86bc73bd770b49fb5a2deab8e63ce43541c4692c
[releases.git] / quota.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
5  */
6
7 /*
8  * Quota change tags are associated with each transaction that allocates or
9  * deallocates space.  Those changes are accumulated locally to each node (in a
10  * per-node file) and then are periodically synced to the quota file.  This
11  * avoids the bottleneck of constantly touching the quota file, but introduces
12  * fuzziness in the current usage value of IDs that are being used on different
13  * nodes in the cluster simultaneously.  So, it is possible for a user on
14  * multiple nodes to overrun their quota, but that overrun is controlable.
15  * Since quota tags are part of transactions, there is no need for a quota check
16  * program to be run on node crashes or anything like that.
17  *
18  * There are couple of knobs that let the administrator manage the quota
19  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
20  * sitting on one node before being synced to the quota file.  (The default is
21  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
22  * of quota file syncs increases as the user moves closer to their limit.  The
23  * more frequent the syncs, the more accurate the quota enforcement, but that
24  * means that there is more contention between the nodes for the quota file.
25  * The default value is one.  This sets the maximum theoretical quota overrun
26  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
27  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
28  * number greater than one makes quota syncs more frequent and reduces the
29  * maximum overrun.  Numbers less than one (but greater than zero) make quota
30  * syncs less frequent.
31  *
32  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
33  * the quota file, so it is not being constantly read.
34  */
35
36 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37
38 #include <linux/sched.h>
39 #include <linux/slab.h>
40 #include <linux/mm.h>
41 #include <linux/spinlock.h>
42 #include <linux/completion.h>
43 #include <linux/buffer_head.h>
44 #include <linux/sort.h>
45 #include <linux/fs.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/kthread.h>
49 #include <linux/freezer.h>
50 #include <linux/quota.h>
51 #include <linux/dqblk_xfs.h>
52 #include <linux/lockref.h>
53 #include <linux/list_lru.h>
54 #include <linux/rcupdate.h>
55 #include <linux/rculist_bl.h>
56 #include <linux/bit_spinlock.h>
57 #include <linux/jhash.h>
58 #include <linux/vmalloc.h>
59
60 #include "gfs2.h"
61 #include "incore.h"
62 #include "bmap.h"
63 #include "glock.h"
64 #include "glops.h"
65 #include "log.h"
66 #include "meta_io.h"
67 #include "quota.h"
68 #include "rgrp.h"
69 #include "super.h"
70 #include "trans.h"
71 #include "inode.h"
72 #include "util.h"
73
74 #define GFS2_QD_HASH_SHIFT      12
75 #define GFS2_QD_HASH_SIZE       BIT(GFS2_QD_HASH_SHIFT)
76 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
77
78 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
79 /*                     -> sd_bitmap_lock                              */
80 static DEFINE_SPINLOCK(qd_lock);
81 struct list_lru gfs2_qd_lru;
82
83 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
84
85 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
86                                  const struct kqid qid)
87 {
88         unsigned int h;
89
90         h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
91         h = jhash(&qid, sizeof(struct kqid), h);
92
93         return h & GFS2_QD_HASH_MASK;
94 }
95
96 static inline void spin_lock_bucket(unsigned int hash)
97 {
98         hlist_bl_lock(&qd_hash_table[hash]);
99 }
100
101 static inline void spin_unlock_bucket(unsigned int hash)
102 {
103         hlist_bl_unlock(&qd_hash_table[hash]);
104 }
105
106 static void gfs2_qd_dealloc(struct rcu_head *rcu)
107 {
108         struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
109         kmem_cache_free(gfs2_quotad_cachep, qd);
110 }
111
112 static void gfs2_qd_dispose(struct list_head *list)
113 {
114         struct gfs2_quota_data *qd;
115         struct gfs2_sbd *sdp;
116
117         while (!list_empty(list)) {
118                 qd = list_first_entry(list, struct gfs2_quota_data, qd_lru);
119                 sdp = qd->qd_gl->gl_name.ln_sbd;
120
121                 list_del(&qd->qd_lru);
122
123                 /* Free from the filesystem-specific list */
124                 spin_lock(&qd_lock);
125                 list_del(&qd->qd_list);
126                 spin_unlock(&qd_lock);
127
128                 spin_lock_bucket(qd->qd_hash);
129                 hlist_bl_del_rcu(&qd->qd_hlist);
130                 spin_unlock_bucket(qd->qd_hash);
131
132                 gfs2_assert_warn(sdp, !qd->qd_change);
133                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
134                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
135
136                 gfs2_glock_put(qd->qd_gl);
137                 atomic_dec(&sdp->sd_quota_count);
138
139                 /* Delete it from the common reclaim list */
140                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
141         }
142 }
143
144
145 static enum lru_status gfs2_qd_isolate(struct list_head *item,
146                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
147 {
148         struct list_head *dispose = arg;
149         struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
150
151         if (!spin_trylock(&qd->qd_lockref.lock))
152                 return LRU_SKIP;
153
154         if (qd->qd_lockref.count == 0) {
155                 lockref_mark_dead(&qd->qd_lockref);
156                 list_lru_isolate_move(lru, &qd->qd_lru, dispose);
157         }
158
159         spin_unlock(&qd->qd_lockref.lock);
160         return LRU_REMOVED;
161 }
162
163 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
164                                          struct shrink_control *sc)
165 {
166         LIST_HEAD(dispose);
167         unsigned long freed;
168
169         if (!(sc->gfp_mask & __GFP_FS))
170                 return SHRINK_STOP;
171
172         freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
173                                      gfs2_qd_isolate, &dispose);
174
175         gfs2_qd_dispose(&dispose);
176
177         return freed;
178 }
179
180 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
181                                           struct shrink_control *sc)
182 {
183         return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
184 }
185
186 struct shrinker gfs2_qd_shrinker = {
187         .count_objects = gfs2_qd_shrink_count,
188         .scan_objects = gfs2_qd_shrink_scan,
189         .seeks = DEFAULT_SEEKS,
190         .flags = SHRINKER_NUMA_AWARE,
191 };
192
193
194 static u64 qd2index(struct gfs2_quota_data *qd)
195 {
196         struct kqid qid = qd->qd_id;
197         return (2 * (u64)from_kqid(&init_user_ns, qid)) +
198                 ((qid.type == USRQUOTA) ? 0 : 1);
199 }
200
201 static u64 qd2offset(struct gfs2_quota_data *qd)
202 {
203         u64 offset;
204
205         offset = qd2index(qd);
206         offset *= sizeof(struct gfs2_quota);
207
208         return offset;
209 }
210
211 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
212 {
213         struct gfs2_quota_data *qd;
214         int error;
215
216         qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
217         if (!qd)
218                 return NULL;
219
220         qd->qd_sbd = sdp;
221         qd->qd_lockref.count = 1;
222         spin_lock_init(&qd->qd_lockref.lock);
223         qd->qd_id = qid;
224         qd->qd_slot = -1;
225         INIT_LIST_HEAD(&qd->qd_lru);
226         qd->qd_hash = hash;
227
228         error = gfs2_glock_get(sdp, qd2index(qd),
229                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
230         if (error)
231                 goto fail;
232
233         return qd;
234
235 fail:
236         kmem_cache_free(gfs2_quotad_cachep, qd);
237         return NULL;
238 }
239
240 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
241                                                      const struct gfs2_sbd *sdp,
242                                                      struct kqid qid)
243 {
244         struct gfs2_quota_data *qd;
245         struct hlist_bl_node *h;
246
247         hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
248                 if (!qid_eq(qd->qd_id, qid))
249                         continue;
250                 if (qd->qd_sbd != sdp)
251                         continue;
252                 if (lockref_get_not_dead(&qd->qd_lockref)) {
253                         list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
254                         return qd;
255                 }
256         }
257
258         return NULL;
259 }
260
261
262 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
263                   struct gfs2_quota_data **qdp)
264 {
265         struct gfs2_quota_data *qd, *new_qd;
266         unsigned int hash = gfs2_qd_hash(sdp, qid);
267
268         rcu_read_lock();
269         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
270         rcu_read_unlock();
271
272         if (qd)
273                 return 0;
274
275         new_qd = qd_alloc(hash, sdp, qid);
276         if (!new_qd)
277                 return -ENOMEM;
278
279         spin_lock(&qd_lock);
280         spin_lock_bucket(hash);
281         *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
282         if (qd == NULL) {
283                 *qdp = new_qd;
284                 list_add(&new_qd->qd_list, &sdp->sd_quota_list);
285                 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
286                 atomic_inc(&sdp->sd_quota_count);
287         }
288         spin_unlock_bucket(hash);
289         spin_unlock(&qd_lock);
290
291         if (qd) {
292                 gfs2_glock_put(new_qd->qd_gl);
293                 kmem_cache_free(gfs2_quotad_cachep, new_qd);
294         }
295
296         return 0;
297 }
298
299
300 static void qd_hold(struct gfs2_quota_data *qd)
301 {
302         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
303         gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
304         lockref_get(&qd->qd_lockref);
305 }
306
307 static void qd_put(struct gfs2_quota_data *qd)
308 {
309         if (lockref_put_or_lock(&qd->qd_lockref))
310                 return;
311
312         qd->qd_lockref.count = 0;
313         list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
314         spin_unlock(&qd->qd_lockref.lock);
315
316 }
317
318 static int slot_get(struct gfs2_quota_data *qd)
319 {
320         struct gfs2_sbd *sdp = qd->qd_sbd;
321         unsigned int bit;
322         int error = 0;
323
324         spin_lock(&sdp->sd_bitmap_lock);
325         if (qd->qd_slot_count != 0)
326                 goto out;
327
328         error = -ENOSPC;
329         bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
330         if (bit < sdp->sd_quota_slots) {
331                 set_bit(bit, sdp->sd_quota_bitmap);
332                 qd->qd_slot = bit;
333                 error = 0;
334 out:
335                 qd->qd_slot_count++;
336         }
337         spin_unlock(&sdp->sd_bitmap_lock);
338
339         return error;
340 }
341
342 static void slot_hold(struct gfs2_quota_data *qd)
343 {
344         struct gfs2_sbd *sdp = qd->qd_sbd;
345
346         spin_lock(&sdp->sd_bitmap_lock);
347         gfs2_assert(sdp, qd->qd_slot_count);
348         qd->qd_slot_count++;
349         spin_unlock(&sdp->sd_bitmap_lock);
350 }
351
352 static void slot_put(struct gfs2_quota_data *qd)
353 {
354         struct gfs2_sbd *sdp = qd->qd_sbd;
355
356         spin_lock(&sdp->sd_bitmap_lock);
357         gfs2_assert(sdp, qd->qd_slot_count);
358         if (!--qd->qd_slot_count) {
359                 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
360                 qd->qd_slot = -1;
361         }
362         spin_unlock(&sdp->sd_bitmap_lock);
363 }
364
365 static int bh_get(struct gfs2_quota_data *qd)
366 {
367         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
368         struct inode *inode = sdp->sd_qc_inode;
369         struct gfs2_inode *ip = GFS2_I(inode);
370         unsigned int block, offset;
371         struct buffer_head *bh;
372         struct iomap iomap = { };
373         int error;
374
375         mutex_lock(&sdp->sd_quota_mutex);
376
377         if (qd->qd_bh_count++) {
378                 mutex_unlock(&sdp->sd_quota_mutex);
379                 return 0;
380         }
381
382         block = qd->qd_slot / sdp->sd_qc_per_block;
383         offset = qd->qd_slot % sdp->sd_qc_per_block;
384
385         error = gfs2_iomap_get(inode,
386                                (loff_t)block << inode->i_blkbits,
387                                i_blocksize(inode), &iomap);
388         if (error)
389                 goto fail;
390         error = -ENOENT;
391         if (iomap.type != IOMAP_MAPPED)
392                 goto fail;
393
394         error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
395                                DIO_WAIT, 0, &bh);
396         if (error)
397                 goto fail;
398         error = -EIO;
399         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
400                 goto fail_brelse;
401
402         qd->qd_bh = bh;
403         qd->qd_bh_qc = (struct gfs2_quota_change *)
404                 (bh->b_data + sizeof(struct gfs2_meta_header) +
405                  offset * sizeof(struct gfs2_quota_change));
406
407         mutex_unlock(&sdp->sd_quota_mutex);
408
409         return 0;
410
411 fail_brelse:
412         brelse(bh);
413 fail:
414         qd->qd_bh_count--;
415         mutex_unlock(&sdp->sd_quota_mutex);
416         return error;
417 }
418
419 static void bh_put(struct gfs2_quota_data *qd)
420 {
421         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
422
423         mutex_lock(&sdp->sd_quota_mutex);
424         gfs2_assert(sdp, qd->qd_bh_count);
425         if (!--qd->qd_bh_count) {
426                 brelse(qd->qd_bh);
427                 qd->qd_bh = NULL;
428                 qd->qd_bh_qc = NULL;
429         }
430         mutex_unlock(&sdp->sd_quota_mutex);
431 }
432
433 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
434                          u64 *sync_gen)
435 {
436         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
437             !test_bit(QDF_CHANGE, &qd->qd_flags) ||
438             (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
439                 return 0;
440
441         /*
442          * If qd_change is 0 it means a pending quota change was negated.
443          * We should not sync it, but we still have a qd reference and slot
444          * reference taken by gfs2_quota_change -> do_qc that need to be put.
445          */
446         if (!qd->qd_change && test_and_clear_bit(QDF_CHANGE, &qd->qd_flags)) {
447                 slot_put(qd);
448                 qd_put(qd);
449                 return 0;
450         }
451
452         if (!lockref_get_not_dead(&qd->qd_lockref))
453                 return 0;
454
455         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
456         set_bit(QDF_LOCKED, &qd->qd_flags);
457         qd->qd_change_sync = qd->qd_change;
458         slot_hold(qd);
459         return 1;
460 }
461
462 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
463 {
464         struct gfs2_quota_data *qd = NULL, *iter;
465         int error;
466
467         *qdp = NULL;
468
469         if (sb_rdonly(sdp->sd_vfs))
470                 return 0;
471
472         spin_lock(&qd_lock);
473
474         list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
475                 if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
476                         qd = iter;
477                         break;
478                 }
479         }
480
481         spin_unlock(&qd_lock);
482
483         if (qd) {
484                 gfs2_assert_warn(sdp, qd->qd_change_sync);
485                 error = bh_get(qd);
486                 if (error) {
487                         clear_bit(QDF_LOCKED, &qd->qd_flags);
488                         slot_put(qd);
489                         qd_put(qd);
490                         return error;
491                 }
492         }
493
494         *qdp = qd;
495
496         return 0;
497 }
498
499 static void qd_unlock(struct gfs2_quota_data *qd)
500 {
501         gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
502                          test_bit(QDF_LOCKED, &qd->qd_flags));
503         clear_bit(QDF_LOCKED, &qd->qd_flags);
504         bh_put(qd);
505         slot_put(qd);
506         qd_put(qd);
507 }
508
509 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
510                     struct gfs2_quota_data **qdp)
511 {
512         int error;
513
514         error = qd_get(sdp, qid, qdp);
515         if (error)
516                 return error;
517
518         error = slot_get(*qdp);
519         if (error)
520                 goto fail;
521
522         error = bh_get(*qdp);
523         if (error)
524                 goto fail_slot;
525
526         return 0;
527
528 fail_slot:
529         slot_put(*qdp);
530 fail:
531         qd_put(*qdp);
532         return error;
533 }
534
535 static void qdsb_put(struct gfs2_quota_data *qd)
536 {
537         bh_put(qd);
538         slot_put(qd);
539         qd_put(qd);
540 }
541
542 /**
543  * gfs2_qa_get - make sure we have a quota allocations data structure,
544  *               if necessary
545  * @ip: the inode for this reservation
546  */
547 int gfs2_qa_get(struct gfs2_inode *ip)
548 {
549         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
550         struct inode *inode = &ip->i_inode;
551
552         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
553                 return 0;
554
555         spin_lock(&inode->i_lock);
556         if (ip->i_qadata == NULL) {
557                 struct gfs2_qadata *tmp;
558
559                 spin_unlock(&inode->i_lock);
560                 tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
561                 if (!tmp)
562                         return -ENOMEM;
563
564                 spin_lock(&inode->i_lock);
565                 if (ip->i_qadata == NULL)
566                         ip->i_qadata = tmp;
567                 else
568                         kmem_cache_free(gfs2_qadata_cachep, tmp);
569         }
570         ip->i_qadata->qa_ref++;
571         spin_unlock(&inode->i_lock);
572         return 0;
573 }
574
575 void gfs2_qa_put(struct gfs2_inode *ip)
576 {
577         struct inode *inode = &ip->i_inode;
578
579         spin_lock(&inode->i_lock);
580         if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
581                 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
582                 ip->i_qadata = NULL;
583         }
584         spin_unlock(&inode->i_lock);
585 }
586
587 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
588 {
589         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
590         struct gfs2_quota_data **qd;
591         int error;
592
593         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
594                 return 0;
595
596         error = gfs2_qa_get(ip);
597         if (error)
598                 return error;
599
600         qd = ip->i_qadata->qa_qd;
601
602         if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
603             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
604                 error = -EIO;
605                 goto out;
606         }
607
608         error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
609         if (error)
610                 goto out_unhold;
611         ip->i_qadata->qa_qd_num++;
612         qd++;
613
614         error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
615         if (error)
616                 goto out_unhold;
617         ip->i_qadata->qa_qd_num++;
618         qd++;
619
620         if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
621             !uid_eq(uid, ip->i_inode.i_uid)) {
622                 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
623                 if (error)
624                         goto out_unhold;
625                 ip->i_qadata->qa_qd_num++;
626                 qd++;
627         }
628
629         if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
630             !gid_eq(gid, ip->i_inode.i_gid)) {
631                 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
632                 if (error)
633                         goto out_unhold;
634                 ip->i_qadata->qa_qd_num++;
635                 qd++;
636         }
637
638 out_unhold:
639         if (error)
640                 gfs2_quota_unhold(ip);
641 out:
642         return error;
643 }
644
645 void gfs2_quota_unhold(struct gfs2_inode *ip)
646 {
647         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
648         u32 x;
649
650         if (ip->i_qadata == NULL)
651                 return;
652
653         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
654
655         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
656                 qdsb_put(ip->i_qadata->qa_qd[x]);
657                 ip->i_qadata->qa_qd[x] = NULL;
658         }
659         ip->i_qadata->qa_qd_num = 0;
660         gfs2_qa_put(ip);
661 }
662
663 static int sort_qd(const void *a, const void *b)
664 {
665         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
666         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
667
668         if (qid_lt(qd_a->qd_id, qd_b->qd_id))
669                 return -1;
670         if (qid_lt(qd_b->qd_id, qd_a->qd_id))
671                 return 1;
672         return 0;
673 }
674
675 static void do_qc(struct gfs2_quota_data *qd, s64 change)
676 {
677         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
678         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
679         struct gfs2_quota_change *qc = qd->qd_bh_qc;
680         s64 x;
681
682         mutex_lock(&sdp->sd_quota_mutex);
683         gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
684
685         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
686                 qc->qc_change = 0;
687                 qc->qc_flags = 0;
688                 if (qd->qd_id.type == USRQUOTA)
689                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
690                 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
691         }
692
693         x = be64_to_cpu(qc->qc_change) + change;
694         qc->qc_change = cpu_to_be64(x);
695
696         spin_lock(&qd_lock);
697         qd->qd_change = x;
698         spin_unlock(&qd_lock);
699
700         if (!x) {
701                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
702                 clear_bit(QDF_CHANGE, &qd->qd_flags);
703                 qc->qc_flags = 0;
704                 qc->qc_id = 0;
705                 slot_put(qd);
706                 qd_put(qd);
707         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
708                 qd_hold(qd);
709                 slot_hold(qd);
710         }
711
712         if (change < 0) /* Reset quiet flag if we freed some blocks */
713                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
714         mutex_unlock(&sdp->sd_quota_mutex);
715 }
716
717 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
718                                   unsigned off, void *buf, unsigned bytes)
719 {
720         struct inode *inode = &ip->i_inode;
721         struct gfs2_sbd *sdp = GFS2_SB(inode);
722         struct address_space *mapping = inode->i_mapping;
723         struct page *page;
724         struct buffer_head *bh;
725         void *kaddr;
726         u64 blk;
727         unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
728         unsigned to_write = bytes, pg_off = off;
729         int done = 0;
730
731         blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
732         boff = off % bsize;
733
734         page = find_or_create_page(mapping, index, GFP_NOFS);
735         if (!page)
736                 return -ENOMEM;
737         if (!page_has_buffers(page))
738                 create_empty_buffers(page, bsize, 0);
739
740         bh = page_buffers(page);
741         while (!done) {
742                 /* Find the beginning block within the page */
743                 if (pg_off >= ((bnum * bsize) + bsize)) {
744                         bh = bh->b_this_page;
745                         bnum++;
746                         blk++;
747                         continue;
748                 }
749                 if (!buffer_mapped(bh)) {
750                         gfs2_block_map(inode, blk, bh, 1);
751                         if (!buffer_mapped(bh))
752                                 goto unlock_out;
753                         /* If it's a newly allocated disk block, zero it */
754                         if (buffer_new(bh))
755                                 zero_user(page, bnum * bsize, bh->b_size);
756                 }
757                 if (PageUptodate(page))
758                         set_buffer_uptodate(bh);
759                 if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
760                         goto unlock_out;
761                 if (gfs2_is_jdata(ip))
762                         gfs2_trans_add_data(ip->i_gl, bh);
763                 else
764                         gfs2_ordered_add_inode(ip);
765
766                 /* If we need to write to the next block as well */
767                 if (to_write > (bsize - boff)) {
768                         pg_off += (bsize - boff);
769                         to_write -= (bsize - boff);
770                         boff = pg_off % bsize;
771                         continue;
772                 }
773                 done = 1;
774         }
775
776         /* Write to the page, now that we have setup the buffer(s) */
777         kaddr = kmap_atomic(page);
778         memcpy(kaddr + off, buf, bytes);
779         flush_dcache_page(page);
780         kunmap_atomic(kaddr);
781         unlock_page(page);
782         put_page(page);
783
784         return 0;
785
786 unlock_out:
787         unlock_page(page);
788         put_page(page);
789         return -EIO;
790 }
791
792 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
793                                  loff_t loc)
794 {
795         unsigned long pg_beg;
796         unsigned pg_off, nbytes, overflow = 0;
797         int pg_oflow = 0, error;
798         void *ptr;
799
800         nbytes = sizeof(struct gfs2_quota);
801
802         pg_beg = loc >> PAGE_SHIFT;
803         pg_off = offset_in_page(loc);
804
805         /* If the quota straddles a page boundary, split the write in two */
806         if ((pg_off + nbytes) > PAGE_SIZE) {
807                 pg_oflow = 1;
808                 overflow = (pg_off + nbytes) - PAGE_SIZE;
809         }
810
811         ptr = qp;
812         error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
813                                        nbytes - overflow);
814         /* If there's an overflow, write the remaining bytes to the next page */
815         if (!error && pg_oflow)
816                 error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
817                                                ptr + nbytes - overflow,
818                                                overflow);
819         return error;
820 }
821
822 /**
823  * gfs2_adjust_quota - adjust record of current block usage
824  * @ip: The quota inode
825  * @loc: Offset of the entry in the quota file
826  * @change: The amount of usage change to record
827  * @qd: The quota data
828  * @fdq: The updated limits to record
829  *
830  * This function was mostly borrowed from gfs2_block_truncate_page which was
831  * in turn mostly borrowed from ext3
832  *
833  * Returns: 0 or -ve on error
834  */
835
836 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
837                              s64 change, struct gfs2_quota_data *qd,
838                              struct qc_dqblk *fdq)
839 {
840         struct inode *inode = &ip->i_inode;
841         struct gfs2_sbd *sdp = GFS2_SB(inode);
842         struct gfs2_quota q;
843         int err;
844         u64 size;
845
846         if (gfs2_is_stuffed(ip)) {
847                 err = gfs2_unstuff_dinode(ip);
848                 if (err)
849                         return err;
850         }
851
852         memset(&q, 0, sizeof(struct gfs2_quota));
853         err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
854         if (err < 0)
855                 return err;
856
857         loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
858         err = -EIO;
859         be64_add_cpu(&q.qu_value, change);
860         if (((s64)be64_to_cpu(q.qu_value)) < 0)
861                 q.qu_value = 0; /* Never go negative on quota usage */
862         qd->qd_qb.qb_value = q.qu_value;
863         if (fdq) {
864                 if (fdq->d_fieldmask & QC_SPC_SOFT) {
865                         q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
866                         qd->qd_qb.qb_warn = q.qu_warn;
867                 }
868                 if (fdq->d_fieldmask & QC_SPC_HARD) {
869                         q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
870                         qd->qd_qb.qb_limit = q.qu_limit;
871                 }
872                 if (fdq->d_fieldmask & QC_SPACE) {
873                         q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
874                         qd->qd_qb.qb_value = q.qu_value;
875                 }
876         }
877
878         err = gfs2_write_disk_quota(ip, &q, loc);
879         if (!err) {
880                 size = loc + sizeof(struct gfs2_quota);
881                 if (size > inode->i_size)
882                         i_size_write(inode, size);
883                 inode->i_mtime = inode->i_atime = current_time(inode);
884                 mark_inode_dirty(inode);
885                 set_bit(QDF_REFRESH, &qd->qd_flags);
886         }
887
888         return err;
889 }
890
891 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
892 {
893         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
894         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
895         struct gfs2_alloc_parms ap = { .aflags = 0, };
896         unsigned int data_blocks, ind_blocks;
897         struct gfs2_holder *ghs, i_gh;
898         unsigned int qx, x;
899         struct gfs2_quota_data *qd;
900         unsigned reserved;
901         loff_t offset;
902         unsigned int nalloc = 0, blocks;
903         int error;
904
905         error = gfs2_qa_get(ip);
906         if (error)
907                 return error;
908
909         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
910                               &data_blocks, &ind_blocks);
911
912         ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
913         if (!ghs) {
914                 error = -ENOMEM;
915                 goto out;
916         }
917
918         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
919         inode_lock(&ip->i_inode);
920         for (qx = 0; qx < num_qd; qx++) {
921                 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
922                                            GL_NOCACHE, &ghs[qx]);
923                 if (error)
924                         goto out_dq;
925         }
926
927         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
928         if (error)
929                 goto out_dq;
930
931         for (x = 0; x < num_qd; x++) {
932                 offset = qd2offset(qda[x]);
933                 if (gfs2_write_alloc_required(ip, offset,
934                                               sizeof(struct gfs2_quota)))
935                         nalloc++;
936         }
937
938         /* 
939          * 1 blk for unstuffing inode if stuffed. We add this extra
940          * block to the reservation unconditionally. If the inode
941          * doesn't need unstuffing, the block will be released to the 
942          * rgrp since it won't be allocated during the transaction
943          */
944         /* +3 in the end for unstuffing block, inode size update block
945          * and another block in case quota straddles page boundary and 
946          * two blocks need to be updated instead of 1 */
947         blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
948
949         reserved = 1 + (nalloc * (data_blocks + ind_blocks));
950         ap.target = reserved;
951         error = gfs2_inplace_reserve(ip, &ap);
952         if (error)
953                 goto out_alloc;
954
955         if (nalloc)
956                 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
957
958         error = gfs2_trans_begin(sdp, blocks, 0);
959         if (error)
960                 goto out_ipres;
961
962         for (x = 0; x < num_qd; x++) {
963                 qd = qda[x];
964                 offset = qd2offset(qd);
965                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
966                 if (error)
967                         goto out_end_trans;
968
969                 do_qc(qd, -qd->qd_change_sync);
970                 set_bit(QDF_REFRESH, &qd->qd_flags);
971         }
972
973         error = 0;
974
975 out_end_trans:
976         gfs2_trans_end(sdp);
977 out_ipres:
978         gfs2_inplace_release(ip);
979 out_alloc:
980         gfs2_glock_dq_uninit(&i_gh);
981 out_dq:
982         while (qx--)
983                 gfs2_glock_dq_uninit(&ghs[qx]);
984         inode_unlock(&ip->i_inode);
985         kfree(ghs);
986         gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
987                        GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
988 out:
989         gfs2_qa_put(ip);
990         return error;
991 }
992
993 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
994 {
995         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
996         struct gfs2_quota q;
997         struct gfs2_quota_lvb *qlvb;
998         loff_t pos;
999         int error;
1000
1001         memset(&q, 0, sizeof(struct gfs2_quota));
1002         pos = qd2offset(qd);
1003         error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
1004         if (error < 0)
1005                 return error;
1006
1007         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1008         qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1009         qlvb->__pad = 0;
1010         qlvb->qb_limit = q.qu_limit;
1011         qlvb->qb_warn = q.qu_warn;
1012         qlvb->qb_value = q.qu_value;
1013         qd->qd_qb = *qlvb;
1014
1015         return 0;
1016 }
1017
1018 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1019                     struct gfs2_holder *q_gh)
1020 {
1021         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1022         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1023         struct gfs2_holder i_gh;
1024         int error;
1025
1026 restart:
1027         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1028         if (error)
1029                 return error;
1030
1031         if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1032                 force_refresh = FORCE;
1033
1034         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1035
1036         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1037                 gfs2_glock_dq_uninit(q_gh);
1038                 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1039                                            GL_NOCACHE, q_gh);
1040                 if (error)
1041                         return error;
1042
1043                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1044                 if (error)
1045                         goto fail;
1046
1047                 error = update_qd(sdp, qd);
1048                 if (error)
1049                         goto fail_gunlock;
1050
1051                 gfs2_glock_dq_uninit(&i_gh);
1052                 gfs2_glock_dq_uninit(q_gh);
1053                 force_refresh = 0;
1054                 goto restart;
1055         }
1056
1057         return 0;
1058
1059 fail_gunlock:
1060         gfs2_glock_dq_uninit(&i_gh);
1061 fail:
1062         gfs2_glock_dq_uninit(q_gh);
1063         return error;
1064 }
1065
1066 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1067 {
1068         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1069         struct gfs2_quota_data *qd;
1070         u32 x;
1071         int error = 0;
1072
1073         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1074                 return 0;
1075
1076         error = gfs2_quota_hold(ip, uid, gid);
1077         if (error)
1078                 return error;
1079
1080         sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1081              sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1082
1083         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1084                 qd = ip->i_qadata->qa_qd[x];
1085                 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1086                 if (error)
1087                         break;
1088         }
1089
1090         if (!error)
1091                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1092         else {
1093                 while (x--)
1094                         gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1095                 gfs2_quota_unhold(ip);
1096         }
1097
1098         return error;
1099 }
1100
1101 static int need_sync(struct gfs2_quota_data *qd)
1102 {
1103         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1104         struct gfs2_tune *gt = &sdp->sd_tune;
1105         s64 value;
1106         unsigned int num, den;
1107         int do_sync = 1;
1108
1109         if (!qd->qd_qb.qb_limit)
1110                 return 0;
1111
1112         spin_lock(&qd_lock);
1113         value = qd->qd_change;
1114         spin_unlock(&qd_lock);
1115
1116         spin_lock(&gt->gt_spin);
1117         num = gt->gt_quota_scale_num;
1118         den = gt->gt_quota_scale_den;
1119         spin_unlock(&gt->gt_spin);
1120
1121         if (value < 0)
1122                 do_sync = 0;
1123         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1124                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1125                 do_sync = 0;
1126         else {
1127                 value *= gfs2_jindex_size(sdp) * num;
1128                 value = div_s64(value, den);
1129                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1130                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1131                         do_sync = 0;
1132         }
1133
1134         return do_sync;
1135 }
1136
1137 void gfs2_quota_unlock(struct gfs2_inode *ip)
1138 {
1139         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1140         struct gfs2_quota_data *qda[4];
1141         unsigned int count = 0;
1142         u32 x;
1143         int found;
1144
1145         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1146                 return;
1147
1148         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1149                 struct gfs2_quota_data *qd;
1150                 int sync;
1151
1152                 qd = ip->i_qadata->qa_qd[x];
1153                 sync = need_sync(qd);
1154
1155                 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1156                 if (!sync)
1157                         continue;
1158
1159                 spin_lock(&qd_lock);
1160                 found = qd_check_sync(sdp, qd, NULL);
1161                 spin_unlock(&qd_lock);
1162
1163                 if (!found)
1164                         continue;
1165
1166                 gfs2_assert_warn(sdp, qd->qd_change_sync);
1167                 if (bh_get(qd)) {
1168                         clear_bit(QDF_LOCKED, &qd->qd_flags);
1169                         slot_put(qd);
1170                         qd_put(qd);
1171                         continue;
1172                 }
1173
1174                 qda[count++] = qd;
1175         }
1176
1177         if (count) {
1178                 do_sync(count, qda);
1179                 for (x = 0; x < count; x++)
1180                         qd_unlock(qda[x]);
1181         }
1182
1183         gfs2_quota_unhold(ip);
1184 }
1185
1186 #define MAX_LINE 256
1187
1188 static int print_message(struct gfs2_quota_data *qd, char *type)
1189 {
1190         struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1191
1192         fs_info(sdp, "quota %s for %s %u\n",
1193                 type,
1194                 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1195                 from_kqid(&init_user_ns, qd->qd_id));
1196
1197         return 0;
1198 }
1199
1200 /**
1201  * gfs2_quota_check - check if allocating new blocks will exceed quota
1202  * @ip:  The inode for which this check is being performed
1203  * @uid: The uid to check against
1204  * @gid: The gid to check against
1205  * @ap:  The allocation parameters. ap->target contains the requested
1206  *       blocks. ap->min_target, if set, contains the minimum blks
1207  *       requested.
1208  *
1209  * Returns: 0 on success.
1210  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1211  *                  quota must allow at least min_req blks for success and
1212  *                  ap->allowed is set to the number of blocks allowed
1213  *
1214  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1215  *                  of blocks available.
1216  */
1217 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1218                      struct gfs2_alloc_parms *ap)
1219 {
1220         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1221         struct gfs2_quota_data *qd;
1222         s64 value, warn, limit;
1223         u32 x;
1224         int error = 0;
1225
1226         ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1227         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1228                 return 0;
1229
1230         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1231                 qd = ip->i_qadata->qa_qd[x];
1232
1233                 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1234                       qid_eq(qd->qd_id, make_kqid_gid(gid))))
1235                         continue;
1236
1237                 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1238                 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1239                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1240                 spin_lock(&qd_lock);
1241                 value += qd->qd_change;
1242                 spin_unlock(&qd_lock);
1243
1244                 if (limit > 0 && (limit - value) < ap->allowed)
1245                         ap->allowed = limit - value;
1246                 /* If we can't meet the target */
1247                 if (limit && limit < (value + (s64)ap->target)) {
1248                         /* If no min_target specified or we don't meet
1249                          * min_target, return -EDQUOT */
1250                         if (!ap->min_target || ap->min_target > ap->allowed) {
1251                                 if (!test_and_set_bit(QDF_QMSG_QUIET,
1252                                                       &qd->qd_flags)) {
1253                                         print_message(qd, "exceeded");
1254                                         quota_send_warning(qd->qd_id,
1255                                                            sdp->sd_vfs->s_dev,
1256                                                            QUOTA_NL_BHARDWARN);
1257                                 }
1258                                 error = -EDQUOT;
1259                                 break;
1260                         }
1261                 } else if (warn && warn < value &&
1262                            time_after_eq(jiffies, qd->qd_last_warn +
1263                                          gfs2_tune_get(sdp, gt_quota_warn_period)
1264                                          * HZ)) {
1265                         quota_send_warning(qd->qd_id,
1266                                            sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1267                         error = print_message(qd, "warning");
1268                         qd->qd_last_warn = jiffies;
1269                 }
1270         }
1271         return error;
1272 }
1273
1274 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1275                        kuid_t uid, kgid_t gid)
1276 {
1277         struct gfs2_quota_data *qd;
1278         u32 x;
1279         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1280
1281         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1282             gfs2_assert_warn(sdp, change))
1283                 return;
1284         if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1285                 return;
1286
1287         if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1288                                  ip->i_qadata->qa_ref > 0))
1289                 return;
1290         for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1291                 qd = ip->i_qadata->qa_qd[x];
1292
1293                 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1294                     qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1295                         do_qc(qd, change);
1296                 }
1297         }
1298 }
1299
1300 int gfs2_quota_sync(struct super_block *sb, int type)
1301 {
1302         struct gfs2_sbd *sdp = sb->s_fs_info;
1303         struct gfs2_quota_data **qda;
1304         unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1305         unsigned int num_qd;
1306         unsigned int x;
1307         int error = 0;
1308
1309         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1310         if (!qda)
1311                 return -ENOMEM;
1312
1313         mutex_lock(&sdp->sd_quota_sync_mutex);
1314         sdp->sd_quota_sync_gen++;
1315
1316         do {
1317                 num_qd = 0;
1318
1319                 for (;;) {
1320                         error = qd_fish(sdp, qda + num_qd);
1321                         if (error || !qda[num_qd])
1322                                 break;
1323                         if (++num_qd == max_qd)
1324                                 break;
1325                 }
1326
1327                 if (num_qd) {
1328                         if (!error)
1329                                 error = do_sync(num_qd, qda);
1330                         if (!error)
1331                                 for (x = 0; x < num_qd; x++)
1332                                         qda[x]->qd_sync_gen =
1333                                                 sdp->sd_quota_sync_gen;
1334
1335                         for (x = 0; x < num_qd; x++)
1336                                 qd_unlock(qda[x]);
1337                 }
1338         } while (!error && num_qd == max_qd);
1339
1340         mutex_unlock(&sdp->sd_quota_sync_mutex);
1341         kfree(qda);
1342
1343         return error;
1344 }
1345
1346 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1347 {
1348         struct gfs2_quota_data *qd;
1349         struct gfs2_holder q_gh;
1350         int error;
1351
1352         error = qd_get(sdp, qid, &qd);
1353         if (error)
1354                 return error;
1355
1356         error = do_glock(qd, FORCE, &q_gh);
1357         if (!error)
1358                 gfs2_glock_dq_uninit(&q_gh);
1359
1360         qd_put(qd);
1361         return error;
1362 }
1363
1364 int gfs2_quota_init(struct gfs2_sbd *sdp)
1365 {
1366         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1367         u64 size = i_size_read(sdp->sd_qc_inode);
1368         unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1369         unsigned int x, slot = 0;
1370         unsigned int found = 0;
1371         unsigned int hash;
1372         unsigned int bm_size;
1373         u64 dblock;
1374         u32 extlen = 0;
1375         int error;
1376
1377         if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1378                 return -EIO;
1379
1380         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1381         bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1382         bm_size *= sizeof(unsigned long);
1383         error = -ENOMEM;
1384         sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1385         if (sdp->sd_quota_bitmap == NULL)
1386                 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1387                                                  __GFP_ZERO);
1388         if (!sdp->sd_quota_bitmap)
1389                 return error;
1390
1391         for (x = 0; x < blocks; x++) {
1392                 struct buffer_head *bh;
1393                 const struct gfs2_quota_change *qc;
1394                 unsigned int y;
1395
1396                 if (!extlen) {
1397                         extlen = 32;
1398                         error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1399                         if (error)
1400                                 goto fail;
1401                 }
1402                 error = -EIO;
1403                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1404                 if (!bh)
1405                         goto fail;
1406                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1407                         brelse(bh);
1408                         goto fail;
1409                 }
1410
1411                 qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1412                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1413                      y++, slot++) {
1414                         struct gfs2_quota_data *qd;
1415                         s64 qc_change = be64_to_cpu(qc->qc_change);
1416                         u32 qc_flags = be32_to_cpu(qc->qc_flags);
1417                         enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1418                                                 USRQUOTA : GRPQUOTA;
1419                         struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1420                                                       be32_to_cpu(qc->qc_id));
1421                         qc++;
1422                         if (!qc_change)
1423                                 continue;
1424
1425                         hash = gfs2_qd_hash(sdp, qc_id);
1426                         qd = qd_alloc(hash, sdp, qc_id);
1427                         if (qd == NULL) {
1428                                 brelse(bh);
1429                                 goto fail;
1430                         }
1431
1432                         set_bit(QDF_CHANGE, &qd->qd_flags);
1433                         qd->qd_change = qc_change;
1434                         qd->qd_slot = slot;
1435                         qd->qd_slot_count = 1;
1436
1437                         spin_lock(&qd_lock);
1438                         BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1439                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1440                         atomic_inc(&sdp->sd_quota_count);
1441                         spin_unlock(&qd_lock);
1442
1443                         spin_lock_bucket(hash);
1444                         hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1445                         spin_unlock_bucket(hash);
1446
1447                         found++;
1448                 }
1449
1450                 brelse(bh);
1451                 dblock++;
1452                 extlen--;
1453         }
1454
1455         if (found)
1456                 fs_info(sdp, "found %u quota changes\n", found);
1457
1458         return 0;
1459
1460 fail:
1461         gfs2_quota_cleanup(sdp);
1462         return error;
1463 }
1464
1465 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1466 {
1467         struct list_head *head = &sdp->sd_quota_list;
1468         struct gfs2_quota_data *qd;
1469
1470         spin_lock(&qd_lock);
1471         while (!list_empty(head)) {
1472                 qd = list_last_entry(head, struct gfs2_quota_data, qd_list);
1473
1474                 list_del(&qd->qd_list);
1475
1476                 /* Also remove if this qd exists in the reclaim list */
1477                 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1478                 atomic_dec(&sdp->sd_quota_count);
1479                 spin_unlock(&qd_lock);
1480
1481                 spin_lock_bucket(qd->qd_hash);
1482                 hlist_bl_del_rcu(&qd->qd_hlist);
1483                 spin_unlock_bucket(qd->qd_hash);
1484
1485                 gfs2_assert_warn(sdp, !qd->qd_change);
1486                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1487                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1488
1489                 gfs2_glock_put(qd->qd_gl);
1490                 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1491
1492                 spin_lock(&qd_lock);
1493         }
1494         spin_unlock(&qd_lock);
1495
1496         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_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         DEFINE_WAIT(wait);
1549
1550         while (!kthread_should_stop()) {
1551
1552                 if (gfs2_withdrawn(sdp))
1553                         goto bypass;
1554                 /* Update the master statfs file */
1555                 if (sdp->sd_statfs_force_sync) {
1556                         int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1557                         quotad_error(sdp, "statfs", error);
1558                         statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1559                 }
1560                 else
1561                         quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1562                                            &statfs_timeo,
1563                                            &tune->gt_statfs_quantum);
1564
1565                 /* Update quota file */
1566                 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1567                                    &quotad_timeo, &tune->gt_quota_quantum);
1568
1569                 try_to_freeze();
1570
1571 bypass:
1572                 t = min(quotad_timeo, statfs_timeo);
1573
1574                 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1575                 if (!sdp->sd_statfs_force_sync)
1576                         t -= schedule_timeout(t);
1577                 else
1578                         t = 0;
1579                 finish_wait(&sdp->sd_quota_wait, &wait);
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_ON:
1593                 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1594                 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1595                 fallthrough;
1596         case GFS2_QUOTA_ACCOUNT:
1597                 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1598                                                   QCI_SYSFILE;
1599                 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1600                                                   QCI_SYSFILE;
1601                 break;
1602         case GFS2_QUOTA_OFF:
1603                 break;
1604         }
1605         if (sdp->sd_quota_inode) {
1606                 state->s_state[USRQUOTA].ino =
1607                                         GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1608                 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1609         }
1610         state->s_state[USRQUOTA].nextents = 1;  /* unsupported */
1611         state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1612         state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1613         return 0;
1614 }
1615
1616 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1617                           struct qc_dqblk *fdq)
1618 {
1619         struct gfs2_sbd *sdp = sb->s_fs_info;
1620         struct gfs2_quota_lvb *qlvb;
1621         struct gfs2_quota_data *qd;
1622         struct gfs2_holder q_gh;
1623         int error;
1624
1625         memset(fdq, 0, sizeof(*fdq));
1626
1627         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1628                 return -ESRCH; /* Crazy XFS error code */
1629
1630         if ((qid.type != USRQUOTA) &&
1631             (qid.type != GRPQUOTA))
1632                 return -EINVAL;
1633
1634         error = qd_get(sdp, qid, &qd);
1635         if (error)
1636                 return error;
1637         error = do_glock(qd, FORCE, &q_gh);
1638         if (error)
1639                 goto out;
1640
1641         qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1642         fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1643         fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1644         fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1645
1646         gfs2_glock_dq_uninit(&q_gh);
1647 out:
1648         qd_put(qd);
1649         return error;
1650 }
1651
1652 /* GFS2 only supports a subset of the XFS fields */
1653 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1654
1655 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1656                           struct qc_dqblk *fdq)
1657 {
1658         struct gfs2_sbd *sdp = sb->s_fs_info;
1659         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1660         struct gfs2_quota_data *qd;
1661         struct gfs2_holder q_gh, i_gh;
1662         unsigned int data_blocks, ind_blocks;
1663         unsigned int blocks = 0;
1664         int alloc_required;
1665         loff_t offset;
1666         int error;
1667
1668         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1669                 return -ESRCH; /* Crazy XFS error code */
1670
1671         if ((qid.type != USRQUOTA) &&
1672             (qid.type != GRPQUOTA))
1673                 return -EINVAL;
1674
1675         if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1676                 return -EINVAL;
1677
1678         error = qd_get(sdp, qid, &qd);
1679         if (error)
1680                 return error;
1681
1682         error = gfs2_qa_get(ip);
1683         if (error)
1684                 goto out_put;
1685
1686         inode_lock(&ip->i_inode);
1687         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1688         if (error)
1689                 goto out_unlockput;
1690         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1691         if (error)
1692                 goto out_q;
1693
1694         /* Check for existing entry, if none then alloc new blocks */
1695         error = update_qd(sdp, qd);
1696         if (error)
1697                 goto out_i;
1698
1699         /* If nothing has changed, this is a no-op */
1700         if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1701             ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1702                 fdq->d_fieldmask ^= QC_SPC_SOFT;
1703
1704         if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1705             ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1706                 fdq->d_fieldmask ^= QC_SPC_HARD;
1707
1708         if ((fdq->d_fieldmask & QC_SPACE) &&
1709             ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1710                 fdq->d_fieldmask ^= QC_SPACE;
1711
1712         if (fdq->d_fieldmask == 0)
1713                 goto out_i;
1714
1715         offset = qd2offset(qd);
1716         alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1717         if (gfs2_is_stuffed(ip))
1718                 alloc_required = 1;
1719         if (alloc_required) {
1720                 struct gfs2_alloc_parms ap = { .aflags = 0, };
1721                 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1722                                        &data_blocks, &ind_blocks);
1723                 blocks = 1 + data_blocks + ind_blocks;
1724                 ap.target = blocks;
1725                 error = gfs2_inplace_reserve(ip, &ap);
1726                 if (error)
1727                         goto out_i;
1728                 blocks += gfs2_rg_blocks(ip, blocks);
1729         }
1730
1731         /* Some quotas span block boundaries and can update two blocks,
1732            adding an extra block to the transaction to handle such quotas */
1733         error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1734         if (error)
1735                 goto out_release;
1736
1737         /* Apply changes */
1738         error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1739         if (!error)
1740                 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1741
1742         gfs2_trans_end(sdp);
1743 out_release:
1744         if (alloc_required)
1745                 gfs2_inplace_release(ip);
1746 out_i:
1747         gfs2_glock_dq_uninit(&i_gh);
1748 out_q:
1749         gfs2_glock_dq_uninit(&q_gh);
1750 out_unlockput:
1751         gfs2_qa_put(ip);
1752         inode_unlock(&ip->i_inode);
1753 out_put:
1754         qd_put(qd);
1755         return error;
1756 }
1757
1758 const struct quotactl_ops gfs2_quotactl_ops = {
1759         .quota_sync     = gfs2_quota_sync,
1760         .get_state      = gfs2_quota_get_state,
1761         .get_dqblk      = gfs2_get_dqblk,
1762         .set_dqblk      = gfs2_set_dqblk,
1763 };
1764
1765 void __init gfs2_quota_hash_init(void)
1766 {
1767         unsigned i;
1768
1769         for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1770                 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1771 }