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         if (!lockref_get_not_dead(&qd->qd_lockref))
 438                 return 0;
 439
 440         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 441         set_bit(QDF_LOCKED, &qd->qd_flags);
 442         qd->qd_change_sync = qd->qd_change;
 443         slot_hold(qd);
 444         return 1;
 445 }
 446
 447 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 448 {
 449         struct gfs2_quota_data *qd = NULL;
 450         int error;
 451         int found = 0;
 452
 453         *qdp = NULL;
 454
 455         if (sb_rdonly(sdp->sd_vfs))
 456                 return 0;
 457
 458         spin_lock(&qd_lock);
 459
 460         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 461                 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
 462                 if (found)
 463                         break;
 464         }
 465
 466         if (!found)
 467                 qd = NULL;
 468
 469         spin_unlock(&qd_lock);
 470
 471         if (qd) {
 472                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 473                 error = bh_get(qd);
 474                 if (error) {
 475                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 476                         slot_put(qd);
 477                         qd_put(qd);
 478                         return error;
 479                 }
 480         }
 481
 482         *qdp = qd;
 483
 484         return 0;
 485 }
 486
 487 static void qd_unlock(struct gfs2_quota_data *qd)
 488 {
 489         gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
 490                          test_bit(QDF_LOCKED, &qd->qd_flags));
 491         clear_bit(QDF_LOCKED, &qd->qd_flags);
 492         bh_put(qd);
 493         slot_put(qd);
 494         qd_put(qd);
 495 }
 496
 497 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
 498                     struct gfs2_quota_data **qdp)
 499 {
 500         int error;
 501
 502         error = qd_get(sdp, qid, qdp);
 503         if (error)
 504                 return error;
 505
 506         error = slot_get(*qdp);
 507         if (error)
 508                 goto fail;
 509
 510         error = bh_get(*qdp);
 511         if (error)
 512                 goto fail_slot;
 513
 514         return 0;
 515
 516 fail_slot:
 517         slot_put(*qdp);
 518 fail:
 519         qd_put(*qdp);
 520         return error;
 521 }
 522
 523 static void qdsb_put(struct gfs2_quota_data *qd)
 524 {
 525         bh_put(qd);
 526         slot_put(qd);
 527         qd_put(qd);
 528 }
 529
 530 /**
 531  * gfs2_qa_alloc - make sure we have a quota allocations data structure,
 532  *                 if necessary
 533  * @ip: the inode for this reservation
 534  */
 535 int gfs2_qa_alloc(struct gfs2_inode *ip)
 536 {
 537         int error = 0;
 538         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 539
 540         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 541                 return 0;
 542
 543         down_write(&ip->i_rw_mutex);
 544         if (ip->i_qadata == NULL) {
 545                 ip->i_qadata = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
 546                 if (!ip->i_qadata)
 547                         error = -ENOMEM;
 548         }
 549         up_write(&ip->i_rw_mutex);
 550         return error;
 551 }
 552
 553 void gfs2_qa_delete(struct gfs2_inode *ip, atomic_t *wcount)
 554 {
 555         down_write(&ip->i_rw_mutex);
 556         if (ip->i_qadata && ((wcount == NULL) || (atomic_read(wcount) <= 1))) {
 557                 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
 558                 ip->i_qadata = NULL;
 559         }
 560         up_write(&ip->i_rw_mutex);
 561 }
 562
 563 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
 564 {
 565         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 566         struct gfs2_quota_data **qd;
 567         int error;
 568
 569         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 570                 return 0;
 571
 572         if (ip->i_qadata == NULL) {
 573                 error = gfs2_rsqa_alloc(ip);
 574                 if (error)
 575                         return error;
 576         }
 577
 578         qd = ip->i_qadata->qa_qd;
 579
 580         if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
 581             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 582                 return -EIO;
 583
 584         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
 585         if (error)
 586                 goto out;
 587         ip->i_qadata->qa_qd_num++;
 588         qd++;
 589
 590         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
 591         if (error)
 592                 goto out;
 593         ip->i_qadata->qa_qd_num++;
 594         qd++;
 595
 596         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
 597             !uid_eq(uid, ip->i_inode.i_uid)) {
 598                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
 599                 if (error)
 600                         goto out;
 601                 ip->i_qadata->qa_qd_num++;
 602                 qd++;
 603         }
 604
 605         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
 606             !gid_eq(gid, ip->i_inode.i_gid)) {
 607                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
 608                 if (error)
 609                         goto out;
 610                 ip->i_qadata->qa_qd_num++;
 611                 qd++;
 612         }
 613
 614 out:
 615         if (error)
 616                 gfs2_quota_unhold(ip);
 617         return error;
 618 }
 619
 620 void gfs2_quota_unhold(struct gfs2_inode *ip)
 621 {
 622         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 623         u32 x;
 624
 625         if (ip->i_qadata == NULL)
 626                 return;
 627         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
 628
 629         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
 630                 qdsb_put(ip->i_qadata->qa_qd[x]);
 631                 ip->i_qadata->qa_qd[x] = NULL;
 632         }
 633         ip->i_qadata->qa_qd_num = 0;
 634 }
 635
 636 static int sort_qd(const void *a, const void *b)
 637 {
 638         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
 639         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
 640
 641         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
 642                 return -1;
 643         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
 644                 return 1;
 645         return 0;
 646 }
 647
 648 static void do_qc(struct gfs2_quota_data *qd, s64 change)
 649 {
 650         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 651         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 652         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 653         s64 x;
 654
 655         mutex_lock(&sdp->sd_quota_mutex);
 656         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
 657
 658         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
 659                 qc->qc_change = 0;
 660                 qc->qc_flags = 0;
 661                 if (qd->qd_id.type == USRQUOTA)
 662                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
 663                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
 664         }
 665
 666         x = be64_to_cpu(qc->qc_change) + change;
 667         qc->qc_change = cpu_to_be64(x);
 668
 669         spin_lock(&qd_lock);
 670         qd->qd_change = x;
 671         spin_unlock(&qd_lock);
 672
 673         if (!x) {
 674                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
 675                 clear_bit(QDF_CHANGE, &qd->qd_flags);
 676                 qc->qc_flags = 0;
 677                 qc->qc_id = 0;
 678                 slot_put(qd);
 679                 qd_put(qd);
 680         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
 681                 qd_hold(qd);
 682                 slot_hold(qd);
 683         }
 684
 685         if (change < 0) /* Reset quiet flag if we freed some blocks */
 686                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
 687         mutex_unlock(&sdp->sd_quota_mutex);
 688 }
 689
 690 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
 691                                   unsigned off, void *buf, unsigned bytes)
 692 {
 693         struct inode *inode = &ip->i_inode;
 694         struct gfs2_sbd *sdp = GFS2_SB(inode);
 695         struct address_space *mapping = inode->i_mapping;
 696         struct page *page;
 697         struct buffer_head *bh;
 698         void *kaddr;
 699         u64 blk;
 700         unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
 701         unsigned to_write = bytes, pg_off = off;
 702         int done = 0;
 703
 704         blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
 705         boff = off % bsize;
 706
 707         page = find_or_create_page(mapping, index, GFP_NOFS);
 708         if (!page)
 709                 return -ENOMEM;
 710         if (!page_has_buffers(page))
 711                 create_empty_buffers(page, bsize, 0);
 712
 713         bh = page_buffers(page);
 714         while (!done) {
 715                 /* Find the beginning block within the page */
 716                 if (pg_off >= ((bnum * bsize) + bsize)) {
 717                         bh = bh->b_this_page;
 718                         bnum++;
 719                         blk++;
 720                         continue;
 721                 }
 722                 if (!buffer_mapped(bh)) {
 723                         gfs2_block_map(inode, blk, bh, 1);
 724                         if (!buffer_mapped(bh))
 725                                 goto unlock_out;
 726                         /* If it's a newly allocated disk block, zero it */
 727                         if (buffer_new(bh))
 728                                 zero_user(page, bnum * bsize, bh->b_size);
 729                 }
 730                 if (PageUptodate(page))
 731                         set_buffer_uptodate(bh);
 732                 if (!buffer_uptodate(bh)) {
 733                         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
 734                         wait_on_buffer(bh);
 735                         if (!buffer_uptodate(bh))
 736                                 goto unlock_out;
 737                 }
 738                 if (gfs2_is_jdata(ip))
 739                         gfs2_trans_add_data(ip->i_gl, bh);
 740                 else
 741                         gfs2_ordered_add_inode(ip);
 742
 743                 /* If we need to write to the next block as well */
 744                 if (to_write > (bsize - boff)) {
 745                         pg_off += (bsize - boff);
 746                         to_write -= (bsize - boff);
 747                         boff = pg_off % bsize;
 748                         continue;
 749                 }
 750                 done = 1;
 751         }
 752
 753         /* Write to the page, now that we have setup the buffer(s) */
 754         kaddr = kmap_atomic(page);
 755         memcpy(kaddr + off, buf, bytes);
 756         flush_dcache_page(page);
 757         kunmap_atomic(kaddr);
 758         unlock_page(page);
 759         put_page(page);
 760
 761         return 0;
 762
 763 unlock_out:
 764         unlock_page(page);
 765         put_page(page);
 766         return -EIO;
 767 }
 768
 769 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
 770                                  loff_t loc)
 771 {
 772         unsigned long pg_beg;
 773         unsigned pg_off, nbytes, overflow = 0;
 774         int pg_oflow = 0, error;
 775         void *ptr;
 776
 777         nbytes = sizeof(struct gfs2_quota);
 778
 779         pg_beg = loc >> PAGE_SHIFT;
 780         pg_off = loc % PAGE_SIZE;
 781
 782         /* If the quota straddles a page boundary, split the write in two */
 783         if ((pg_off + nbytes) > PAGE_SIZE) {
 784                 pg_oflow = 1;
 785                 overflow = (pg_off + nbytes) - PAGE_SIZE;
 786         }
 787
 788         ptr = qp;
 789         error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
 790                                        nbytes - overflow);
 791         /* If there's an overflow, write the remaining bytes to the next page */
 792         if (!error && pg_oflow)
 793                 error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
 794                                                ptr + nbytes - overflow,
 795                                                overflow);
 796         return error;
 797 }
 798
 799 /**
 800  * gfs2_adjust_quota - adjust record of current block usage
 801  * @ip: The quota inode
 802  * @loc: Offset of the entry in the quota file
 803  * @change: The amount of usage change to record
 804  * @qd: The quota data
 805  * @fdq: The updated limits to record
 806  *
 807  * This function was mostly borrowed from gfs2_block_truncate_page which was
 808  * in turn mostly borrowed from ext3
 809  *
 810  * Returns: 0 or -ve on error
 811  */
 812
 813 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
 814                              s64 change, struct gfs2_quota_data *qd,
 815                              struct qc_dqblk *fdq)
 816 {
 817         struct inode *inode = &ip->i_inode;
 818         struct gfs2_sbd *sdp = GFS2_SB(inode);
 819         struct gfs2_quota q;
 820         int err;
 821         u64 size;
 822
 823         if (gfs2_is_stuffed(ip)) {
 824                 err = gfs2_unstuff_dinode(ip, NULL);
 825                 if (err)
 826                         return err;
 827         }
 828
 829         memset(&q, 0, sizeof(struct gfs2_quota));
 830         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
 831         if (err < 0)
 832                 return err;
 833
 834         loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
 835         err = -EIO;
 836         be64_add_cpu(&q.qu_value, change);
 837         if (((s64)be64_to_cpu(q.qu_value)) < 0)
 838                 q.qu_value = 0; /* Never go negative on quota usage */
 839         qd->qd_qb.qb_value = q.qu_value;
 840         if (fdq) {
 841                 if (fdq->d_fieldmask & QC_SPC_SOFT) {
 842                         q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
 843                         qd->qd_qb.qb_warn = q.qu_warn;
 844                 }
 845                 if (fdq->d_fieldmask & QC_SPC_HARD) {
 846                         q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
 847                         qd->qd_qb.qb_limit = q.qu_limit;
 848                 }
 849                 if (fdq->d_fieldmask & QC_SPACE) {
 850                         q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
 851                         qd->qd_qb.qb_value = q.qu_value;
 852                 }
 853         }
 854
 855         err = gfs2_write_disk_quota(ip, &q, loc);
 856         if (!err) {
 857                 size = loc + sizeof(struct gfs2_quota);
 858                 if (size > inode->i_size)
 859                         i_size_write(inode, size);
 860                 inode->i_mtime = inode->i_atime = current_time(inode);
 861                 mark_inode_dirty(inode);
 862                 set_bit(QDF_REFRESH, &qd->qd_flags);
 863         }
 864
 865         return err;
 866 }
 867
 868 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 869 {
 870         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
 871         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 872         struct gfs2_alloc_parms ap = { .aflags = 0, };
 873         unsigned int data_blocks, ind_blocks;
 874         struct gfs2_holder *ghs, i_gh;
 875         unsigned int qx, x;
 876         struct gfs2_quota_data *qd;
 877         unsigned reserved;
 878         loff_t offset;
 879         unsigned int nalloc = 0, blocks;
 880         int error;
 881
 882         error = gfs2_rsqa_alloc(ip);
 883         if (error)
 884                 return error;
 885
 886         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
 887                               &data_blocks, &ind_blocks);
 888
 889         ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
 890         if (!ghs)
 891                 return -ENOMEM;
 892
 893         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 894         inode_lock(&ip->i_inode);
 895         for (qx = 0; qx < num_qd; qx++) {
 896                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
 897                                            GL_NOCACHE, &ghs[qx]);
 898                 if (error)
 899                         goto out;
 900         }
 901
 902         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
 903         if (error)
 904                 goto out;
 905
 906         for (x = 0; x < num_qd; x++) {
 907                 offset = qd2offset(qda[x]);
 908                 if (gfs2_write_alloc_required(ip, offset,
 909                                               sizeof(struct gfs2_quota)))
 910                         nalloc++;
 911         }
 912
 913         /*
 914          * 1 blk for unstuffing inode if stuffed. We add this extra
 915          * block to the reservation unconditionally. If the inode
 916          * doesn't need unstuffing, the block will be released to the
 917          * rgrp since it won't be allocated during the transaction
 918          */
 919         /* +3 in the end for unstuffing block, inode size update block
 920          * and another block in case quota straddles page boundary and
 921          * two blocks need to be updated instead of 1 */
 922         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
 923
 924         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
 925         ap.target = reserved;
 926         error = gfs2_inplace_reserve(ip, &ap);
 927         if (error)
 928                 goto out_alloc;
 929
 930         if (nalloc)
 931                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
 932
 933         error = gfs2_trans_begin(sdp, blocks, 0);
 934         if (error)
 935                 goto out_ipres;
 936
 937         for (x = 0; x < num_qd; x++) {
 938                 qd = qda[x];
 939                 offset = qd2offset(qd);
 940                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
 941                 if (error)
 942                         goto out_end_trans;
 943
 944                 do_qc(qd, -qd->qd_change_sync);
 945                 set_bit(QDF_REFRESH, &qd->qd_flags);
 946         }
 947
 948         error = 0;
 949
 950 out_end_trans:
 951         gfs2_trans_end(sdp);
 952 out_ipres:
 953         gfs2_inplace_release(ip);
 954 out_alloc:
 955         gfs2_glock_dq_uninit(&i_gh);
 956 out:
 957         while (qx--)
 958                 gfs2_glock_dq_uninit(&ghs[qx]);
 959         inode_unlock(&ip->i_inode);
 960         kfree(ghs);
 961         gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
 962                        GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
 963         return error;
 964 }
 965
 966 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
 967 {
 968         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 969         struct gfs2_quota q;
 970         struct gfs2_quota_lvb *qlvb;
 971         loff_t pos;
 972         int error;
 973
 974         memset(&q, 0, sizeof(struct gfs2_quota));
 975         pos = qd2offset(qd);
 976         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
 977         if (error < 0)
 978                 return error;
 979
 980         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
 981         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
 982         qlvb->__pad = 0;
 983         qlvb->qb_limit = q.qu_limit;
 984         qlvb->qb_warn = q.qu_warn;
 985         qlvb->qb_value = q.qu_value;
 986         qd->qd_qb = *qlvb;
 987
 988         return 0;
 989 }
 990
 991 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
 992                     struct gfs2_holder *q_gh)
 993 {
 994         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 995         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 996         struct gfs2_holder i_gh;
 997         int error;
 998
 999 restart:
1000         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1001         if (error)
1002                 return error;
1003
1004         if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1005                 force_refresh = FORCE;
1006
1007         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1008
1009         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1010                 gfs2_glock_dq_uninit(q_gh);
1011                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1012                                            GL_NOCACHE, q_gh);
1013                 if (error)
1014                         return error;
1015
1016                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1017                 if (error)
1018                         goto fail;
1019
1020                 error = update_qd(sdp, qd);
1021                 if (error)
1022                         goto fail_gunlock;
1023
1024                 gfs2_glock_dq_uninit(&i_gh);
1025                 gfs2_glock_dq_uninit(q_gh);
1026                 force_refresh = 0;
1027                 goto restart;
1028         }
1029
1030         return 0;
1031
1032 fail_gunlock:
1033         gfs2_glock_dq_uninit(&i_gh);
1034 fail:
1035         gfs2_glock_dq_uninit(q_gh);
1036         return error;
1037 }
1038
1039 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1040 {
1041         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1042         struct gfs2_quota_data *qd;
1043         u32 x;
1044         int error = 0;
1045
1046         if (capable(CAP_SYS_RESOURCE) ||
1047             sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1048                 return 0;
1049
1050         error = gfs2_quota_hold(ip, uid, gid);
1051         if (error)
1052                 return error;
1053
1054         sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1055              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1056
1057         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1058                 qd = ip->i_qadata->qa_qd[x];
1059                 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1060                 if (error)
1061                         break;
1062         }
1063
1064         if (!error)
1065                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1066         else {
1067                 while (x--)
1068                         gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1069                 gfs2_quota_unhold(ip);
1070         }
1071
1072         return error;
1073 }
1074
1075 static int need_sync(struct gfs2_quota_data *qd)
1076 {
1077         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1078         struct gfs2_tune *gt = &sdp->sd_tune;
1079         s64 value;
1080         unsigned int num, den;
1081         int do_sync = 1;
1082
1083         if (!qd->qd_qb.qb_limit)
1084                 return 0;
1085
1086         spin_lock(&qd_lock);
1087         value = qd->qd_change;
1088         spin_unlock(&qd_lock);
1089
1090         spin_lock(&gt->gt_spin);
1091         num = gt->gt_quota_scale_num;
1092         den = gt->gt_quota_scale_den;
1093         spin_unlock(&gt->gt_spin);
1094
1095         if (value < 0)
1096                 do_sync = 0;
1097         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1098                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1099                 do_sync = 0;
1100         else {
1101                 value *= gfs2_jindex_size(sdp) * num;
1102                 value = div_s64(value, den);
1103                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1104                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1105                         do_sync = 0;
1106         }
1107
1108         return do_sync;
1109 }
1110
1111 void gfs2_quota_unlock(struct gfs2_inode *ip)
1112 {
1113         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1114         struct gfs2_quota_data *qda[4];
1115         unsigned int count = 0;
1116         u32 x;
1117         int found;
1118
1119         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1120                 goto out;
1121
1122         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1123                 struct gfs2_quota_data *qd;
1124                 int sync;
1125
1126                 qd = ip->i_qadata->qa_qd[x];
1127                 sync = need_sync(qd);
1128
1129                 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1130                 if (!sync)
1131                         continue;
1132
1133                 spin_lock(&qd_lock);
1134                 found = qd_check_sync(sdp, qd, NULL);
1135                 spin_unlock(&qd_lock);
1136
1137                 if (!found)
1138                         continue;
1139
1140                 gfs2_assert_warn(sdp, qd->qd_change_sync);
1141                 if (bh_get(qd)) {
1142                         clear_bit(QDF_LOCKED, &qd->qd_flags);
1143                         slot_put(qd);
1144                         qd_put(qd);
1145                         continue;
1146                 }
1147
1148                 qda[count++] = qd;
1149         }
1150
1151         if (count) {
1152                 do_sync(count, qda);
1153                 for (x = 0; x < count; x++)
1154                         qd_unlock(qda[x]);
1155         }
1156
1157 out:
1158         gfs2_quota_unhold(ip);
1159 }
1160
1161 #define MAX_LINE 256
1162
1163 static int print_message(struct gfs2_quota_data *qd, char *type)
1164 {
1165         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1166
1167         fs_info(sdp, "quota %s for %s %u\n",
1168                 type,
1169                 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1170                 from_kqid(&init_user_ns, qd->qd_id));
1171
1172         return 0;
1173 }
1174
1175 /**
1176  * gfs2_quota_check - check if allocating new blocks will exceed quota
1177  * @ip:  The inode for which this check is being performed
1178  * @uid: The uid to check against
1179  * @gid: The gid to check against
1180  * @ap:  The allocation parameters. ap->target contains the requested
1181  *       blocks. ap->min_target, if set, contains the minimum blks
1182  *       requested.
1183  *
1184  * Returns: 0 on success.
1185  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1186  *                  quota must allow at least min_req blks for success and
1187  *                  ap->allowed is set to the number of blocks allowed
1188  *
1189  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1190  *                  of blocks available.
1191  */
1192 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1193                      struct gfs2_alloc_parms *ap)
1194 {
1195         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1196         struct gfs2_quota_data *qd;
1197         s64 value, warn, limit;
1198         u32 x;
1199         int error = 0;
1200
1201         ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1202         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1203                 return 0;
1204
1205         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1206                 return 0;
1207
1208         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1209                 qd = ip->i_qadata->qa_qd[x];
1210
1211                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1212                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1213                         continue;
1214
1215                 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1216                 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1217                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1218                 spin_lock(&qd_lock);
1219                 value += qd->qd_change;
1220                 spin_unlock(&qd_lock);
1221
1222                 if (limit > 0 && (limit - value) < ap->allowed)
1223                         ap->allowed = limit - value;
1224                 /* If we can't meet the target */
1225                 if (limit && limit < (value + (s64)ap->target)) {
1226                         /* If no min_target specified or we don't meet
1227                          * min_target, return -EDQUOT */
1228                         if (!ap->min_target || ap->min_target > ap->allowed) {
1229                                 if (!test_and_set_bit(QDF_QMSG_QUIET,
1230                                                       &qd->qd_flags)) {
1231                                         print_message(qd, "exceeded");
1232                                         quota_send_warning(qd->qd_id,
1233                                                            sdp->sd_vfs->s_dev,
1234                                                            QUOTA_NL_BHARDWARN);
1235                                 }
1236                                 error = -EDQUOT;
1237                                 break;
1238                         }
1239                 } else if (warn && warn < value &&
1240                            time_after_eq(jiffies, qd->qd_last_warn +
1241                                          gfs2_tune_get(sdp, gt_quota_warn_period)
1242                                          * HZ)) {
1243                         quota_send_warning(qd->qd_id,
1244                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1245                         error = print_message(qd, "warning");
1246                         qd->qd_last_warn = jiffies;
1247                 }
1248         }
1249         return error;
1250 }
1251
1252 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1253                        kuid_t uid, kgid_t gid)
1254 {
1255         struct gfs2_quota_data *qd;
1256         u32 x;
1257         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1258
1259         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1260             gfs2_assert_warn(sdp, change))
1261                 return;
1262         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1263                 return;
1264
1265         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1266                 qd = ip->i_qadata->qa_qd[x];
1267
1268                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1269                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1270                         do_qc(qd, change);
1271                 }
1272         }
1273 }
1274
1275 int gfs2_quota_sync(struct super_block *sb, int type)
1276 {
1277         struct gfs2_sbd *sdp = sb->s_fs_info;
1278         struct gfs2_quota_data **qda;
1279         unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1280         unsigned int num_qd;
1281         unsigned int x;
1282         int error = 0;
1283
1284         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1285         if (!qda)
1286                 return -ENOMEM;
1287
1288         mutex_lock(&sdp->sd_quota_sync_mutex);
1289         sdp->sd_quota_sync_gen++;
1290
1291         do {
1292                 num_qd = 0;
1293
1294                 for (;;) {
1295                         error = qd_fish(sdp, qda + num_qd);
1296                         if (error || !qda[num_qd])
1297                                 break;
1298                         if (++num_qd == max_qd)
1299                                 break;
1300                 }
1301
1302                 if (num_qd) {
1303                         if (!error)
1304                                 error = do_sync(num_qd, qda);
1305                         if (!error)
1306                                 for (x = 0; x < num_qd; x++)
1307                                         qda[x]->qd_sync_gen =
1308                                                 sdp->sd_quota_sync_gen;
1309
1310                         for (x = 0; x < num_qd; x++)
1311                                 qd_unlock(qda[x]);
1312                 }
1313         } while (!error && num_qd == max_qd);
1314
1315         mutex_unlock(&sdp->sd_quota_sync_mutex);
1316         kfree(qda);
1317
1318         return error;
1319 }
1320
1321 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1322 {
1323         struct gfs2_quota_data *qd;
1324         struct gfs2_holder q_gh;
1325         int error;
1326
1327         error = qd_get(sdp, qid, &qd);
1328         if (error)
1329                 return error;
1330
1331         error = do_glock(qd, FORCE, &q_gh);
1332         if (!error)
1333                 gfs2_glock_dq_uninit(&q_gh);
1334
1335         qd_put(qd);
1336         return error;
1337 }
1338
1339 int gfs2_quota_init(struct gfs2_sbd *sdp)
1340 {
1341         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1342         u64 size = i_size_read(sdp->sd_qc_inode);
1343         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1344         unsigned int x, slot = 0;
1345         unsigned int found = 0;
1346         unsigned int hash;
1347         unsigned int bm_size;
1348         u64 dblock;
1349         u32 extlen = 0;
1350         int error;
1351
1352         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1353                 return -EIO;
1354
1355         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1356         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1357         bm_size *= sizeof(unsigned long);
1358         error = -ENOMEM;
1359         sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1360         if (sdp->sd_quota_bitmap == NULL)
1361                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1362                                                  __GFP_ZERO, PAGE_KERNEL);
1363         if (!sdp->sd_quota_bitmap)
1364                 return error;
1365
1366         for (x = 0; x < blocks; x++) {
1367                 struct buffer_head *bh;
1368                 const struct gfs2_quota_change *qc;
1369                 unsigned int y;
1370
1371                 if (!extlen) {
1372                         int new = 0;
1373                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1374                         if (error)
1375                                 goto fail;
1376                 }
1377                 error = -EIO;
1378                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1379                 if (!bh)
1380                         goto fail;
1381                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1382                         brelse(bh);
1383                         goto fail;
1384                 }
1385
1386                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1387                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1388                      y++, slot++) {
1389                         struct gfs2_quota_data *qd;
1390                         s64 qc_change = be64_to_cpu(qc->qc_change);
1391                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1392                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1393                                                 USRQUOTA : GRPQUOTA;
1394                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1395                                                       be32_to_cpu(qc->qc_id));
1396                         qc++;
1397                         if (!qc_change)
1398                                 continue;
1399
1400                         hash = gfs2_qd_hash(sdp, qc_id);
1401                         qd = qd_alloc(hash, sdp, qc_id);
1402                         if (qd == NULL) {
1403                                 brelse(bh);
1404                                 goto fail;
1405                         }
1406
1407                         set_bit(QDF_CHANGE, &qd->qd_flags);
1408                         qd->qd_change = qc_change;
1409                         qd->qd_slot = slot;
1410                         qd->qd_slot_count = 1;
1411
1412                         spin_lock(&qd_lock);
1413                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1414                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1415                         atomic_inc(&sdp->sd_quota_count);
1416                         spin_unlock(&qd_lock);
1417
1418                         spin_lock_bucket(hash);
1419                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1420                         spin_unlock_bucket(hash);
1421
1422                         found++;
1423                 }
1424
1425                 brelse(bh);
1426                 dblock++;
1427                 extlen--;
1428         }
1429
1430         if (found)
1431                 fs_info(sdp, "found %u quota changes\n", found);
1432
1433         return 0;
1434
1435 fail:
1436         gfs2_quota_cleanup(sdp);
1437         return error;
1438 }
1439
1440 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1441 {
1442         struct list_head *head = &sdp->sd_quota_list;
1443         struct gfs2_quota_data *qd;
1444
1445         spin_lock(&qd_lock);
1446         while (!list_empty(head)) {
1447                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1448
1449                 list_del(&qd->qd_list);
1450
1451                 /* Also remove if this qd exists in the reclaim list */
1452                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1453                 atomic_dec(&sdp->sd_quota_count);
1454                 spin_unlock(&qd_lock);
1455
1456                 spin_lock_bucket(qd->qd_hash);
1457                 hlist_bl_del_rcu(&qd->qd_hlist);
1458                 spin_unlock_bucket(qd->qd_hash);
1459
1460                 gfs2_assert_warn(sdp, !qd->qd_change);
1461                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1462                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1463
1464                 gfs2_glock_put(qd->qd_gl);
1465                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1466
1467                 spin_lock(&qd_lock);
1468         }
1469         spin_unlock(&qd_lock);
1470
1471         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1472
1473         kvfree(sdp->sd_quota_bitmap);
1474         sdp->sd_quota_bitmap = NULL;
1475 }
1476
1477 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1478 {
1479         if (error == 0 || error == -EROFS)
1480                 return;
1481         if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
1482                 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1483                 sdp->sd_log_error = error;
1484                 wake_up(&sdp->sd_logd_waitq);
1485         }
1486 }
1487
1488 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1489                                int (*fxn)(struct super_block *sb, int type),
1490                                unsigned long t, unsigned long *timeo,
1491                                unsigned int *new_timeo)
1492 {
1493         if (t >= *timeo) {
1494                 int error = fxn(sdp->sd_vfs, 0);
1495                 quotad_error(sdp, msg, error);
1496                 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1497         } else {
1498                 *timeo -= t;
1499         }
1500 }
1501
1502 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1503 {
1504         struct gfs2_inode *ip;
1505
1506         while(1) {
1507                 ip = NULL;
1508                 spin_lock(&sdp->sd_trunc_lock);
1509                 if (!list_empty(&sdp->sd_trunc_list)) {
1510                         ip = list_entry(sdp->sd_trunc_list.next,
1511                                         struct gfs2_inode, i_trunc_list);
1512                         list_del_init(&ip->i_trunc_list);
1513                 }
1514                 spin_unlock(&sdp->sd_trunc_lock);
1515                 if (ip == NULL)
1516                         return;
1517                 gfs2_glock_finish_truncate(ip);
1518         }
1519 }
1520
1521 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1522         if (!sdp->sd_statfs_force_sync) {
1523                 sdp->sd_statfs_force_sync = 1;
1524                 wake_up(&sdp->sd_quota_wait);
1525         }
1526 }
1527
1528
1529 /**
1530  * gfs2_quotad - Write cached quota changes into the quota file
1531  * @sdp: Pointer to GFS2 superblock
1532  *
1533  */
1534
1535 int gfs2_quotad(void *data)
1536 {
1537         struct gfs2_sbd *sdp = data;
1538         struct gfs2_tune *tune = &sdp->sd_tune;
1539         unsigned long statfs_timeo = 0;
1540         unsigned long quotad_timeo = 0;
1541         unsigned long t = 0;
1542         DEFINE_WAIT(wait);
1543         int empty;
1544
1545         while (!kthread_should_stop()) {
1546
1547                 /* Update the master statfs file */
1548                 if (sdp->sd_statfs_force_sync) {
1549                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1550                         quotad_error(sdp, "statfs", error);
1551                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1552                 }
1553                 else
1554                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1555                                            &statfs_timeo,
1556                                            &tune->gt_statfs_quantum);
1557
1558                 /* Update quota file */
1559                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1560                                    &quotad_timeo, &tune->gt_quota_quantum);
1561
1562                 /* Check for & recover partially truncated inodes */
1563                 quotad_check_trunc_list(sdp);
1564
1565                 try_to_freeze();
1566
1567                 t = min(quotad_timeo, statfs_timeo);
1568
1569                 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1570                 spin_lock(&sdp->sd_trunc_lock);
1571                 empty = list_empty(&sdp->sd_trunc_list);
1572                 spin_unlock(&sdp->sd_trunc_lock);
1573                 if (empty && !sdp->sd_statfs_force_sync)
1574                         t -= schedule_timeout(t);
1575                 else
1576                         t = 0;
1577                 finish_wait(&sdp->sd_quota_wait, &wait);
1578         }
1579
1580         return 0;
1581 }
1582
1583 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1584 {
1585         struct gfs2_sbd *sdp = sb->s_fs_info;
1586
1587         memset(state, 0, sizeof(*state));
1588
1589         switch (sdp->sd_args.ar_quota) {
1590         case GFS2_QUOTA_ON:
1591                 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1592                 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1593                 /*FALLTHRU*/
1594         case GFS2_QUOTA_ACCOUNT:
1595                 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1596                                                   QCI_SYSFILE;
1597                 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1598                                                   QCI_SYSFILE;
1599                 break;
1600         case GFS2_QUOTA_OFF:
1601                 break;
1602         }
1603         if (sdp->sd_quota_inode) {
1604                 state->s_state[USRQUOTA].ino =
1605                                         GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1606                 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1607         }
1608         state->s_state[USRQUOTA].nextents = 1;  /* unsupported */
1609         state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1610         state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1611         return 0;
1612 }
1613
1614 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1615                           struct qc_dqblk *fdq)
1616 {
1617         struct gfs2_sbd *sdp = sb->s_fs_info;
1618         struct gfs2_quota_lvb *qlvb;
1619         struct gfs2_quota_data *qd;
1620         struct gfs2_holder q_gh;
1621         int error;
1622
1623         memset(fdq, 0, sizeof(*fdq));
1624
1625         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1626                 return -ESRCH; /* Crazy XFS error code */
1627
1628         if ((qid.type != USRQUOTA) &&
1629             (qid.type != GRPQUOTA))
1630                 return -EINVAL;
1631
1632         error = qd_get(sdp, qid, &qd);
1633         if (error)
1634                 return error;
1635         error = do_glock(qd, FORCE, &q_gh);
1636         if (error)
1637                 goto out;
1638
1639         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1640         fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1641         fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1642         fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1643
1644         gfs2_glock_dq_uninit(&q_gh);
1645 out:
1646         qd_put(qd);
1647         return error;
1648 }
1649
1650 /* GFS2 only supports a subset of the XFS fields */
1651 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1652
1653 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1654                           struct qc_dqblk *fdq)
1655 {
1656         struct gfs2_sbd *sdp = sb->s_fs_info;
1657         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1658         struct gfs2_quota_data *qd;
1659         struct gfs2_holder q_gh, i_gh;
1660         unsigned int data_blocks, ind_blocks;
1661         unsigned int blocks = 0;
1662         int alloc_required;
1663         loff_t offset;
1664         int error;
1665
1666         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1667                 return -ESRCH; /* Crazy XFS error code */
1668
1669         if ((qid.type != USRQUOTA) &&
1670             (qid.type != GRPQUOTA))
1671                 return -EINVAL;
1672
1673         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1674                 return -EINVAL;
1675
1676         error = qd_get(sdp, qid, &qd);
1677         if (error)
1678                 return error;
1679
1680         error = gfs2_rsqa_alloc(ip);
1681         if (error)
1682                 goto out_put;
1683
1684         inode_lock(&ip->i_inode);
1685         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1686         if (error)
1687                 goto out_unlockput;
1688         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1689         if (error)
1690                 goto out_q;
1691
1692         /* Check for existing entry, if none then alloc new blocks */
1693         error = update_qd(sdp, qd);
1694         if (error)
1695                 goto out_i;
1696
1697         /* If nothing has changed, this is a no-op */
1698         if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1699             ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1700                 fdq->d_fieldmask ^= QC_SPC_SOFT;
1701
1702         if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1703             ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1704                 fdq->d_fieldmask ^= QC_SPC_HARD;
1705
1706         if ((fdq->d_fieldmask & QC_SPACE) &&
1707             ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1708                 fdq->d_fieldmask ^= QC_SPACE;
1709
1710         if (fdq->d_fieldmask == 0)
1711                 goto out_i;
1712
1713         offset = qd2offset(qd);
1714         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1715         if (gfs2_is_stuffed(ip))
1716                 alloc_required = 1;
1717         if (alloc_required) {
1718                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1719                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1720                                        &data_blocks, &ind_blocks);
1721                 blocks = 1 + data_blocks + ind_blocks;
1722                 ap.target = blocks;
1723                 error = gfs2_inplace_reserve(ip, &ap);
1724                 if (error)
1725                         goto out_i;
1726                 blocks += gfs2_rg_blocks(ip, blocks);
1727         }
1728
1729         /* Some quotas span block boundaries and can update two blocks,
1730            adding an extra block to the transaction to handle such quotas */
1731         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1732         if (error)
1733                 goto out_release;
1734
1735         /* Apply changes */
1736         error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1737         if (!error)
1738                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1739
1740         gfs2_trans_end(sdp);
1741 out_release:
1742         if (alloc_required)
1743                 gfs2_inplace_release(ip);
1744 out_i:
1745         gfs2_glock_dq_uninit(&i_gh);
1746 out_q:
1747         gfs2_glock_dq_uninit(&q_gh);
1748 out_unlockput:
1749         inode_unlock(&ip->i_inode);
1750 out_put:
1751         qd_put(qd);
1752         return error;
1753 }
1754
1755 const struct quotactl_ops gfs2_quotactl_ops = {
1756         .quota_sync     = gfs2_quota_sync,
1757         .get_state      = gfs2_quota_get_state,
1758         .get_dqblk      = gfs2_get_dqblk,
1759         .set_dqblk      = gfs2_set_dqblk,
1760 };
1761
1762 void __init gfs2_quota_hash_init(void)
1763 {
1764         unsigned i;
1765
1766         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1767                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1768 }