock(&qinfo_list_lock);
 963         for (i = 0; i < NR_DQHASH; i++) {
 964                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 965                         remove_qinfo_hash(oqi);
 966                         free_qinfo(oqi);
 967                 }
 968         }
 969         spin_unlock(&qinfo_list_lock);
 970
 971         rc = cfs_mem_cache_destroy(qinfo_cachep);
 972         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 973         qinfo_cachep = NULL;
 974
 975         RETURN(0);
 976 }
 977
 978 #ifdef __KERNEL__
 979 quota_interface_t mds_quota_interface = {
 980         .quota_init     = mds_quota_init,
 981         .quota_exit     = mds_quota_exit,
 982         .quota_setup    = mds_quota_setup,
 983         .quota_cleanup  = mds_quota_cleanup,
 984         .quota_check    = target_quota_check,
 985         .quota_ctl      = mds_quota_ctl,
 986         .quota_fs_cleanup       =mds_quota_fs_cleanup,
 987         .quota_recovery = mds_quota_recovery,
 988         .quota_adjust   = mds_quota_adjust,
 989         .quota_chkquota = mds_quota_check,
 990         .quota_acquire  = mds_quota_acquire,
 991         .quota_pending_commit = mds_quota_pending_commit,
 992 };
 993
 994 quota_interface_t filter_quota_interface = {
 995         .quota_setup    = filter_quota_setup,
 996         .quota_cleanup  = filter_quota_cleanup,
 997         .quota_check    = target_quota_check,
 998         .quota_ctl      = filter_quota_ctl,
 999         .quota_setinfo  = filter_quota_setinfo,
1000         .quota_clearinfo = filter_quota_clearinfo,
1001         .quota_enforce  = filter_quota_enforce,
1002         .quota_getflag  = filter_quota_getflag,
1003         .quota_acquire  = filter_quota_acquire,
1004         .quota_adjust   = filter_quota_adjust,
1005         .quota_chkquota = filter_quota_check,
1006         .quota_adjust_qunit   = filter_quota_adjust_qunit,
1007         .quota_pending_commit = filter_quota_pending_commit,
1008 };
1009 #endif /* __KERNEL__ */
1010
1011 quota_interface_t mdc_quota_interface = {
1012         .quota_ctl      = client_quota_ctl,
1013         .quota_check    = client_quota_check,
1014         .quota_poll_check = client_quota_poll_check,
1015 };
1016
1017 quota_interface_t osc_quota_interface = {
1018         .quota_ctl      = client_quota_ctl,
1019         .quota_check    = client_quota_check,
1020         .quota_poll_check = client_quota_poll_check,
1021         .quota_init     = osc_quota_init,
1022         .quota_exit     = osc_quota_exit,
1023         .quota_chkdq    = osc_quota_chkdq,
1024         .quota_setdq    = osc_quota_setdq,
1025         .quota_cleanup  = osc_quota_cleanup,
1026         .quota_adjust_qunit = client_quota_adjust_qunit,
1027 };
1028
1029 quota_interface_t lov_quota_interface = {
1030         .quota_check    = lov_quota_check,
1031         .quota_ctl      = lov_quota_ctl,
1032         .quota_adjust_qunit = lov_quota_adjust_qunit,
1033 };
1034
1035 #ifdef __KERNEL__
1036 static int __init init_lustre_quota(void)
1037 {
1038         int rc = qunit_cache_init();
1039         if (rc)
1040                 return rc;
1041         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
1042         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
1043         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
1044         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
1045         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
1046         return 0;
1047 }
1048
1049 static void /*__exit*/ exit_lustre_quota(void)
1050 {
1051         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
1052         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
1053         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
1054         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
1055         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
1056
1057         qunit_cache_cleanup();
1058 }
1059
1060 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1061 MODULE_DESCRIPTION("Lustre Quota");
1062 MODULE_LICENSE("GPL");
1063
1064 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1065
1066 EXPORT_SYMBOL(mds_quota_interface);
1067 EXPORT_SYMBOL(filter_quota_interface);
1068 EXPORT_SYMBOL(mdc_quota_interface);
1069 EXPORT_SYMBOL(osc_quota_interface);
1070 EXPORT_SYMBOL(lov_quota_interface);
1071 #endif /* __KERNEL */