fs/gfs2/quota.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 /*
  11  * Quota change tags are associated with each transaction that allocates or
  12  * deallocates space.  Those changes are accumulated locally to each node (in a
  13  * per-node file) and then are periodically synced to the quota file.  This
  14  * avoids the bottleneck of constantly touching the quota file, but introduces
  15  * fuzziness in the current usage value of IDs that are being used on different
  16  * nodes in the cluster simultaneously.  So, it is possible for a user on
  17  * multiple nodes to overrun their quota, but that overrun is controlable.
  18  * Since quota tags are part of transactions, there is no need for a quota check
  19  * program to be run on node crashes or anything like that.
  20  *
  21  * There are couple of knobs that let the administrator manage the quota
  22  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
  23  * sitting on one node before being synced to the quota file.  (The default is
  24  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
  25  * of quota file syncs increases as the user moves closer to their limit.  The
  26  * more frequent the syncs, the more accurate the quota enforcement, but that
  27  * means that there is more contention between the nodes for the quota file.
  28  * The default value is one.  This sets the maximum theoretical quota overrun
  29  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
  30  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
  31  * number greater than one makes quota syncs more frequent and reduces the
  32  * maximum overrun.  Numbers less than one (but greater than zero) make quota
  33  * syncs less frequent.
  34  *
  35  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
  36  * the quota file, so it is not being constantly read.
  37  */
  38
  39 #include <linux/sched.h>
  40 #include <linux/slab.h>
  41 #include <linux/mm.h>
  42 #include <linux/spinlock.h>
  43 #include <linux/completion.h>
  44 #include <linux/buffer_head.h>
  45 #include <linux/sort.h>
  46 #include <linux/fs.h>
  47 #include <linux/bio.h>
  48 #include <linux/gfs2_ondisk.h>
  49 #include <linux/kthread.h>
  50 #include <linux/freezer.h>
  51 #include <linux/quota.h>
  52 #include <linux/dqblk_xfs.h>
  53 #include <linux/lockref.h>
  54 #include <linux/list_lru.h>
  55 #include <linux/rcupdate.h>
  56 #include <linux/rculist_bl.h>
  57 #include <linux/bit_spinlock.h>
  58 #include <linux/jhash.h>
  59
  60 #include "gfs2.h"
  61 #include "incore.h"
  62 #include "bmap.h"
  63 #include "glock.h"
  64 #include "glops.h"
  65 #include "log.h"
  66 #include "meta_io.h"
  67 #include "quota.h"
  68 #include "rgrp.h"
  69 #include "super.h"
  70 #include "trans.h"
  71 #include "inode.h"
  72 #include "util.h"
  73
  74 #define GFS2_QD_HASH_SHIFT      12
  75 #define GFS2_QD_HASH_SIZE       (1 << GFS2_QD_HASH_SHIFT)
  76 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
  77
  78 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
  79 static DEFINE_SPINLOCK(qd_lock);
  80 struct list_lru gfs2_qd_lru;
  81
  82 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
  83
  84 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
  85                                  const struct kqid qid)
  86 {
  87         unsigned int h;
  88
  89         h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
  90         h = jhash(&qid, sizeof(struct kqid), h);
  91
  92         return h & GFS2_QD_HASH_MASK;
  93 }
  94
  95 static inline void spin_lock_bucket(unsigned int hash)
  96 {
  97         hlist_bl_lock(&qd_hash_table[hash]);
  98 }
  99
 100 static inline void spin_unlock_bucket(unsigned int hash)
 101 {
 102         hlist_bl_unlock(&qd_hash_table[hash]);
 103 }
 104
 105 static void gfs2_qd_dealloc(struct rcu_head *rcu)
 106 {
 107         struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
 108         kmem_cache_free(gfs2_quotad_cachep, qd);
 109 }
 110
 111 static void gfs2_qd_dispose(struct list_head *list)
 112 {
 113         struct gfs2_quota_data *qd;
 114         struct gfs2_sbd *sdp;
 115
 116         while (!list_empty(list)) {
 117                 qd = list_entry(list->next, struct gfs2_quota_data, qd_lru);
 118                 sdp = qd->qd_gl->gl_sbd;
 119
 120                 list_del(&qd->qd_lru);
 121
 122                 /* Free from the filesystem-specific list */
 123                 spin_lock(&qd_lock);
 124                 list_del(&qd->qd_list);
 125                 spin_unlock(&qd_lock);
 126
 127                 spin_lock_bucket(qd->qd_hash);
 128                 hlist_bl_del_rcu(&qd->qd_hlist);
 129                 spin_unlock_bucket(qd->qd_hash);
 130
 131                 gfs2_assert_warn(sdp, !qd->qd_change);
 132                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
 133                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
 134
 135                 gfs2_glock_put(qd->qd_gl);
 136                 atomic_dec(&sdp->sd_quota_count);
 137
 138                 /* Delete it from the common reclaim list */
 139                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
 140         }
 141 }
 142
 143
 144 static enum lru_status gfs2_qd_isolate(struct list_head *item, spinlock_t *lock, void *arg)
 145 {
 146         struct list_head *dispose = arg;
 147         struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
 148
 149         if (!spin_trylock(&qd->qd_lockref.lock))
 150                 return LRU_SKIP;
 151
 152         if (qd->qd_lockref.count == 0) {
 153                 lockref_mark_dead(&qd->qd_lockref);
 154                 list_move(&qd->qd_lru, dispose);
 155         }
 156
 157         spin_unlock(&qd->qd_lockref.lock);
 158         return LRU_REMOVED;
 159 }
 160
 161 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
 162                                          struct shrink_control *sc)
 163 {
 164         LIST_HEAD(dispose);
 165         unsigned long freed;
 166
 167         if (!(sc->gfp_mask & __GFP_FS))
 168                 return SHRINK_STOP;
 169
 170         freed = list_lru_walk_node(&gfs2_qd_lru, sc->nid, gfs2_qd_isolate,
 171                                    &dispose, &sc->nr_to_scan);
 172
 173         gfs2_qd_dispose(&dispose);
 174
 175         return freed;
 176 }
 177
 178 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
 179                                           struct shrink_control *sc)
 180 {
 181         return vfs_pressure_ratio(list_lru_count_node(&gfs2_qd_lru, sc->nid));
 182 }
 183
 184 struct shrinker gfs2_qd_shrinker = {
 185         .count_objects = gfs2_qd_shrink_count,
 186         .scan_objects = gfs2_qd_shrink_scan,
 187         .seeks = DEFAULT_SEEKS,
 188         .flags = SHRINKER_NUMA_AWARE,
 189 };
 190
 191
 192 static u64 qd2index(struct gfs2_quota_data *qd)
 193 {
 194         struct kqid qid = qd->qd_id;
 195         return (2 * (u64)from_kqid(&init_user_ns, qid)) +
 196                 ((qid.type == USRQUOTA) ? 0 : 1);
 197 }
 198
 199 static u64 qd2offset(struct gfs2_quota_data *qd)
 200 {
 201         u64 offset;
 202
 203         offset = qd2index(qd);
 204         offset *= sizeof(struct gfs2_quota);
 205
 206         return offset;
 207 }
 208
 209 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
 210 {
 211         struct gfs2_quota_data *qd;
 212         int error;
 213
 214         qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
 215         if (!qd)
 216                 return NULL;
 217
 218         qd->qd_sbd = sdp;
 219         qd->qd_lockref.count = 1;
 220         spin_lock_init(&qd->qd_lockref.lock);
 221         qd->qd_id = qid;
 222         qd->qd_slot = -1;
 223         INIT_LIST_HEAD(&qd->qd_lru);
 224         qd->qd_hash = hash;
 225
 226         error = gfs2_glock_get(sdp, qd2index(qd),
 227                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
 228         if (error)
 229                 goto fail;
 230
 231         return qd;
 232
 233 fail:
 234         kmem_cache_free(gfs2_quotad_cachep, qd);
 235         return NULL;
 236 }
 237
 238 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
 239                                                      const struct gfs2_sbd *sdp,
 240                                                      struct kqid qid)
 241 {
 242         struct gfs2_quota_data *qd;
 243         struct hlist_bl_node *h;
 244
 245         hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
 246                 if (!qid_eq(qd->qd_id, qid))
 247                         continue;
 248                 if (qd->qd_sbd != sdp)
 249                         continue;
 250                 if (lockref_get_not_dead(&qd->qd_lockref)) {
 251                         list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
 252                         return qd;
 253                 }
 254         }
 255
 256         return NULL;
 257 }
 258
 259
 260 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
 261                   struct gfs2_quota_data **qdp)
 262 {
 263         struct gfs2_quota_data *qd, *new_qd;
 264         unsigned int hash = gfs2_qd_hash(sdp, qid);
 265
 266         rcu_read_lock();
 267         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
 268         rcu_read_unlock();
 269
 270         if (qd)
 271                 return 0;
 272
 273         new_qd = qd_alloc(hash, sdp, qid);
 274         if (!new_qd)
 275                 return -ENOMEM;
 276
 277         spin_lock(&qd_lock);
 278         spin_lock_bucket(hash);
 279         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
 280         if (qd == NULL) {
 281                 *qdp = new_qd;
 282                 list_add(&new_qd->qd_list, &sdp->sd_quota_list);
 283                 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
 284                 atomic_inc(&sdp->sd_quota_count);
 285         }
 286         spin_unlock_bucket(hash);
 287         spin_unlock(&qd_lock);
 288
 289         if (qd) {
 290                 gfs2_glock_put(new_qd->qd_gl);
 291                 kmem_cache_free(gfs2_quotad_cachep, new_qd);
 292         }
 293
 294         return 0;
 295 }
 296
 297
 298 static void qd_hold(struct gfs2_quota_data *qd)
 299 {
 300         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 301         gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
 302         lockref_get(&qd->qd_lockref);
 303 }
 304
 305 static void qd_put(struct gfs2_quota_data *qd)
 306 {
 307         if (lockref_put_or_lock(&qd->qd_lockref))
 308                 return;
 309
 310         qd->qd_lockref.count = 0;
 311         list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
 312         spin_unlock(&qd->qd_lockref.lock);
 313
 314 }
 315
 316 static int slot_get(struct gfs2_quota_data *qd)
 317 {
 318         struct gfs2_sbd *sdp = qd->qd_sbd;
 319         unsigned int bit;
 320         int error = 0;
 321
 322         spin_lock(&qd_lock);
 323         if (qd->qd_slot_count != 0)
 324                 goto out;
 325
 326         error = -ENOSPC;
 327         bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
 328         if (bit < sdp->sd_quota_slots) {
 329                 set_bit(bit, sdp->sd_quota_bitmap);
 330                 qd->qd_slot = bit;
 331 out:
 332                 qd->qd_slot_count++;
 333         }
 334         spin_unlock(&qd_lock);
 335
 336         return error;
 337 }
 338
 339 static void slot_hold(struct gfs2_quota_data *qd)
 340 {
 341         struct gfs2_sbd *sdp = qd->qd_sbd;
 342
 343         spin_lock(&qd_lock);
 344         gfs2_assert(sdp, qd->qd_slot_count);
 345         qd->qd_slot_count++;
 346         spin_unlock(&qd_lock);
 347 }
 348
 349 static void slot_put(struct gfs2_quota_data *qd)
 350 {
 351         struct gfs2_sbd *sdp = qd->qd_sbd;
 352
 353         spin_lock(&qd_lock);
 354         gfs2_assert(sdp, qd->qd_slot_count);
 355         if (!--qd->qd_slot_count) {
 356                 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
 357                 qd->qd_slot = -1;
 358         }
 359         spin_unlock(&qd_lock);
 360 }
 361
 362 static int bh_get(struct gfs2_quota_data *qd)
 363 {
 364         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 365         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 366         unsigned int block, offset;
 367         struct buffer_head *bh;
 368         int error;
 369         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 370
 371         mutex_lock(&sdp->sd_quota_mutex);
 372
 373         if (qd->qd_bh_count++) {
 374                 mutex_unlock(&sdp->sd_quota_mutex);
 375                 return 0;
 376         }
 377
 378         block = qd->qd_slot / sdp->sd_qc_per_block;
 379         offset = qd->qd_slot % sdp->sd_qc_per_block;
 380
 381         bh_map.b_size = 1 << ip->i_inode.i_blkbits;
 382         error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
 383         if (error)
 384                 goto fail;
 385         error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
 386         if (error)
 387                 goto fail;
 388         error = -EIO;
 389         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
 390                 goto fail_brelse;
 391
 392         qd->qd_bh = bh;
 393         qd->qd_bh_qc = (struct gfs2_quota_change *)
 394                 (bh->b_data + sizeof(struct gfs2_meta_header) +
 395                  offset * sizeof(struct gfs2_quota_change));
 396
 397         mutex_unlock(&sdp->sd_quota_mutex);
 398
 399         return 0;
 400
 401 fail_brelse:
 402         brelse(bh);
 403 fail:
 404         qd->qd_bh_count--;
 405         mutex_unlock(&sdp->sd_quota_mutex);
 406         return error;
 407 }
 408
 409 static void bh_put(struct gfs2_quota_data *qd)
 410 {
 411         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 412
 413         mutex_lock(&sdp->sd_quota_mutex);
 414         gfs2_assert(sdp, qd->qd_bh_count);
 415         if (!--qd->qd_bh_count) {
 416                 brelse(qd->qd_bh);
 417                 qd->qd_bh = NULL;
 418                 qd->qd_bh_qc = NULL;
 419         }
 420         mutex_unlock(&sdp->sd_quota_mutex);
 421 }
 422
 423 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
 424                          u64 *sync_gen)
 425 {
 426         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 427             !test_bit(QDF_CHANGE, &qd->qd_flags) ||
 428             (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
 429                 return 0;
 430
 431         if (!lockref_get_not_dead(&qd->qd_lockref))
 432                 return 0;
 433
 434         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 435         set_bit(QDF_LOCKED, &qd->qd_flags);
 436         qd->qd_change_sync = qd->qd_change;
 437         gfs2_assert_warn(sdp, qd->qd_slot_count);
 438         qd->qd_slot_count++;
 439         return 1;
 440 }
 441
 442 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 443 {
 444         struct gfs2_quota_data *qd = NULL;
 445         int error;
 446         int found = 0;
 447
 448         *qdp = NULL;
 449
 450         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 451                 return 0;
 452
 453         spin_lock(&qd_lock);
 454
 455         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 456                 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
 457                 if (found)
 458                         break;
 459         }
 460
 461         if (!found)
 462                 qd = NULL;
 463
 464         spin_unlock(&qd_lock);
 465
 466         if (qd) {
 467                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 468                 error = bh_get(qd);
 469                 if (error) {
 470                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 471                         slot_put(qd);
 472                         qd_put(qd);
 473                         return error;
 474                 }
 475         }
 476
 477         *qdp = qd;
 478
 479         return 0;
 480 }
 481
 482 static void qd_unlock(struct gfs2_quota_data *qd)
 483 {
 484         gfs2_assert_warn(qd->qd_gl->gl_sbd,
 485                          test_bit(QDF_LOCKED, &qd->qd_flags));
 486         clear_bit(QDF_LOCKED, &qd->qd_flags);
 487         bh_put(qd);
 488         slot_put(qd);
 489         qd_put(qd);
 490 }
 491
 492 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
 493                     struct gfs2_quota_data **qdp)
 494 {
 495         int error;
 496
 497         error = qd_get(sdp, qid, qdp);
 498         if (error)
 499                 return error;
 500
 501         error = slot_get(*qdp);
 502         if (error)
 503                 goto fail;
 504
 505         error = bh_get(*qdp);
 506         if (error)
 507                 goto fail_slot;
 508
 509         return 0;
 510
 511 fail_slot:
 512         slot_put(*qdp);
 513 fail:
 514         qd_put(*qdp);
 515         return error;
 516 }
 517
 518 static void qdsb_put(struct gfs2_quota_data *qd)
 519 {
 520         bh_put(qd);
 521         slot_put(qd);
 522         qd_put(qd);
 523 }
 524
 525 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
 526 {
 527         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 528         struct gfs2_quota_data **qd;
 529         int error;
 530
 531         if (ip->i_res == NULL) {
 532                 error = gfs2_rs_alloc(ip);
 533                 if (error)
 534                         return error;
 535         }
 536
 537         qd = ip->i_res->rs_qa_qd;
 538
 539         if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
 540             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 541                 return -EIO;
 542
 543         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 544                 return 0;
 545
 546         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
 547         if (error)
 548                 goto out;
 549         ip->i_res->rs_qa_qd_num++;
 550         qd++;
 551
 552         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
 553         if (error)
 554                 goto out;
 555         ip->i_res->rs_qa_qd_num++;
 556         qd++;
 557
 558         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
 559             !uid_eq(uid, ip->i_inode.i_uid)) {
 560                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
 561                 if (error)
 562                         goto out;
 563                 ip->i_res->rs_qa_qd_num++;
 564                 qd++;
 565         }
 566
 567         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
 568             !gid_eq(gid, ip->i_inode.i_gid)) {
 569                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
 570                 if (error)
 571                         goto out;
 572                 ip->i_res->rs_qa_qd_num++;
 573                 qd++;
 574         }
 575
 576 out:
 577         if (error)
 578                 gfs2_quota_unhold(ip);
 579         return error;
 580 }
 581
 582 void gfs2_quota_unhold(struct gfs2_inode *ip)
 583 {
 584         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 585         unsigned int x;
 586
 587         if (ip->i_res == NULL)
 588                 return;
 589         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
 590
 591         for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
 592                 qdsb_put(ip->i_res->rs_qa_qd[x]);
 593                 ip->i_res->rs_qa_qd[x] = NULL;
 594         }
 595         ip->i_res->rs_qa_qd_num = 0;
 596 }
 597
 598 static int sort_qd(const void *a, const void *b)
 599 {
 600         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
 601         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
 602
 603         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
 604                 return -1;
 605         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
 606                 return 1;
 607         return 0;
 608 }
 609
 610 static void do_qc(struct gfs2_quota_data *qd, s64 change)
 611 {
 612         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 613         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 614         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 615         s64 x;
 616
 617         mutex_lock(&sdp->sd_quota_mutex);
 618         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
 619
 620         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
 621                 qc->qc_change = 0;
 622                 qc->qc_flags = 0;
 623                 if (qd->qd_id.type == USRQUOTA)
 624                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
 625                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
 626         }
 627
 628         x = be64_to_cpu(qc->qc_change) + change;
 629         qc->qc_change = cpu_to_be64(x);
 630
 631         spin_lock(&qd_lock);
 632         qd->qd_change = x;
 633         spin_unlock(&qd_lock);
 634
 635         if (!x) {
 636                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
 637                 clear_bit(QDF_CHANGE, &qd->qd_flags);
 638                 qc->qc_flags = 0;
 639                 qc->qc_id = 0;
 640                 slot_put(qd);
 641                 qd_put(qd);
 642         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
 643                 qd_hold(qd);
 644                 slot_hold(qd);
 645         }
 646
 647         mutex_unlock(&sdp->sd_quota_mutex);
 648 }
 649
 650 /**
 651  * gfs2_adjust_quota - adjust record of current block usage
 652  * @ip: The quota inode
 653  * @loc: Offset of the entry in the quota file
 654  * @change: The amount of usage change to record
 655  * @qd: The quota data
 656  * @fdq: The updated limits to record
 657  *
 658  * This function was mostly borrowed from gfs2_block_truncate_page which was
 659  * in turn mostly borrowed from ext3
 660  *
 661  * Returns: 0 or -ve on error
 662  */
 663
 664 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
 665                              s64 change, struct gfs2_quota_data *qd,
 666                              struct fs_disk_quota *fdq)
 667 {
 668         struct inode *inode = &ip->i_inode;
 669         struct gfs2_sbd *sdp = GFS2_SB(inode);
 670         struct address_space *mapping = inode->i_mapping;
 671         unsigned long index = loc >> PAGE_CACHE_SHIFT;
 672         unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
 673         unsigned blocksize, iblock, pos;
 674         struct buffer_head *bh;
 675         struct page *page;
 676         void *kaddr, *ptr;
 677         struct gfs2_quota q;
 678         int err, nbytes;
 679         u64 size;
 680
 681         if (gfs2_is_stuffed(ip)) {
 682                 err = gfs2_unstuff_dinode(ip, NULL);
 683                 if (err)
 684                         return err;
 685         }
 686
 687         memset(&q, 0, sizeof(struct gfs2_quota));
 688         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
 689         if (err < 0)
 690                 return err;
 691
 692         err = -EIO;
 693         be64_add_cpu(&q.qu_value, change);
 694         qd->qd_qb.qb_value = q.qu_value;
 695         if (fdq) {
 696                 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
 697                         q.qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
 698                         qd->qd_qb.qb_warn = q.qu_warn;
 699                 }
 700                 if (fdq->d_fieldmask & FS_DQ_BHARD) {
 701                         q.qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
 702                         qd->qd_qb.qb_limit = q.qu_limit;
 703                 }
 704                 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
 705                         q.qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
 706                         qd->qd_qb.qb_value = q.qu_value;
 707                 }
 708         }
 709
 710         /* Write the quota into the quota file on disk */
 711         ptr = &q;
 712         nbytes = sizeof(struct gfs2_quota);
 713 get_a_page:
 714         page = find_or_create_page(mapping, index, GFP_NOFS);
 715         if (!page)
 716                 return -ENOMEM;
 717
 718         blocksize = inode->i_sb->s_blocksize;
 719         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 720
 721         if (!page_has_buffers(page))
 722                 create_empty_buffers(page, blocksize, 0);
 723
 724         bh = page_buffers(page);
 725         pos = blocksize;
 726         while (offset >= pos) {
 727                 bh = bh->b_this_page;
 728                 iblock++;
 729                 pos += blocksize;
 730         }
 731
 732         if (!buffer_mapped(bh)) {
 733                 gfs2_block_map(inode, iblock, bh, 1);
 734                 if (!buffer_mapped(bh))
 735                         goto unlock_out;
 736                 /* If it's a newly allocated disk block for quota, zero it */
 737                 if (buffer_new(bh))
 738                         zero_user(page, pos - blocksize, bh->b_size);
 739         }
 740
 741         if (PageUptodate(page))
 742                 set_buffer_uptodate(bh);
 743
 744         if (!buffer_uptodate(bh)) {
 745                 ll_rw_block(READ | REQ_META, 1, &bh);
 746                 wait_on_buffer(bh);
 747                 if (!buffer_uptodate(bh))
 748                         goto unlock_out;
 749         }
 750
 751         gfs2_trans_add_data(ip->i_gl, bh);
 752
 753         kaddr = kmap_atomic(page);
 754         if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
 755                 nbytes = PAGE_CACHE_SIZE - offset;
 756         memcpy(kaddr + offset, ptr, nbytes);
 757         flush_dcache_page(page);
 758         kunmap_atomic(kaddr);
 759         unlock_page(page);
 760         page_cache_release(page);
 761
 762         /* If quota straddles page boundary, we need to update the rest of the
 763          * quota at the beginning of the next page */
 764         if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
 765                 ptr = ptr + nbytes;
 766                 nbytes = sizeof(struct gfs2_quota) - nbytes;
 767                 offset = 0;
 768                 index++;
 769                 goto get_a_page;
 770         }
 771
 772         size = loc + sizeof(struct gfs2_quota);
 773         if (size > inode->i_size)
 774                 i_size_write(inode, size);
 775         inode->i_mtime = inode->i_atime = CURRENT_TIME;
 776         mark_inode_dirty(inode);
 777         return 0;
 778
 779 unlock_out:
 780         unlock_page(page);
 781         page_cache_release(page);
 782         return err;
 783 }
 784
 785 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 786 {
 787         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
 788         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 789         struct gfs2_alloc_parms ap = { .aflags = 0, };
 790         unsigned int data_blocks, ind_blocks;
 791         struct gfs2_holder *ghs, i_gh;
 792         unsigned int qx, x;
 793         struct gfs2_quota_data *qd;
 794         unsigned reserved;
 795         loff_t offset;
 796         unsigned int nalloc = 0, blocks;
 797         int error;
 798
 799         error = gfs2_rs_alloc(ip);
 800         if (error)
 801                 return error;
 802
 803         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
 804                               &data_blocks, &ind_blocks);
 805
 806         ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
 807         if (!ghs)
 808                 return -ENOMEM;
 809
 810         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 811         mutex_lock(&ip->i_inode.i_mutex);
 812         for (qx = 0; qx < num_qd; qx++) {
 813                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
 814                                            GL_NOCACHE, &ghs[qx]);
 815                 if (error)
 816                         goto out;
 817         }
 818
 819         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
 820         if (error)
 821                 goto out;
 822
 823         for (x = 0; x < num_qd; x++) {
 824                 offset = qd2offset(qda[x]);
 825                 if (gfs2_write_alloc_required(ip, offset,
 826                                               sizeof(struct gfs2_quota)))
 827                         nalloc++;
 828         }
 829
 830         /*
 831          * 1 blk for unstuffing inode if stuffed. We add this extra
 832          * block to the reservation unconditionally. If the inode
 833          * doesn't need unstuffing, the block will be released to the
 834          * rgrp since it won't be allocated during the transaction
 835          */
 836         /* +3 in the end for unstuffing block, inode size update block
 837          * and another block in case quota straddles page boundary and
 838          * two blocks need to be updated instead of 1 */
 839         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
 840
 841         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
 842         ap.target = reserved;
 843         error = gfs2_inplace_reserve(ip, &ap);
 844         if (error)
 845                 goto out_alloc;
 846
 847         if (nalloc)
 848                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
 849
 850         error = gfs2_trans_begin(sdp, blocks, 0);
 851         if (error)
 852                 goto out_ipres;
 853
 854         for (x = 0; x < num_qd; x++) {
 855                 qd = qda[x];
 856                 offset = qd2offset(qd);
 857                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
 858                 if (error)
 859                         goto out_end_trans;
 860
 861                 do_qc(qd, -qd->qd_change_sync);
 862                 set_bit(QDF_REFRESH, &qd->qd_flags);
 863         }
 864
 865         error = 0;
 866
 867 out_end_trans:
 868         gfs2_trans_end(sdp);
 869 out_ipres:
 870         gfs2_inplace_release(ip);
 871 out_alloc:
 872         gfs2_glock_dq_uninit(&i_gh);
 873 out:
 874         while (qx--)
 875                 gfs2_glock_dq_uninit(&ghs[qx]);
 876         mutex_unlock(&ip->i_inode.i_mutex);
 877         kfree(ghs);
 878         gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
 879         return error;
 880 }
 881
 882 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
 883 {
 884         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 885         struct gfs2_quota q;
 886         struct gfs2_quota_lvb *qlvb;
 887         loff_t pos;
 888         int error;
 889
 890         memset(&q, 0, sizeof(struct gfs2_quota));
 891         pos = qd2offset(qd);
 892         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
 893         if (error < 0)
 894                 return error;
 895
 896         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
 897         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
 898         qlvb->__pad = 0;
 899         qlvb->qb_limit = q.qu_limit;
 900         qlvb->qb_warn = q.qu_warn;
 901         qlvb->qb_value = q.qu_value;
 902         qd->qd_qb = *qlvb;
 903
 904         return 0;
 905 }
 906
 907 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
 908                     struct gfs2_holder *q_gh)
 909 {
 910         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 911         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 912         struct gfs2_holder i_gh;
 913         int error;
 914
 915 restart:
 916         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
 917         if (error)
 918                 return error;
 919
 920         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
 921
 922         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
 923                 gfs2_glock_dq_uninit(q_gh);
 924                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
 925                                            GL_NOCACHE, q_gh);
 926                 if (error)
 927                         return error;
 928
 929                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
 930                 if (error)
 931                         goto fail;
 932
 933                 error = update_qd(sdp, qd);
 934                 if (error)
 935                         goto fail_gunlock;
 936
 937                 gfs2_glock_dq_uninit(&i_gh);
 938                 gfs2_glock_dq_uninit(q_gh);
 939                 force_refresh = 0;
 940                 goto restart;
 941         }
 942
 943         return 0;
 944
 945 fail_gunlock:
 946         gfs2_glock_dq_uninit(&i_gh);
 947 fail:
 948         gfs2_glock_dq_uninit(q_gh);
 949         return error;
 950 }
 951
 952 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
 953 {
 954         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 955         struct gfs2_quota_data *qd;
 956         unsigned int x;
 957         int error = 0;
 958
 959         error = gfs2_quota_hold(ip, uid, gid);
 960         if (error)
 961                 return error;
 962
 963         if (capable(CAP_SYS_RESOURCE) ||
 964             sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
 965                 return 0;
 966
 967         sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
 968              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 969
 970         for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
 971                 int force = NO_FORCE;
 972                 qd = ip->i_res->rs_qa_qd[x];
 973                 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
 974                         force = FORCE;
 975                 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
 976                 if (error)
 977                         break;
 978         }
 979
 980         if (!error)
 981                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
 982         else {
 983                 while (x--)
 984                         gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
 985                 gfs2_quota_unhold(ip);
 986         }
 987
 988         return error;
 989 }
 990
 991 static int need_sync(struct gfs2_quota_data *qd)
 992 {
 993         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 994         struct gfs2_tune *gt = &sdp->sd_tune;
 995         s64 value;
 996         unsigned int num, den;
 997         int do_sync = 1;
 998
 999         if (!qd->qd_qb.qb_limit)
1000                 return 0;
1001
1002         spin_lock(&qd_lock);
1003         value = qd->qd_change;
1004         spin_unlock(&qd_lock);
1005
1006         spin_lock(&gt->gt_spin);
1007         num = gt->gt_quota_scale_num;
1008         den = gt->gt_quota_scale_den;
1009         spin_unlock(&gt->gt_spin);
1010
1011         if (value < 0)
1012                 do_sync = 0;
1013         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1014                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1015                 do_sync = 0;
1016         else {
1017                 value *= gfs2_jindex_size(sdp) * num;
1018                 value = div_s64(value, den);
1019                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1020                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1021                         do_sync = 0;
1022         }
1023
1024         return do_sync;
1025 }
1026
1027 void gfs2_quota_unlock(struct gfs2_inode *ip)
1028 {
1029         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1030         struct gfs2_quota_data *qda[4];
1031         unsigned int count = 0;
1032         unsigned int x;
1033         int found;
1034
1035         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1036                 goto out;
1037
1038         for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1039                 struct gfs2_quota_data *qd;
1040                 int sync;
1041
1042                 qd = ip->i_res->rs_qa_qd[x];
1043                 sync = need_sync(qd);
1044
1045                 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1046                 if (!sync)
1047                         continue;
1048
1049                 spin_lock(&qd_lock);
1050                 found = qd_check_sync(sdp, qd, NULL);
1051                 spin_unlock(&qd_lock);
1052
1053                 if (!found)
1054                         continue;
1055
1056                 gfs2_assert_warn(sdp, qd->qd_change_sync);
1057                 if (bh_get(qd)) {
1058                         clear_bit(QDF_LOCKED, &qd->qd_flags);
1059                         slot_put(qd);
1060                         qd_put(qd);
1061                         continue;
1062                 }
1063
1064                 qda[count++] = qd;
1065         }
1066
1067         if (count) {
1068                 do_sync(count, qda);
1069                 for (x = 0; x < count; x++)
1070                         qd_unlock(qda[x]);
1071         }
1072
1073 out:
1074         gfs2_quota_unhold(ip);
1075 }
1076
1077 #define MAX_LINE 256
1078
1079 static int print_message(struct gfs2_quota_data *qd, char *type)
1080 {
1081         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1082
1083         printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1084                sdp->sd_fsname, type,
1085                (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1086                from_kqid(&init_user_ns, qd->qd_id));
1087
1088         return 0;
1089 }
1090
1091 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1092 {
1093         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1094         struct gfs2_quota_data *qd;
1095         s64 value;
1096         unsigned int x;
1097         int error = 0;
1098
1099         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1100                 return 0;
1101
1102         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1103                 return 0;
1104
1105         for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1106                 qd = ip->i_res->rs_qa_qd[x];
1107
1108                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1109                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1110                         continue;
1111
1112                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1113                 spin_lock(&qd_lock);
1114                 value += qd->qd_change;
1115                 spin_unlock(&qd_lock);
1116
1117                 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1118                         print_message(qd, "exceeded");
1119                         quota_send_warning(qd->qd_id,
1120                                            sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1121
1122                         error = -EDQUOT;
1123                         break;
1124                 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1125                            (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1126                            time_after_eq(jiffies, qd->qd_last_warn +
1127                                          gfs2_tune_get(sdp,
1128                                                 gt_quota_warn_period) * HZ)) {
1129                         quota_send_warning(qd->qd_id,
1130                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1131                         error = print_message(qd, "warning");
1132                         qd->qd_last_warn = jiffies;
1133                 }
1134         }
1135
1136         return error;
1137 }
1138
1139 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1140                        kuid_t uid, kgid_t gid)
1141 {
1142         struct gfs2_quota_data *qd;
1143         unsigned int x;
1144
1145         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1146                 return;
1147         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1148                 return;
1149
1150         for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1151                 qd = ip->i_res->rs_qa_qd[x];
1152
1153                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1154                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1155                         do_qc(qd, change);
1156                 }
1157         }
1158 }
1159
1160 int gfs2_quota_sync(struct super_block *sb, int type)
1161 {
1162         struct gfs2_sbd *sdp = sb->s_fs_info;
1163         struct gfs2_quota_data **qda;
1164         unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1165         unsigned int num_qd;
1166         unsigned int x;
1167         int error = 0;
1168
1169         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1170         if (!qda)
1171                 return -ENOMEM;
1172
1173         mutex_lock(&sdp->sd_quota_sync_mutex);
1174         sdp->sd_quota_sync_gen++;
1175
1176         do {
1177                 num_qd = 0;
1178
1179                 for (;;) {
1180                         error = qd_fish(sdp, qda + num_qd);
1181                         if (error || !qda[num_qd])
1182                                 break;
1183                         if (++num_qd == max_qd)
1184                                 break;
1185                 }
1186
1187                 if (num_qd) {
1188                         if (!error)
1189                                 error = do_sync(num_qd, qda);
1190                         if (!error)
1191                                 for (x = 0; x < num_qd; x++)
1192                                         qda[x]->qd_sync_gen =
1193                                                 sdp->sd_quota_sync_gen;
1194
1195                         for (x = 0; x < num_qd; x++)
1196                                 qd_unlock(qda[x]);
1197                 }
1198         } while (!error && num_qd == max_qd);
1199
1200         mutex_unlock(&sdp->sd_quota_sync_mutex);
1201         kfree(qda);
1202
1203         return error;
1204 }
1205
1206 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1207 {
1208         struct gfs2_quota_data *qd;
1209         struct gfs2_holder q_gh;
1210         int error;
1211
1212         error = qd_get(sdp, qid, &qd);
1213         if (error)
1214                 return error;
1215
1216         error = do_glock(qd, FORCE, &q_gh);
1217         if (!error)
1218                 gfs2_glock_dq_uninit(&q_gh);
1219
1220         qd_put(qd);
1221         return error;
1222 }
1223
1224 int gfs2_quota_init(struct gfs2_sbd *sdp)
1225 {
1226         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1227         u64 size = i_size_read(sdp->sd_qc_inode);
1228         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1229         unsigned int x, slot = 0;
1230         unsigned int found = 0;
1231         unsigned int hash;
1232         unsigned int bm_size;
1233         u64 dblock;
1234         u32 extlen = 0;
1235         int error;
1236
1237         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1238                 return -EIO;
1239
1240         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1241         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1242         bm_size *= sizeof(unsigned long);
1243         error = -ENOMEM;
1244         sdp->sd_quota_bitmap = kmalloc(bm_size, GFP_NOFS|__GFP_NOWARN);
1245         if (sdp->sd_quota_bitmap == NULL)
1246                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS, PAGE_KERNEL);
1247         if (!sdp->sd_quota_bitmap)
1248                 return error;
1249
1250         memset(sdp->sd_quota_bitmap, 0, bm_size);
1251
1252         for (x = 0; x < blocks; x++) {
1253                 struct buffer_head *bh;
1254                 const struct gfs2_quota_change *qc;
1255                 unsigned int y;
1256
1257                 if (!extlen) {
1258                         int new = 0;
1259                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1260                         if (error)
1261                                 goto fail;
1262                 }
1263                 error = -EIO;
1264                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1265                 if (!bh)
1266                         goto fail;
1267                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1268                         brelse(bh);
1269                         goto fail;
1270                 }
1271
1272                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1273                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1274                      y++, slot++) {
1275                         struct gfs2_quota_data *qd;
1276                         s64 qc_change = be64_to_cpu(qc->qc_change);
1277                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1278                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1279                                                 USRQUOTA : GRPQUOTA;
1280                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1281                                                       be32_to_cpu(qc->qc_id));
1282                         qc++;
1283                         if (!qc_change)
1284                                 continue;
1285
1286                         hash = gfs2_qd_hash(sdp, qc_id);
1287                         qd = qd_alloc(hash, sdp, qc_id);
1288                         if (qd == NULL) {
1289                                 brelse(bh);
1290                                 goto fail;
1291                         }
1292
1293                         set_bit(QDF_CHANGE, &qd->qd_flags);
1294                         qd->qd_change = qc_change;
1295                         qd->qd_slot = slot;
1296                         qd->qd_slot_count = 1;
1297
1298                         spin_lock(&qd_lock);
1299                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1300                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1301                         atomic_inc(&sdp->sd_quota_count);
1302                         spin_unlock(&qd_lock);
1303
1304                         spin_lock_bucket(hash);
1305                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1306                         spin_unlock_bucket(hash);
1307
1308                         found++;
1309                 }
1310
1311                 brelse(bh);
1312                 dblock++;
1313                 extlen--;
1314         }
1315
1316         if (found)
1317                 fs_info(sdp, "found %u quota changes\n", found);
1318
1319         return 0;
1320
1321 fail:
1322         gfs2_quota_cleanup(sdp);
1323         return error;
1324 }
1325
1326 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1327 {
1328         struct list_head *head = &sdp->sd_quota_list;
1329         struct gfs2_quota_data *qd;
1330
1331         spin_lock(&qd_lock);
1332         while (!list_empty(head)) {
1333                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1334
1335                 list_del(&qd->qd_list);
1336
1337                 /* Also remove if this qd exists in the reclaim list */
1338                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1339                 atomic_dec(&sdp->sd_quota_count);
1340                 spin_unlock(&qd_lock);
1341
1342                 spin_lock_bucket(qd->qd_hash);
1343                 hlist_bl_del_rcu(&qd->qd_hlist);
1344                 spin_unlock_bucket(qd->qd_hash);
1345
1346                 gfs2_assert_warn(sdp, !qd->qd_change);
1347                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1348                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1349
1350                 gfs2_glock_put(qd->qd_gl);
1351                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1352
1353                 spin_lock(&qd_lock);
1354         }
1355         spin_unlock(&qd_lock);
1356
1357         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1358
1359         if (sdp->sd_quota_bitmap) {
1360                 if (is_vmalloc_addr(sdp->sd_quota_bitmap))
1361                         vfree(sdp->sd_quota_bitmap);
1362                 else
1363                         kfree(sdp->sd_quota_bitmap);
1364                 sdp->sd_quota_bitmap = NULL;
1365         }
1366 }
1367
1368 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1369 {
1370         if (error == 0 || error == -EROFS)
1371                 return;
1372         if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1373                 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1374 }
1375
1376 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1377                                int (*fxn)(struct super_block *sb, int type),
1378                                unsigned long t, unsigned long *timeo,
1379                                unsigned int *new_timeo)
1380 {
1381         if (t >= *timeo) {
1382                 int error = fxn(sdp->sd_vfs, 0);
1383                 quotad_error(sdp, msg, error);
1384                 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1385         } else {
1386                 *timeo -= t;
1387         }
1388 }
1389
1390 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1391 {
1392         struct gfs2_inode *ip;
1393
1394         while(1) {
1395                 ip = NULL;
1396                 spin_lock(&sdp->sd_trunc_lock);
1397                 if (!list_empty(&sdp->sd_trunc_list)) {
1398                         ip = list_entry(sdp->sd_trunc_list.next,
1399                                         struct gfs2_inode, i_trunc_list);
1400                         list_del_init(&ip->i_trunc_list);
1401                 }
1402                 spin_unlock(&sdp->sd_trunc_lock);
1403                 if (ip == NULL)
1404                         return;
1405                 gfs2_glock_finish_truncate(ip);
1406         }
1407 }
1408
1409 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1410         if (!sdp->sd_statfs_force_sync) {
1411                 sdp->sd_statfs_force_sync = 1;
1412                 wake_up(&sdp->sd_quota_wait);
1413         }
1414 }
1415
1416
1417 /**
1418  * gfs2_quotad - Write cached quota changes into the quota file
1419  * @sdp: Pointer to GFS2 superblock
1420  *
1421  */
1422
1423 int gfs2_quotad(void *data)
1424 {
1425         struct gfs2_sbd *sdp = data;
1426         struct gfs2_tune *tune = &sdp->sd_tune;
1427         unsigned long statfs_timeo = 0;
1428         unsigned long quotad_timeo = 0;
1429         unsigned long t = 0;
1430         DEFINE_WAIT(wait);
1431         int empty;
1432
1433         while (!kthread_should_stop()) {
1434
1435                 /* Update the master statfs file */
1436                 if (sdp->sd_statfs_force_sync) {
1437                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1438                         quotad_error(sdp, "statfs", error);
1439                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1440                 }
1441                 else
1442                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1443                                            &statfs_timeo,
1444                                            &tune->gt_statfs_quantum);
1445
1446                 /* Update quota file */
1447                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1448                                    &quotad_timeo, &tune->gt_quota_quantum);
1449
1450                 /* Check for & recover partially truncated inodes */
1451                 quotad_check_trunc_list(sdp);
1452
1453                 try_to_freeze();
1454
1455                 t = min(quotad_timeo, statfs_timeo);
1456
1457                 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1458                 spin_lock(&sdp->sd_trunc_lock);
1459                 empty = list_empty(&sdp->sd_trunc_list);
1460                 spin_unlock(&sdp->sd_trunc_lock);
1461                 if (empty && !sdp->sd_statfs_force_sync)
1462                         t -= schedule_timeout(t);
1463                 else
1464                         t = 0;
1465                 finish_wait(&sdp->sd_quota_wait, &wait);
1466         }
1467
1468         return 0;
1469 }
1470
1471 static int gfs2_quota_get_xstate(struct super_block *sb,
1472                                  struct fs_quota_stat *fqs)
1473 {
1474         struct gfs2_sbd *sdp = sb->s_fs_info;
1475
1476         memset(fqs, 0, sizeof(struct fs_quota_stat));
1477         fqs->qs_version = FS_QSTAT_VERSION;
1478
1479         switch (sdp->sd_args.ar_quota) {
1480         case GFS2_QUOTA_ON:
1481                 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1482                 /*FALLTHRU*/
1483         case GFS2_QUOTA_ACCOUNT:
1484                 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1485                 break;
1486         case GFS2_QUOTA_OFF:
1487                 break;
1488         }
1489
1490         if (sdp->sd_quota_inode) {
1491                 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1492                 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1493         }
1494         fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1495         fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1496         fqs->qs_incoredqs = list_lru_count(&gfs2_qd_lru);
1497         return 0;
1498 }
1499
1500 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1501                           struct fs_disk_quota *fdq)
1502 {
1503         struct gfs2_sbd *sdp = sb->s_fs_info;
1504         struct gfs2_quota_lvb *qlvb;
1505         struct gfs2_quota_data *qd;
1506         struct gfs2_holder q_gh;
1507         int error;
1508
1509         memset(fdq, 0, sizeof(struct fs_disk_quota));
1510
1511         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1512                 return -ESRCH; /* Crazy XFS error code */
1513
1514         if ((qid.type != USRQUOTA) &&
1515             (qid.type != GRPQUOTA))
1516                 return -EINVAL;
1517
1518         error = qd_get(sdp, qid, &qd);
1519         if (error)
1520                 return error;
1521         error = do_glock(qd, FORCE, &q_gh);
1522         if (error)
1523                 goto out;
1524
1525         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1526         fdq->d_version = FS_DQUOT_VERSION;
1527         fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1528         fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1529         fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1530         fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1531         fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1532
1533         gfs2_glock_dq_uninit(&q_gh);
1534 out:
1535         qd_put(qd);
1536         return error;
1537 }
1538
1539 /* GFS2 only supports a subset of the XFS fields */
1540 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1541
1542 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1543                           struct fs_disk_quota *fdq)
1544 {
1545         struct gfs2_sbd *sdp = sb->s_fs_info;
1546         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1547         struct gfs2_quota_data *qd;
1548         struct gfs2_holder q_gh, i_gh;
1549         unsigned int data_blocks, ind_blocks;
1550         unsigned int blocks = 0;
1551         int alloc_required;
1552         loff_t offset;
1553         int error;
1554
1555         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1556                 return -ESRCH; /* Crazy XFS error code */
1557
1558         if ((qid.type != USRQUOTA) &&
1559             (qid.type != GRPQUOTA))
1560                 return -EINVAL;
1561
1562         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1563                 return -EINVAL;
1564
1565         error = qd_get(sdp, qid, &qd);
1566         if (error)
1567                 return error;
1568
1569         error = gfs2_rs_alloc(ip);
1570         if (error)
1571                 goto out_put;
1572
1573         mutex_lock(&ip->i_inode.i_mutex);
1574         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1575         if (error)
1576                 goto out_unlockput;
1577         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1578         if (error)
1579                 goto out_q;
1580
1581         /* Check for existing entry, if none then alloc new blocks */
1582         error = update_qd(sdp, qd);
1583         if (error)
1584                 goto out_i;
1585
1586         /* If nothing has changed, this is a no-op */
1587         if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1588             ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1589                 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1590
1591         if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1592             ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1593                 fdq->d_fieldmask ^= FS_DQ_BHARD;
1594
1595         if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1596             ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1597                 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1598
1599         if (fdq->d_fieldmask == 0)
1600                 goto out_i;
1601
1602         offset = qd2offset(qd);
1603         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1604         if (gfs2_is_stuffed(ip))
1605                 alloc_required = 1;
1606         if (alloc_required) {
1607                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1608                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1609                                        &data_blocks, &ind_blocks);
1610                 blocks = 1 + data_blocks + ind_blocks;
1611                 ap.target = blocks;
1612                 error = gfs2_inplace_reserve(ip, &ap);
1613                 if (error)
1614                         goto out_i;
1615                 blocks += gfs2_rg_blocks(ip, blocks);
1616         }
1617
1618         /* Some quotas span block boundaries and can update two blocks,
1619            adding an extra block to the transaction to handle such quotas */
1620         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1621         if (error)
1622                 goto out_release;
1623
1624         /* Apply changes */
1625         error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1626
1627         gfs2_trans_end(sdp);
1628 out_release:
1629         if (alloc_required)
1630                 gfs2_inplace_release(ip);
1631 out_i:
1632         gfs2_glock_dq_uninit(&i_gh);
1633 out_q:
1634         gfs2_glock_dq_uninit(&q_gh);
1635 out_unlockput:
1636         mutex_unlock(&ip->i_inode.i_mutex);
1637 out_put:
1638         qd_put(qd);
1639         return error;
1640 }
1641
1642 const struct quotactl_ops gfs2_quotactl_ops = {
1643         .quota_sync     = gfs2_quota_sync,
1644         .get_xstate     = gfs2_quota_get_xstate,
1645         .get_dqblk      = gfs2_get_dqblk,
1646         .set_dqblk      = gfs2_set_dqblk,
1647 };
1648
1649 void __init gfs2_quota_hash_init(void)
1650 {
1651         unsigned i;
1652
1653         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1654                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1655 }