gfs2: introduce qd_bh_get_or_undo
[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_bh_get_or_undo(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
477 {
478         int error;
479
480         error = bh_get(qd);
481         if (!error)
482                 return 0;
483
484         clear_bit(QDF_LOCKED, &qd->qd_flags);
485         slot_put(qd);
486         qd_put(qd);
487         return error;
488 }
489
490 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
491 {
492         struct gfs2_quota_data *qd = NULL, *iter;
493         int error;
494
495         *qdp = NULL;
496
497         if (sb_rdonly(sdp->sd_vfs))
498                 return 0;
499
500         spin_lock(&qd_lock);
501
502         list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
503                 if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
504                         qd = iter;
505                         break;
506                 }
507         }
508
509         spin_unlock(&qd_lock);
510
511         if (qd) {
512                 error = qd_bh_get_or_undo(sdp, qd);
513                 if (error)
514                         return error;
515                 *qdp = qd;
516         }
517
518         return 0;
519 }
520
521 static void qdsb_put(struct gfs2_quota_data *qd)
522 {
523         bh_put(qd);
524         slot_put(qd);
525         qd_put(qd);
526 }
527
528 static void qd_unlock(struct gfs2_quota_data *qd)
529 {
530         gfs2_assert_warn(qd->qd_sbd, test_bit(QDF_LOCKED, &qd->qd_flags));
531         clear_bit(QDF_LOCKED, &qd->qd_flags);
532         qdsb_put(qd);
533 }
534
535 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
536                     struct gfs2_quota_data **qdp)
537 {
538         int error;
539
540         error = qd_get(sdp, qid, qdp);
541         if (error)
542                 return error;
543
544         error = slot_get(*qdp);
545         if (error)
546                 goto fail;
547
548         error = bh_get(*qdp);
549         if (error)
550                 goto fail_slot;
551
552         return 0;
553
554 fail_slot:
555         slot_put(*qdp);
556 fail:
557         qd_put(*qdp);
558         return error;
559 }
560
561 /**
562  * gfs2_qa_get - make sure we have a quota allocations data structure,
563  *               if necessary
564  * @ip: the inode for this reservation
565  */
566 int gfs2_qa_get(struct gfs2_inode *ip)
567 {
568         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
569         struct inode *inode = &ip->i_inode;
570
571         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
572                 return 0;
573
574         spin_lock(&inode->i_lock);
575         if (ip->i_qadata == NULL) {
576                 struct gfs2_qadata *tmp;
577
578                 spin_unlock(&inode->i_lock);
579                 tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
580                 if (!tmp)
581                         return -ENOMEM;
582
583                 spin_lock(&inode->i_lock);
584                 if (ip->i_qadata == NULL)
585                         ip->i_qadata = tmp;
586                 else
587                         kmem_cache_free(gfs2_qadata_cachep, tmp);
588         }
589         ip->i_qadata->qa_ref++;
590         spin_unlock(&inode->i_lock);
591         return 0;
592 }
593
594 void gfs2_qa_put(struct gfs2_inode *ip)
595 {
596         struct inode *inode = &ip->i_inode;
597
598         spin_lock(&inode->i_lock);
599         if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
600                 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
601                 ip->i_qadata = NULL;
602         }
603         spin_unlock(&inode->i_lock);
604 }
605
606 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
607 {
608         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
609         struct gfs2_quota_data **qd;
610         int error;
611
612         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
613                 return 0;
614
615         error = gfs2_qa_get(ip);
616         if (error)
617                 return error;
618
619         qd = ip->i_qadata->qa_qd;
620
621         if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
622             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
623                 error = -EIO;
624                 gfs2_qa_put(ip);
625                 goto out;
626         }
627
628         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
629         if (error)
630                 goto out_unhold;
631         ip->i_qadata->qa_qd_num++;
632         qd++;
633
634         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
635         if (error)
636                 goto out_unhold;
637         ip->i_qadata->qa_qd_num++;
638         qd++;
639
640         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
641             !uid_eq(uid, ip->i_inode.i_uid)) {
642                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
643                 if (error)
644                         goto out_unhold;
645                 ip->i_qadata->qa_qd_num++;
646                 qd++;
647         }
648
649         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
650             !gid_eq(gid, ip->i_inode.i_gid)) {
651                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
652                 if (error)
653                         goto out_unhold;
654                 ip->i_qadata->qa_qd_num++;
655                 qd++;
656         }
657
658 out_unhold:
659         if (error)
660                 gfs2_quota_unhold(ip);
661 out:
662         return error;
663 }
664
665 void gfs2_quota_unhold(struct gfs2_inode *ip)
666 {
667         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
668         u32 x;
669
670         if (ip->i_qadata == NULL)
671                 return;
672
673         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
674
675         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
676                 qdsb_put(ip->i_qadata->qa_qd[x]);
677                 ip->i_qadata->qa_qd[x] = NULL;
678         }
679         ip->i_qadata->qa_qd_num = 0;
680         gfs2_qa_put(ip);
681 }
682
683 static int sort_qd(const void *a, const void *b)
684 {
685         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
686         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
687
688         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
689                 return -1;
690         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
691                 return 1;
692         return 0;
693 }
694
695 static void do_qc(struct gfs2_quota_data *qd, s64 change, int qc_type)
696 {
697         struct gfs2_sbd *sdp = qd->qd_sbd;
698         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
699         struct gfs2_quota_change *qc = qd->qd_bh_qc;
700         s64 x;
701
702         mutex_lock(&sdp->sd_quota_mutex);
703         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
704
705         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
706                 qc->qc_change = 0;
707                 qc->qc_flags = 0;
708                 if (qd->qd_id.type == USRQUOTA)
709                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
710                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
711         }
712
713         x = be64_to_cpu(qc->qc_change) + change;
714         qc->qc_change = cpu_to_be64(x);
715
716         spin_lock(&qd_lock);
717         qd->qd_change = x;
718         spin_unlock(&qd_lock);
719
720         if (qc_type == QC_CHANGE) {
721                 if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
722                         qd_hold(qd);
723                         slot_hold(qd);
724                 }
725         } else {
726                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
727                 clear_bit(QDF_CHANGE, &qd->qd_flags);
728                 qc->qc_flags = 0;
729                 qc->qc_id = 0;
730                 slot_put(qd);
731                 qd_put(qd);
732         }
733
734         if (change < 0) /* Reset quiet flag if we freed some blocks */
735                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
736         mutex_unlock(&sdp->sd_quota_mutex);
737 }
738
739 static int gfs2_write_buf_to_page(struct gfs2_sbd *sdp, unsigned long index,
740                                   unsigned off, void *buf, unsigned bytes)
741 {
742         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
743         struct inode *inode = &ip->i_inode;
744         struct address_space *mapping = inode->i_mapping;
745         struct page *page;
746         struct buffer_head *bh;
747         u64 blk;
748         unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
749         unsigned to_write = bytes, pg_off = off;
750
751         blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
752         boff = off % bsize;
753
754         page = grab_cache_page(mapping, index);
755         if (!page)
756                 return -ENOMEM;
757         if (!page_has_buffers(page))
758                 create_empty_buffers(page, bsize, 0);
759
760         bh = page_buffers(page);
761         for(;;) {
762                 /* Find the beginning block within the page */
763                 if (pg_off >= ((bnum * bsize) + bsize)) {
764                         bh = bh->b_this_page;
765                         bnum++;
766                         blk++;
767                         continue;
768                 }
769                 if (!buffer_mapped(bh)) {
770                         gfs2_block_map(inode, blk, bh, 1);
771                         if (!buffer_mapped(bh))
772                                 goto unlock_out;
773                         /* If it's a newly allocated disk block, zero it */
774                         if (buffer_new(bh))
775                                 zero_user(page, bnum * bsize, bh->b_size);
776                 }
777                 if (PageUptodate(page))
778                         set_buffer_uptodate(bh);
779                 if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
780                         goto unlock_out;
781                 gfs2_trans_add_data(ip->i_gl, bh);
782
783                 /* If we need to write to the next block as well */
784                 if (to_write > (bsize - boff)) {
785                         pg_off += (bsize - boff);
786                         to_write -= (bsize - boff);
787                         boff = pg_off % bsize;
788                         continue;
789                 }
790                 break;
791         }
792
793         /* Write to the page, now that we have setup the buffer(s) */
794         memcpy_to_page(page, off, buf, bytes);
795         flush_dcache_page(page);
796         unlock_page(page);
797         put_page(page);
798
799         return 0;
800
801 unlock_out:
802         unlock_page(page);
803         put_page(page);
804         return -EIO;
805 }
806
807 static int gfs2_write_disk_quota(struct gfs2_sbd *sdp, struct gfs2_quota *qp,
808                                  loff_t loc)
809 {
810         unsigned long pg_beg;
811         unsigned pg_off, nbytes, overflow = 0;
812         int error;
813         void *ptr;
814
815         nbytes = sizeof(struct gfs2_quota);
816
817         pg_beg = loc >> PAGE_SHIFT;
818         pg_off = offset_in_page(loc);
819
820         /* If the quota straddles a page boundary, split the write in two */
821         if ((pg_off + nbytes) > PAGE_SIZE)
822                 overflow = (pg_off + nbytes) - PAGE_SIZE;
823
824         ptr = qp;
825         error = gfs2_write_buf_to_page(sdp, pg_beg, pg_off, ptr,
826                                        nbytes - overflow);
827         /* If there's an overflow, write the remaining bytes to the next page */
828         if (!error && overflow)
829                 error = gfs2_write_buf_to_page(sdp, pg_beg + 1, 0,
830                                                ptr + nbytes - overflow,
831                                                overflow);
832         return error;
833 }
834
835 /**
836  * gfs2_adjust_quota - adjust record of current block usage
837  * @sdp: The superblock
838  * @loc: Offset of the entry in the quota file
839  * @change: The amount of usage change to record
840  * @qd: The quota data
841  * @fdq: The updated limits to record
842  *
843  * This function was mostly borrowed from gfs2_block_truncate_page which was
844  * in turn mostly borrowed from ext3
845  *
846  * Returns: 0 or -ve on error
847  */
848
849 static int gfs2_adjust_quota(struct gfs2_sbd *sdp, loff_t loc,
850                              s64 change, struct gfs2_quota_data *qd,
851                              struct qc_dqblk *fdq)
852 {
853         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
854         struct inode *inode = &ip->i_inode;
855         struct gfs2_quota q;
856         int err;
857         u64 size;
858
859         if (gfs2_is_stuffed(ip)) {
860                 err = gfs2_unstuff_dinode(ip);
861                 if (err)
862                         return err;
863         }
864
865         memset(&q, 0, sizeof(struct gfs2_quota));
866         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
867         if (err < 0)
868                 return err;
869
870         loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
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(sdp, &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         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
918                               &data_blocks, &ind_blocks);
919
920         ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
921         if (!ghs)
922                 return -ENOMEM;
923
924         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
925         inode_lock(&ip->i_inode);
926         for (qx = 0; qx < num_qd; qx++) {
927                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
928                                            GL_NOCACHE, &ghs[qx]);
929                 if (error)
930                         goto out_dq;
931         }
932
933         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
934         if (error)
935                 goto out_dq;
936
937         for (x = 0; x < num_qd; x++) {
938                 offset = qd2offset(qda[x]);
939                 if (gfs2_write_alloc_required(ip, offset,
940                                               sizeof(struct gfs2_quota)))
941                         nalloc++;
942         }
943
944         /* 
945          * 1 blk for unstuffing inode if stuffed. We add this extra
946          * block to the reservation unconditionally. If the inode
947          * doesn't need unstuffing, the block will be released to the 
948          * rgrp since it won't be allocated during the transaction
949          */
950         /* +3 in the end for unstuffing block, inode size update block
951          * and another block in case quota straddles page boundary and 
952          * two blocks need to be updated instead of 1 */
953         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
954
955         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
956         ap.target = reserved;
957         error = gfs2_inplace_reserve(ip, &ap);
958         if (error)
959                 goto out_alloc;
960
961         if (nalloc)
962                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
963
964         error = gfs2_trans_begin(sdp, blocks, 0);
965         if (error)
966                 goto out_ipres;
967
968         for (x = 0; x < num_qd; x++) {
969                 qd = qda[x];
970                 offset = qd2offset(qd);
971                 error = gfs2_adjust_quota(sdp, offset, qd->qd_change_sync, qd,
972                                                         NULL);
973                 if (error)
974                         goto out_end_trans;
975
976                 do_qc(qd, -qd->qd_change_sync, QC_SYNC);
977                 set_bit(QDF_REFRESH, &qd->qd_flags);
978         }
979
980         error = 0;
981
982 out_end_trans:
983         gfs2_trans_end(sdp);
984 out_ipres:
985         gfs2_inplace_release(ip);
986 out_alloc:
987         gfs2_glock_dq_uninit(&i_gh);
988 out_dq:
989         while (qx--)
990                 gfs2_glock_dq_uninit(&ghs[qx]);
991         inode_unlock(&ip->i_inode);
992         kfree(ghs);
993         gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
994                        GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
995         if (!error) {
996                 for (x = 0; x < num_qd; x++)
997                         qda[x]->qd_sync_gen = sdp->sd_quota_sync_gen;
998         }
999         return error;
1000 }
1001
1002 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
1003 {
1004         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1005         struct gfs2_quota q;
1006         struct gfs2_quota_lvb *qlvb;
1007         loff_t pos;
1008         int error;
1009
1010         memset(&q, 0, sizeof(struct gfs2_quota));
1011         pos = qd2offset(qd);
1012         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
1013         if (error < 0)
1014                 return error;
1015
1016         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1017         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1018         qlvb->__pad = 0;
1019         qlvb->qb_limit = q.qu_limit;
1020         qlvb->qb_warn = q.qu_warn;
1021         qlvb->qb_value = q.qu_value;
1022         qd->qd_qb = *qlvb;
1023
1024         return 0;
1025 }
1026
1027 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1028                     struct gfs2_holder *q_gh)
1029 {
1030         struct gfs2_sbd *sdp = qd->qd_sbd;
1031         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1032         struct gfs2_holder i_gh;
1033         int error;
1034
1035         gfs2_assert_warn(sdp, sdp == qd->qd_gl->gl_name.ln_sbd);
1036 restart:
1037         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1038         if (error)
1039                 return error;
1040
1041         if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1042                 force_refresh = FORCE;
1043
1044         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1045
1046         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1047                 gfs2_glock_dq_uninit(q_gh);
1048                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1049                                            GL_NOCACHE, q_gh);
1050                 if (error)
1051                         return error;
1052
1053                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1054                 if (error)
1055                         goto fail;
1056
1057                 error = update_qd(sdp, qd);
1058                 if (error)
1059                         goto fail_gunlock;
1060
1061                 gfs2_glock_dq_uninit(&i_gh);
1062                 gfs2_glock_dq_uninit(q_gh);
1063                 force_refresh = 0;
1064                 goto restart;
1065         }
1066
1067         return 0;
1068
1069 fail_gunlock:
1070         gfs2_glock_dq_uninit(&i_gh);
1071 fail:
1072         gfs2_glock_dq_uninit(q_gh);
1073         return error;
1074 }
1075
1076 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1077 {
1078         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1079         struct gfs2_quota_data *qd;
1080         u32 x;
1081         int error;
1082
1083         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON &&
1084             sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET)
1085                 return 0;
1086
1087         error = gfs2_quota_hold(ip, uid, gid);
1088         if (error)
1089                 return error;
1090
1091         sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1092              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1093
1094         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1095                 qd = ip->i_qadata->qa_qd[x];
1096                 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1097                 if (error)
1098                         break;
1099         }
1100
1101         if (!error)
1102                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1103         else {
1104                 while (x--)
1105                         gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1106                 gfs2_quota_unhold(ip);
1107         }
1108
1109         return error;
1110 }
1111
1112 static bool need_sync(struct gfs2_quota_data *qd)
1113 {
1114         struct gfs2_sbd *sdp = qd->qd_sbd;
1115         struct gfs2_tune *gt = &sdp->sd_tune;
1116         s64 value;
1117         unsigned int num, den;
1118
1119         if (!qd->qd_qb.qb_limit)
1120                 return false;
1121
1122         spin_lock(&qd_lock);
1123         value = qd->qd_change;
1124         spin_unlock(&qd_lock);
1125
1126         spin_lock(&gt->gt_spin);
1127         num = gt->gt_quota_scale_num;
1128         den = gt->gt_quota_scale_den;
1129         spin_unlock(&gt->gt_spin);
1130
1131         if (value <= 0)
1132                 return false;
1133         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1134                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1135                 return false;
1136         else {
1137                 value *= gfs2_jindex_size(sdp) * num;
1138                 value = div_s64(value, den);
1139                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1140                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1141                         return false;
1142         }
1143
1144         return true;
1145 }
1146
1147 void gfs2_quota_unlock(struct gfs2_inode *ip)
1148 {
1149         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1150         struct gfs2_quota_data *qda[2 * GFS2_MAXQUOTAS];
1151         unsigned int count = 0;
1152         u32 x;
1153         int found;
1154
1155         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1156                 return;
1157
1158         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1159                 struct gfs2_quota_data *qd;
1160                 bool sync;
1161
1162                 qd = ip->i_qadata->qa_qd[x];
1163                 sync = need_sync(qd);
1164
1165                 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1166                 if (!sync)
1167                         continue;
1168
1169                 spin_lock(&qd_lock);
1170                 found = qd_check_sync(sdp, qd, NULL);
1171                 spin_unlock(&qd_lock);
1172
1173                 if (!found)
1174                         continue;
1175
1176                 if (!qd_bh_get_or_undo(sdp, qd))
1177                         qda[count++] = qd;
1178         }
1179
1180         if (count) {
1181                 do_sync(count, qda);
1182                 for (x = 0; x < count; x++)
1183                         qd_unlock(qda[x]);
1184         }
1185
1186         gfs2_quota_unhold(ip);
1187 }
1188
1189 #define MAX_LINE 256
1190
1191 static int print_message(struct gfs2_quota_data *qd, char *type)
1192 {
1193         struct gfs2_sbd *sdp = qd->qd_sbd;
1194
1195         if (sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET)
1196                 fs_info(sdp, "quota %s for %s %u\n",
1197                         type,
1198                         (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1199                         from_kqid(&init_user_ns, qd->qd_id));
1200
1201         return 0;
1202 }
1203
1204 /**
1205  * gfs2_quota_check - check if allocating new blocks will exceed quota
1206  * @ip:  The inode for which this check is being performed
1207  * @uid: The uid to check against
1208  * @gid: The gid to check against
1209  * @ap:  The allocation parameters. ap->target contains the requested
1210  *       blocks. ap->min_target, if set, contains the minimum blks
1211  *       requested.
1212  *
1213  * Returns: 0 on success.
1214  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1215  *                  quota must allow at least min_req blks for success and
1216  *                  ap->allowed is set to the number of blocks allowed
1217  *
1218  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1219  *                  of blocks available.
1220  */
1221 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1222                      struct gfs2_alloc_parms *ap)
1223 {
1224         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1225         struct gfs2_quota_data *qd;
1226         s64 value, warn, limit;
1227         u32 x;
1228         int error = 0;
1229
1230         ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1231         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1232                 return 0;
1233
1234         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1235                 qd = ip->i_qadata->qa_qd[x];
1236
1237                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1238                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1239                         continue;
1240
1241                 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1242                 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1243                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1244                 spin_lock(&qd_lock);
1245                 value += qd->qd_change;
1246                 spin_unlock(&qd_lock);
1247
1248                 if (limit > 0 && (limit - value) < ap->allowed)
1249                         ap->allowed = limit - value;
1250                 /* If we can't meet the target */
1251                 if (limit && limit < (value + (s64)ap->target)) {
1252                         /* If no min_target specified or we don't meet
1253                          * min_target, return -EDQUOT */
1254                         if (!ap->min_target || ap->min_target > ap->allowed) {
1255                                 if (!test_and_set_bit(QDF_QMSG_QUIET,
1256                                                       &qd->qd_flags)) {
1257                                         print_message(qd, "exceeded");
1258                                         quota_send_warning(qd->qd_id,
1259                                                            sdp->sd_vfs->s_dev,
1260                                                            QUOTA_NL_BHARDWARN);
1261                                 }
1262                                 error = -EDQUOT;
1263                                 break;
1264                         }
1265                 } else if (warn && warn < value &&
1266                            time_after_eq(jiffies, qd->qd_last_warn +
1267                                          gfs2_tune_get(sdp, gt_quota_warn_period)
1268                                          * HZ)) {
1269                         quota_send_warning(qd->qd_id,
1270                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1271                         error = print_message(qd, "warning");
1272                         qd->qd_last_warn = jiffies;
1273                 }
1274         }
1275         return error;
1276 }
1277
1278 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1279                        kuid_t uid, kgid_t gid)
1280 {
1281         struct gfs2_quota_data *qd;
1282         u32 x;
1283         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1284
1285         if ((sdp->sd_args.ar_quota != GFS2_QUOTA_ON &&
1286             sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET) ||
1287             gfs2_assert_warn(sdp, change))
1288                 return;
1289         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1290                 return;
1291
1292         if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1293                                  ip->i_qadata->qa_ref > 0))
1294                 return;
1295         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1296                 qd = ip->i_qadata->qa_qd[x];
1297
1298                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1299                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1300                         do_qc(qd, change, QC_CHANGE);
1301                 }
1302         }
1303 }
1304
1305 int gfs2_quota_sync(struct super_block *sb, int type)
1306 {
1307         struct gfs2_sbd *sdp = sb->s_fs_info;
1308         struct gfs2_quota_data **qda;
1309         unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1310         unsigned int num_qd;
1311         unsigned int x;
1312         int error = 0;
1313
1314         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1315         if (!qda)
1316                 return -ENOMEM;
1317
1318         mutex_lock(&sdp->sd_quota_sync_mutex);
1319         sdp->sd_quota_sync_gen++;
1320
1321         do {
1322                 num_qd = 0;
1323
1324                 for (;;) {
1325                         error = qd_fish(sdp, qda + num_qd);
1326                         if (error || !qda[num_qd])
1327                                 break;
1328                         if (++num_qd == max_qd)
1329                                 break;
1330                 }
1331
1332                 if (num_qd) {
1333                         if (!error)
1334                                 error = do_sync(num_qd, qda);
1335
1336                         for (x = 0; x < num_qd; x++)
1337                                 qd_unlock(qda[x]);
1338                 }
1339         } while (!error && num_qd == max_qd);
1340
1341         mutex_unlock(&sdp->sd_quota_sync_mutex);
1342         kfree(qda);
1343
1344         return error;
1345 }
1346
1347 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1348 {
1349         struct gfs2_quota_data *qd;
1350         struct gfs2_holder q_gh;
1351         int error;
1352
1353         error = qd_get(sdp, qid, &qd);
1354         if (error)
1355                 return error;
1356
1357         error = do_glock(qd, FORCE, &q_gh);
1358         if (!error)
1359                 gfs2_glock_dq_uninit(&q_gh);
1360
1361         qd_put(qd);
1362         return error;
1363 }
1364
1365 int gfs2_quota_init(struct gfs2_sbd *sdp)
1366 {
1367         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1368         u64 size = i_size_read(sdp->sd_qc_inode);
1369         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1370         unsigned int x, slot = 0;
1371         unsigned int found = 0;
1372         unsigned int hash;
1373         unsigned int bm_size;
1374         u64 dblock;
1375         u32 extlen = 0;
1376         int error;
1377
1378         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1379                 return -EIO;
1380
1381         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1382         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1383         bm_size *= sizeof(unsigned long);
1384         error = -ENOMEM;
1385         sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1386         if (sdp->sd_quota_bitmap == NULL)
1387                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1388                                                  __GFP_ZERO);
1389         if (!sdp->sd_quota_bitmap)
1390                 return error;
1391
1392         for (x = 0; x < blocks; x++) {
1393                 struct buffer_head *bh;
1394                 const struct gfs2_quota_change *qc;
1395                 unsigned int y;
1396
1397                 if (!extlen) {
1398                         extlen = 32;
1399                         error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1400                         if (error)
1401                                 goto fail;
1402                 }
1403                 error = -EIO;
1404                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1405                 if (!bh)
1406                         goto fail;
1407                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1408                         brelse(bh);
1409                         goto fail;
1410                 }
1411
1412                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1413                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1414                      y++, slot++) {
1415                         struct gfs2_quota_data *qd;
1416                         s64 qc_change = be64_to_cpu(qc->qc_change);
1417                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1418                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1419                                                 USRQUOTA : GRPQUOTA;
1420                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1421                                                       be32_to_cpu(qc->qc_id));
1422                         qc++;
1423                         if (!qc_change)
1424                                 continue;
1425
1426                         hash = gfs2_qd_hash(sdp, qc_id);
1427                         qd = qd_alloc(hash, sdp, qc_id);
1428                         if (qd == NULL) {
1429                                 brelse(bh);
1430                                 goto fail;
1431                         }
1432
1433                         set_bit(QDF_CHANGE, &qd->qd_flags);
1434                         qd->qd_change = qc_change;
1435                         qd->qd_slot = slot;
1436                         qd->qd_slot_count = 1;
1437
1438                         spin_lock(&qd_lock);
1439                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1440                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1441                         atomic_inc(&sdp->sd_quota_count);
1442                         spin_unlock(&qd_lock);
1443
1444                         spin_lock_bucket(hash);
1445                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1446                         spin_unlock_bucket(hash);
1447
1448                         found++;
1449                 }
1450
1451                 brelse(bh);
1452                 dblock++;
1453                 extlen--;
1454         }
1455
1456         if (found)
1457                 fs_info(sdp, "found %u quota changes\n", found);
1458
1459         return 0;
1460
1461 fail:
1462         gfs2_quota_cleanup(sdp);
1463         return error;
1464 }
1465
1466 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1467 {
1468         struct gfs2_quota_data *qd;
1469         LIST_HEAD(dispose);
1470         int count;
1471
1472         BUG_ON(test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
1473
1474         spin_lock(&qd_lock);
1475         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1476                 spin_lock(&qd->qd_lockref.lock);
1477                 if (qd->qd_lockref.count != 0) {
1478                         spin_unlock(&qd->qd_lockref.lock);
1479                         continue;
1480                 }
1481                 lockref_mark_dead(&qd->qd_lockref);
1482                 spin_unlock(&qd->qd_lockref.lock);
1483
1484                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1485                 list_add(&qd->qd_lru, &dispose);
1486         }
1487         spin_unlock(&qd_lock);
1488
1489         gfs2_qd_list_dispose(&dispose);
1490
1491         wait_event_timeout(sdp->sd_kill_wait,
1492                 (count = atomic_read(&sdp->sd_quota_count)) == 0,
1493                 HZ * 60);
1494
1495         if (count != 0)
1496                 fs_err(sdp, "%d left-over quota data objects\n", count);
1497
1498         kvfree(sdp->sd_quota_bitmap);
1499         sdp->sd_quota_bitmap = NULL;
1500 }
1501
1502 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1503 {
1504         if (error == 0 || error == -EROFS)
1505                 return;
1506         if (!gfs2_withdrawn(sdp)) {
1507                 if (!cmpxchg(&sdp->sd_log_error, 0, error))
1508                         fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1509                 wake_up(&sdp->sd_logd_waitq);
1510         }
1511 }
1512
1513 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1514                                int (*fxn)(struct super_block *sb, int type),
1515                                unsigned long t, unsigned long *timeo,
1516                                unsigned int *new_timeo)
1517 {
1518         if (t >= *timeo) {
1519                 int error = fxn(sdp->sd_vfs, 0);
1520                 quotad_error(sdp, msg, error);
1521                 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1522         } else {
1523                 *timeo -= t;
1524         }
1525 }
1526
1527 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1528         if (!sdp->sd_statfs_force_sync) {
1529                 sdp->sd_statfs_force_sync = 1;
1530                 wake_up(&sdp->sd_quota_wait);
1531         }
1532 }
1533
1534
1535 /**
1536  * gfs2_quotad - Write cached quota changes into the quota file
1537  * @data: Pointer to GFS2 superblock
1538  *
1539  */
1540
1541 int gfs2_quotad(void *data)
1542 {
1543         struct gfs2_sbd *sdp = data;
1544         struct gfs2_tune *tune = &sdp->sd_tune;
1545         unsigned long statfs_timeo = 0;
1546         unsigned long quotad_timeo = 0;
1547         unsigned long t = 0;
1548
1549         while (!kthread_should_stop()) {
1550                 if (gfs2_withdrawn(sdp))
1551                         break;
1552
1553                 /* Update the master statfs file */
1554                 if (sdp->sd_statfs_force_sync) {
1555                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1556                         quotad_error(sdp, "statfs", error);
1557                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1558                 }
1559                 else
1560                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1561                                            &statfs_timeo,
1562                                            &tune->gt_statfs_quantum);
1563
1564                 /* Update quota file */
1565                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1566                                    &quotad_timeo, &tune->gt_quota_quantum);
1567
1568                 try_to_freeze();
1569
1570                 t = min(quotad_timeo, statfs_timeo);
1571
1572                 t = wait_event_interruptible_timeout(sdp->sd_quota_wait,
1573                                 sdp->sd_statfs_force_sync ||
1574                                 gfs2_withdrawn(sdp) ||
1575                                 kthread_should_stop(),
1576                                 t);
1577
1578                 if (sdp->sd_statfs_force_sync)
1579                         t = 0;
1580         }
1581
1582         return 0;
1583 }
1584
1585 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1586 {
1587         struct gfs2_sbd *sdp = sb->s_fs_info;
1588
1589         memset(state, 0, sizeof(*state));
1590
1591         switch (sdp->sd_args.ar_quota) {
1592         case GFS2_QUOTA_QUIET:
1593                 fallthrough;
1594         case GFS2_QUOTA_ON:
1595                 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1596                 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1597                 fallthrough;
1598         case GFS2_QUOTA_ACCOUNT:
1599                 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1600                                                   QCI_SYSFILE;
1601                 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1602                                                   QCI_SYSFILE;
1603                 break;
1604         case GFS2_QUOTA_OFF:
1605                 break;
1606         }
1607         if (sdp->sd_quota_inode) {
1608                 state->s_state[USRQUOTA].ino =
1609                                         GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1610                 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1611         }
1612         state->s_state[USRQUOTA].nextents = 1;  /* unsupported */
1613         state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1614         state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1615         return 0;
1616 }
1617
1618 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1619                           struct qc_dqblk *fdq)
1620 {
1621         struct gfs2_sbd *sdp = sb->s_fs_info;
1622         struct gfs2_quota_lvb *qlvb;
1623         struct gfs2_quota_data *qd;
1624         struct gfs2_holder q_gh;
1625         int error;
1626
1627         memset(fdq, 0, sizeof(*fdq));
1628
1629         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1630                 return -ESRCH; /* Crazy XFS error code */
1631
1632         if ((qid.type != USRQUOTA) &&
1633             (qid.type != GRPQUOTA))
1634                 return -EINVAL;
1635
1636         error = qd_get(sdp, qid, &qd);
1637         if (error)
1638                 return error;
1639         error = do_glock(qd, FORCE, &q_gh);
1640         if (error)
1641                 goto out;
1642
1643         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1644         fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1645         fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1646         fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1647
1648         gfs2_glock_dq_uninit(&q_gh);
1649 out:
1650         qd_put(qd);
1651         return error;
1652 }
1653
1654 /* GFS2 only supports a subset of the XFS fields */
1655 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1656
1657 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1658                           struct qc_dqblk *fdq)
1659 {
1660         struct gfs2_sbd *sdp = sb->s_fs_info;
1661         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1662         struct gfs2_quota_data *qd;
1663         struct gfs2_holder q_gh, i_gh;
1664         unsigned int data_blocks, ind_blocks;
1665         unsigned int blocks = 0;
1666         int alloc_required;
1667         loff_t offset;
1668         int error;
1669
1670         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1671                 return -ESRCH; /* Crazy XFS error code */
1672
1673         if ((qid.type != USRQUOTA) &&
1674             (qid.type != GRPQUOTA))
1675                 return -EINVAL;
1676
1677         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1678                 return -EINVAL;
1679
1680         error = qd_get(sdp, qid, &qd);
1681         if (error)
1682                 return error;
1683
1684         error = gfs2_qa_get(ip);
1685         if (error)
1686                 goto out_put;
1687
1688         inode_lock(&ip->i_inode);
1689         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1690         if (error)
1691                 goto out_unlockput;
1692         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1693         if (error)
1694                 goto out_q;
1695
1696         /* Check for existing entry, if none then alloc new blocks */
1697         error = update_qd(sdp, qd);
1698         if (error)
1699                 goto out_i;
1700
1701         /* If nothing has changed, this is a no-op */
1702         if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1703             ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1704                 fdq->d_fieldmask ^= QC_SPC_SOFT;
1705
1706         if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1707             ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1708                 fdq->d_fieldmask ^= QC_SPC_HARD;
1709
1710         if ((fdq->d_fieldmask & QC_SPACE) &&
1711             ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1712                 fdq->d_fieldmask ^= QC_SPACE;
1713
1714         if (fdq->d_fieldmask == 0)
1715                 goto out_i;
1716
1717         offset = qd2offset(qd);
1718         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1719         if (gfs2_is_stuffed(ip))
1720                 alloc_required = 1;
1721         if (alloc_required) {
1722                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1723                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1724                                        &data_blocks, &ind_blocks);
1725                 blocks = 1 + data_blocks + ind_blocks;
1726                 ap.target = blocks;
1727                 error = gfs2_inplace_reserve(ip, &ap);
1728                 if (error)
1729                         goto out_i;
1730                 blocks += gfs2_rg_blocks(ip, blocks);
1731         }
1732
1733         /* Some quotas span block boundaries and can update two blocks,
1734            adding an extra block to the transaction to handle such quotas */
1735         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1736         if (error)
1737                 goto out_release;
1738
1739         /* Apply changes */
1740         error = gfs2_adjust_quota(sdp, offset, 0, qd, fdq);
1741         if (!error)
1742                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1743
1744         gfs2_trans_end(sdp);
1745 out_release:
1746         if (alloc_required)
1747                 gfs2_inplace_release(ip);
1748 out_i:
1749         gfs2_glock_dq_uninit(&i_gh);
1750 out_q:
1751         gfs2_glock_dq_uninit(&q_gh);
1752 out_unlockput:
1753         gfs2_qa_put(ip);
1754         inode_unlock(&ip->i_inode);
1755 out_put:
1756         qd_put(qd);
1757         return error;
1758 }
1759
1760 const struct quotactl_ops gfs2_quotactl_ops = {
1761         .quota_sync     = gfs2_quota_sync,
1762         .get_state      = gfs2_quota_get_state,
1763         .get_dqblk      = gfs2_get_dqblk,
1764         .set_dqblk      = gfs2_set_dqblk,
1765 };
1766
1767 void __init gfs2_quota_hash_init(void)
1768 {
1769         unsigned i;
1770
1771         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1772                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1773 }