fs/gfs2/quota.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 /*
   8  * Quota change tags are associated with each transaction that allocates or
   9  * deallocates space.  Those changes are accumulated locally to each node (in a
  10  * per-node file) and then are periodically synced to the quota file.  This
  11  * avoids the bottleneck of constantly touching the quota file, but introduces
  12  * fuzziness in the current usage value of IDs that are being used on different
  13  * nodes in the cluster simultaneously.  So, it is possible for a user on
  14  * multiple nodes to overrun their quota, but that overrun is controlable.
  15  * Since quota tags are part of transactions, there is no need for a quota check
  16  * program to be run on node crashes or anything like that.
  17  *
  18  * There are couple of knobs that let the administrator manage the quota
  19  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
  20  * sitting on one node before being synced to the quota file.  (The default is
  21  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
  22  * of quota file syncs increases as the user moves closer to their limit.  The
  23  * more frequent the syncs, the more accurate the quota enforcement, but that
  24  * means that there is more contention between the nodes for the quota file.
  25  * The default value is one.  This sets the maximum theoretical quota overrun
  26  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
  27  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
  28  * number greater than one makes quota syncs more frequent and reduces the
  29  * maximum overrun.  Numbers less than one (but greater than zero) make quota
  30  * syncs less frequent.
  31  *
  32  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
  33  * the quota file, so it is not being constantly read.
  34  */
  35
  36 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  37
  38 #include <linux/sched.h>
  39 #include <linux/slab.h>
  40 #include <linux/mm.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/completion.h>
  43 #include <linux/buffer_head.h>
  44 #include <linux/sort.h>
  45 #include <linux/fs.h>
  46 #include <linux/bio.h>
  47 #include <linux/gfs2_ondisk.h>
  48 #include <linux/kthread.h>
  49 #include <linux/freezer.h>
  50 #include <linux/quota.h>
  51 #include <linux/dqblk_xfs.h>
  52 #include <linux/lockref.h>
  53 #include <linux/list_lru.h>
  54 #include <linux/rcupdate.h>
  55 #include <linux/rculist_bl.h>
  56 #include <linux/bit_spinlock.h>
  57 #include <linux/jhash.h>
  58 #include <linux/vmalloc.h>
  59
  60 #include "gfs2.h"
  61 #include "incore.h"
  62 #include "bmap.h"
  63 #include "glock.h"
  64 #include "glops.h"
  65 #include "log.h"
  66 #include "meta_io.h"
  67 #include "quota.h"
  68 #include "rgrp.h"
  69 #include "super.h"
  70 #include "trans.h"
  71 #include "inode.h"
  72 #include "util.h"
  73
  74 #define GFS2_QD_HASH_SHIFT      12
  75 #define GFS2_QD_HASH_SIZE       BIT(GFS2_QD_HASH_SHIFT)
  76 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
  77
  78 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
  79 /*                     -> sd_bitmap_lock                              */
  80 static DEFINE_SPINLOCK(qd_lock);
  81 struct list_lru gfs2_qd_lru;
  82
  83 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
  84
  85 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
  86                                  const struct kqid qid)
  87 {
  88         unsigned int h;
  89
  90         h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
  91         h = jhash(&qid, sizeof(struct kqid), h);
  92
  93         return h & GFS2_QD_HASH_MASK;
  94 }
  95
  96 static inline void spin_lock_bucket(unsigned int hash)
  97 {
  98         hlist_bl_lock(&qd_hash_table[hash]);
  99 }
 100
 101 static inline void spin_unlock_bucket(unsigned int hash)
 102 {
 103         hlist_bl_unlock(&qd_hash_table[hash]);
 104 }
 105
 106 static void gfs2_qd_dealloc(struct rcu_head *rcu)
 107 {
 108         struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
 109         kmem_cache_free(gfs2_quotad_cachep, qd);
 110 }
 111
 112 static void gfs2_qd_dispose(struct list_head *list)
 113 {
 114         struct gfs2_quota_data *qd;
 115         struct gfs2_sbd *sdp;
 116
 117         while (!list_empty(list)) {
 118                 qd = list_entry(list->next, struct gfs2_quota_data, qd_lru);
 119                 sdp = qd->qd_gl->gl_name.ln_sbd;
 120
 121                 list_del(&qd->qd_lru);
 122
 123                 /* Free from the filesystem-specific list */
 124                 spin_lock(&qd_lock);
 125                 list_del(&qd->qd_list);
 126                 spin_unlock(&qd_lock);
 127
 128                 spin_lock_bucket(qd->qd_hash);
 129                 hlist_bl_del_rcu(&qd->qd_hlist);
 130                 spin_unlock_bucket(qd->qd_hash);
 131
 132                 gfs2_assert_warn(sdp, !qd->qd_change);
 133                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
 134                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
 135
 136                 gfs2_glock_put(qd->qd_gl);
 137                 atomic_dec(&sdp->sd_quota_count);
 138
 139                 /* Delete it from the common reclaim list */
 140                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
 141         }
 142 }
 143
 144
 145 static enum lru_status gfs2_qd_isolate(struct list_head *item,
 146                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
 147 {
 148         struct list_head *dispose = arg;
 149         struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
 150
 151         if (!spin_trylock(&qd->qd_lockref.lock))
 152                 return LRU_SKIP;
 153
 154         if (qd->qd_lockref.count == 0) {
 155                 lockref_mark_dead(&qd->qd_lockref);
 156                 list_lru_isolate_move(lru, &qd->qd_lru, dispose);
 157         }
 158
 159         spin_unlock(&qd->qd_lockref.lock);
 160         return LRU_REMOVED;
 161 }
 162
 163 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
 164                                          struct shrink_control *sc)
 165 {
 166         LIST_HEAD(dispose);
 167         unsigned long freed;
 168
 169         if (!(sc->gfp_mask & __GFP_FS))
 170                 return SHRINK_STOP;
 171
 172         freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
 173                                      gfs2_qd_isolate, &dispose);
 174
 175         gfs2_qd_dispose(&dispose);
 176
 177         return freed;
 178 }
 179
 180 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
 181                                           struct shrink_control *sc)
 182 {
 183         return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
 184 }
 185
 186 struct shrinker gfs2_qd_shrinker = {
 187         .count_objects = gfs2_qd_shrink_count,
 188         .scan_objects = gfs2_qd_shrink_scan,
 189         .seeks = DEFAULT_SEEKS,
 190         .flags = SHRINKER_NUMA_AWARE,
 191 };
 192
 193
 194 static u64 qd2index(struct gfs2_quota_data *qd)
 195 {
 196         struct kqid qid = qd->qd_id;
 197         return (2 * (u64)from_kqid(&init_user_ns, qid)) +
 198                 ((qid.type == USRQUOTA) ? 0 : 1);
 199 }
 200
 201 static u64 qd2offset(struct gfs2_quota_data *qd)
 202 {
 203         u64 offset;
 204
 205         offset = qd2index(qd);
 206         offset *= sizeof(struct gfs2_quota);
 207
 208         return offset;
 209 }
 210
 211 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
 212 {
 213         struct gfs2_quota_data *qd;
 214         int error;
 215
 216         qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
 217         if (!qd)
 218                 return NULL;
 219
 220         qd->qd_sbd = sdp;
 221         qd->qd_lockref.count = 1;
 222         spin_lock_init(&qd->qd_lockref.lock);
 223         qd->qd_id = qid;
 224         qd->qd_slot = -1;
 225         INIT_LIST_HEAD(&qd->qd_lru);
 226         qd->qd_hash = hash;
 227
 228         error = gfs2_glock_get(sdp, qd2index(qd),
 229                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
 230         if (error)
 231                 goto fail;
 232
 233         return qd;
 234
 235 fail:
 236         kmem_cache_free(gfs2_quotad_cachep, qd);
 237         return NULL;
 238 }
 239
 240 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
 241                                                      const struct gfs2_sbd *sdp,
 242                                                      struct kqid qid)
 243 {
 244         struct gfs2_quota_data *qd;
 245         struct hlist_bl_node *h;
 246
 247         hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
 248                 if (!qid_eq(qd->qd_id, qid))
 249                         continue;
 250                 if (qd->qd_sbd != sdp)
 251                         continue;
 252                 if (lockref_get_not_dead(&qd->qd_lockref)) {
 253                         list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
 254                         return qd;
 255                 }
 256         }
 257
 258         return NULL;
 259 }
 260
 261
 262 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
 263                   struct gfs2_quota_data **qdp)
 264 {
 265         struct gfs2_quota_data *qd, *new_qd;
 266         unsigned int hash = gfs2_qd_hash(sdp, qid);
 267
 268         rcu_read_lock();
 269         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
 270         rcu_read_unlock();
 271
 272         if (qd)
 273                 return 0;
 274
 275         new_qd = qd_alloc(hash, sdp, qid);
 276         if (!new_qd)
 277                 return -ENOMEM;
 278
 279         spin_lock(&qd_lock);
 280         spin_lock_bucket(hash);
 281         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
 282         if (qd == NULL) {
 283                 *qdp = new_qd;
 284                 list_add(&new_qd->qd_list, &sdp->sd_quota_list);
 285                 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
 286                 atomic_inc(&sdp->sd_quota_count);
 287         }
 288         spin_unlock_bucket(hash);
 289         spin_unlock(&qd_lock);
 290
 291         if (qd) {
 292                 gfs2_glock_put(new_qd->qd_gl);
 293                 kmem_cache_free(gfs2_quotad_cachep, new_qd);
 294         }
 295
 296         return 0;
 297 }
 298
 299
 300 static void qd_hold(struct gfs2_quota_data *qd)
 301 {
 302         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 303         gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
 304         lockref_get(&qd->qd_lockref);
 305 }
 306
 307 static void qd_put(struct gfs2_quota_data *qd)
 308 {
 309         if (lockref_put_or_lock(&qd->qd_lockref))
 310                 return;
 311
 312         qd->qd_lockref.count = 0;
 313         list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
 314         spin_unlock(&qd->qd_lockref.lock);
 315
 316 }
 317
 318 static int slot_get(struct gfs2_quota_data *qd)
 319 {
 320         struct gfs2_sbd *sdp = qd->qd_sbd;
 321         unsigned int bit;
 322         int error = 0;
 323
 324         spin_lock(&sdp->sd_bitmap_lock);
 325         if (qd->qd_slot_count != 0)
 326                 goto out;
 327
 328         error = -ENOSPC;
 329         bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
 330         if (bit < sdp->sd_quota_slots) {
 331                 set_bit(bit, sdp->sd_quota_bitmap);
 332                 qd->qd_slot = bit;
 333                 error = 0;
 334 out:
 335                 qd->qd_slot_count++;
 336         }
 337         spin_unlock(&sdp->sd_bitmap_lock);
 338
 339         return error;
 340 }
 341
 342 static void slot_hold(struct gfs2_quota_data *qd)
 343 {
 344         struct gfs2_sbd *sdp = qd->qd_sbd;
 345
 346         spin_lock(&sdp->sd_bitmap_lock);
 347         gfs2_assert(sdp, qd->qd_slot_count);
 348         qd->qd_slot_count++;
 349         spin_unlock(&sdp->sd_bitmap_lock);
 350 }
 351
 352 static void slot_put(struct gfs2_quota_data *qd)
 353 {
 354         struct gfs2_sbd *sdp = qd->qd_sbd;
 355
 356         spin_lock(&sdp->sd_bitmap_lock);
 357         gfs2_assert(sdp, qd->qd_slot_count);
 358         if (!--qd->qd_slot_count) {
 359                 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
 360                 qd->qd_slot = -1;
 361         }
 362         spin_unlock(&sdp->sd_bitmap_lock);
 363 }
 364
 365 static int bh_get(struct gfs2_quota_data *qd)
 366 {
 367         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 368         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 369         unsigned int block, offset;
 370         struct buffer_head *bh;
 371         int error;
 372         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 373
 374         mutex_lock(&sdp->sd_quota_mutex);
 375
 376         if (qd->qd_bh_count++) {
 377                 mutex_unlock(&sdp->sd_quota_mutex);
 378                 return 0;
 379         }
 380
 381         block = qd->qd_slot / sdp->sd_qc_per_block;
 382         offset = qd->qd_slot % sdp->sd_qc_per_block;
 383
 384         bh_map.b_size = BIT(ip->i_inode.i_blkbits);
 385         error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
 386         if (error)
 387                 goto fail;
 388         error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, 0, &bh);
 389         if (error)
 390                 goto fail;
 391         error = -EIO;
 392         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
 393                 goto fail_brelse;
 394
 395         qd->qd_bh = bh;
 396         qd->qd_bh_qc = (struct gfs2_quota_change *)
 397                 (bh->b_data + sizeof(struct gfs2_meta_header) +
 398                  offset * sizeof(struct gfs2_quota_change));
 399
 400         mutex_unlock(&sdp->sd_quota_mutex);
 401
 402         return 0;
 403
 404 fail_brelse:
 405         brelse(bh);
 406 fail:
 407         qd->qd_bh_count--;
 408         mutex_unlock(&sdp->sd_quota_mutex);
 409         return error;
 410 }
 411
 412 static void bh_put(struct gfs2_quota_data *qd)
 413 {
 414         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 415
 416         mutex_lock(&sdp->sd_quota_mutex);
 417         gfs2_assert(sdp, qd->qd_bh_count);
 418         if (!--qd->qd_bh_count) {
 419                 brelse(qd->qd_bh);
 420                 qd->qd_bh = NULL;
 421                 qd->qd_bh_qc = NULL;
 422         }
 423         mutex_unlock(&sdp->sd_quota_mutex);
 424 }
 425
 426 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
 427                          u64 *sync_gen)
 428 {
 429         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 430             !test_bit(QDF_CHANGE, &qd->qd_flags) ||
 431             (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
 432                 return 0;
 433
 434         if (!lockref_get_not_dead(&qd->qd_lockref))
 435                 return 0;
 436
 437         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 438         set_bit(QDF_LOCKED, &qd->qd_flags);
 439         qd->qd_change_sync = qd->qd_change;
 440         slot_hold(qd);
 441         return 1;
 442 }
 443
 444 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 445 {
 446         struct gfs2_quota_data *qd = NULL;
 447         int error;
 448         int found = 0;
 449
 450         *qdp = NULL;
 451
 452         if (sb_rdonly(sdp->sd_vfs))
 453                 return 0;
 454
 455         spin_lock(&qd_lock);
 456
 457         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 458                 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
 459                 if (found)
 460                         break;
 461         }
 462
 463         if (!found)
 464                 qd = NULL;
 465
 466         spin_unlock(&qd_lock);
 467
 468         if (qd) {
 469                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 470                 error = bh_get(qd);
 471                 if (error) {
 472                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 473                         slot_put(qd);
 474                         qd_put(qd);
 475                         return error;
 476                 }
 477         }
 478
 479         *qdp = qd;
 480
 481         return 0;
 482 }
 483
 484 static void qd_unlock(struct gfs2_quota_data *qd)
 485 {
 486         gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
 487                          test_bit(QDF_LOCKED, &qd->qd_flags));
 488         clear_bit(QDF_LOCKED, &qd->qd_flags);
 489         bh_put(qd);
 490         slot_put(qd);
 491         qd_put(qd);
 492 }
 493
 494 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
 495                     struct gfs2_quota_data **qdp)
 496 {
 497         int error;
 498
 499         error = qd_get(sdp, qid, qdp);
 500         if (error)
 501                 return error;
 502
 503         error = slot_get(*qdp);
 504         if (error)
 505                 goto fail;
 506
 507         error = bh_get(*qdp);
 508         if (error)
 509                 goto fail_slot;
 510
 511         return 0;
 512
 513 fail_slot:
 514         slot_put(*qdp);
 515 fail:
 516         qd_put(*qdp);
 517         return error;
 518 }
 519
 520 static void qdsb_put(struct gfs2_quota_data *qd)
 521 {
 522         bh_put(qd);
 523         slot_put(qd);
 524         qd_put(qd);
 525 }
 526
 527 /**
 528  * gfs2_qa_alloc - make sure we have a quota allocations data structure,
 529  *                 if necessary
 530  * @ip: the inode for this reservation
 531  */
 532 int gfs2_qa_alloc(struct gfs2_inode *ip)
 533 {
 534         int error = 0;
 535         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 536
 537         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 538                 return 0;
 539
 540         down_write(&ip->i_rw_mutex);
 541         if (ip->i_qadata == NULL) {
 542                 ip->i_qadata = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
 543                 if (!ip->i_qadata)
 544                         error = -ENOMEM;
 545         }
 546         up_write(&ip->i_rw_mutex);
 547         return error;
 548 }
 549
 550 void gfs2_qa_delete(struct gfs2_inode *ip, atomic_t *wcount)
 551 {
 552         down_write(&ip->i_rw_mutex);
 553         if (ip->i_qadata && ((wcount == NULL) || (atomic_read(wcount) <= 1))) {
 554                 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
 555                 ip->i_qadata = NULL;
 556         }
 557         up_write(&ip->i_rw_mutex);
 558 }
 559
 560 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
 561 {
 562         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 563         struct gfs2_quota_data **qd;
 564         int error;
 565
 566         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 567                 return 0;
 568
 569         if (ip->i_qadata == NULL) {
 570                 error = gfs2_rsqa_alloc(ip);
 571                 if (error)
 572                         return error;
 573         }
 574
 575         qd = ip->i_qadata->qa_qd;
 576
 577         if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
 578             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 579                 return -EIO;
 580
 581         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
 582         if (error)
 583                 goto out;
 584         ip->i_qadata->qa_qd_num++;
 585         qd++;
 586
 587         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
 588         if (error)
 589                 goto out;
 590         ip->i_qadata->qa_qd_num++;
 591         qd++;
 592
 593         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
 594             !uid_eq(uid, ip->i_inode.i_uid)) {
 595                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
 596                 if (error)
 597                         goto out;
 598                 ip->i_qadata->qa_qd_num++;
 599                 qd++;
 600         }
 601
 602         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
 603             !gid_eq(gid, ip->i_inode.i_gid)) {
 604                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
 605                 if (error)
 606                         goto out;
 607                 ip->i_qadata->qa_qd_num++;
 608                 qd++;
 609         }
 610
 611 out:
 612         if (error)
 613                 gfs2_quota_unhold(ip);
 614         return error;
 615 }
 616
 617 void gfs2_quota_unhold(struct gfs2_inode *ip)
 618 {
 619         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 620         u32 x;
 621
 622         if (ip->i_qadata == NULL)
 623                 return;
 624         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
 625
 626         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
 627                 qdsb_put(ip->i_qadata->qa_qd[x]);
 628                 ip->i_qadata->qa_qd[x] = NULL;
 629         }
 630         ip->i_qadata->qa_qd_num = 0;
 631 }
 632
 633 static int sort_qd(const void *a, const void *b)
 634 {
 635         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
 636         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
 637
 638         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
 639                 return -1;
 640         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
 641                 return 1;
 642         return 0;
 643 }
 644
 645 static void do_qc(struct gfs2_quota_data *qd, s64 change)
 646 {
 647         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 648         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 649         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 650         s64 x;
 651
 652         mutex_lock(&sdp->sd_quota_mutex);
 653         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
 654
 655         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
 656                 qc->qc_change = 0;
 657                 qc->qc_flags = 0;
 658                 if (qd->qd_id.type == USRQUOTA)
 659                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
 660                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
 661         }
 662
 663         x = be64_to_cpu(qc->qc_change) + change;
 664         qc->qc_change = cpu_to_be64(x);
 665
 666         spin_lock(&qd_lock);
 667         qd->qd_change = x;
 668         spin_unlock(&qd_lock);
 669
 670         if (!x) {
 671                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
 672                 clear_bit(QDF_CHANGE, &qd->qd_flags);
 673                 qc->qc_flags = 0;
 674                 qc->qc_id = 0;
 675                 slot_put(qd);
 676                 qd_put(qd);
 677         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
 678                 qd_hold(qd);
 679                 slot_hold(qd);
 680         }
 681
 682         if (change < 0) /* Reset quiet flag if we freed some blocks */
 683                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
 684         mutex_unlock(&sdp->sd_quota_mutex);
 685 }
 686
 687 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
 688                                   unsigned off, void *buf, unsigned bytes)
 689 {
 690         struct inode *inode = &ip->i_inode;
 691         struct gfs2_sbd *sdp = GFS2_SB(inode);
 692         struct address_space *mapping = inode->i_mapping;
 693         struct page *page;
 694         struct buffer_head *bh;
 695         void *kaddr;
 696         u64 blk;
 697         unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
 698         unsigned to_write = bytes, pg_off = off;
 699         int done = 0;
 700
 701         blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
 702         boff = off % bsize;
 703
 704         page = find_or_create_page(mapping, index, GFP_NOFS);
 705         if (!page)
 706                 return -ENOMEM;
 707         if (!page_has_buffers(page))
 708                 create_empty_buffers(page, bsize, 0);
 709
 710         bh = page_buffers(page);
 711         while (!done) {
 712                 /* Find the beginning block within the page */
 713                 if (pg_off >= ((bnum * bsize) + bsize)) {
 714                         bh = bh->b_this_page;
 715                         bnum++;
 716                         blk++;
 717                         continue;
 718                 }
 719                 if (!buffer_mapped(bh)) {
 720                         gfs2_block_map(inode, blk, bh, 1);
 721                         if (!buffer_mapped(bh))
 722                                 goto unlock_out;
 723                         /* If it's a newly allocated disk block, zero it */
 724                         if (buffer_new(bh))
 725                                 zero_user(page, bnum * bsize, bh->b_size);
 726                 }
 727                 if (PageUptodate(page))
 728                         set_buffer_uptodate(bh);
 729                 if (!buffer_uptodate(bh)) {
 730                         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
 731                         wait_on_buffer(bh);
 732                         if (!buffer_uptodate(bh))
 733                                 goto unlock_out;
 734                 }
 735                 if (gfs2_is_jdata(ip))
 736                         gfs2_trans_add_data(ip->i_gl, bh);
 737                 else
 738                         gfs2_ordered_add_inode(ip);
 739
 740                 /* If we need to write to the next block as well */
 741                 if (to_write > (bsize - boff)) {
 742                         pg_off += (bsize - boff);
 743                         to_write -= (bsize - boff);
 744                         boff = pg_off % bsize;
 745                         continue;
 746                 }
 747                 done = 1;
 748         }
 749
 750         /* Write to the page, now that we have setup the buffer(s) */
 751         kaddr = kmap_atomic(page);
 752         memcpy(kaddr + off, buf, bytes);
 753         flush_dcache_page(page);
 754         kunmap_atomic(kaddr);
 755         unlock_page(page);
 756         put_page(page);
 757
 758         return 0;
 759
 760 unlock_out:
 761         unlock_page(page);
 762         put_page(page);
 763         return -EIO;
 764 }
 765
 766 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
 767                                  loff_t loc)
 768 {
 769         unsigned long pg_beg;
 770         unsigned pg_off, nbytes, overflow = 0;
 771         int pg_oflow = 0, error;
 772         void *ptr;
 773
 774         nbytes = sizeof(struct gfs2_quota);
 775
 776         pg_beg = loc >> PAGE_SHIFT;
 777         pg_off = offset_in_page(loc);
 778
 779         /* If the quota straddles a page boundary, split the write in two */
 780         if ((pg_off + nbytes) > PAGE_SIZE) {
 781                 pg_oflow = 1;
 782                 overflow = (pg_off + nbytes) - PAGE_SIZE;
 783         }
 784
 785         ptr = qp;
 786         error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
 787                                        nbytes - overflow);
 788         /* If there's an overflow, write the remaining bytes to the next page */
 789         if (!error && pg_oflow)
 790                 error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
 791                                                ptr + nbytes - overflow,
 792                                                overflow);
 793         return error;
 794 }
 795
 796 /**
 797  * gfs2_adjust_quota - adjust record of current block usage
 798  * @ip: The quota inode
 799  * @loc: Offset of the entry in the quota file
 800  * @change: The amount of usage change to record
 801  * @qd: The quota data
 802  * @fdq: The updated limits to record
 803  *
 804  * This function was mostly borrowed from gfs2_block_truncate_page which was
 805  * in turn mostly borrowed from ext3
 806  *
 807  * Returns: 0 or -ve on error
 808  */
 809
 810 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
 811                              s64 change, struct gfs2_quota_data *qd,
 812                              struct qc_dqblk *fdq)
 813 {
 814         struct inode *inode = &ip->i_inode;
 815         struct gfs2_sbd *sdp = GFS2_SB(inode);
 816         struct gfs2_quota q;
 817         int err;
 818         u64 size;
 819
 820         if (gfs2_is_stuffed(ip)) {
 821                 err = gfs2_unstuff_dinode(ip, NULL);
 822                 if (err)
 823                         return err;
 824         }
 825
 826         memset(&q, 0, sizeof(struct gfs2_quota));
 827         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
 828         if (err < 0)
 829                 return err;
 830
 831         loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
 832         err = -EIO;
 833         be64_add_cpu(&q.qu_value, change);
 834         if (((s64)be64_to_cpu(q.qu_value)) < 0)
 835                 q.qu_value = 0; /* Never go negative on quota usage */
 836         qd->qd_qb.qb_value = q.qu_value;
 837         if (fdq) {
 838                 if (fdq->d_fieldmask & QC_SPC_SOFT) {
 839                         q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
 840                         qd->qd_qb.qb_warn = q.qu_warn;
 841                 }
 842                 if (fdq->d_fieldmask & QC_SPC_HARD) {
 843                         q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
 844                         qd->qd_qb.qb_limit = q.qu_limit;
 845                 }
 846                 if (fdq->d_fieldmask & QC_SPACE) {
 847                         q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
 848                         qd->qd_qb.qb_value = q.qu_value;
 849                 }
 850         }
 851
 852         err = gfs2_write_disk_quota(ip, &q, loc);
 853         if (!err) {
 854                 size = loc + sizeof(struct gfs2_quota);
 855                 if (size > inode->i_size)
 856                         i_size_write(inode, size);
 857                 inode->i_mtime = inode->i_atime = current_time(inode);
 858                 mark_inode_dirty(inode);
 859                 set_bit(QDF_REFRESH, &qd->qd_flags);
 860         }
 861
 862         return err;
 863 }
 864
 865 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 866 {
 867         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
 868         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 869         struct gfs2_alloc_parms ap = { .aflags = 0, };
 870         unsigned int data_blocks, ind_blocks;
 871         struct gfs2_holder *ghs, i_gh;
 872         unsigned int qx, x;
 873         struct gfs2_quota_data *qd;
 874         unsigned reserved;
 875         loff_t offset;
 876         unsigned int nalloc = 0, blocks;
 877         int error;
 878
 879         error = gfs2_rsqa_alloc(ip);
 880         if (error)
 881                 return error;
 882
 883         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
 884                               &data_blocks, &ind_blocks);
 885
 886         ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
 887         if (!ghs)
 888                 return -ENOMEM;
 889
 890         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 891         inode_lock(&ip->i_inode);
 892         for (qx = 0; qx < num_qd; qx++) {
 893                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
 894                                            GL_NOCACHE, &ghs[qx]);
 895                 if (error)
 896                         goto out;
 897         }
 898
 899         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
 900         if (error)
 901                 goto out;
 902
 903         for (x = 0; x < num_qd; x++) {
 904                 offset = qd2offset(qda[x]);
 905                 if (gfs2_write_alloc_required(ip, offset,
 906                                               sizeof(struct gfs2_quota)))
 907                         nalloc++;
 908         }
 909
 910         /*
 911          * 1 blk for unstuffing inode if stuffed. We add this extra
 912          * block to the reservation unconditionally. If the inode
 913          * doesn't need unstuffing, the block will be released to the
 914          * rgrp since it won't be allocated during the transaction
 915          */
 916         /* +3 in the end for unstuffing block, inode size update block
 917          * and another block in case quota straddles page boundary and
 918          * two blocks need to be updated instead of 1 */
 919         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
 920
 921         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
 922         ap.target = reserved;
 923         error = gfs2_inplace_reserve(ip, &ap);
 924         if (error)
 925                 goto out_alloc;
 926
 927         if (nalloc)
 928                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
 929
 930         error = gfs2_trans_begin(sdp, blocks, 0);
 931         if (error)
 932                 goto out_ipres;
 933
 934         for (x = 0; x < num_qd; x++) {
 935                 qd = qda[x];
 936                 offset = qd2offset(qd);
 937                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
 938                 if (error)
 939                         goto out_end_trans;
 940
 941                 do_qc(qd, -qd->qd_change_sync);
 942                 set_bit(QDF_REFRESH, &qd->qd_flags);
 943         }
 944
 945         error = 0;
 946
 947 out_end_trans:
 948         gfs2_trans_end(sdp);
 949 out_ipres:
 950         gfs2_inplace_release(ip);
 951 out_alloc:
 952         gfs2_glock_dq_uninit(&i_gh);
 953 out:
 954         while (qx--)
 955                 gfs2_glock_dq_uninit(&ghs[qx]);
 956         inode_unlock(&ip->i_inode);
 957         kfree(ghs);
 958         gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
 959                        GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
 960         return error;
 961 }
 962
 963 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
 964 {
 965         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 966         struct gfs2_quota q;
 967         struct gfs2_quota_lvb *qlvb;
 968         loff_t pos;
 969         int error;
 970
 971         memset(&q, 0, sizeof(struct gfs2_quota));
 972         pos = qd2offset(qd);
 973         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
 974         if (error < 0)
 975                 return error;
 976
 977         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
 978         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
 979         qlvb->__pad = 0;
 980         qlvb->qb_limit = q.qu_limit;
 981         qlvb->qb_warn = q.qu_warn;
 982         qlvb->qb_value = q.qu_value;
 983         qd->qd_qb = *qlvb;
 984
 985         return 0;
 986 }
 987
 988 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
 989                     struct gfs2_holder *q_gh)
 990 {
 991         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
 992         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 993         struct gfs2_holder i_gh;
 994         int error;
 995
 996 restart:
 997         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
 998         if (error)
 999                 return error;
1000
1001         if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1002                 force_refresh = FORCE;
1003
1004         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1005
1006         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1007                 gfs2_glock_dq_uninit(q_gh);
1008                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1009                                            GL_NOCACHE, q_gh);
1010                 if (error)
1011                         return error;
1012
1013                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1014                 if (error)
1015                         goto fail;
1016
1017                 error = update_qd(sdp, qd);
1018                 if (error)
1019                         goto fail_gunlock;
1020
1021                 gfs2_glock_dq_uninit(&i_gh);
1022                 gfs2_glock_dq_uninit(q_gh);
1023                 force_refresh = 0;
1024                 goto restart;
1025         }
1026
1027         return 0;
1028
1029 fail_gunlock:
1030         gfs2_glock_dq_uninit(&i_gh);
1031 fail:
1032         gfs2_glock_dq_uninit(q_gh);
1033         return error;
1034 }
1035
1036 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1037 {
1038         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1039         struct gfs2_quota_data *qd;
1040         u32 x;
1041         int error = 0;
1042
1043         if (capable(CAP_SYS_RESOURCE) ||
1044             sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1045                 return 0;
1046
1047         error = gfs2_quota_hold(ip, uid, gid);
1048         if (error)
1049                 return error;
1050
1051         sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1052              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1053
1054         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1055                 qd = ip->i_qadata->qa_qd[x];
1056                 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1057                 if (error)
1058                         break;
1059         }
1060
1061         if (!error)
1062                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1063         else {
1064                 while (x--)
1065                         gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1066                 gfs2_quota_unhold(ip);
1067         }
1068
1069         return error;
1070 }
1071
1072 static int need_sync(struct gfs2_quota_data *qd)
1073 {
1074         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1075         struct gfs2_tune *gt = &sdp->sd_tune;
1076         s64 value;
1077         unsigned int num, den;
1078         int do_sync = 1;
1079
1080         if (!qd->qd_qb.qb_limit)
1081                 return 0;
1082
1083         spin_lock(&qd_lock);
1084         value = qd->qd_change;
1085         spin_unlock(&qd_lock);
1086
1087         spin_lock(&gt->gt_spin);
1088         num = gt->gt_quota_scale_num;
1089         den = gt->gt_quota_scale_den;
1090         spin_unlock(&gt->gt_spin);
1091
1092         if (value < 0)
1093                 do_sync = 0;
1094         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1095                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1096                 do_sync = 0;
1097         else {
1098                 value *= gfs2_jindex_size(sdp) * num;
1099                 value = div_s64(value, den);
1100                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1101                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1102                         do_sync = 0;
1103         }
1104
1105         return do_sync;
1106 }
1107
1108 void gfs2_quota_unlock(struct gfs2_inode *ip)
1109 {
1110         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1111         struct gfs2_quota_data *qda[4];
1112         unsigned int count = 0;
1113         u32 x;
1114         int found;
1115
1116         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1117                 goto out;
1118
1119         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1120                 struct gfs2_quota_data *qd;
1121                 int sync;
1122
1123                 qd = ip->i_qadata->qa_qd[x];
1124                 sync = need_sync(qd);
1125
1126                 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1127                 if (!sync)
1128                         continue;
1129
1130                 spin_lock(&qd_lock);
1131                 found = qd_check_sync(sdp, qd, NULL);
1132                 spin_unlock(&qd_lock);
1133
1134                 if (!found)
1135                         continue;
1136
1137                 gfs2_assert_warn(sdp, qd->qd_change_sync);
1138                 if (bh_get(qd)) {
1139                         clear_bit(QDF_LOCKED, &qd->qd_flags);
1140                         slot_put(qd);
1141                         qd_put(qd);
1142                         continue;
1143                 }
1144
1145                 qda[count++] = qd;
1146         }
1147
1148         if (count) {
1149                 do_sync(count, qda);
1150                 for (x = 0; x < count; x++)
1151                         qd_unlock(qda[x]);
1152         }
1153
1154 out:
1155         gfs2_quota_unhold(ip);
1156 }
1157
1158 #define MAX_LINE 256
1159
1160 static int print_message(struct gfs2_quota_data *qd, char *type)
1161 {
1162         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1163
1164         fs_info(sdp, "quota %s for %s %u\n",
1165                 type,
1166                 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1167                 from_kqid(&init_user_ns, qd->qd_id));
1168
1169         return 0;
1170 }
1171
1172 /**
1173  * gfs2_quota_check - check if allocating new blocks will exceed quota
1174  * @ip:  The inode for which this check is being performed
1175  * @uid: The uid to check against
1176  * @gid: The gid to check against
1177  * @ap:  The allocation parameters. ap->target contains the requested
1178  *       blocks. ap->min_target, if set, contains the minimum blks
1179  *       requested.
1180  *
1181  * Returns: 0 on success.
1182  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1183  *                  quota must allow at least min_req blks for success and
1184  *                  ap->allowed is set to the number of blocks allowed
1185  *
1186  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1187  *                  of blocks available.
1188  */
1189 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1190                      struct gfs2_alloc_parms *ap)
1191 {
1192         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1193         struct gfs2_quota_data *qd;
1194         s64 value, warn, limit;
1195         u32 x;
1196         int error = 0;
1197
1198         ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1199         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1200                 return 0;
1201
1202         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1203                 return 0;
1204
1205         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1206                 qd = ip->i_qadata->qa_qd[x];
1207
1208                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1209                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1210                         continue;
1211
1212                 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1213                 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1214                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1215                 spin_lock(&qd_lock);
1216                 value += qd->qd_change;
1217                 spin_unlock(&qd_lock);
1218
1219                 if (limit > 0 && (limit - value) < ap->allowed)
1220                         ap->allowed = limit - value;
1221                 /* If we can't meet the target */
1222                 if (limit && limit < (value + (s64)ap->target)) {
1223                         /* If no min_target specified or we don't meet
1224                          * min_target, return -EDQUOT */
1225                         if (!ap->min_target || ap->min_target > ap->allowed) {
1226                                 if (!test_and_set_bit(QDF_QMSG_QUIET,
1227                                                       &qd->qd_flags)) {
1228                                         print_message(qd, "exceeded");
1229                                         quota_send_warning(qd->qd_id,
1230                                                            sdp->sd_vfs->s_dev,
1231                                                            QUOTA_NL_BHARDWARN);
1232                                 }
1233                                 error = -EDQUOT;
1234                                 break;
1235                         }
1236                 } else if (warn && warn < value &&
1237                            time_after_eq(jiffies, qd->qd_last_warn +
1238                                          gfs2_tune_get(sdp, gt_quota_warn_period)
1239                                          * HZ)) {
1240                         quota_send_warning(qd->qd_id,
1241                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1242                         error = print_message(qd, "warning");
1243                         qd->qd_last_warn = jiffies;
1244                 }
1245         }
1246         return error;
1247 }
1248
1249 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1250                        kuid_t uid, kgid_t gid)
1251 {
1252         struct gfs2_quota_data *qd;
1253         u32 x;
1254         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1255
1256         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1257             gfs2_assert_warn(sdp, change))
1258                 return;
1259         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1260                 return;
1261
1262         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1263                 qd = ip->i_qadata->qa_qd[x];
1264
1265                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1266                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1267                         do_qc(qd, change);
1268                 }
1269         }
1270 }
1271
1272 int gfs2_quota_sync(struct super_block *sb, int type)
1273 {
1274         struct gfs2_sbd *sdp = sb->s_fs_info;
1275         struct gfs2_quota_data **qda;
1276         unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1277         unsigned int num_qd;
1278         unsigned int x;
1279         int error = 0;
1280
1281         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1282         if (!qda)
1283                 return -ENOMEM;
1284
1285         mutex_lock(&sdp->sd_quota_sync_mutex);
1286         sdp->sd_quota_sync_gen++;
1287
1288         do {
1289                 num_qd = 0;
1290
1291                 for (;;) {
1292                         error = qd_fish(sdp, qda + num_qd);
1293                         if (error || !qda[num_qd])
1294                                 break;
1295                         if (++num_qd == max_qd)
1296                                 break;
1297                 }
1298
1299                 if (num_qd) {
1300                         if (!error)
1301                                 error = do_sync(num_qd, qda);
1302                         if (!error)
1303                                 for (x = 0; x < num_qd; x++)
1304                                         qda[x]->qd_sync_gen =
1305                                                 sdp->sd_quota_sync_gen;
1306
1307                         for (x = 0; x < num_qd; x++)
1308                                 qd_unlock(qda[x]);
1309                 }
1310         } while (!error && num_qd == max_qd);
1311
1312         mutex_unlock(&sdp->sd_quota_sync_mutex);
1313         kfree(qda);
1314
1315         return error;
1316 }
1317
1318 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1319 {
1320         struct gfs2_quota_data *qd;
1321         struct gfs2_holder q_gh;
1322         int error;
1323
1324         error = qd_get(sdp, qid, &qd);
1325         if (error)
1326                 return error;
1327
1328         error = do_glock(qd, FORCE, &q_gh);
1329         if (!error)
1330                 gfs2_glock_dq_uninit(&q_gh);
1331
1332         qd_put(qd);
1333         return error;
1334 }
1335
1336 int gfs2_quota_init(struct gfs2_sbd *sdp)
1337 {
1338         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1339         u64 size = i_size_read(sdp->sd_qc_inode);
1340         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1341         unsigned int x, slot = 0;
1342         unsigned int found = 0;
1343         unsigned int hash;
1344         unsigned int bm_size;
1345         u64 dblock;
1346         u32 extlen = 0;
1347         int error;
1348
1349         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1350                 return -EIO;
1351
1352         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1353         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1354         bm_size *= sizeof(unsigned long);
1355         error = -ENOMEM;
1356         sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1357         if (sdp->sd_quota_bitmap == NULL)
1358                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1359                                                  __GFP_ZERO, PAGE_KERNEL);
1360         if (!sdp->sd_quota_bitmap)
1361                 return error;
1362
1363         for (x = 0; x < blocks; x++) {
1364                 struct buffer_head *bh;
1365                 const struct gfs2_quota_change *qc;
1366                 unsigned int y;
1367
1368                 if (!extlen) {
1369                         int new = 0;
1370                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1371                         if (error)
1372                                 goto fail;
1373                 }
1374                 error = -EIO;
1375                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1376                 if (!bh)
1377                         goto fail;
1378                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1379                         brelse(bh);
1380                         goto fail;
1381                 }
1382
1383                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1384                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1385                      y++, slot++) {
1386                         struct gfs2_quota_data *qd;
1387                         s64 qc_change = be64_to_cpu(qc->qc_change);
1388                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1389                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1390                                                 USRQUOTA : GRPQUOTA;
1391                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1392                                                       be32_to_cpu(qc->qc_id));
1393                         qc++;
1394                         if (!qc_change)
1395                                 continue;
1396
1397                         hash = gfs2_qd_hash(sdp, qc_id);
1398                         qd = qd_alloc(hash, sdp, qc_id);
1399                         if (qd == NULL) {
1400                                 brelse(bh);
1401                                 goto fail;
1402                         }
1403
1404                         set_bit(QDF_CHANGE, &qd->qd_flags);
1405                         qd->qd_change = qc_change;
1406                         qd->qd_slot = slot;
1407                         qd->qd_slot_count = 1;
1408
1409                         spin_lock(&qd_lock);
1410                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1411                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1412                         atomic_inc(&sdp->sd_quota_count);
1413                         spin_unlock(&qd_lock);
1414
1415                         spin_lock_bucket(hash);
1416                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1417                         spin_unlock_bucket(hash);
1418
1419                         found++;
1420                 }
1421
1422                 brelse(bh);
1423                 dblock++;
1424                 extlen--;
1425         }
1426
1427         if (found)
1428                 fs_info(sdp, "found %u quota changes\n", found);
1429
1430         return 0;
1431
1432 fail:
1433         gfs2_quota_cleanup(sdp);
1434         return error;
1435 }
1436
1437 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1438 {
1439         struct list_head *head = &sdp->sd_quota_list;
1440         struct gfs2_quota_data *qd;
1441
1442         spin_lock(&qd_lock);
1443         while (!list_empty(head)) {
1444                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1445
1446                 list_del(&qd->qd_list);
1447
1448                 /* Also remove if this qd exists in the reclaim list */
1449                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1450                 atomic_dec(&sdp->sd_quota_count);
1451                 spin_unlock(&qd_lock);
1452
1453                 spin_lock_bucket(qd->qd_hash);
1454                 hlist_bl_del_rcu(&qd->qd_hlist);
1455                 spin_unlock_bucket(qd->qd_hash);
1456
1457                 gfs2_assert_warn(sdp, !qd->qd_change);
1458                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1459                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1460
1461                 gfs2_glock_put(qd->qd_gl);
1462                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1463
1464                 spin_lock(&qd_lock);
1465         }
1466         spin_unlock(&qd_lock);
1467
1468         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1469
1470         kvfree(sdp->sd_quota_bitmap);
1471         sdp->sd_quota_bitmap = NULL;
1472 }
1473
1474 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1475 {
1476         if (error == 0 || error == -EROFS)
1477                 return;
1478         if (!gfs2_withdrawn(sdp)) {
1479                 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1480                 sdp->sd_log_error = error;
1481                 wake_up(&sdp->sd_logd_waitq);
1482         }
1483 }
1484
1485 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1486                                int (*fxn)(struct super_block *sb, int type),
1487                                unsigned long t, unsigned long *timeo,
1488                                unsigned int *new_timeo)
1489 {
1490         if (t >= *timeo) {
1491                 int error = fxn(sdp->sd_vfs, 0);
1492                 quotad_error(sdp, msg, error);
1493                 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1494         } else {
1495                 *timeo -= t;
1496         }
1497 }
1498
1499 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1500 {
1501         struct gfs2_inode *ip;
1502
1503         while(1) {
1504                 ip = NULL;
1505                 spin_lock(&sdp->sd_trunc_lock);
1506                 if (!list_empty(&sdp->sd_trunc_list)) {
1507                         ip = list_entry(sdp->sd_trunc_list.next,
1508                                         struct gfs2_inode, i_trunc_list);
1509                         list_del_init(&ip->i_trunc_list);
1510                 }
1511                 spin_unlock(&sdp->sd_trunc_lock);
1512                 if (ip == NULL)
1513                         return;
1514                 gfs2_glock_finish_truncate(ip);
1515         }
1516 }
1517
1518 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1519         if (!sdp->sd_statfs_force_sync) {
1520                 sdp->sd_statfs_force_sync = 1;
1521                 wake_up(&sdp->sd_quota_wait);
1522         }
1523 }
1524
1525
1526 /**
1527  * gfs2_quotad - Write cached quota changes into the quota file
1528  * @sdp: Pointer to GFS2 superblock
1529  *
1530  */
1531
1532 int gfs2_quotad(void *data)
1533 {
1534         struct gfs2_sbd *sdp = data;
1535         struct gfs2_tune *tune = &sdp->sd_tune;
1536         unsigned long statfs_timeo = 0;
1537         unsigned long quotad_timeo = 0;
1538         unsigned long t = 0;
1539         DEFINE_WAIT(wait);
1540         int empty;
1541
1542         while (!kthread_should_stop()) {
1543
1544                 /* Update the master statfs file */
1545                 if (sdp->sd_statfs_force_sync) {
1546                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1547                         quotad_error(sdp, "statfs", error);
1548                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1549                 }
1550                 else
1551                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1552                                            &statfs_timeo,
1553                                            &tune->gt_statfs_quantum);
1554
1555                 /* Update quota file */
1556                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1557                                    &quotad_timeo, &tune->gt_quota_quantum);
1558
1559                 /* Check for & recover partially truncated inodes */
1560                 quotad_check_trunc_list(sdp);
1561
1562                 try_to_freeze();
1563
1564                 t = min(quotad_timeo, statfs_timeo);
1565
1566                 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1567                 spin_lock(&sdp->sd_trunc_lock);
1568                 empty = list_empty(&sdp->sd_trunc_list);
1569                 spin_unlock(&sdp->sd_trunc_lock);
1570                 if (empty && !sdp->sd_statfs_force_sync)
1571                         t -= schedule_timeout(t);
1572                 else
1573                         t = 0;
1574                 finish_wait(&sdp->sd_quota_wait, &wait);
1575         }
1576
1577         return 0;
1578 }
1579
1580 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1581 {
1582         struct gfs2_sbd *sdp = sb->s_fs_info;
1583
1584         memset(state, 0, sizeof(*state));
1585
1586         switch (sdp->sd_args.ar_quota) {
1587         case GFS2_QUOTA_ON:
1588                 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1589                 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1590                 /*FALLTHRU*/
1591         case GFS2_QUOTA_ACCOUNT:
1592                 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1593                                                   QCI_SYSFILE;
1594                 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1595                                                   QCI_SYSFILE;
1596                 break;
1597         case GFS2_QUOTA_OFF:
1598                 break;
1599         }
1600         if (sdp->sd_quota_inode) {
1601                 state->s_state[USRQUOTA].ino =
1602                                         GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1603                 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1604         }
1605         state->s_state[USRQUOTA].nextents = 1;  /* unsupported */
1606         state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1607         state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1608         return 0;
1609 }
1610
1611 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1612                           struct qc_dqblk *fdq)
1613 {
1614         struct gfs2_sbd *sdp = sb->s_fs_info;
1615         struct gfs2_quota_lvb *qlvb;
1616         struct gfs2_quota_data *qd;
1617         struct gfs2_holder q_gh;
1618         int error;
1619
1620         memset(fdq, 0, sizeof(*fdq));
1621
1622         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1623                 return -ESRCH; /* Crazy XFS error code */
1624
1625         if ((qid.type != USRQUOTA) &&
1626             (qid.type != GRPQUOTA))
1627                 return -EINVAL;
1628
1629         error = qd_get(sdp, qid, &qd);
1630         if (error)
1631                 return error;
1632         error = do_glock(qd, FORCE, &q_gh);
1633         if (error)
1634                 goto out;
1635
1636         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1637         fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1638         fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1639         fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1640
1641         gfs2_glock_dq_uninit(&q_gh);
1642 out:
1643         qd_put(qd);
1644         return error;
1645 }
1646
1647 /* GFS2 only supports a subset of the XFS fields */
1648 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1649
1650 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1651                           struct qc_dqblk *fdq)
1652 {
1653         struct gfs2_sbd *sdp = sb->s_fs_info;
1654         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1655         struct gfs2_quota_data *qd;
1656         struct gfs2_holder q_gh, i_gh;
1657         unsigned int data_blocks, ind_blocks;
1658         unsigned int blocks = 0;
1659         int alloc_required;
1660         loff_t offset;
1661         int error;
1662
1663         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1664                 return -ESRCH; /* Crazy XFS error code */
1665
1666         if ((qid.type != USRQUOTA) &&
1667             (qid.type != GRPQUOTA))
1668                 return -EINVAL;
1669
1670         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1671                 return -EINVAL;
1672
1673         error = qd_get(sdp, qid, &qd);
1674         if (error)
1675                 return error;
1676
1677         error = gfs2_rsqa_alloc(ip);
1678         if (error)
1679                 goto out_put;
1680
1681         inode_lock(&ip->i_inode);
1682         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1683         if (error)
1684                 goto out_unlockput;
1685         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1686         if (error)
1687                 goto out_q;
1688
1689         /* Check for existing entry, if none then alloc new blocks */
1690         error = update_qd(sdp, qd);
1691         if (error)
1692                 goto out_i;
1693
1694         /* If nothing has changed, this is a no-op */
1695         if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1696             ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1697                 fdq->d_fieldmask ^= QC_SPC_SOFT;
1698
1699         if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1700             ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1701                 fdq->d_fieldmask ^= QC_SPC_HARD;
1702
1703         if ((fdq->d_fieldmask & QC_SPACE) &&
1704             ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1705                 fdq->d_fieldmask ^= QC_SPACE;
1706
1707         if (fdq->d_fieldmask == 0)
1708                 goto out_i;
1709
1710         offset = qd2offset(qd);
1711         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1712         if (gfs2_is_stuffed(ip))
1713                 alloc_required = 1;
1714         if (alloc_required) {
1715                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1716                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1717                                        &data_blocks, &ind_blocks);
1718                 blocks = 1 + data_blocks + ind_blocks;
1719                 ap.target = blocks;
1720                 error = gfs2_inplace_reserve(ip, &ap);
1721                 if (error)
1722                         goto out_i;
1723                 blocks += gfs2_rg_blocks(ip, blocks);
1724         }
1725
1726         /* Some quotas span block boundaries and can update two blocks,
1727            adding an extra block to the transaction to handle such quotas */
1728         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1729         if (error)
1730                 goto out_release;
1731
1732         /* Apply changes */
1733         error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1734         if (!error)
1735                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1736
1737         gfs2_trans_end(sdp);
1738 out_release:
1739         if (alloc_required)
1740                 gfs2_inplace_release(ip);
1741 out_i:
1742         gfs2_glock_dq_uninit(&i_gh);
1743 out_q:
1744         gfs2_glock_dq_uninit(&q_gh);
1745 out_unlockput:
1746         inode_unlock(&ip->i_inode);
1747 out_put:
1748         qd_put(qd);
1749         return error;
1750 }
1751
1752 const struct quotactl_ops gfs2_quotactl_ops = {
1753         .quota_sync     = gfs2_quota_sync,
1754         .get_state      = gfs2_quota_get_state,
1755         .get_dqblk      = gfs2_get_dqblk,
1756         .set_dqblk      = gfs2_set_dqblk,
1757 };
1758
1759 void __init gfs2_quota_hash_init(void)
1760 {
1761         unsigned i;
1762
1763         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1764                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1765 }