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 #ifdef __KERNEL__
  48
  49 /* quota proc file handling functions */
  50 #ifdef LPROCFS
  51
  52 #define USER_QUOTA      1
  53 #define GROUP_QUOTA     2
  54
  55 #define MAX_STYPE_SIZE  5
  56 int lprocfs_quota_rd_type(char *page, char **start, off_t off, int count,
  57                           int *eof, void *data)
  58 {
  59         struct obd_device *obd = (struct obd_device *)data;
  60         char stype[MAX_STYPE_SIZE + 1] = "";
  61         int type = obd->u.obt.obt_qctxt.lqc_atype;
  62         LASSERT(obd != NULL);
  63
  64         if (type == 0) {
  65                 strcpy(stype, "off");
  66         } else {
  67                 if (type & USER_QUOTA)
  68                         strcat(stype, "u");
  69                 if (type & GROUP_QUOTA)
  70                         strcat(stype, "g");
  71         }
  72
  73         /* append with quota version on MDS */
  74         if (!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME)) {
  75                 int rc;
  76                 lustre_quota_version_t version;
  77
  78                 rc = mds_quota_get_version(obd, &version);
  79                 if (rc)
  80                         return rc;
  81
  82                 switch (version) {
  83                         case LUSTRE_QUOTA_V1:
  84                                 strcat(stype, "1");
  85                                 break;
  86                         case LUSTRE_QUOTA_V2:
  87                                 strcat(stype, "2");
  88                                 break;
  89                         default:
  90                                 return -ENOSYS;
  91                 }
  92         }
  93
  94         return snprintf(page, count, "%s\n", stype);
  95 }
  96 EXPORT_SYMBOL(lprocfs_quota_rd_type);
  97
  98 static int auto_quota_on(struct obd_device *obd, int type,
  99                          struct super_block *sb, int is_master)
 100 {
 101         struct obd_quotactl *oqctl;
 102         struct lvfs_run_ctxt saved;
 103         int rc;
 104         ENTRY;
 105
 106         LASSERT(type == USRQUOTA || type == GRPQUOTA || type == UGQUOTA);
 107
 108         /* quota already turned on */
 109         if (obd->u.obt.obt_qctxt.lqc_status)
 110                 RETURN(0);
 111
 112         OBD_ALLOC_PTR(oqctl);
 113         if (!oqctl)
 114                 RETURN(-ENOMEM);
 115
 116         oqctl->qc_type = type;
 117         oqctl->qc_cmd = Q_QUOTAON;
 118         oqctl->qc_id = QFMT_LDISKFS;
 119
 120         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 121
 122         if (!is_master)
 123                 goto local_quota;
 124
 125         /* turn on cluster wide quota */
 126         rc = mds_admin_quota_on(obd, oqctl);
 127         if (rc) {
 128                 CDEBUG(rc == -ENOENT ? D_QUOTA : D_ERROR,
 129                        "auto-enable admin quota failed. rc=%d\n", rc);
 130                 GOTO(out_pop, rc);
 131         }
 132 local_quota:
 133         /* turn on local quota */
 134         rc = fsfilt_quotactl(obd, sb, oqctl);
 135         if (rc) {
 136                 CDEBUG(rc == -ENOENT ? D_QUOTA : D_ERROR,
 137                        "auto-enable local quota failed. rc=%d\n", rc);
 138                 if (is_master)
 139                         mds_quota_off(obd, oqctl);
 140         } else {
 141                 obd->u.obt.obt_qctxt.lqc_status = 1;
 142         }
 143 out_pop:
 144         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 145
 146         OBD_FREE_PTR(oqctl);
 147         RETURN(rc);
 148 }
 149
 150 int lprocfs_quota_wr_type(struct file *file, const char *buffer,
 151                           unsigned long count, void *data)
 152 {
 153         struct obd_device *obd = (struct obd_device *)data;
 154         struct obd_device_target *obt = &obd->u.obt;
 155         int type = 0;
 156         unsigned long i;
 157         char stype[MAX_STYPE_SIZE + 1] = "";
 158         LASSERT(obd != NULL);
 159
 160         if (count > MAX_STYPE_SIZE)
 161                 return -EINVAL;
 162
 163         if (copy_from_user(stype, buffer, count))
 164                 return -EFAULT;
 165
 166         for (i = 0 ; i < count ; i++) {
 167                 int rc;
 168
 169                 switch (stype[i]) {
 170                 case 'u' :
 171                         type |= USER_QUOTA;
 172                         break;
 173                 case 'g' :
 174                         type |= GROUP_QUOTA;
 175                         break;
 176                 /* quota version specifiers */
 177                 case '1' :
 178                         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
 179                                 break;
 180
 181                         rc = mds_quota_set_version(obd, LUSTRE_QUOTA_V1);
 182                         if (rc) {
 183                                 CDEBUG(D_QUOTA, "failed to set quota v1! %d\n", rc);
 184                                 return rc;
 185                         }
 186                         break;
 187                 case '2' :
 188                         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
 189                                 break;
 190
 191                         rc = mds_quota_set_version(obd, LUSTRE_QUOTA_V2);
 192                         if (rc) {
 193                                 CDEBUG(D_QUOTA, "could not set quota v2! %d\n", rc);
 194                                 return rc;
 195                         }
 196                         break;
 197                 default  : /* just skip stray symbols like \n */
 198                         break;
 199                 }
 200         }
 201
 202         obt->obt_qctxt.lqc_atype = type;
 203
 204         if (type == 0)
 205                 return count;
 206
 207         if (!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
 208                 auto_quota_on(obd, type - 1, obt->obt_sb, 1);
 209         else if (!strcmp(obd->obd_type->typ_name, LUSTRE_OST_NAME))
 210                 auto_quota_on(obd, type - 1, obt->obt_sb, 0);
 211         else
 212                 return -EFAULT;
 213
 214         return count;
 215 }
 216 EXPORT_SYMBOL(lprocfs_quota_wr_type);
 217
 218 #endif /* LPROCFS */
 219
 220 static int filter_quota_setup(struct obd_device *obd)
 221 {
 222         int rc = 0;
 223         struct obd_device_target *obt = &obd->u.obt;
 224         ENTRY;
 225
 226         atomic_set(&obt->obt_quotachecking, 1);
 227         rc = qctxt_init(&obt->obt_qctxt, obt->obt_sb, NULL);
 228         if (rc) {
 229                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 230                 RETURN(rc);
 231         }
 232         RETURN(rc);
 233 }
 234
 235 static int filter_quota_cleanup(struct obd_device *obd)
 236 {
 237         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 238         return 0;
 239 }
 240
 241 static int filter_quota_setinfo(struct obd_export *exp, struct obd_device *obd)
 242 {
 243         struct obd_import *imp;
 244
 245         /* setup the quota context import */
 246         spin_lock(&obd->u.obt.obt_qctxt.lqc_lock);
 247         obd->u.obt.obt_qctxt.lqc_import = exp->exp_imp_reverse;
 248         spin_unlock(&obd->u.obt.obt_qctxt.lqc_lock);
 249
 250         /* make imp's connect flags equal relative exp's connect flags
 251          * adding it to avoid the scan export list
 252          */
 253         imp = exp->exp_imp_reverse;
 254         if (imp)
 255                 imp->imp_connect_data.ocd_connect_flags |=
 256                         (exp->exp_connect_flags &
 257                          (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 258
 259         /* start quota slave recovery thread. (release high limits) */
 260         qslave_start_recovery(obd, &obd->u.obt.obt_qctxt);
 261         return 0;
 262 }
 263
 264 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 265 {
 266         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 267
 268         /* lquota may be not set up before destroying export, b=14896 */
 269         if (!obd->obd_set_up)
 270                 return 0;
 271
 272         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 273          * should be invalid b=12374 */
 274         if (qctxt->lqc_import == exp->exp_imp_reverse) {
 275                 spin_lock(&qctxt->lqc_lock);
 276                 qctxt->lqc_import = NULL;
 277                 spin_unlock(&qctxt->lqc_lock);
 278         }
 279
 280         return 0;
 281 }
 282
 283 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 284 {
 285         ENTRY;
 286
 287         if (!sb_any_quota_enabled(obd->u.obt.obt_sb))
 288                 RETURN(0);
 289
 290         if (ignore)
 291                 cap_raise(current->cap_effective, CAP_SYS_RESOURCE);
 292         else
 293                 cap_lower(current->cap_effective, CAP_SYS_RESOURCE);
 294
 295         RETURN(0);
 296 }
 297
 298 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 299 {
 300         struct obd_device_target *obt = &obd->u.obt;
 301         int err, cnt, rc = 0;
 302         struct obd_quotactl *oqctl;
 303         ENTRY;
 304
 305         if (!sb_any_quota_enabled(obt->obt_sb))
 306                 RETURN(0);
 307
 308         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 309
 310         OBD_ALLOC_PTR(oqctl);
 311         if (!oqctl) {
 312                 CERROR("Not enough memory!");
 313                 RETURN(-ENOMEM);
 314         }
 315
 316         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 317                 memset(oqctl, 0, sizeof(*oqctl));
 318
 319                 oqctl->qc_cmd = Q_GETQUOTA;
 320                 oqctl->qc_type = cnt;
 321                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 322                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 323                 if (err) {
 324                         if (!rc)
 325                                 rc = err;
 326                         continue;
 327                 }
 328
 329                 /* set over quota flags for a uid/gid */
 330                 oa->o_valid |= (cnt == USRQUOTA) ?
 331                                OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA;
 332                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 333                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 334                     oqctl->qc_dqblk.dqb_bhardlimit))
 335                         oa->o_flags |= (cnt == USRQUOTA) ?
 336                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 337         }
 338         OBD_FREE_PTR(oqctl);
 339         RETURN(rc);
 340 }
 341
 342 static int filter_quota_acquire(struct obd_device *obd, unsigned int uid,
 343                                 unsigned int gid)
 344 {
 345         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 346         int rc;
 347         ENTRY;
 348
 349         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, LQUOTA_FLAGS_BLK, 1);
 350         RETURN(rc);
 351 }
 352
 353 /* check whether the left quota of certain uid and gid can satisfy a block_write
 354  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA */
 355 static int quota_check_common(struct obd_device *obd, unsigned int uid,
 356                               unsigned int gid, int count, int cycle, int isblk)
 357 {
 358         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 359         int i;
 360         __u32 id[MAXQUOTAS] = { uid, gid };
 361         struct qunit_data qdata[MAXQUOTAS];
 362         int rc = 0, rc2[2] = { 0, 0 };
 363         ENTRY;
 364
 365         CLASSERT(MAXQUOTAS < 4);
 366         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 367                 RETURN(rc);
 368
 369         for (i = 0; i < MAXQUOTAS; i++) {
 370                 struct lustre_qunit_size *lqs = NULL;
 371
 372                 qdata[i].qd_id = id[i];
 373                 qdata[i].qd_flags = i;
 374                 if (isblk)
 375                         QDATA_SET_BLK(&qdata[i]);
 376                 qdata[i].qd_count = 0;
 377
 378                 /* ignore root user */
 379                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 380                         continue;
 381
 382                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 383                 if (!lqs)
 384                         continue;
 385
 386                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 387                 spin_lock(&lqs->lqs_lock);
 388                 if (!cycle) {
 389                         rc = QUOTA_RET_INC_PENDING;
 390                         if (isblk)
 391                                 lqs->lqs_bwrite_pending += count;
 392                         else
 393                                 lqs->lqs_iwrite_pending += count;
 394                 }
 395
 396                 CDEBUG(D_QUOTA, "write pending: %lu, qd_count: "LPU64".\n",
 397                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 398                        qdata[i].qd_count);
 399                 if (rc2[i] == QUOTA_RET_OK) {
 400                         if (isblk && qdata[i].qd_count <
 401                             lqs->lqs_bwrite_pending * CFS_PAGE_SIZE)
 402                                 rc2[i] = QUOTA_RET_ACQUOTA;
 403                         if (!isblk && qdata[i].qd_count <
 404                             lqs->lqs_iwrite_pending)
 405                                 rc2[i] = QUOTA_RET_ACQUOTA;
 406                 }
 407
 408                 spin_unlock(&lqs->lqs_lock);
 409
 410                 /* When cycle is zero, lqs_*_pending will be changed. We will
 411                  * get reference of the lqs here and put reference of lqs in
 412                  * quota_pending_commit b=14784 */
 413                 if (!cycle)
 414                         lqs_getref(lqs);
 415
 416                 /* this is for quota_search_lqs */
 417                 lqs_putref(lqs);
 418         }
 419
 420         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 421                 RETURN(rc | QUOTA_RET_ACQUOTA);
 422         else
 423                 RETURN(rc);
 424 }
 425
 426 static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid,
 427                                 unsigned int gid, int count, int *pending,
 428                                 int isblk, quota_acquire acquire)
 429 {
 430         int rc = 0, cycle = 0, count_err = 0;
 431         ENTRY;
 432
 433         /* Unfortunately, if quota master is too busy to handle the
 434          * pre-dqacq in time and quota hash on ost is used up, we
 435          * have to wait for the completion of in flight dqacq/dqrel,
 436          * in order to get enough quota for write b=12588 */
 437         while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk)) &
 438                QUOTA_RET_ACQUOTA) {
 439
 440                 if (rc & QUOTA_RET_INC_PENDING)
 441                         *pending = 1;
 442
 443                 cycle++;
 444                 if (isblk)
 445                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 446                 /* after acquire(), we should run quota_check_common again
 447                  * so that we confirm there are enough quota to finish write */
 448                 rc = acquire(obd, uid, gid);
 449
 450                 /* please reference to dqacq_completion for the below */
 451                 /* a new request is finished, try again */
 452                 if (rc == -EAGAIN) {
 453                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 454                         continue;
 455                 }
 456
 457                 /* it is out of quota already */
 458                 if (rc == -EDQUOT) {
 459                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 460                         break;
 461                 }
 462
 463                 /* -EBUSY and others, try 10 times */
 464                 if (rc < 0 && count_err < 10) {
 465                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc, count_err++);
 466                         cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE, HZ);
 467                         continue;
 468                 }
 469
 470                 if (count_err >= 10 || cycle >= 1000) {
 471                         CDEBUG(D_ERROR, "we meet 10 errors or run too many"
 472                                " cycles when acquiring quota, quit checking with"
 473                                " rc: %d, cycle: %d.\n", rc, cycle);
 474                         break;
 475                 }
 476
 477                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 478                        cycle);
 479         }
 480
 481         if (!cycle && rc & QUOTA_RET_INC_PENDING)
 482                 *pending = 1;
 483
 484         RETURN(rc);
 485 }
 486
 487
 488 static int filter_quota_check(struct obd_device *obd, unsigned int uid,
 489                               unsigned int gid, int npage, int *flag,
 490                               quota_acquire acquire)
 491 {
 492         return quota_chk_acq_common(obd, uid, gid, npage, flag, LQUOTA_FLAGS_BLK,
 493                                     acquire);
 494 }
 495
 496 /* when a block_write or inode_create rpc is finished, adjust the record for
 497  * pending blocks and inodes*/
 498 static int quota_pending_commit(struct obd_device *obd, unsigned int uid,
 499                                 unsigned int gid, int count, int isblk)
 500 {
 501         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 502         int i;
 503         __u32 id[MAXQUOTAS] = { uid, gid };
 504         struct qunit_data qdata[MAXQUOTAS];
 505         ENTRY;
 506
 507         CLASSERT(MAXQUOTAS < 4);
 508         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 509                 RETURN(0);
 510
 511         for (i = 0; i < MAXQUOTAS; i++) {
 512                 struct lustre_qunit_size *lqs = NULL;
 513
 514                 qdata[i].qd_id = id[i];
 515                 qdata[i].qd_flags = i;
 516                 if (isblk)
 517                         QDATA_SET_BLK(&qdata[i]);
 518                 qdata[i].qd_count = 0;
 519
 520                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 521                         continue;
 522
 523                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 524                 if (lqs) {
 525                         int flag = 0;
 526                         spin_lock(&lqs->lqs_lock);
 527                         CDEBUG(D_QUOTA, "pending: %lu, count: %d.\n",
 528                                isblk ? lqs->lqs_bwrite_pending :
 529                                lqs->lqs_iwrite_pending, count);
 530
 531                         if (isblk) {
 532                                 if (lqs->lqs_bwrite_pending >= count) {
 533                                         lqs->lqs_bwrite_pending -= count;
 534                                         flag = 1;
 535                                 } else {
 536                                         CDEBUG(D_ERROR,
 537                                                "there are too many blocks!\n");
 538                                 }
 539                         } else {
 540                                 if (lqs->lqs_iwrite_pending >= count) {
 541                                         lqs->lqs_iwrite_pending -= count;
 542                                         flag = 1;
 543                                 } else {
 544                                         CDEBUG(D_ERROR,
 545                                                "there are too many files!\n");
 546                                 }
 547                         }
 548
 549                         spin_unlock(&lqs->lqs_lock);
 550                         lqs_putref(lqs);
 551                         /* When lqs_*_pening is changed back, we'll putref lqs
 552                          * here b=14784 */
 553                         if (flag)
 554                                 lqs_putref(lqs);
 555                 }
 556         }
 557
 558         RETURN(0);
 559 }
 560
 561 static int filter_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 562                                        unsigned int gid, int npage)
 563 {
 564         return quota_pending_commit(obd, uid, gid, npage, LQUOTA_FLAGS_BLK);
 565 }
 566
 567 static int mds_quota_init(void)
 568 {
 569         return lustre_dquot_init();
 570 }
 571
 572 static int mds_quota_exit(void)
 573 {
 574         lustre_dquot_exit();
 575         return 0;
 576 }
 577
 578 static int mds_quota_setup(struct obd_device *obd)
 579 {
 580         struct obd_device_target *obt = &obd->u.obt;
 581         struct mds_obd *mds = &obd->u.mds;
 582         int rc;
 583         ENTRY;
 584
 585         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 586         atomic_set(&obt->obt_quotachecking, 1);
 587         /* initialize quota master and quota context */
 588         sema_init(&mds->mds_qonoff_sem, 1);
 589         rc = qctxt_init(&obt->obt_qctxt, obt->obt_sb, dqacq_handler);
 590         if (rc) {
 591                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 592                 RETURN(rc);
 593         }
 594         RETURN(rc);
 595 }
 596
 597 static int mds_quota_cleanup(struct obd_device *obd)
 598 {
 599         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 600         RETURN(0);
 601 }
 602
 603 static int mds_quota_fs_cleanup(struct obd_device *obd)
 604 {
 605         struct mds_obd *mds = &obd->u.mds;
 606         struct obd_quotactl oqctl;
 607         ENTRY;
 608
 609         memset(&oqctl, 0, sizeof(oqctl));
 610         oqctl.qc_type = UGQUOTA;
 611
 612         down(&mds->mds_qonoff_sem);
 613         mds_admin_quota_off(obd, &oqctl);
 614         up(&mds->mds_qonoff_sem);
 615         RETURN(0);
 616 }
 617
 618 static int mds_quota_check(struct obd_device *obd, unsigned int uid,
 619                            unsigned int gid, int inodes, int *flag,
 620                            quota_acquire acquire)
 621 {
 622         return quota_chk_acq_common(obd, uid, gid, inodes, flag, 0, acquire);
 623 }
 624
 625 static int mds_quota_acquire(struct obd_device *obd, unsigned int uid,
 626                              unsigned int gid)
 627 {
 628         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 629         int rc;
 630         ENTRY;
 631
 632         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, 0, 1);
 633         RETURN(rc);
 634 }
 635
 636 static int mds_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 637                                     unsigned int gid, int inodes)
 638 {
 639         return quota_pending_commit(obd, uid, gid, inodes, 0);
 640 }
 641 #endif /* __KERNEL__ */
 642
 643 struct osc_quota_info {
 644         struct list_head        oqi_hash;       /* hash list */
 645         struct client_obd      *oqi_cli;        /* osc obd */
 646         unsigned int            oqi_id;         /* uid/gid of a file */
 647         short                   oqi_type;       /* quota type */
 648 };
 649
 650 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 651
 652 static struct list_head qinfo_hash[NR_DQHASH];
 653 /* SLAB cache for client quota context */
 654 cfs_mem_cache_t *qinfo_cachep = NULL;
 655
 656 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 657                          __attribute__((__const__));
 658
 659 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 660 {
 661         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 662         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 663         return tmp;
 664 }
 665
 666 /* caller must hold qinfo_list_lock */
 667 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 668 {
 669         struct list_head *head = qinfo_hash +
 670                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 671
 672         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 673         list_add(&oqi->oqi_hash, head);
 674 }
 675
 676 /* caller must hold qinfo_list_lock */
 677 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 678 {
 679         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 680         list_del_init(&oqi->oqi_hash);
 681 }
 682
 683 /* caller must hold qinfo_list_lock */
 684 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 685                                                 unsigned int id, int type)
 686 {
 687         unsigned int hashent = hashfn(cli, id, type);
 688         struct osc_quota_info *oqi;
 689
 690         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 691         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 692                 if (oqi->oqi_cli == cli &&
 693                     oqi->oqi_id == id && oqi->oqi_type == type)
 694                         return oqi;
 695         }
 696         return NULL;
 697 }
 698
 699 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 700                                           unsigned int id, int type)
 701 {
 702         struct osc_quota_info *oqi;
 703         ENTRY;
 704
 705         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 706         if(!oqi)
 707                 RETURN(NULL);
 708
 709         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 710         oqi->oqi_cli = cli;
 711         oqi->oqi_id = id;
 712         oqi->oqi_type = type;
 713
 714         RETURN(oqi);
 715 }
 716
 717 static void free_qinfo(struct osc_quota_info *oqi)
 718 {
 719         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 720 }
 721
 722 int osc_quota_chkdq(struct client_obd *cli, unsigned int uid, unsigned int gid)
 723 {
 724         unsigned int id;
 725         int cnt, rc = QUOTA_OK;
 726         ENTRY;
 727
 728         spin_lock(&qinfo_list_lock);
 729         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 730                 struct osc_quota_info *oqi = NULL;
 731
 732                 id = (cnt == USRQUOTA) ? uid : gid;
 733                 oqi = find_qinfo(cli, id, cnt);
 734                 if (oqi) {
 735                         rc = NO_QUOTA;
 736                         break;
 737                 }
 738         }
 739         spin_unlock(&qinfo_list_lock);
 740
 741         RETURN(rc);
 742 }
 743
 744 int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid,
 745                     obd_flag valid, obd_flag flags)
 746 {
 747         unsigned int id;
 748         obd_flag noquota;
 749         int cnt, rc = 0;
 750         ENTRY;
 751
 752
 753         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 754                 struct osc_quota_info *oqi, *old;
 755
 756                 if (!(valid & ((cnt == USRQUOTA) ?
 757                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 758                         continue;
 759
 760                 id = (cnt == USRQUOTA) ? uid : gid;
 761                 noquota = (cnt == USRQUOTA) ?
 762                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 763
 764                 oqi = alloc_qinfo(cli, id, cnt);
 765                 if (oqi) {
 766                         spin_lock(&qinfo_list_lock);
 767
 768                         old = find_qinfo(cli, id, cnt);
 769                         if (old && !noquota)
 770                                 remove_qinfo_hash(old);
 771                         else if (!old && noquota)
 772                                 insert_qinfo_hash(oqi);
 773
 774                         spin_unlock(&qinfo_list_lock);
 775
 776                         if (old || !noquota)
 777                                 free_qinfo(oqi);
 778                         if (old && !noquota)
 779                                 free_qinfo(old);
 780                 } else {
 781                         CERROR("not enough mem!\n");
 782                         rc = -ENOMEM;
 783                         break;
 784                 }
 785         }
 786
 787         RETURN(rc);
 788 }
 789
 790 int osc_quota_cleanup(struct obd_device *obd)
 791 {
 792         struct client_obd *cli = &obd->u.cli;
 793         struct osc_quota_info *oqi, *n;
 794         int i;
 795         ENTRY;
 796
 797         spin_lock(&qinfo_list_lock);
 798         for (i = 0; i < NR_DQHASH; i++) {
 799                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 800                         if (oqi->oqi_cli != cli)
 801                                 continue;
 802                         remove_qinfo_hash(oqi);
 803                         free_qinfo(oqi);
 804                 }
 805         }
 806         spin_unlock(&qinfo_list_lock);
 807
 808         RETURN(0);
 809 }
 810
 811 int osc_quota_init(void)
 812 {
 813         int i;
 814         ENTRY;
 815
 816         LASSERT(qinfo_cachep == NULL);
 817         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 818                                             sizeof(struct osc_quota_info),
 819                                             0, 0);
 820         if (!qinfo_cachep)
 821                 RETURN(-ENOMEM);
 822
 823         for (i = 0; i < NR_DQHASH; i++)
 824                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 825
 826         RETURN(0);
 827 }
 828
 829 int osc_quota_exit(void)
 830 {
 831         struct osc_quota_info *oqi, *n;
 832         int i, rc;
 833         ENTRY;
 834
 835         spin_lock(&qinfo_list_lock);
 836         for (i = 0; i < NR_DQHASH; i++) {
 837                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 838                         remove_qinfo_hash(oqi);
 839                         free_qinfo(oqi);
 840                 }
 841         }
 842         spin_unlock(&qinfo_list_lock);
 843
 844         rc = cfs_mem_cache_destroy(qinfo_cachep);
 845         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 846         qinfo_cachep = NULL;
 847
 848         RETURN(0);
 849 }
 850
 851 #ifdef __KERNEL__
 852 quota_interface_t mds_quota_interface = {
 853         .quota_init     = mds_quota_init,
 854         .quota_exit     = mds_quota_exit,
 855         .quota_setup    = mds_quota_setup,
 856         .quota_cleanup  = mds_quota_cleanup,
 857         .quota_check    = target_quota_check,
 858         .quota_ctl      = mds_quota_ctl,
 859         .quota_fs_cleanup       =mds_quota_fs_cleanup,
 860         .quota_recovery = mds_quota_recovery,
 861         .quota_adjust   = mds_quota_adjust,
 862         .quota_chkquota = mds_quota_check,
 863         .quota_acquire  = mds_quota_acquire,
 864         .quota_pending_commit = mds_quota_pending_commit,
 865 };
 866
 867 quota_interface_t filter_quota_interface = {
 868         .quota_setup    = filter_quota_setup,
 869         .quota_cleanup  = filter_quota_cleanup,
 870         .quota_check    = target_quota_check,
 871         .quota_ctl      = filter_quota_ctl,
 872         .quota_setinfo  = filter_quota_setinfo,
 873         .quota_clearinfo = filter_quota_clearinfo,
 874         .quota_enforce  = filter_quota_enforce,
 875         .quota_getflag  = filter_quota_getflag,
 876         .quota_acquire  = filter_quota_acquire,
 877         .quota_adjust   = filter_quota_adjust,
 878         .quota_chkquota = filter_quota_check,
 879         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 880         .quota_pending_commit = filter_quota_pending_commit,
 881 };
 882 #endif /* __KERNEL__ */
 883
 884 quota_interface_t mdc_quota_interface = {
 885         .quota_ctl      = client_quota_ctl,
 886         .quota_check    = client_quota_check,
 887         .quota_poll_check = client_quota_poll_check,
 888 };
 889
 890 quota_interface_t osc_quota_interface = {
 891         .quota_ctl      = client_quota_ctl,
 892         .quota_check    = client_quota_check,
 893         .quota_poll_check = client_quota_poll_check,
 894         .quota_init     = osc_quota_init,
 895         .quota_exit     = osc_quota_exit,
 896         .quota_chkdq    = osc_quota_chkdq,
 897         .quota_setdq    = osc_quota_setdq,
 898         .quota_cleanup  = osc_quota_cleanup,
 899         .quota_adjust_qunit = client_quota_adjust_qunit,
 900 };
 901
 902 quota_interface_t lov_quota_interface = {
 903         .quota_check    = lov_quota_check,
 904         .quota_ctl      = lov_quota_ctl,
 905         .quota_adjust_qunit = lov_quota_adjust_qunit,
 906 };
 907
 908 #ifdef __KERNEL__
 909 static int __init init_lustre_quota(void)
 910 {
 911         int rc = qunit_cache_init();
 912         if (rc)
 913                 return rc;
 914         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 915         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 916         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
 917         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
 918         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
 919         return 0;
 920 }
 921
 922 static void /*__exit*/ exit_lustre_quota(void)
 923 {
 924         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 925         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 926         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
 927         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
 928         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
 929
 930         qunit_cache_cleanup();
 931 }
 932
 933 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 934 MODULE_DESCRIPTION("Lustre Quota");
 935 MODULE_LICENSE("GPL");
 936
 937 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 938
 939 EXPORT_SYMBOL(mds_quota_interface);
 940 EXPORT_SYMBOL(filter_quota_interface);
 941 EXPORT_SYMBOL(mdc_quota_interface);
 942 EXPORT_SYMBOL(osc_quota_interface);
 943 EXPORT_SYMBOL(lov_quota_interface);
 944 #endif /* __KERNEL */