0cc585064eae5de97db9b40e8571e4cb7cf74ca9
[platform/kernel/linux-rpi.git] / 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 #define QC_CHANGE 0
79 #define QC_SYNC 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         struct gfs2_sbd *sdp = qd->qd_sbd;
113
114         kmem_cache_free(gfs2_quotad_cachep, qd);
115         if (atomic_dec_and_test(&sdp->sd_quota_count))
116                 wake_up(&sdp->sd_kill_wait);
117 }
118
119 static void gfs2_qd_dispose(struct gfs2_quota_data *qd)
120 {
121         struct gfs2_sbd *sdp = qd->qd_sbd;
122
123         spin_lock(&qd_lock);
124         list_del(&qd->qd_list);
125         spin_unlock(&qd_lock);
126
127         spin_lock_bucket(qd->qd_hash);
128         hlist_bl_del_rcu(&qd->qd_hlist);
129         spin_unlock_bucket(qd->qd_hash);
130
131         if (!gfs2_withdrawn(sdp)) {
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
137         gfs2_glock_put(qd->qd_gl);
138         call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
139 }
140
141 static void gfs2_qd_list_dispose(struct list_head *list)
142 {
143         struct gfs2_quota_data *qd;
144
145         while (!list_empty(list)) {
146                 qd = list_first_entry(list, struct gfs2_quota_data, qd_lru);
147                 list_del(&qd->qd_lru);
148
149                 gfs2_qd_dispose(qd);
150         }
151 }
152
153
154 static enum lru_status gfs2_qd_isolate(struct list_head *item,
155                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
156 {
157         struct list_head *dispose = arg;
158         struct gfs2_quota_data *qd =
159                 list_entry(item, struct gfs2_quota_data, qd_lru);
160         enum lru_status status;
161
162         if (!spin_trylock(&qd->qd_lockref.lock))
163                 return LRU_SKIP;
164
165         status = LRU_SKIP;
166         if (qd->qd_lockref.count == 0) {
167                 lockref_mark_dead(&qd->qd_lockref);
168                 list_lru_isolate_move(lru, &qd->qd_lru, dispose);
169                 status = LRU_REMOVED;
170         }
171
172         spin_unlock(&qd->qd_lockref.lock);
173         return status;
174 }
175
176 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
177                                          struct shrink_control *sc)
178 {
179         LIST_HEAD(dispose);
180         unsigned long freed;
181
182         if (!(sc->gfp_mask & __GFP_FS))
183                 return SHRINK_STOP;
184
185         freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
186                                      gfs2_qd_isolate, &dispose);
187
188         gfs2_qd_list_dispose(&dispose);
189
190         return freed;
191 }
192
193 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
194                                           struct shrink_control *sc)
195 {
196         return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
197 }
198
199 struct shrinker gfs2_qd_shrinker = {
200         .count_objects = gfs2_qd_shrink_count,
201         .scan_objects = gfs2_qd_shrink_scan,
202         .seeks = DEFAULT_SEEKS,
203         .flags = SHRINKER_NUMA_AWARE,
204 };
205
206
207 static u64 qd2index(struct gfs2_quota_data *qd)
208 {
209         struct kqid qid = qd->qd_id;
210         return (2 * (u64)from_kqid(&init_user_ns, qid)) +
211                 ((qid.type == USRQUOTA) ? 0 : 1);
212 }
213
214 static u64 qd2offset(struct gfs2_quota_data *qd)
215 {
216         u64 offset;
217
218         offset = qd2index(qd);
219         offset *= sizeof(struct gfs2_quota);
220
221         return offset;
222 }
223
224 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
225 {
226         struct gfs2_quota_data *qd;
227         int error;
228
229         qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
230         if (!qd)
231                 return NULL;
232
233         qd->qd_sbd = sdp;
234         qd->qd_lockref.count = 0;
235         spin_lock_init(&qd->qd_lockref.lock);
236         qd->qd_id = qid;
237         qd->qd_slot = -1;
238         INIT_LIST_HEAD(&qd->qd_lru);
239         qd->qd_hash = hash;
240
241         error = gfs2_glock_get(sdp, qd2index(qd),
242                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
243         if (error)
244                 goto fail;
245
246         return qd;
247
248 fail:
249         kmem_cache_free(gfs2_quotad_cachep, qd);
250         return NULL;
251 }
252
253 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
254                                                      const struct gfs2_sbd *sdp,
255                                                      struct kqid qid)
256 {
257         struct gfs2_quota_data *qd;
258         struct hlist_bl_node *h;
259
260         hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
261                 if (!qid_eq(qd->qd_id, qid))
262                         continue;
263                 if (qd->qd_sbd != sdp)
264                         continue;
265                 if (lockref_get_not_dead(&qd->qd_lockref)) {
266                         list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
267                         return qd;
268                 }
269         }
270
271         return NULL;
272 }
273
274
275 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
276                   struct gfs2_quota_data **qdp)
277 {
278         struct gfs2_quota_data *qd, *new_qd;
279         unsigned int hash = gfs2_qd_hash(sdp, qid);
280
281         rcu_read_lock();
282         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
283         rcu_read_unlock();
284
285         if (qd)
286                 return 0;
287
288         new_qd = qd_alloc(hash, sdp, qid);
289         if (!new_qd)
290                 return -ENOMEM;
291
292         spin_lock(&qd_lock);
293         spin_lock_bucket(hash);
294         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
295         if (qd == NULL) {
296                 new_qd->qd_lockref.count++;
297                 *qdp = new_qd;
298                 list_add(&new_qd->qd_list, &sdp->sd_quota_list);
299                 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
300                 atomic_inc(&sdp->sd_quota_count);
301         }
302         spin_unlock_bucket(hash);
303         spin_unlock(&qd_lock);
304
305         if (qd) {
306                 gfs2_glock_put(new_qd->qd_gl);
307                 kmem_cache_free(gfs2_quotad_cachep, new_qd);
308         }
309
310         return 0;
311 }
312
313
314 static void qd_hold(struct gfs2_quota_data *qd)
315 {
316         struct gfs2_sbd *sdp = qd->qd_sbd;
317         gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
318         lockref_get(&qd->qd_lockref);
319 }
320
321 static void qd_put(struct gfs2_quota_data *qd)
322 {
323         struct gfs2_sbd *sdp;
324
325         if (lockref_put_or_lock(&qd->qd_lockref))
326                 return;
327
328         BUG_ON(__lockref_is_dead(&qd->qd_lockref));
329         sdp = qd->qd_sbd;
330         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
331                 lockref_mark_dead(&qd->qd_lockref);
332                 spin_unlock(&qd->qd_lockref.lock);
333
334                 gfs2_qd_dispose(qd);
335                 return;
336         }
337
338         qd->qd_lockref.count = 0;
339         list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
340         spin_unlock(&qd->qd_lockref.lock);
341 }
342
343 static int slot_get(struct gfs2_quota_data *qd)
344 {
345         struct gfs2_sbd *sdp = qd->qd_sbd;
346         unsigned int bit;
347         int error = 0;
348
349         spin_lock(&sdp->sd_bitmap_lock);
350         if (qd->qd_slot_count != 0)
351                 goto out;
352
353         error = -ENOSPC;
354         bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
355         if (bit < sdp->sd_quota_slots) {
356                 set_bit(bit, sdp->sd_quota_bitmap);
357                 qd->qd_slot = bit;
358                 error = 0;
359 out:
360                 qd->qd_slot_count++;
361         }
362         spin_unlock(&sdp->sd_bitmap_lock);
363
364         return error;
365 }
366
367 static void slot_hold(struct gfs2_quota_data *qd)
368 {
369         struct gfs2_sbd *sdp = qd->qd_sbd;
370
371         spin_lock(&sdp->sd_bitmap_lock);
372         gfs2_assert(sdp, qd->qd_slot_count);
373         qd->qd_slot_count++;
374         spin_unlock(&sdp->sd_bitmap_lock);
375 }
376
377 static void slot_put(struct gfs2_quota_data *qd)
378 {
379         struct gfs2_sbd *sdp = qd->qd_sbd;
380
381         spin_lock(&sdp->sd_bitmap_lock);
382         gfs2_assert(sdp, qd->qd_slot_count);
383         if (!--qd->qd_slot_count) {
384                 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
385                 qd->qd_slot = -1;
386         }
387         spin_unlock(&sdp->sd_bitmap_lock);
388 }
389
390 static int bh_get(struct gfs2_quota_data *qd)
391 {
392         struct gfs2_sbd *sdp = qd->qd_sbd;
393         struct inode *inode = sdp->sd_qc_inode;
394         struct gfs2_inode *ip = GFS2_I(inode);
395         unsigned int block, offset;
396         struct buffer_head *bh;
397         struct iomap iomap = { };
398         int error;
399
400         mutex_lock(&sdp->sd_quota_mutex);
401
402         if (qd->qd_bh_count++) {
403                 mutex_unlock(&sdp->sd_quota_mutex);
404                 return 0;
405         }
406
407         block = qd->qd_slot / sdp->sd_qc_per_block;
408         offset = qd->qd_slot % sdp->sd_qc_per_block;
409
410         error = gfs2_iomap_get(inode,
411                                (loff_t)block << inode->i_blkbits,
412                                i_blocksize(inode), &iomap);
413         if (error)
414                 goto fail;
415         error = -ENOENT;
416         if (iomap.type != IOMAP_MAPPED)
417                 goto fail;
418
419         error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
420                                DIO_WAIT, 0, &bh);
421         if (error)
422                 goto fail;
423         error = -EIO;
424         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
425                 goto fail_brelse;
426
427         qd->qd_bh = bh;
428         qd->qd_bh_qc = (struct gfs2_quota_change *)
429                 (bh->b_data + sizeof(struct gfs2_meta_header) +
430                  offset * sizeof(struct gfs2_quota_change));
431
432         mutex_unlock(&sdp->sd_quota_mutex);
433
434         return 0;
435
436 fail_brelse:
437         brelse(bh);
438 fail:
439         qd->qd_bh_count--;
440         mutex_unlock(&sdp->sd_quota_mutex);
441         return error;
442 }
443
444 static void bh_put(struct gfs2_quota_data *qd)
445 {
446         struct gfs2_sbd *sdp = qd->qd_sbd;
447
448         mutex_lock(&sdp->sd_quota_mutex);
449         gfs2_assert(sdp, qd->qd_bh_count);
450         if (!--qd->qd_bh_count) {
451                 brelse(qd->qd_bh);
452                 qd->qd_bh = NULL;
453                 qd->qd_bh_qc = NULL;
454         }
455         mutex_unlock(&sdp->sd_quota_mutex);
456 }
457
458 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
459                          u64 *sync_gen)
460 {
461         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
462             !test_bit(QDF_CHANGE, &qd->qd_flags) ||
463             (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
464                 return 0;
465
466         if (!lockref_get_not_dead(&qd->qd_lockref))
467                 return 0;
468
469         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
470         set_bit(QDF_LOCKED, &qd->qd_flags);
471         qd->qd_change_sync = qd->qd_change;
472         slot_hold(qd);
473         return 1;
474 }
475
476 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
477 {
478         struct gfs2_quota_data *qd = NULL, *iter;
479         int error;
480
481         *qdp = NULL;
482
483         if (sb_rdonly(sdp->sd_vfs))
484                 return 0;
485
486         spin_lock(&qd_lock);
487
488         list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
489                 if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
490                         qd = iter;
491                         break;
492                 }
493         }
494
495         spin_unlock(&qd_lock);
496
497         if (qd) {
498                 error = bh_get(qd);
499                 if (error) {
500                         clear_bit(QDF_LOCKED, &qd->qd_flags);
501                         slot_put(qd);
502                         qd_put(qd);
503                         return error;
504                 }
505         }
506
507         *qdp = qd;
508
509         return 0;
510 }
511
512 static void qd_unlock(struct gfs2_quota_data *qd)
513 {
514         gfs2_assert_warn(qd->qd_sbd, test_bit(QDF_LOCKED, &qd->qd_flags));
515         clear_bit(QDF_LOCKED, &qd->qd_flags);
516         bh_put(qd);
517         slot_put(qd);
518         qd_put(qd);
519 }
520
521 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
522                     struct gfs2_quota_data **qdp)
523 {
524         int error;
525
526         error = qd_get(sdp, qid, qdp);
527         if (error)
528                 return error;
529
530         error = slot_get(*qdp);
531         if (error)
532                 goto fail;
533
534         error = bh_get(*qdp);
535         if (error)
536                 goto fail_slot;
537
538         return 0;
539
540 fail_slot:
541         slot_put(*qdp);
542 fail:
543         qd_put(*qdp);
544         return error;
545 }
546
547 static void qdsb_put(struct gfs2_quota_data *qd)
548 {
549         bh_put(qd);
550         slot_put(qd);
551         qd_put(qd);
552 }
553
554 /**
555  * gfs2_qa_get - make sure we have a quota allocations data structure,
556  *               if necessary
557  * @ip: the inode for this reservation
558  */
559 int gfs2_qa_get(struct gfs2_inode *ip)
560 {
561         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
562         struct inode *inode = &ip->i_inode;
563
564         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
565                 return 0;
566
567         spin_lock(&inode->i_lock);
568         if (ip->i_qadata == NULL) {
569                 struct gfs2_qadata *tmp;
570
571                 spin_unlock(&inode->i_lock);
572                 tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
573                 if (!tmp)
574                         return -ENOMEM;
575
576                 spin_lock(&inode->i_lock);
577                 if (ip->i_qadata == NULL)
578                         ip->i_qadata = tmp;
579                 else
580                         kmem_cache_free(gfs2_qadata_cachep, tmp);
581         }
582         ip->i_qadata->qa_ref++;
583         spin_unlock(&inode->i_lock);
584         return 0;
585 }
586
587 void gfs2_qa_put(struct gfs2_inode *ip)
588 {
589         struct inode *inode = &ip->i_inode;
590
591         spin_lock(&inode->i_lock);
592         if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
593                 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
594                 ip->i_qadata = NULL;
595         }
596         spin_unlock(&inode->i_lock);
597 }
598
599 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
600 {
601         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
602         struct gfs2_quota_data **qd;
603         int error;
604
605         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
606                 return 0;
607
608         error = gfs2_qa_get(ip);
609         if (error)
610                 return error;
611
612         qd = ip->i_qadata->qa_qd;
613
614         if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
615             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
616                 error = -EIO;
617                 gfs2_qa_put(ip);
618                 goto out;
619         }
620
621         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
622         if (error)
623                 goto out_unhold;
624         ip->i_qadata->qa_qd_num++;
625         qd++;
626
627         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
628         if (error)
629                 goto out_unhold;
630         ip->i_qadata->qa_qd_num++;
631         qd++;
632
633         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
634             !uid_eq(uid, ip->i_inode.i_uid)) {
635                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
636                 if (error)
637                         goto out_unhold;
638                 ip->i_qadata->qa_qd_num++;
639                 qd++;
640         }
641
642         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
643             !gid_eq(gid, ip->i_inode.i_gid)) {
644                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
645                 if (error)
646                         goto out_unhold;
647                 ip->i_qadata->qa_qd_num++;
648                 qd++;
649         }
650
651 out_unhold:
652         if (error)
653                 gfs2_quota_unhold(ip);
654 out:
655         return error;
656 }
657
658 void gfs2_quota_unhold(struct gfs2_inode *ip)
659 {
660         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
661         u32 x;
662
663         if (ip->i_qadata == NULL)
664                 return;
665
666         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
667
668         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
669                 qdsb_put(ip->i_qadata->qa_qd[x]);
670                 ip->i_qadata->qa_qd[x] = NULL;
671         }
672         ip->i_qadata->qa_qd_num = 0;
673         gfs2_qa_put(ip);
674 }
675
676 static int sort_qd(const void *a, const void *b)
677 {
678         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
679         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
680
681         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
682                 return -1;
683         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
684                 return 1;
685         return 0;
686 }
687
688 static void do_qc(struct gfs2_quota_data *qd, s64 change, int qc_type)
689 {
690         struct gfs2_sbd *sdp = qd->qd_sbd;
691         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
692         struct gfs2_quota_change *qc = qd->qd_bh_qc;
693         s64 x;
694
695         mutex_lock(&sdp->sd_quota_mutex);
696         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
697
698         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
699                 qc->qc_change = 0;
700                 qc->qc_flags = 0;
701                 if (qd->qd_id.type == USRQUOTA)
702                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
703                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
704         }
705
706         x = be64_to_cpu(qc->qc_change) + change;
707         qc->qc_change = cpu_to_be64(x);
708
709         spin_lock(&qd_lock);
710         qd->qd_change = x;
711         spin_unlock(&qd_lock);
712
713         if (qc_type == QC_CHANGE) {
714                 if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
715                         qd_hold(qd);
716                         slot_hold(qd);
717                 }
718         } else {
719                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
720                 clear_bit(QDF_CHANGE, &qd->qd_flags);
721                 qc->qc_flags = 0;
722                 qc->qc_id = 0;
723                 slot_put(qd);
724                 qd_put(qd);
725         }
726
727         if (change < 0) /* Reset quiet flag if we freed some blocks */
728                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
729         mutex_unlock(&sdp->sd_quota_mutex);
730 }
731
732 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
733                                   unsigned off, void *buf, unsigned bytes)
734 {
735         struct inode *inode = &ip->i_inode;
736         struct gfs2_sbd *sdp = GFS2_SB(inode);
737         struct address_space *mapping = inode->i_mapping;
738         struct page *page;
739         struct buffer_head *bh;
740         u64 blk;
741         unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
742         unsigned to_write = bytes, pg_off = off;
743         int done = 0;
744
745         blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
746         boff = off % bsize;
747
748         page = grab_cache_page(mapping, index);
749         if (!page)
750                 return -ENOMEM;
751         if (!page_has_buffers(page))
752                 create_empty_buffers(page, bsize, 0);
753
754         bh = page_buffers(page);
755         while (!done) {
756                 /* Find the beginning block within the page */
757                 if (pg_off >= ((bnum * bsize) + bsize)) {
758                         bh = bh->b_this_page;
759                         bnum++;
760                         blk++;
761                         continue;
762                 }
763                 if (!buffer_mapped(bh)) {
764                         gfs2_block_map(inode, blk, bh, 1);
765                         if (!buffer_mapped(bh))
766                                 goto unlock_out;
767                         /* If it's a newly allocated disk block, zero it */
768                         if (buffer_new(bh))
769                                 zero_user(page, bnum * bsize, bh->b_size);
770                 }
771                 if (PageUptodate(page))
772                         set_buffer_uptodate(bh);
773                 if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
774                         goto unlock_out;
775                 if (gfs2_is_jdata(ip))
776                         gfs2_trans_add_data(ip->i_gl, bh);
777                 else
778                         gfs2_ordered_add_inode(ip);
779
780                 /* If we need to write to the next block as well */
781                 if (to_write > (bsize - boff)) {
782                         pg_off += (bsize - boff);
783                         to_write -= (bsize - boff);
784                         boff = pg_off % bsize;
785                         continue;
786                 }
787                 done = 1;
788         }
789
790         /* Write to the page, now that we have setup the buffer(s) */
791         memcpy_to_page(page, off, buf, bytes);
792         flush_dcache_page(page);
793         unlock_page(page);
794         put_page(page);
795
796         return 0;
797
798 unlock_out:
799         unlock_page(page);
800         put_page(page);
801         return -EIO;
802 }
803
804 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
805                                  loff_t loc)
806 {
807         unsigned long pg_beg;
808         unsigned pg_off, nbytes, overflow = 0;
809         int pg_oflow = 0, error;
810         void *ptr;
811
812         nbytes = sizeof(struct gfs2_quota);
813
814         pg_beg = loc >> PAGE_SHIFT;
815         pg_off = offset_in_page(loc);
816
817         /* If the quota straddles a page boundary, split the write in two */
818         if ((pg_off + nbytes) > PAGE_SIZE) {
819                 pg_oflow = 1;
820                 overflow = (pg_off + nbytes) - PAGE_SIZE;
821         }
822
823         ptr = qp;
824         error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
825                                        nbytes - overflow);
826         /* If there's an overflow, write the remaining bytes to the next page */
827         if (!error && pg_oflow)
828                 error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
829                                                ptr + nbytes - overflow,
830                                                overflow);
831         return error;
832 }
833
834 /**
835  * gfs2_adjust_quota - adjust record of current block usage
836  * @ip: The quota inode
837  * @loc: Offset of the entry in the quota file
838  * @change: The amount of usage change to record
839  * @qd: The quota data
840  * @fdq: The updated limits to record
841  *
842  * This function was mostly borrowed from gfs2_block_truncate_page which was
843  * in turn mostly borrowed from ext3
844  *
845  * Returns: 0 or -ve on error
846  */
847
848 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
849                              s64 change, struct gfs2_quota_data *qd,
850                              struct qc_dqblk *fdq)
851 {
852         struct inode *inode = &ip->i_inode;
853         struct gfs2_sbd *sdp = GFS2_SB(inode);
854         struct gfs2_quota q;
855         int err;
856         u64 size;
857
858         if (gfs2_is_stuffed(ip)) {
859                 err = gfs2_unstuff_dinode(ip);
860                 if (err)
861                         return err;
862         }
863
864         memset(&q, 0, sizeof(struct gfs2_quota));
865         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
866         if (err < 0)
867                 return err;
868
869         loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
870         err = -EIO;
871         be64_add_cpu(&q.qu_value, change);
872         if (((s64)be64_to_cpu(q.qu_value)) < 0)
873                 q.qu_value = 0; /* Never go negative on quota usage */
874         qd->qd_qb.qb_value = q.qu_value;
875         if (fdq) {
876                 if (fdq->d_fieldmask & QC_SPC_SOFT) {
877                         q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
878                         qd->qd_qb.qb_warn = q.qu_warn;
879                 }
880                 if (fdq->d_fieldmask & QC_SPC_HARD) {
881                         q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
882                         qd->qd_qb.qb_limit = q.qu_limit;
883                 }
884                 if (fdq->d_fieldmask & QC_SPACE) {
885                         q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
886                         qd->qd_qb.qb_value = q.qu_value;
887                 }
888         }
889
890         err = gfs2_write_disk_quota(ip, &q, loc);
891         if (!err) {
892                 size = loc + sizeof(struct gfs2_quota);
893                 if (size > inode->i_size)
894                         i_size_write(inode, size);
895                 inode->i_mtime = inode->i_atime = current_time(inode);
896                 mark_inode_dirty(inode);
897                 set_bit(QDF_REFRESH, &qd->qd_flags);
898         }
899
900         return err;
901 }
902
903 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
904 {
905         struct gfs2_sbd *sdp = (*qda)->qd_sbd;
906         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
907         struct gfs2_alloc_parms ap = { .aflags = 0, };
908         unsigned int data_blocks, ind_blocks;
909         struct gfs2_holder *ghs, i_gh;
910         unsigned int qx, x;
911         struct gfs2_quota_data *qd;
912         unsigned reserved;
913         loff_t offset;
914         unsigned int nalloc = 0, blocks;
915         int error;
916
917         error = gfs2_qa_get(ip);
918         if (error)
919                 return error;
920
921         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
922                               &data_blocks, &ind_blocks);
923
924         ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
925         if (!ghs) {
926                 error = -ENOMEM;
927                 goto out;
928         }
929
930         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
931         inode_lock(&ip->i_inode);
932         for (qx = 0; qx < num_qd; qx++) {
933                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
934                                            GL_NOCACHE, &ghs[qx]);
935                 if (error)
936                         goto out_dq;
937         }
938
939         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
940         if (error)
941                 goto out_dq;
942
943         for (x = 0; x < num_qd; x++) {
944                 offset = qd2offset(qda[x]);
945                 if (gfs2_write_alloc_required(ip, offset,
946                                               sizeof(struct gfs2_quota)))
947                         nalloc++;
948         }
949
950         /* 
951          * 1 blk for unstuffing inode if stuffed. We add this extra
952          * block to the reservation unconditionally. If the inode
953          * doesn't need unstuffing, the block will be released to the 
954          * rgrp since it won't be allocated during the transaction
955          */
956         /* +3 in the end for unstuffing block, inode size update block
957          * and another block in case quota straddles page boundary and 
958          * two blocks need to be updated instead of 1 */
959         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
960
961         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
962         ap.target = reserved;
963         error = gfs2_inplace_reserve(ip, &ap);
964         if (error)
965                 goto out_alloc;
966
967         if (nalloc)
968                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
969
970         error = gfs2_trans_begin(sdp, blocks, 0);
971         if (error)
972                 goto out_ipres;
973
974         for (x = 0; x < num_qd; x++) {
975                 qd = qda[x];
976                 offset = qd2offset(qd);
977                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
978                 if (error)
979                         goto out_end_trans;
980
981                 do_qc(qd, -qd->qd_change_sync, QC_SYNC);
982                 set_bit(QDF_REFRESH, &qd->qd_flags);
983         }
984
985         error = 0;
986
987 out_end_trans:
988         gfs2_trans_end(sdp);
989 out_ipres:
990         gfs2_inplace_release(ip);
991 out_alloc:
992         gfs2_glock_dq_uninit(&i_gh);
993 out_dq:
994         while (qx--)
995                 gfs2_glock_dq_uninit(&ghs[qx]);
996         inode_unlock(&ip->i_inode);
997         kfree(ghs);
998         gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
999                        GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
1000 out:
1001         gfs2_qa_put(ip);
1002         return error;
1003 }
1004
1005 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
1006 {
1007         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1008         struct gfs2_quota q;
1009         struct gfs2_quota_lvb *qlvb;
1010         loff_t pos;
1011         int error;
1012
1013         memset(&q, 0, sizeof(struct gfs2_quota));
1014         pos = qd2offset(qd);
1015         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
1016         if (error < 0)
1017                 return error;
1018
1019         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1020         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1021         qlvb->__pad = 0;
1022         qlvb->qb_limit = q.qu_limit;
1023         qlvb->qb_warn = q.qu_warn;
1024         qlvb->qb_value = q.qu_value;
1025         qd->qd_qb = *qlvb;
1026
1027         return 0;
1028 }
1029
1030 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1031                     struct gfs2_holder *q_gh)
1032 {
1033         struct gfs2_sbd *sdp = qd->qd_sbd;
1034         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1035         struct gfs2_holder i_gh;
1036         int error;
1037
1038         gfs2_assert_warn(sdp, sdp == qd->qd_gl->gl_name.ln_sbd);
1039 restart:
1040         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1041         if (error)
1042                 return error;
1043
1044         if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1045                 force_refresh = FORCE;
1046
1047         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1048
1049         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1050                 gfs2_glock_dq_uninit(q_gh);
1051                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1052                                            GL_NOCACHE, q_gh);
1053                 if (error)
1054                         return error;
1055
1056                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1057                 if (error)
1058                         goto fail;
1059
1060                 error = update_qd(sdp, qd);
1061                 if (error)
1062                         goto fail_gunlock;
1063
1064                 gfs2_glock_dq_uninit(&i_gh);
1065                 gfs2_glock_dq_uninit(q_gh);
1066                 force_refresh = 0;
1067                 goto restart;
1068         }
1069
1070         return 0;
1071
1072 fail_gunlock:
1073         gfs2_glock_dq_uninit(&i_gh);
1074 fail:
1075         gfs2_glock_dq_uninit(q_gh);
1076         return error;
1077 }
1078
1079 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1080 {
1081         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1082         struct gfs2_quota_data *qd;
1083         u32 x;
1084         int error = 0;
1085
1086         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON &&
1087             sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET)
1088                 return 0;
1089
1090         error = gfs2_quota_hold(ip, uid, gid);
1091         if (error)
1092                 return error;
1093
1094         sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1095              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1096
1097         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1098                 qd = ip->i_qadata->qa_qd[x];
1099                 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1100                 if (error)
1101                         break;
1102         }
1103
1104         if (!error)
1105                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1106         else {
1107                 while (x--)
1108                         gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1109                 gfs2_quota_unhold(ip);
1110         }
1111
1112         return error;
1113 }
1114
1115 static int need_sync(struct gfs2_quota_data *qd)
1116 {
1117         struct gfs2_sbd *sdp = qd->qd_sbd;
1118         struct gfs2_tune *gt = &sdp->sd_tune;
1119         s64 value;
1120         unsigned int num, den;
1121         int do_sync = 1;
1122
1123         if (!qd->qd_qb.qb_limit)
1124                 return 0;
1125
1126         spin_lock(&qd_lock);
1127         value = qd->qd_change;
1128         spin_unlock(&qd_lock);
1129
1130         spin_lock(&gt->gt_spin);
1131         num = gt->gt_quota_scale_num;
1132         den = gt->gt_quota_scale_den;
1133         spin_unlock(&gt->gt_spin);
1134
1135         if (value < 0)
1136                 do_sync = 0;
1137         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1138                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1139                 do_sync = 0;
1140         else {
1141                 value *= gfs2_jindex_size(sdp) * num;
1142                 value = div_s64(value, den);
1143                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1144                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1145                         do_sync = 0;
1146         }
1147
1148         return do_sync;
1149 }
1150
1151 void gfs2_quota_unlock(struct gfs2_inode *ip)
1152 {
1153         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1154         struct gfs2_quota_data *qda[4];
1155         unsigned int count = 0;
1156         u32 x;
1157         int found;
1158
1159         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1160                 return;
1161
1162         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1163                 struct gfs2_quota_data *qd;
1164                 int sync;
1165
1166                 qd = ip->i_qadata->qa_qd[x];
1167                 sync = need_sync(qd);
1168
1169                 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1170                 if (!sync)
1171                         continue;
1172
1173                 spin_lock(&qd_lock);
1174                 found = qd_check_sync(sdp, qd, NULL);
1175                 spin_unlock(&qd_lock);
1176
1177                 if (!found)
1178                         continue;
1179
1180                 gfs2_assert_warn(sdp, qd->qd_change_sync);
1181                 if (bh_get(qd)) {
1182                         clear_bit(QDF_LOCKED, &qd->qd_flags);
1183                         slot_put(qd);
1184                         qd_put(qd);
1185                         continue;
1186                 }
1187
1188                 qda[count++] = qd;
1189         }
1190
1191         if (count) {
1192                 do_sync(count, qda);
1193                 for (x = 0; x < count; x++)
1194                         qd_unlock(qda[x]);
1195         }
1196
1197         gfs2_quota_unhold(ip);
1198 }
1199
1200 #define MAX_LINE 256
1201
1202 static int print_message(struct gfs2_quota_data *qd, char *type)
1203 {
1204         struct gfs2_sbd *sdp = qd->qd_sbd;
1205
1206         if (sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET)
1207                 fs_info(sdp, "quota %s for %s %u\n",
1208                         type,
1209                         (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1210                         from_kqid(&init_user_ns, qd->qd_id));
1211
1212         return 0;
1213 }
1214
1215 /**
1216  * gfs2_quota_check - check if allocating new blocks will exceed quota
1217  * @ip:  The inode for which this check is being performed
1218  * @uid: The uid to check against
1219  * @gid: The gid to check against
1220  * @ap:  The allocation parameters. ap->target contains the requested
1221  *       blocks. ap->min_target, if set, contains the minimum blks
1222  *       requested.
1223  *
1224  * Returns: 0 on success.
1225  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1226  *                  quota must allow at least min_req blks for success and
1227  *                  ap->allowed is set to the number of blocks allowed
1228  *
1229  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1230  *                  of blocks available.
1231  */
1232 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1233                      struct gfs2_alloc_parms *ap)
1234 {
1235         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1236         struct gfs2_quota_data *qd;
1237         s64 value, warn, limit;
1238         u32 x;
1239         int error = 0;
1240
1241         ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1242         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1243                 return 0;
1244
1245         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1246                 qd = ip->i_qadata->qa_qd[x];
1247
1248                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1249                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1250                         continue;
1251
1252                 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1253                 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1254                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1255                 spin_lock(&qd_lock);
1256                 value += qd->qd_change;
1257                 spin_unlock(&qd_lock);
1258
1259                 if (limit > 0 && (limit - value) < ap->allowed)
1260                         ap->allowed = limit - value;
1261                 /* If we can't meet the target */
1262                 if (limit && limit < (value + (s64)ap->target)) {
1263                         /* If no min_target specified or we don't meet
1264                          * min_target, return -EDQUOT */
1265                         if (!ap->min_target || ap->min_target > ap->allowed) {
1266                                 if (!test_and_set_bit(QDF_QMSG_QUIET,
1267                                                       &qd->qd_flags)) {
1268                                         print_message(qd, "exceeded");
1269                                         quota_send_warning(qd->qd_id,
1270                                                            sdp->sd_vfs->s_dev,
1271                                                            QUOTA_NL_BHARDWARN);
1272                                 }
1273                                 error = -EDQUOT;
1274                                 break;
1275                         }
1276                 } else if (warn && warn < value &&
1277                            time_after_eq(jiffies, qd->qd_last_warn +
1278                                          gfs2_tune_get(sdp, gt_quota_warn_period)
1279                                          * HZ)) {
1280                         quota_send_warning(qd->qd_id,
1281                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1282                         error = print_message(qd, "warning");
1283                         qd->qd_last_warn = jiffies;
1284                 }
1285         }
1286         return error;
1287 }
1288
1289 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1290                        kuid_t uid, kgid_t gid)
1291 {
1292         struct gfs2_quota_data *qd;
1293         u32 x;
1294         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1295
1296         if ((sdp->sd_args.ar_quota != GFS2_QUOTA_ON &&
1297             sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET) ||
1298             gfs2_assert_warn(sdp, change))
1299                 return;
1300         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1301                 return;
1302
1303         if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1304                                  ip->i_qadata->qa_ref > 0))
1305                 return;
1306         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1307                 qd = ip->i_qadata->qa_qd[x];
1308
1309                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1310                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1311                         do_qc(qd, change, QC_CHANGE);
1312                 }
1313         }
1314 }
1315
1316 int gfs2_quota_sync(struct super_block *sb, int type)
1317 {
1318         struct gfs2_sbd *sdp = sb->s_fs_info;
1319         struct gfs2_quota_data **qda;
1320         unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1321         unsigned int num_qd;
1322         unsigned int x;
1323         int error = 0;
1324
1325         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1326         if (!qda)
1327                 return -ENOMEM;
1328
1329         mutex_lock(&sdp->sd_quota_sync_mutex);
1330         sdp->sd_quota_sync_gen++;
1331
1332         do {
1333                 num_qd = 0;
1334
1335                 for (;;) {
1336                         error = qd_fish(sdp, qda + num_qd);
1337                         if (error || !qda[num_qd])
1338                                 break;
1339                         if (++num_qd == max_qd)
1340                                 break;
1341                 }
1342
1343                 if (num_qd) {
1344                         if (!error)
1345                                 error = do_sync(num_qd, qda);
1346                         if (!error)
1347                                 for (x = 0; x < num_qd; x++)
1348                                         qda[x]->qd_sync_gen =
1349                                                 sdp->sd_quota_sync_gen;
1350
1351                         for (x = 0; x < num_qd; x++)
1352                                 qd_unlock(qda[x]);
1353                 }
1354         } while (!error && num_qd == max_qd);
1355
1356         mutex_unlock(&sdp->sd_quota_sync_mutex);
1357         kfree(qda);
1358
1359         return error;
1360 }
1361
1362 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1363 {
1364         struct gfs2_quota_data *qd;
1365         struct gfs2_holder q_gh;
1366         int error;
1367
1368         error = qd_get(sdp, qid, &qd);
1369         if (error)
1370                 return error;
1371
1372         error = do_glock(qd, FORCE, &q_gh);
1373         if (!error)
1374                 gfs2_glock_dq_uninit(&q_gh);
1375
1376         qd_put(qd);
1377         return error;
1378 }
1379
1380 int gfs2_quota_init(struct gfs2_sbd *sdp)
1381 {
1382         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1383         u64 size = i_size_read(sdp->sd_qc_inode);
1384         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1385         unsigned int x, slot = 0;
1386         unsigned int found = 0;
1387         unsigned int hash;
1388         unsigned int bm_size;
1389         u64 dblock;
1390         u32 extlen = 0;
1391         int error;
1392
1393         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1394                 return -EIO;
1395
1396         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1397         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1398         bm_size *= sizeof(unsigned long);
1399         error = -ENOMEM;
1400         sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1401         if (sdp->sd_quota_bitmap == NULL)
1402                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1403                                                  __GFP_ZERO);
1404         if (!sdp->sd_quota_bitmap)
1405                 return error;
1406
1407         for (x = 0; x < blocks; x++) {
1408                 struct buffer_head *bh;
1409                 const struct gfs2_quota_change *qc;
1410                 unsigned int y;
1411
1412                 if (!extlen) {
1413                         extlen = 32;
1414                         error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1415                         if (error)
1416                                 goto fail;
1417                 }
1418                 error = -EIO;
1419                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1420                 if (!bh)
1421                         goto fail;
1422                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1423                         brelse(bh);
1424                         goto fail;
1425                 }
1426
1427                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1428                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1429                      y++, slot++) {
1430                         struct gfs2_quota_data *qd;
1431                         s64 qc_change = be64_to_cpu(qc->qc_change);
1432                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1433                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1434                                                 USRQUOTA : GRPQUOTA;
1435                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1436                                                       be32_to_cpu(qc->qc_id));
1437                         qc++;
1438                         if (!qc_change)
1439                                 continue;
1440
1441                         hash = gfs2_qd_hash(sdp, qc_id);
1442                         qd = qd_alloc(hash, sdp, qc_id);
1443                         if (qd == NULL) {
1444                                 brelse(bh);
1445                                 goto fail;
1446                         }
1447
1448                         set_bit(QDF_CHANGE, &qd->qd_flags);
1449                         qd->qd_change = qc_change;
1450                         qd->qd_slot = slot;
1451                         qd->qd_slot_count = 1;
1452
1453                         spin_lock(&qd_lock);
1454                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1455                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1456                         atomic_inc(&sdp->sd_quota_count);
1457                         spin_unlock(&qd_lock);
1458
1459                         spin_lock_bucket(hash);
1460                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1461                         spin_unlock_bucket(hash);
1462
1463                         found++;
1464                 }
1465
1466                 brelse(bh);
1467                 dblock++;
1468                 extlen--;
1469         }
1470
1471         if (found)
1472                 fs_info(sdp, "found %u quota changes\n", found);
1473
1474         return 0;
1475
1476 fail:
1477         gfs2_quota_cleanup(sdp);
1478         return error;
1479 }
1480
1481 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1482 {
1483         struct gfs2_quota_data *qd;
1484         LIST_HEAD(dispose);
1485         int count;
1486
1487         BUG_ON(test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
1488
1489         spin_lock(&qd_lock);
1490         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1491                 spin_lock(&qd->qd_lockref.lock);
1492                 if (qd->qd_lockref.count != 0) {
1493                         spin_unlock(&qd->qd_lockref.lock);
1494                         continue;
1495                 }
1496                 lockref_mark_dead(&qd->qd_lockref);
1497                 spin_unlock(&qd->qd_lockref.lock);
1498
1499                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1500                 list_add(&qd->qd_lru, &dispose);
1501         }
1502         spin_unlock(&qd_lock);
1503
1504         gfs2_qd_list_dispose(&dispose);
1505
1506         wait_event_timeout(sdp->sd_kill_wait,
1507                 (count = atomic_read(&sdp->sd_quota_count)) == 0,
1508                 HZ * 60);
1509
1510         if (count != 0)
1511                 fs_err(sdp, "%d left-over quota data objects\n", count);
1512
1513         kvfree(sdp->sd_quota_bitmap);
1514         sdp->sd_quota_bitmap = NULL;
1515 }
1516
1517 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1518 {
1519         if (error == 0 || error == -EROFS)
1520                 return;
1521         if (!gfs2_withdrawn(sdp)) {
1522                 if (!cmpxchg(&sdp->sd_log_error, 0, error))
1523                         fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1524                 wake_up(&sdp->sd_logd_waitq);
1525         }
1526 }
1527
1528 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1529                                int (*fxn)(struct super_block *sb, int type),
1530                                unsigned long t, unsigned long *timeo,
1531                                unsigned int *new_timeo)
1532 {
1533         if (t >= *timeo) {
1534                 int error = fxn(sdp->sd_vfs, 0);
1535                 quotad_error(sdp, msg, error);
1536                 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1537         } else {
1538                 *timeo -= t;
1539         }
1540 }
1541
1542 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1543         if (!sdp->sd_statfs_force_sync) {
1544                 sdp->sd_statfs_force_sync = 1;
1545                 wake_up(&sdp->sd_quota_wait);
1546         }
1547 }
1548
1549
1550 /**
1551  * gfs2_quotad - Write cached quota changes into the quota file
1552  * @data: Pointer to GFS2 superblock
1553  *
1554  */
1555
1556 int gfs2_quotad(void *data)
1557 {
1558         struct gfs2_sbd *sdp = data;
1559         struct gfs2_tune *tune = &sdp->sd_tune;
1560         unsigned long statfs_timeo = 0;
1561         unsigned long quotad_timeo = 0;
1562         unsigned long t = 0;
1563
1564         while (!kthread_should_stop()) {
1565                 if (gfs2_withdrawn(sdp))
1566                         break;
1567
1568                 /* Update the master statfs file */
1569                 if (sdp->sd_statfs_force_sync) {
1570                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1571                         quotad_error(sdp, "statfs", error);
1572                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1573                 }
1574                 else
1575                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1576                                            &statfs_timeo,
1577                                            &tune->gt_statfs_quantum);
1578
1579                 /* Update quota file */
1580                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1581                                    &quotad_timeo, &tune->gt_quota_quantum);
1582
1583                 try_to_freeze();
1584
1585                 t = min(quotad_timeo, statfs_timeo);
1586
1587                 t = wait_event_interruptible_timeout(sdp->sd_quota_wait,
1588                                 sdp->sd_statfs_force_sync ||
1589                                 gfs2_withdrawn(sdp) ||
1590                                 kthread_should_stop(),
1591                                 t);
1592
1593                 if (sdp->sd_statfs_force_sync)
1594                         t = 0;
1595         }
1596
1597         return 0;
1598 }
1599
1600 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1601 {
1602         struct gfs2_sbd *sdp = sb->s_fs_info;
1603
1604         memset(state, 0, sizeof(*state));
1605
1606         switch (sdp->sd_args.ar_quota) {
1607         case GFS2_QUOTA_QUIET:
1608                 fallthrough;
1609         case GFS2_QUOTA_ON:
1610                 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1611                 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1612                 fallthrough;
1613         case GFS2_QUOTA_ACCOUNT:
1614                 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1615                                                   QCI_SYSFILE;
1616                 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1617                                                   QCI_SYSFILE;
1618                 break;
1619         case GFS2_QUOTA_OFF:
1620                 break;
1621         }
1622         if (sdp->sd_quota_inode) {
1623                 state->s_state[USRQUOTA].ino =
1624                                         GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1625                 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1626         }
1627         state->s_state[USRQUOTA].nextents = 1;  /* unsupported */
1628         state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1629         state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1630         return 0;
1631 }
1632
1633 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1634                           struct qc_dqblk *fdq)
1635 {
1636         struct gfs2_sbd *sdp = sb->s_fs_info;
1637         struct gfs2_quota_lvb *qlvb;
1638         struct gfs2_quota_data *qd;
1639         struct gfs2_holder q_gh;
1640         int error;
1641
1642         memset(fdq, 0, sizeof(*fdq));
1643
1644         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1645                 return -ESRCH; /* Crazy XFS error code */
1646
1647         if ((qid.type != USRQUOTA) &&
1648             (qid.type != GRPQUOTA))
1649                 return -EINVAL;
1650
1651         error = qd_get(sdp, qid, &qd);
1652         if (error)
1653                 return error;
1654         error = do_glock(qd, FORCE, &q_gh);
1655         if (error)
1656                 goto out;
1657
1658         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1659         fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1660         fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1661         fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1662
1663         gfs2_glock_dq_uninit(&q_gh);
1664 out:
1665         qd_put(qd);
1666         return error;
1667 }
1668
1669 /* GFS2 only supports a subset of the XFS fields */
1670 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1671
1672 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1673                           struct qc_dqblk *fdq)
1674 {
1675         struct gfs2_sbd *sdp = sb->s_fs_info;
1676         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1677         struct gfs2_quota_data *qd;
1678         struct gfs2_holder q_gh, i_gh;
1679         unsigned int data_blocks, ind_blocks;
1680         unsigned int blocks = 0;
1681         int alloc_required;
1682         loff_t offset;
1683         int error;
1684
1685         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1686                 return -ESRCH; /* Crazy XFS error code */
1687
1688         if ((qid.type != USRQUOTA) &&
1689             (qid.type != GRPQUOTA))
1690                 return -EINVAL;
1691
1692         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1693                 return -EINVAL;
1694
1695         error = qd_get(sdp, qid, &qd);
1696         if (error)
1697                 return error;
1698
1699         error = gfs2_qa_get(ip);
1700         if (error)
1701                 goto out_put;
1702
1703         inode_lock(&ip->i_inode);
1704         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1705         if (error)
1706                 goto out_unlockput;
1707         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1708         if (error)
1709                 goto out_q;
1710
1711         /* Check for existing entry, if none then alloc new blocks */
1712         error = update_qd(sdp, qd);
1713         if (error)
1714                 goto out_i;
1715
1716         /* If nothing has changed, this is a no-op */
1717         if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1718             ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1719                 fdq->d_fieldmask ^= QC_SPC_SOFT;
1720
1721         if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1722             ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1723                 fdq->d_fieldmask ^= QC_SPC_HARD;
1724
1725         if ((fdq->d_fieldmask & QC_SPACE) &&
1726             ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1727                 fdq->d_fieldmask ^= QC_SPACE;
1728
1729         if (fdq->d_fieldmask == 0)
1730                 goto out_i;
1731
1732         offset = qd2offset(qd);
1733         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1734         if (gfs2_is_stuffed(ip))
1735                 alloc_required = 1;
1736         if (alloc_required) {
1737                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1738                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1739                                        &data_blocks, &ind_blocks);
1740                 blocks = 1 + data_blocks + ind_blocks;
1741                 ap.target = blocks;
1742                 error = gfs2_inplace_reserve(ip, &ap);
1743                 if (error)
1744                         goto out_i;
1745                 blocks += gfs2_rg_blocks(ip, blocks);
1746         }
1747
1748         /* Some quotas span block boundaries and can update two blocks,
1749            adding an extra block to the transaction to handle such quotas */
1750         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1751         if (error)
1752                 goto out_release;
1753
1754         /* Apply changes */
1755         error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1756         if (!error)
1757                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1758
1759         gfs2_trans_end(sdp);
1760 out_release:
1761         if (alloc_required)
1762                 gfs2_inplace_release(ip);
1763 out_i:
1764         gfs2_glock_dq_uninit(&i_gh);
1765 out_q:
1766         gfs2_glock_dq_uninit(&q_gh);
1767 out_unlockput:
1768         gfs2_qa_put(ip);
1769         inode_unlock(&ip->i_inode);
1770 out_put:
1771         qd_put(qd);
1772         return error;
1773 }
1774
1775 const struct quotactl_ops gfs2_quotactl_ops = {
1776         .quota_sync     = gfs2_quota_sync,
1777         .get_state      = gfs2_quota_get_state,
1778         .get_dqblk      = gfs2_get_dqblk,
1779         .set_dqblk      = gfs2_set_dqblk,
1780 };
1781
1782 void __init gfs2_quota_hash_init(void)
1783 {
1784         unsigned i;
1785
1786         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1787                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1788 }