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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  40
  41 #include <linux/sched.h>
  42 #include <linux/slab.h>
  43 #include <linux/mm.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/completion.h>
  46 #include <linux/buffer_head.h>
  47 #include <linux/sort.h>
  48 #include <linux/fs.h>
  49 #include <linux/bio.h>
  50 #include <linux/gfs2_ondisk.h>
  51 #include <linux/kthread.h>
  52 #include <linux/freezer.h>
  53 #include <linux/quota.h>
  54 #include <linux/dqblk_xfs.h>
  55 #include <linux/lockref.h>
  56 #include <linux/list_lru.h>
  57 #include <linux/rcupdate.h>
  58 #include <linux/rculist_bl.h>
  59 #include <linux/bit_spinlock.h>
  60 #include <linux/jhash.h>
  61 #include <linux/vmalloc.h>
  62
  63 #include "gfs2.h"
  64 #include "incore.h"
  65 #include "bmap.h"
  66 #include "glock.h"
  67 #include "glops.h"
  68 #include "log.h"
  69 #include "meta_io.h"
  70 #include "quota.h"
  71 #include "rgrp.h"
  72 #include "super.h"
  73 #include "trans.h"
  74 #include "inode.h"
  75 #include "util.h"
  76
  77 #define GFS2_QD_HASH_SHIFT      12
  78 #define GFS2_QD_HASH_SIZE       BIT(GFS2_QD_HASH_SHIFT)
  79 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
  80
  81 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
  82 /*                     -> sd_bitmap_lock                              */
  83 static DEFINE_SPINLOCK(qd_lock);
  84 struct list_lru gfs2_qd_lru;
  85
  86 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
  87
  88 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
  89                                  const struct kqid qid)
  90 {
  91         unsigned int h;
  92
  93         h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
  94         h = jhash(&qid, sizeof(struct kqid), h);
  95
  96         return h & GFS2_QD_HASH_MASK;
  97 }
  98
  99 static inline void spin_lock_bucket(unsigned int hash)
 100 {
 101         hlist_bl_lock(&qd_hash_table[hash]);
 102 }
 103
 104 static inline void spin_unlock_bucket(unsigned int hash)
 105 {
 106         hlist_bl_unlock(&qd_hash_table[hash]);
 107 }
 108
 109 static void gfs2_qd_dealloc(struct rcu_head *rcu)
 110 {
 111         struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
 112         kmem_cache_free(gfs2_quotad_cachep, qd);
 113 }
 114
 115 static void gfs2_qd_dispose(struct list_head *list)
 116 {
 117         struct gfs2_quota_data *qd;
 118         struct gfs2_sbd *sdp;
 119
 120         while (!list_empty(list)) {
 121                 qd = list_entry(list->next, struct gfs2_quota_data, qd_lru);
 122                 sdp = qd->qd_gl->gl_name.ln_sbd;
 123
 124                 list_del(&qd->qd_lru);
 125
 126                 /* Free from the filesystem-specific list */
 127                 spin_lock(&qd_lock);
 128                 list_del(&qd->qd_list);
 129                 spin_unlock(&qd_lock);
 130
 131                 spin_lock_bucket(qd->qd_hash);
 132                 hlist_bl_del_rcu(&qd->qd_hlist);
 133                 spin_unlock_bucket(qd->qd_hash);
 134
 135                 gfs2_assert_warn(sdp, !qd->qd_change);
 136                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
 137                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
 138
 139                 gfs2_glock_put(qd->qd_gl);
 140                 atomic_dec(&sdp->sd_quota_count);
 141
 142                 /* Delete it from the common reclaim list */
 143                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
 144         }
 145 }
 146
 147
 148 static enum lru_status gfs2_qd_isolate(struct list_head *item,
 149                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
 150 {
 151         struct list_head *dispose = arg;
 152         struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
 153
 154         if (!spin_trylock(&qd->qd_lockref.lock))
 155                 return LRU_SKIP;
 156
 157         if (qd->qd_lockref.count == 0) {
 158                 lockref_mark_dead(&qd->qd_lockref);
 159                 list_lru_isolate_move(lru, &qd->qd_lru, dispose);
 160         }
 161
 162         spin_unlock(&qd->qd_lockref.lock);
 163         return LRU_REMOVED;
 164 }
 165
 166 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
 167                                          struct shrink_control *sc)
 168 {
 169         LIST_HEAD(dispose);
 170         unsigned long freed;
 171
 172         if (!(sc->gfp_mask & __GFP_FS))
 173                 return SHRINK_STOP;
 174
 175         freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
 176                                      gfs2_qd_isolate, &dispose);
 177
 178         gfs2_qd_dispose(&dispose);
 179
 180         return freed;
 181 }
 182
 183 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
 184                                           struct shrink_control *sc)
 185 {
 186         return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
 187 }
 188
 189 struct shrinker gfs2_qd_shrinker = {
 190         .count_objects = gfs2_qd_shrink_count,
 191         .scan_objects = gfs2_qd_shrink_scan,
 192         .seeks = DEFAULT_SEEKS,
 193         .flags = SHRINKER_NUMA_AWARE,
 194 };
 195
 196
 197 static u64 qd2index(struct gfs2_quota_data *qd)
 198 {
 199         struct kqid qid = qd->qd_id;
 200         return (2 * (u64)from_kqid(&init_user_ns, qid)) +
 201                 ((qid.type == USRQUOTA) ? 0 : 1);
 202 }
 203
 204 static u64 qd2offset(struct gfs2_quota_data *qd)
 205 {
 206         u64 offset;
 207
 208         offset = qd2index(qd);
 209         offset *= sizeof(struct gfs2_quota);
 210
 211         return offset;
 212 }
 213
 214 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
 215 {
 216         struct gfs2_quota_data *qd;
 217         int error;
 218
 219         qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
 220         if (!qd)
 221                 return NULL;
 222
 223         qd->qd_sbd = sdp;
 224         qd->qd_lockref.count = 1;
 225         spin_lock_init(&qd->qd_lockref.lock);
 226         qd->qd_id = qid;
 227         qd->qd_slot = -1;
 228         INIT_LIST_HEAD(&qd->qd_lru);
 229         qd->qd_hash = hash;
 230
 231         error = gfs2_glock_get(sdp, qd2index(qd),
 232                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
 233         if (error)
 234                 goto fail;
 235
 236         return qd;
 237
 238 fail:
 239         kmem_cache_free(gfs2_quotad_cachep, qd);
 240         return NULL;
 241 }
 242
 243 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
 244                                                      const struct gfs2_sbd *sdp,
 245                                                      struct kqid qid)
 246 {
 247         struct gfs2_quota_data *qd;
 248         struct hlist_bl_node *h;
 249
 250         hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
 251                 if (!qid_eq(qd->qd_id, qid))
 252                         continue;
 253                 if (qd->qd_sbd != sdp)
 254                         continue;
 255                 if (lockref_get_not_dead(&qd->qd_lockref)) {
 256                         list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
 257                         return qd;
 258                 }
 259         }
 260
 261         return NULL;
 262 }
 263
 264
 265 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
 266                   struct gfs2_quota_data **qdp)
 267 {
 268         struct gfs2_quota_data *qd, *new_qd;
 269         unsigned int hash = gfs2_qd_hash(sdp, qid);
 270
 271         rcu_read_lock();
 272         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
 273         rcu_read_unlock();
 274
 275         if (qd)
 276                 return 0;
 277
 278         new_qd = qd_alloc(hash, sdp, qid);
 279         if (!new_qd)
 280                 return -ENOMEM;
 281
 282         spin_lock(&qd_lock);
 283         spin_lock_bucket(hash);
 284         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
 285         if (qd == NULL) {
 286                 *qdp = new_qd;
 287                 list_add(&new_qd->qd_list, &sdp->sd_quota_list);
 288                 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
 289                 atomic_inc(&sdp->sd_quota_count);
 290         }
 291         spin_unlock_bucket(hash);
 292         spin_unlock(&qd_lock);
 293
 294         if (qd) {
 295                 gfs2_glock_put(new_qd->qd_gl);
 296                 kmem_cache_free(gfs2_quotad_cachep, new_qd);
 297         }
 298
 299         return 0;
 300 }
 301
 302
 303 static void qd_hold(struct gfs2_quota_data *qd)
 304 {
 305         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 306         gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
 307         lockref_get(&qd->qd_lockref);
 308 }
 309
 310 static void qd_put(struct gfs2_quota_data *qd)
 311 {
 312         if (lockref_put_or_lock(&qd->qd_lockref))
 313                 return;
 314
 315         qd->qd_lockref.count = 0;
 316         list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
 317         spin_unlock(&qd->qd_lockref.lock);
 318
 319 }
 320
 321 static int slot_get(struct gfs2_quota_data *qd)
 322 {
 323         struct gfs2_sbd *sdp = qd->qd_sbd;
 324         unsigned int bit;
 325         int error = 0;
 326
 327         spin_lock(&sdp->sd_bitmap_lock);
 328         if (qd->qd_slot_count != 0)
 329                 goto out;
 330
 331         error = -ENOSPC;
 332         bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
 333         if (bit < sdp->sd_quota_slots) {
 334                 set_bit(bit, sdp->sd_quota_bitmap);
 335                 qd->qd_slot = bit;
 336                 error = 0;
 337 out:
 338                 qd->qd_slot_count++;
 339         }
 340         spin_unlock(&sdp->sd_bitmap_lock);
 341
 342         return error;
 343 }
 344
 345 static void slot_hold(struct gfs2_quota_data *qd)
 346 {
 347         struct gfs2_sbd *sdp = qd->qd_sbd;
 348
 349         spin_lock(&sdp->sd_bitmap_lock);
 350         gfs2_assert(sdp, qd->qd_slot_count);
 351         qd->qd_slot_count++;
 352         spin_unlock(&sdp->sd_bitmap_lock);
 353 }
 354
 355 static void slot_put(struct gfs2_quota_data *qd)
 356 {
 357         struct gfs2_sbd *sdp = qd->qd_sbd;
 358
 359         spin_lock(&sdp->sd_bitmap_lock);
 360         gfs2_assert(sdp, qd->qd_slot_count);
 361         if (!--qd->qd_slot_count) {
 362                 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
 363                 qd->qd_slot = -1;
 364         }
 365         spin_unlock(&sdp->sd_bitmap_lock);
 366 }
 367
 368 static int bh_get(struct gfs2_quota_data *qd)
 369 {
 370         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 371         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 372         unsigned int block, offset;
 373         struct buffer_head *bh;
 374         int error;
 375         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 376
 377         mutex_lock(&sdp->sd_quota_mutex);
 378
 379         if (qd->qd_bh_count++) {
 380                 mutex_unlock(&sdp->sd_quota_mutex);
 381                 return 0;
 382         }
 383
 384         block = qd->qd_slot / sdp->sd_qc_per_block;
 385         offset = qd->qd_slot % sdp->sd_qc_per_block;
 386
 387         bh_map.b_size = BIT(ip->i_inode.i_blkbits);
 388         error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
 389         if (error)
 390                 goto fail;
 391         error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, 0, &bh);
 392         if (error)
 393                 goto fail;
 394         error = -EIO;
 395         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
 396                 goto fail_brelse;
 397
 398         qd->qd_bh = bh;
 399         qd->qd_bh_qc = (struct gfs2_quota_change *)
 400                 (bh->b_data + sizeof(struct gfs2_meta_header) +
 401                  offset * sizeof(struct gfs2_quota_change));
 402
 403         mutex_unlock(&sdp->sd_quota_mutex);
 404
 405         return 0;
 406
 407 fail_brelse:
 408         brelse(bh);
 409 fail:
 410         qd->qd_bh_count--;
 411         mutex_unlock(&sdp->sd_quota_mutex);
 412         return error;
 413 }
 414
 415 static void bh_put(struct gfs2_quota_data *qd)
 416 {
 417         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 418
 419         mutex_lock(&sdp->sd_quota_mutex);
 420         gfs2_assert(sdp, qd->qd_bh_count);
 421         if (!--qd->qd_bh_count) {
 422                 brelse(qd->qd_bh);
 423                 qd->qd_bh = NULL;
 424                 qd->qd_bh_qc = NULL;
 425         }
 426         mutex_unlock(&sdp->sd_quota_mutex);
 427 }
 428
 429 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
 430                          u64 *sync_gen)
 431 {
 432         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 433             !test_bit(QDF_CHANGE, &qd->qd_flags) ||
 434             (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
 435                 return 0;
 436
 437         /*
 438          * If qd_change is 0 it means a pending quota change was negated.
 439          * We should not sync it, but we still have a qd reference and slot
 440          * reference taken by gfs2_quota_change -> do_qc that need to be put.
 441          */
 442         if (!qd->qd_change && test_and_clear_bit(QDF_CHANGE, &qd->qd_flags)) {
 443                 slot_put(qd);
 444                 qd_put(qd);
 445                 return 0;
 446         }
 447
 448         if (!lockref_get_not_dead(&qd->qd_lockref))
 449                 return 0;
 450
 451         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 452         set_bit(QDF_LOCKED, &qd->qd_flags);
 453         qd->qd_change_sync = qd->qd_change;
 454         slot_hold(qd);
 455         return 1;
 456 }
 457
 458 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 459 {
 460         struct gfs2_quota_data *qd = NULL;
 461         int error;
 462         int found = 0;
 463
 464         *qdp = NULL;
 465
 466         if (sb_rdonly(sdp->sd_vfs))
 467                 return 0;
 468
 469         spin_lock(&qd_lock);
 470
 471         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 472                 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
 473                 if (found)
 474                         break;
 475         }
 476
 477         if (!found)
 478                 qd = NULL;
 479
 480         spin_unlock(&qd_lock);
 481
 482         if (qd) {
 483                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 484                 error = bh_get(qd);
 485                 if (error) {
 486                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 487                         slot_put(qd);
 488                         qd_put(qd);
 489                         return error;
 490                 }
 491         }
 492
 493         *qdp = qd;
 494
 495         return 0;
 496 }
 497
 498 static void qd_unlock(struct gfs2_quota_data *qd)
 499 {
 500         gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
 501                          test_bit(QDF_LOCKED, &qd->qd_flags));
 502         clear_bit(QDF_LOCKED, &qd->qd_flags);
 503         bh_put(qd);
 504         slot_put(qd);
 505         qd_put(qd);
 506 }
 507
 508 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
 509                     struct gfs2_quota_data **qdp)
 510 {
 511         int error;
 512
 513         error = qd_get(sdp, qid, qdp);
 514         if (error)
 515                 return error;
 516
 517         error = slot_get(*qdp);
 518         if (error)
 519                 goto fail;
 520
 521         error = bh_get(*qdp);
 522         if (error)
 523                 goto fail_slot;
 524
 525         return 0;
 526
 527 fail_slot:
 528         slot_put(*qdp);
 529 fail:
 530         qd_put(*qdp);
 531         return error;
 532 }
 533
 534 static void qdsb_put(struct gfs2_quota_data *qd)
 535 {
 536         bh_put(qd);
 537         slot_put(qd);
 538         qd_put(qd);
 539 }
 540
 541 /**
 542  * gfs2_qa_alloc - make sure we have a quota allocations data structure,
 543  *                 if necessary
 544  * @ip: the inode for this reservation
 545  */
 546 int gfs2_qa_alloc(struct gfs2_inode *ip)
 547 {
 548         int error = 0;
 549         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 550
 551         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 552                 return 0;
 553
 554         down_write(&ip->i_rw_mutex);
 555         if (ip->i_qadata == NULL) {
 556                 ip->i_qadata = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
 557                 if (!ip->i_qadata)
 558                         error = -ENOMEM;
 559         }
 560         up_write(&ip->i_rw_mutex);
 561         return error;
 562 }
 563
 564 void gfs2_qa_delete(struct gfs2_inode *ip, atomic_t *wcount)
 565 {
 566         down_write(&ip->i_rw_mutex);
 567         if (ip->i_qadata && ((wcount == NULL) || (atomic_read(wcount) <= 1))) {
 568                 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
 569                 ip->i_qadata = NULL;
 570         }
 571         up_write(&ip->i_rw_mutex);
 572 }
 573
 574 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
 575 {
 576         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 577         struct gfs2_quota_data **qd;
 578         int error;
 579
 580         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 581                 return 0;
 582
 583         if (ip->i_qadata == NULL) {
 584                 error = gfs2_rsqa_alloc(ip);
 585                 if (error)
 586                         return error;
 587         }
 588
 589         qd = ip->i_qadata->qa_qd;
 590
 591         if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
 592             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 593                 return -EIO;
 594
 595         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
 596         if (error)
 597                 goto out;
 598         ip->i_qadata->qa_qd_num++;
 599         qd++;
 600
 601         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
 602         if (error)
 603                 goto out;
 604         ip->i_qadata->qa_qd_num++;
 605         qd++;
 606
 607         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
 608             !uid_eq(uid, ip->i_inode.i_uid)) {
 609                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
 610                 if (error)
 611                         goto out;
 612                 ip->i_qadata->qa_qd_num++;
 613                 qd++;
 614         }
 615
 616         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
 617             !gid_eq(gid, ip->i_inode.i_gid)) {
 618                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
 619                 if (error)
 620                         goto out;
 621                 ip->i_qadata->qa_qd_num++;
 622                 qd++;
 623         }
 624
 625 out:
 626         if (error)
 627                 gfs2_quota_unhold(ip);
 628         return error;
 629 }
 630
 631 void gfs2_quota_unhold(struct gfs2_inode *ip)
 632 {
 633         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 634         u32 x;
 635
 636         if (ip->i_qadata == NULL)
 637                 return;
 638         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
 639
 640         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
 641                 qdsb_put(ip->i_qadata->qa_qd[x]);
 642                 ip->i_qadata->qa_qd[x] = NULL;
 643         }
 644         ip->i_qadata->qa_qd_num = 0;
 645 }
 646
 647 static int sort_qd(const void *a, const void *b)
 648 {
 649         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
 650         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
 651
 652         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
 653                 return -1;
 654         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
 655                 return 1;
 656         return 0;
 657 }
 658
 659 static void do_qc(struct gfs2_quota_data *qd, s64 change)
 660 {
 661         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 662         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 663         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 664         s64 x;
 665
 666         mutex_lock(&sdp->sd_quota_mutex);
 667         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
 668
 669         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
 670                 qc->qc_change = 0;
 671                 qc->qc_flags = 0;
 672                 if (qd->qd_id.type == USRQUOTA)
 673                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
 674                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
 675         }
 676
 677         x = be64_to_cpu(qc->qc_change) + change;
 678         qc->qc_change = cpu_to_be64(x);
 679
 680         spin_lock(&qd_lock);
 681         qd->qd_change = x;
 682         spin_unlock(&qd_lock);
 683
 684         if (!x) {
 685                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
 686                 clear_bit(QDF_CHANGE, &qd->qd_flags);
 687                 qc->qc_flags = 0;
 688                 qc->qc_id = 0;
 689                 slot_put(qd);
 690                 qd_put(qd);
 691         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
 692                 qd_hold(qd);
 693                 slot_hold(qd);
 694         }
 695
 696         if (change < 0) /* Reset quiet flag if we freed some blocks */
 697                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
 698         mutex_unlock(&sdp->sd_quota_mutex);
 699 }
 700
 701 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
 702                                   unsigned off, void *buf, unsigned bytes)
 703 {
 704         struct inode *inode = &ip->i_inode;
 705         struct gfs2_sbd *sdp = GFS2_SB(inode);
 706         struct address_space *mapping = inode->i_mapping;
 707         struct page *page;
 708         struct buffer_head *bh;
 709         void *kaddr;
 710         u64 blk;
 711         unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
 712         unsigned to_write = bytes, pg_off = off;
 713         int done = 0;
 714
 715         blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
 716         boff = off % bsize;
 717
 718         page = find_or_create_page(mapping, index, GFP_NOFS);
 719         if (!page)
 720                 return -ENOMEM;
 721         if (!page_has_buffers(page))
 722                 create_empty_buffers(page, bsize, 0);
 723
 724         bh = page_buffers(page);
 725         while (!done) {
 726                 /* Find the beginning block within the page */
 727                 if (pg_off >= ((bnum * bsize) + bsize)) {
 728                         bh = bh->b_this_page;
 729                         bnum++;
 730                         blk++;
 731                         continue;
 732                 }
 733                 if (!buffer_mapped(bh)) {
 734                         gfs2_block_map(inode, blk, bh, 1);
 735                         if (!buffer_mapped(bh))
 736                                 goto unlock_out;
 737                         /* If it's a newly allocated disk block, zero it */
 738                         if (buffer_new(bh))
 739                                 zero_user(page, bnum * bsize, bh->b_size);
 740                 }
 741                 if (PageUptodate(page))
 742                         set_buffer_uptodate(bh);
 743                 if (!buffer_uptodate(bh)) {
 744                         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
 745                         wait_on_buffer(bh);
 746                         if (!buffer_uptodate(bh))
 747                                 goto unlock_out;
 748                 }
 749                 if (gfs2_is_jdata(ip))
 750                         gfs2_trans_add_data(ip->i_gl, bh);
 751                 else
 752                         gfs2_ordered_add_inode(ip);
 753
 754                 /* If we need to write to the next block as well */
 755                 if (to_write > (bsize - boff)) {
 756                         pg_off += (bsize - boff);
 757                         to_write -= (bsize - boff);
 758                         boff = pg_off % bsize;
 759                         continue;
 760                 }
 761                 done = 1;
 762         }
 763
 764         /* Write to the page, now that we have setup the buffer(s) */
 765         kaddr = kmap_atomic(page);
 766         memcpy(kaddr + off, buf, bytes);
 767         flush_dcache_page(page);
 768         kunmap_atomic(kaddr);
 769         unlock_page(page);
 770         put_page(page);
 771
 772         return 0;
 773
 774 unlock_out:
 775         unlock_page(page);
 776         put_page(page);
 777         return -EIO;
 778 }
 779
 780 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
 781                                  loff_t loc)
 782 {
 783         unsigned long pg_beg;
 784         unsigned pg_off, nbytes, overflow = 0;
 785         int pg_oflow = 0, error;
 786         void *ptr;
 787
 788         nbytes = sizeof(struct gfs2_quota);
 789
 790         pg_beg = loc >> PAGE_SHIFT;
 791         pg_off = loc % PAGE_SIZE;
 792
 793         /* If the quota straddles a page boundary, split the write in two */
 794         if ((pg_off + nbytes) > PAGE_SIZE) {
 795                 pg_oflow = 1;
 796                 overflow = (pg_off + nbytes) - PAGE_SIZE;
 797         }
 798
 799         ptr = qp;
 800         error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
 801                                        nbytes - overflow);
 802         /* If there's an overflow, write the remaining bytes to the next page */
 803         if (!error && pg_oflow)
 804                 error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
 805                                                ptr + nbytes - overflow,
 806                                                overflow);
 807         return error;
 808 }
 809
 810 /**
 811  * gfs2_adjust_quota - adjust record of current block usage
 812  * @ip: The quota inode
 813  * @loc: Offset of the entry in the quota file
 814  * @change: The amount of usage change to record
 815  * @qd: The quota data
 816  * @fdq: The updated limits to record
 817  *
 818  * This function was mostly borrowed from gfs2_block_truncate_page which was
 819  * in turn mostly borrowed from ext3
 820  *
 821  * Returns: 0 or -ve on error
 822  */
 823
 824 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
 825                              s64 change, struct gfs2_quota_data *qd,
 826                              struct qc_dqblk *fdq)
 827 {
 828         struct inode *inode = &ip->i_inode;
 829         struct gfs2_sbd *sdp = GFS2_SB(inode);
 830         struct gfs2_quota q;
 831         int err;
 832         u64 size;
 833
 834         if (gfs2_is_stuffed(ip)) {
 835                 err = gfs2_unstuff_dinode(ip, NULL);
 836                 if (err)
 837                         return err;
 838         }
 839
 840         memset(&q, 0, sizeof(struct gfs2_quota));
 841         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
 842         if (err < 0)
 843                 return err;
 844
 845         loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
 846         err = -EIO;
 847         be64_add_cpu(&q.qu_value, change);
 848         if (((s64)be64_to_cpu(q.qu_value)) < 0)
 849                 q.qu_value = 0; /* Never go negative on quota usage */
 850         qd->qd_qb.qb_value = q.qu_value;
 851         if (fdq) {
 852                 if (fdq->d_fieldmask & QC_SPC_SOFT) {
 853                         q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
 854                         qd->qd_qb.qb_warn = q.qu_warn;
 855                 }
 856                 if (fdq->d_fieldmask & QC_SPC_HARD) {
 857                         q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
 858                         qd->qd_qb.qb_limit = q.qu_limit;
 859                 }
 860                 if (fdq->d_fieldmask & QC_SPACE) {
 861                         q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
 862                         qd->qd_qb.qb_value = q.qu_value;
 863                 }
 864         }
 865
 866         err = gfs2_write_disk_quota(ip, &q, loc);
 867         if (!err) {
 868                 size = loc + sizeof(struct gfs2_quota);
 869                 if (size > inode->i_size)
 870                         i_size_write(inode, size);
 871                 inode->i_mtime = inode->i_atime = current_time(inode);
 872                 mark_inode_dirty(inode);
 873                 set_bit(QDF_REFRESH, &qd->qd_flags);
 874         }
 875
 876         return err;
 877 }
 878
 879 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 880 {
 881         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
 882         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 883         struct gfs2_alloc_parms ap = { .aflags = 0, };
 884         unsigned int data_blocks, ind_blocks;
 885         struct gfs2_holder *ghs, i_gh;
 886         unsigned int qx, x;
 887         struct gfs2_quota_data *qd;
 888         unsigned reserved;
 889         loff_t offset;
 890         unsigned int nalloc = 0, blocks;
 891         int error;
 892
 893         error = gfs2_rsqa_alloc(ip);
 894         if (error)
 895                 return error;
 896
 897         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
 898                               &data_blocks, &ind_blocks);
 899
 900         ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
 901         if (!ghs)
 902                 return -ENOMEM;
 903
 904         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 905         inode_lock(&ip->i_inode);
 906         for (qx = 0; qx < num_qd; qx++) {
 907                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
 908                                            GL_NOCACHE, &ghs[qx]);
 909                 if (error)
 910                         goto out;
 911         }
 912
 913         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
 914         if (error)
 915                 goto out;
 916
 917         for (x = 0; x < num_qd; x++) {
 918                 offset = qd2offset(qda[x]);
 919                 if (gfs2_write_alloc_required(ip, offset,
 920                                               sizeof(struct gfs2_quota)))
 921                         nalloc++;
 922         }
 923
 924         /*
 925          * 1 blk for unstuffing inode if stuffed. We add this extra
 926          * block to the reservation unconditionally. If the inode
 927          * doesn't need unstuffing, the block will be released to the
 928          * rgrp since it won't be allocated during the transaction
 929          */
 930         /* +3 in the end for unstuffing block, inode size update block
 931          * and another block in case quota straddles page boundary and
 932          * two blocks need to be updated instead of 1 */
 933         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
 934
 935         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
 936         ap.target = reserved;
 937         error = gfs2_inplace_reserve(ip, &ap);
 938         if (error)
 939                 goto out_alloc;
 940
 941         if (nalloc)
 942                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
 943
 944         error = gfs2_trans_begin(sdp, blocks, 0);
 945         if (error)
 946                 goto out_ipres;
 947
 948         for (x = 0; x < num_qd; x++) {
 949                 qd = qda[x];
 950                 offset = qd2offset(qd);
 951                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
 952                 if (error)
 953                         goto out_end_trans;
 954
 955                 do_qc(qd, -qd->qd_change_sync);
 956                 set_bit(QDF_REFRESH, &qd->qd_flags);
 957         }
 958
 959         error = 0;
 960
 961 out_end_trans:
 962         gfs2_trans_end(sdp);
 963 out_ipres:
 964         gfs2_inplace_release(ip);
 965 out_alloc:
 966         gfs2_glock_dq_uninit(&i_gh);
 967 out:
 968         while (qx--)
 969                 gfs2_glock_dq_uninit(&ghs[qx]);
 970         inode_unlock(&ip->i_inode);
 971         kfree(ghs);
 972         gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
 973                        GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
 974         return error;
 975 }
 976
 977 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
 978 {
 979         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 980         struct gfs2_quota q;
 981         struct gfs2_quota_lvb *qlvb;
 982         loff_t pos;
 983         int error;
 984
 985         memset(&q, 0, sizeof(struct gfs2_quota));
 986         pos = qd2offset(qd);
 987         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
 988         if (error < 0)
 989                 return error;
 990
 991         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
 992         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
 993         qlvb->__pad = 0;
 994         qlvb->qb_limit = q.qu_limit;
 995         qlvb->qb_warn = q.qu_warn;
 996         qlvb->qb_value = q.qu_value;
 997         qd->qd_qb = *qlvb;
 998
 999         return 0;
1000 }
1001
1002 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1003                     struct gfs2_holder *q_gh)
1004 {
1005         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1006         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1007         struct gfs2_holder i_gh;
1008         int error;
1009
1010 restart:
1011         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1012         if (error)
1013                 return error;
1014
1015         if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1016                 force_refresh = FORCE;
1017
1018         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1019
1020         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1021                 gfs2_glock_dq_uninit(q_gh);
1022                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1023                                            GL_NOCACHE, q_gh);
1024                 if (error)
1025                         return error;
1026
1027                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1028                 if (error)
1029                         goto fail;
1030
1031                 error = update_qd(sdp, qd);
1032                 if (error)
1033                         goto fail_gunlock;
1034
1035                 gfs2_glock_dq_uninit(&i_gh);
1036                 gfs2_glock_dq_uninit(q_gh);
1037                 force_refresh = 0;
1038                 goto restart;
1039         }
1040
1041         return 0;
1042
1043 fail_gunlock:
1044         gfs2_glock_dq_uninit(&i_gh);
1045 fail:
1046         gfs2_glock_dq_uninit(q_gh);
1047         return error;
1048 }
1049
1050 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1051 {
1052         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1053         struct gfs2_quota_data *qd;
1054         u32 x;
1055         int error = 0;
1056
1057         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1058                 return 0;
1059
1060         error = gfs2_quota_hold(ip, uid, gid);
1061         if (error)
1062                 return error;
1063
1064         sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1065              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1066
1067         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1068                 qd = ip->i_qadata->qa_qd[x];
1069                 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1070                 if (error)
1071                         break;
1072         }
1073
1074         if (!error)
1075                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1076         else {
1077                 while (x--)
1078                         gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1079                 gfs2_quota_unhold(ip);
1080         }
1081
1082         return error;
1083 }
1084
1085 static int need_sync(struct gfs2_quota_data *qd)
1086 {
1087         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1088         struct gfs2_tune *gt = &sdp->sd_tune;
1089         s64 value;
1090         unsigned int num, den;
1091         int do_sync = 1;
1092
1093         if (!qd->qd_qb.qb_limit)
1094                 return 0;
1095
1096         spin_lock(&qd_lock);
1097         value = qd->qd_change;
1098         spin_unlock(&qd_lock);
1099
1100         spin_lock(&gt->gt_spin);
1101         num = gt->gt_quota_scale_num;
1102         den = gt->gt_quota_scale_den;
1103         spin_unlock(&gt->gt_spin);
1104
1105         if (value < 0)
1106                 do_sync = 0;
1107         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1108                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1109                 do_sync = 0;
1110         else {
1111                 value *= gfs2_jindex_size(sdp) * num;
1112                 value = div_s64(value, den);
1113                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1114                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1115                         do_sync = 0;
1116         }
1117
1118         return do_sync;
1119 }
1120
1121 void gfs2_quota_unlock(struct gfs2_inode *ip)
1122 {
1123         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1124         struct gfs2_quota_data *qda[4];
1125         unsigned int count = 0;
1126         u32 x;
1127         int found;
1128
1129         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1130                 goto out;
1131
1132         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1133                 struct gfs2_quota_data *qd;
1134                 int sync;
1135
1136                 qd = ip->i_qadata->qa_qd[x];
1137                 sync = need_sync(qd);
1138
1139                 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1140                 if (!sync)
1141                         continue;
1142
1143                 spin_lock(&qd_lock);
1144                 found = qd_check_sync(sdp, qd, NULL);
1145                 spin_unlock(&qd_lock);
1146
1147                 if (!found)
1148                         continue;
1149
1150                 gfs2_assert_warn(sdp, qd->qd_change_sync);
1151                 if (bh_get(qd)) {
1152                         clear_bit(QDF_LOCKED, &qd->qd_flags);
1153                         slot_put(qd);
1154                         qd_put(qd);
1155                         continue;
1156                 }
1157
1158                 qda[count++] = qd;
1159         }
1160
1161         if (count) {
1162                 do_sync(count, qda);
1163                 for (x = 0; x < count; x++)
1164                         qd_unlock(qda[x]);
1165         }
1166
1167 out:
1168         gfs2_quota_unhold(ip);
1169 }
1170
1171 #define MAX_LINE 256
1172
1173 static int print_message(struct gfs2_quota_data *qd, char *type)
1174 {
1175         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1176
1177         fs_info(sdp, "quota %s for %s %u\n",
1178                 type,
1179                 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1180                 from_kqid(&init_user_ns, qd->qd_id));
1181
1182         return 0;
1183 }
1184
1185 /**
1186  * gfs2_quota_check - check if allocating new blocks will exceed quota
1187  * @ip:  The inode for which this check is being performed
1188  * @uid: The uid to check against
1189  * @gid: The gid to check against
1190  * @ap:  The allocation parameters. ap->target contains the requested
1191  *       blocks. ap->min_target, if set, contains the minimum blks
1192  *       requested.
1193  *
1194  * Returns: 0 on success.
1195  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1196  *                  quota must allow atleast min_req blks for success and
1197  *                  ap->allowed is set to the number of blocks allowed
1198  *
1199  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1200  *                  of blocks available.
1201  */
1202 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1203                      struct gfs2_alloc_parms *ap)
1204 {
1205         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1206         struct gfs2_quota_data *qd;
1207         s64 value, warn, limit;
1208         u32 x;
1209         int error = 0;
1210
1211         ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1212         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1213                 return 0;
1214
1215         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1216                 return 0;
1217
1218         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1219                 qd = ip->i_qadata->qa_qd[x];
1220
1221                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1222                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1223                         continue;
1224
1225                 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1226                 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1227                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1228                 spin_lock(&qd_lock);
1229                 value += qd->qd_change;
1230                 spin_unlock(&qd_lock);
1231
1232                 if (limit > 0 && (limit - value) < ap->allowed)
1233                         ap->allowed = limit - value;
1234                 /* If we can't meet the target */
1235                 if (limit && limit < (value + (s64)ap->target)) {
1236                         /* If no min_target specified or we don't meet
1237                          * min_target, return -EDQUOT */
1238                         if (!ap->min_target || ap->min_target > ap->allowed) {
1239                                 if (!test_and_set_bit(QDF_QMSG_QUIET,
1240                                                       &qd->qd_flags)) {
1241                                         print_message(qd, "exceeded");
1242                                         quota_send_warning(qd->qd_id,
1243                                                            sdp->sd_vfs->s_dev,
1244                                                            QUOTA_NL_BHARDWARN);
1245                                 }
1246                                 error = -EDQUOT;
1247                                 break;
1248                         }
1249                 } else if (warn && warn < value &&
1250                            time_after_eq(jiffies, qd->qd_last_warn +
1251                                          gfs2_tune_get(sdp, gt_quota_warn_period)
1252                                          * HZ)) {
1253                         quota_send_warning(qd->qd_id,
1254                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1255                         error = print_message(qd, "warning");
1256                         qd->qd_last_warn = jiffies;
1257                 }
1258         }
1259         return error;
1260 }
1261
1262 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1263                        kuid_t uid, kgid_t gid)
1264 {
1265         struct gfs2_quota_data *qd;
1266         u32 x;
1267         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1268
1269         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1270             gfs2_assert_warn(sdp, change))
1271                 return;
1272         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1273                 return;
1274
1275         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1276                 qd = ip->i_qadata->qa_qd[x];
1277
1278                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1279                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1280                         do_qc(qd, change);
1281                 }
1282         }
1283 }
1284
1285 int gfs2_quota_sync(struct super_block *sb, int type)
1286 {
1287         struct gfs2_sbd *sdp = sb->s_fs_info;
1288         struct gfs2_quota_data **qda;
1289         unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1290         unsigned int num_qd;
1291         unsigned int x;
1292         int error = 0;
1293
1294         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1295         if (!qda)
1296                 return -ENOMEM;
1297
1298         mutex_lock(&sdp->sd_quota_sync_mutex);
1299         sdp->sd_quota_sync_gen++;
1300
1301         do {
1302                 num_qd = 0;
1303
1304                 for (;;) {
1305                         error = qd_fish(sdp, qda + num_qd);
1306                         if (error || !qda[num_qd])
1307                                 break;
1308                         if (++num_qd == max_qd)
1309                                 break;
1310                 }
1311
1312                 if (num_qd) {
1313                         if (!error)
1314                                 error = do_sync(num_qd, qda);
1315                         if (!error)
1316                                 for (x = 0; x < num_qd; x++)
1317                                         qda[x]->qd_sync_gen =
1318                                                 sdp->sd_quota_sync_gen;
1319
1320                         for (x = 0; x < num_qd; x++)
1321                                 qd_unlock(qda[x]);
1322                 }
1323         } while (!error && num_qd == max_qd);
1324
1325         mutex_unlock(&sdp->sd_quota_sync_mutex);
1326         kfree(qda);
1327
1328         return error;
1329 }
1330
1331 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1332 {
1333         struct gfs2_quota_data *qd;
1334         struct gfs2_holder q_gh;
1335         int error;
1336
1337         error = qd_get(sdp, qid, &qd);
1338         if (error)
1339                 return error;
1340
1341         error = do_glock(qd, FORCE, &q_gh);
1342         if (!error)
1343                 gfs2_glock_dq_uninit(&q_gh);
1344
1345         qd_put(qd);
1346         return error;
1347 }
1348
1349 int gfs2_quota_init(struct gfs2_sbd *sdp)
1350 {
1351         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1352         u64 size = i_size_read(sdp->sd_qc_inode);
1353         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1354         unsigned int x, slot = 0;
1355         unsigned int found = 0;
1356         unsigned int hash;
1357         unsigned int bm_size;
1358         u64 dblock;
1359         u32 extlen = 0;
1360         int error;
1361
1362         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1363                 return -EIO;
1364
1365         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1366         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1367         bm_size *= sizeof(unsigned long);
1368         error = -ENOMEM;
1369         sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1370         if (sdp->sd_quota_bitmap == NULL)
1371                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1372                                                  __GFP_ZERO, PAGE_KERNEL);
1373         if (!sdp->sd_quota_bitmap)
1374                 return error;
1375
1376         for (x = 0; x < blocks; x++) {
1377                 struct buffer_head *bh;
1378                 const struct gfs2_quota_change *qc;
1379                 unsigned int y;
1380
1381                 if (!extlen) {
1382                         int new = 0;
1383                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1384                         if (error)
1385                                 goto fail;
1386                 }
1387                 error = -EIO;
1388                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1389                 if (!bh)
1390                         goto fail;
1391                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1392                         brelse(bh);
1393                         goto fail;
1394                 }
1395
1396                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1397                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1398                      y++, slot++) {
1399                         struct gfs2_quota_data *qd;
1400                         s64 qc_change = be64_to_cpu(qc->qc_change);
1401                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1402                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1403                                                 USRQUOTA : GRPQUOTA;
1404                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1405                                                       be32_to_cpu(qc->qc_id));
1406                         qc++;
1407                         if (!qc_change)
1408                                 continue;
1409
1410                         hash = gfs2_qd_hash(sdp, qc_id);
1411                         qd = qd_alloc(hash, sdp, qc_id);
1412                         if (qd == NULL) {
1413                                 brelse(bh);
1414                                 goto fail;
1415                         }
1416
1417                         set_bit(QDF_CHANGE, &qd->qd_flags);
1418                         qd->qd_change = qc_change;
1419                         qd->qd_slot = slot;
1420                         qd->qd_slot_count = 1;
1421
1422                         spin_lock(&qd_lock);
1423                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1424                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1425                         atomic_inc(&sdp->sd_quota_count);
1426                         spin_unlock(&qd_lock);
1427
1428                         spin_lock_bucket(hash);
1429                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1430                         spin_unlock_bucket(hash);
1431
1432                         found++;
1433                 }
1434
1435                 brelse(bh);
1436                 dblock++;
1437                 extlen--;
1438         }
1439
1440         if (found)
1441                 fs_info(sdp, "found %u quota changes\n", found);
1442
1443         return 0;
1444
1445 fail:
1446         gfs2_quota_cleanup(sdp);
1447         return error;
1448 }
1449
1450 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1451 {
1452         struct list_head *head = &sdp->sd_quota_list;
1453         struct gfs2_quota_data *qd;
1454
1455         spin_lock(&qd_lock);
1456         while (!list_empty(head)) {
1457                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1458
1459                 list_del(&qd->qd_list);
1460
1461                 /* Also remove if this qd exists in the reclaim list */
1462                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1463                 atomic_dec(&sdp->sd_quota_count);
1464                 spin_unlock(&qd_lock);
1465
1466                 spin_lock_bucket(qd->qd_hash);
1467                 hlist_bl_del_rcu(&qd->qd_hlist);
1468                 spin_unlock_bucket(qd->qd_hash);
1469
1470                 gfs2_assert_warn(sdp, !qd->qd_change);
1471                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1472                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1473
1474                 gfs2_glock_put(qd->qd_gl);
1475                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1476
1477                 spin_lock(&qd_lock);
1478         }
1479         spin_unlock(&qd_lock);
1480
1481         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1482
1483         kvfree(sdp->sd_quota_bitmap);
1484         sdp->sd_quota_bitmap = NULL;
1485 }
1486
1487 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1488 {
1489         if (error == 0 || error == -EROFS)
1490                 return;
1491         if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
1492                 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1493                 sdp->sd_log_error = error;
1494                 wake_up(&sdp->sd_logd_waitq);
1495         }
1496 }
1497
1498 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1499                                int (*fxn)(struct super_block *sb, int type),
1500                                unsigned long t, unsigned long *timeo,
1501                                unsigned int *new_timeo)
1502 {
1503         if (t >= *timeo) {
1504                 int error = fxn(sdp->sd_vfs, 0);
1505                 quotad_error(sdp, msg, error);
1506                 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1507         } else {
1508                 *timeo -= t;
1509         }
1510 }
1511
1512 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1513 {
1514         struct gfs2_inode *ip;
1515
1516         while(1) {
1517                 ip = NULL;
1518                 spin_lock(&sdp->sd_trunc_lock);
1519                 if (!list_empty(&sdp->sd_trunc_list)) {
1520                         ip = list_entry(sdp->sd_trunc_list.next,
1521                                         struct gfs2_inode, i_trunc_list);
1522                         list_del_init(&ip->i_trunc_list);
1523                 }
1524                 spin_unlock(&sdp->sd_trunc_lock);
1525                 if (ip == NULL)
1526                         return;
1527                 gfs2_glock_finish_truncate(ip);
1528         }
1529 }
1530
1531 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1532         if (!sdp->sd_statfs_force_sync) {
1533                 sdp->sd_statfs_force_sync = 1;
1534                 wake_up(&sdp->sd_quota_wait);
1535         }
1536 }
1537
1538
1539 /**
1540  * gfs2_quotad - Write cached quota changes into the quota file
1541  * @sdp: Pointer to GFS2 superblock
1542  *
1543  */
1544
1545 int gfs2_quotad(void *data)
1546 {
1547         struct gfs2_sbd *sdp = data;
1548         struct gfs2_tune *tune = &sdp->sd_tune;
1549         unsigned long statfs_timeo = 0;
1550         unsigned long quotad_timeo = 0;
1551         unsigned long t = 0;
1552         DEFINE_WAIT(wait);
1553         int empty;
1554
1555         while (!kthread_should_stop()) {
1556
1557                 /* Update the master statfs file */
1558                 if (sdp->sd_statfs_force_sync) {
1559                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1560                         quotad_error(sdp, "statfs", error);
1561                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1562                 }
1563                 else
1564                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1565                                            &statfs_timeo,
1566                                            &tune->gt_statfs_quantum);
1567
1568                 /* Update quota file */
1569                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1570                                    &quotad_timeo, &tune->gt_quota_quantum);
1571
1572                 /* Check for & recover partially truncated inodes */
1573                 quotad_check_trunc_list(sdp);
1574
1575                 try_to_freeze();
1576
1577                 t = min(quotad_timeo, statfs_timeo);
1578
1579                 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1580                 spin_lock(&sdp->sd_trunc_lock);
1581                 empty = list_empty(&sdp->sd_trunc_list);
1582                 spin_unlock(&sdp->sd_trunc_lock);
1583                 if (empty && !sdp->sd_statfs_force_sync)
1584                         t -= schedule_timeout(t);
1585                 else
1586                         t = 0;
1587                 finish_wait(&sdp->sd_quota_wait, &wait);
1588         }
1589
1590         return 0;
1591 }
1592
1593 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1594 {
1595         struct gfs2_sbd *sdp = sb->s_fs_info;
1596
1597         memset(state, 0, sizeof(*state));
1598
1599         switch (sdp->sd_args.ar_quota) {
1600         case GFS2_QUOTA_ON:
1601                 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1602                 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1603                 /*FALLTHRU*/
1604         case GFS2_QUOTA_ACCOUNT:
1605                 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1606                                                   QCI_SYSFILE;
1607                 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1608                                                   QCI_SYSFILE;
1609                 break;
1610         case GFS2_QUOTA_OFF:
1611                 break;
1612         }
1613         if (sdp->sd_quota_inode) {
1614                 state->s_state[USRQUOTA].ino =
1615                                         GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1616                 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1617         }
1618         state->s_state[USRQUOTA].nextents = 1;  /* unsupported */
1619         state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1620         state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1621         return 0;
1622 }
1623
1624 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1625                           struct qc_dqblk *fdq)
1626 {
1627         struct gfs2_sbd *sdp = sb->s_fs_info;
1628         struct gfs2_quota_lvb *qlvb;
1629         struct gfs2_quota_data *qd;
1630         struct gfs2_holder q_gh;
1631         int error;
1632
1633         memset(fdq, 0, sizeof(*fdq));
1634
1635         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1636                 return -ESRCH; /* Crazy XFS error code */
1637
1638         if ((qid.type != USRQUOTA) &&
1639             (qid.type != GRPQUOTA))
1640                 return -EINVAL;
1641
1642         error = qd_get(sdp, qid, &qd);
1643         if (error)
1644                 return error;
1645         error = do_glock(qd, FORCE, &q_gh);
1646         if (error)
1647                 goto out;
1648
1649         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1650         fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1651         fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1652         fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1653
1654         gfs2_glock_dq_uninit(&q_gh);
1655 out:
1656         qd_put(qd);
1657         return error;
1658 }
1659
1660 /* GFS2 only supports a subset of the XFS fields */
1661 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1662
1663 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1664                           struct qc_dqblk *fdq)
1665 {
1666         struct gfs2_sbd *sdp = sb->s_fs_info;
1667         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1668         struct gfs2_quota_data *qd;
1669         struct gfs2_holder q_gh, i_gh;
1670         unsigned int data_blocks, ind_blocks;
1671         unsigned int blocks = 0;
1672         int alloc_required;
1673         loff_t offset;
1674         int error;
1675
1676         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1677                 return -ESRCH; /* Crazy XFS error code */
1678
1679         if ((qid.type != USRQUOTA) &&
1680             (qid.type != GRPQUOTA))
1681                 return -EINVAL;
1682
1683         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1684                 return -EINVAL;
1685
1686         error = qd_get(sdp, qid, &qd);
1687         if (error)
1688                 return error;
1689
1690         error = gfs2_rsqa_alloc(ip);
1691         if (error)
1692                 goto out_put;
1693
1694         inode_lock(&ip->i_inode);
1695         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1696         if (error)
1697                 goto out_unlockput;
1698         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1699         if (error)
1700                 goto out_q;
1701
1702         /* Check for existing entry, if none then alloc new blocks */
1703         error = update_qd(sdp, qd);
1704         if (error)
1705                 goto out_i;
1706
1707         /* If nothing has changed, this is a no-op */
1708         if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1709             ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1710                 fdq->d_fieldmask ^= QC_SPC_SOFT;
1711
1712         if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1713             ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1714                 fdq->d_fieldmask ^= QC_SPC_HARD;
1715
1716         if ((fdq->d_fieldmask & QC_SPACE) &&
1717             ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1718                 fdq->d_fieldmask ^= QC_SPACE;
1719
1720         if (fdq->d_fieldmask == 0)
1721                 goto out_i;
1722
1723         offset = qd2offset(qd);
1724         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1725         if (gfs2_is_stuffed(ip))
1726                 alloc_required = 1;
1727         if (alloc_required) {
1728                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1729                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1730                                        &data_blocks, &ind_blocks);
1731                 blocks = 1 + data_blocks + ind_blocks;
1732                 ap.target = blocks;
1733                 error = gfs2_inplace_reserve(ip, &ap);
1734                 if (error)
1735                         goto out_i;
1736                 blocks += gfs2_rg_blocks(ip, blocks);
1737         }
1738
1739         /* Some quotas span block boundaries and can update two blocks,
1740            adding an extra block to the transaction to handle such quotas */
1741         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1742         if (error)
1743                 goto out_release;
1744
1745         /* Apply changes */
1746         error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1747         if (!error)
1748                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1749
1750         gfs2_trans_end(sdp);
1751 out_release:
1752         if (alloc_required)
1753                 gfs2_inplace_release(ip);
1754 out_i:
1755         gfs2_glock_dq_uninit(&i_gh);
1756 out_q:
1757         gfs2_glock_dq_uninit(&q_gh);
1758 out_unlockput:
1759         inode_unlock(&ip->i_inode);
1760 out_put:
1761         qd_put(qd);
1762         return error;
1763 }
1764
1765 const struct quotactl_ops gfs2_quotactl_ops = {
1766         .quota_sync     = gfs2_quota_sync,
1767         .get_state      = gfs2_quota_get_state,
1768         .get_dqblk      = gfs2_get_dqblk,
1769         .set_dqblk      = gfs2_set_dqblk,
1770 };
1771
1772 void __init gfs2_quota_hash_init(void)
1773 {
1774         unsigned i;
1775
1776         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1777                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1778 }