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                              */
83static DEFINE_SPINLOCK(qd_lock);
84struct list_lru gfs2_qd_lru;
85
86static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
87
88static 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
99static inline void spin_lock_bucket(unsigned int hash)
100{
101        hlist_bl_lock(&qd_hash_table[hash]);
102}
103
104static inline void spin_unlock_bucket(unsigned int hash)
105{
106        hlist_bl_unlock(&qd_hash_table[hash]);
107}
108
109static 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
119static 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_withdrawing_or_withdrawn(sdp)) {
132		gfs2_assert_warn(sdp, !qd->qd_change);
133		gfs2_assert_warn(sdp, !qd->qd_slot_ref);
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
141static 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
154static 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
176static 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
193static 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
199static struct shrinker *gfs2_qd_shrinker;
200
201int __init gfs2_qd_shrinker_init(void)
202{
203	gfs2_qd_shrinker = shrinker_alloc(SHRINKER_NUMA_AWARE, "gfs2-qd");
204	if (!gfs2_qd_shrinker)
205		return -ENOMEM;
206
207	gfs2_qd_shrinker->count_objects = gfs2_qd_shrink_count;
208	gfs2_qd_shrinker->scan_objects = gfs2_qd_shrink_scan;
209
210	shrinker_register(gfs2_qd_shrinker);
211
212	return 0;
213}
214
215void gfs2_qd_shrinker_exit(void)
216{
217	shrinker_free(gfs2_qd_shrinker);
218}
219
220static u64 qd2index(struct gfs2_quota_data *qd)
221{
222	struct kqid qid = qd->qd_id;
223	return (2 * (u64)from_kqid(&init_user_ns, qid)) +
224		((qid.type == USRQUOTA) ? 0 : 1);
225}
226
227static u64 qd2offset(struct gfs2_quota_data *qd)
228{
229	return qd2index(qd) * sizeof(struct gfs2_quota);
230}
231
232static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
233{
234	struct gfs2_quota_data *qd;
235	int error;
236
237	qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
238	if (!qd)
239		return NULL;
240
241	qd->qd_sbd = sdp;
242	qd->qd_lockref.count = 0;
243	spin_lock_init(&qd->qd_lockref.lock);
244	qd->qd_id = qid;
245	qd->qd_slot = -1;
246	INIT_LIST_HEAD(&qd->qd_lru);
247	qd->qd_hash = hash;
248
249	error = gfs2_glock_get(sdp, qd2index(qd),
250			      &gfs2_quota_glops, CREATE, &qd->qd_gl);
251	if (error)
252		goto fail;
253
254	return qd;
255
256fail:
257	kmem_cache_free(gfs2_quotad_cachep, qd);
258	return NULL;
259}
260
261static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
262						     const struct gfs2_sbd *sdp,
263						     struct kqid qid)
264{
265	struct gfs2_quota_data *qd;
266	struct hlist_bl_node *h;
267
268	hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
269		if (!qid_eq(qd->qd_id, qid))
270			continue;
271		if (qd->qd_sbd != sdp)
272			continue;
273		if (lockref_get_not_dead(&qd->qd_lockref)) {
274			list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru);
275			return qd;
276		}
277	}
278
279	return NULL;
280}
281
282
283static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
284		  struct gfs2_quota_data **qdp)
285{
286	struct gfs2_quota_data *qd, *new_qd;
287	unsigned int hash = gfs2_qd_hash(sdp, qid);
288
289	rcu_read_lock();
290	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
291	rcu_read_unlock();
292
293	if (qd)
294		return 0;
295
296	new_qd = qd_alloc(hash, sdp, qid);
297	if (!new_qd)
298		return -ENOMEM;
299
300	spin_lock(&qd_lock);
301	spin_lock_bucket(hash);
302	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
303	if (qd == NULL) {
304		new_qd->qd_lockref.count++;
305		*qdp = new_qd;
306		list_add(&new_qd->qd_list, &sdp->sd_quota_list);
307		hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
308		atomic_inc(&sdp->sd_quota_count);
309	}
310	spin_unlock_bucket(hash);
311	spin_unlock(&qd_lock);
312
313	if (qd) {
314		gfs2_glock_put(new_qd->qd_gl);
315		kmem_cache_free(gfs2_quotad_cachep, new_qd);
316	}
317
318	return 0;
319}
320
321
322static void qd_hold(struct gfs2_quota_data *qd)
323{
324	struct gfs2_sbd *sdp = qd->qd_sbd;
325	gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
326	lockref_get(&qd->qd_lockref);
327}
328
329static void qd_put(struct gfs2_quota_data *qd)
330{
331	struct gfs2_sbd *sdp;
332
333	if (lockref_put_or_lock(&qd->qd_lockref))
334		return;
335
336	BUG_ON(__lockref_is_dead(&qd->qd_lockref));
337	sdp = qd->qd_sbd;
338	if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
339		lockref_mark_dead(&qd->qd_lockref);
340		spin_unlock(&qd->qd_lockref.lock);
341
342		gfs2_qd_dispose(qd);
343		return;
344	}
345
346	qd->qd_lockref.count = 0;
347	list_lru_add_obj(&gfs2_qd_lru, &qd->qd_lru);
348	spin_unlock(&qd->qd_lockref.lock);
349}
350
351static int slot_get(struct gfs2_quota_data *qd)
352{
353	struct gfs2_sbd *sdp = qd->qd_sbd;
354	unsigned int bit;
355	int error = 0;
356
357	spin_lock(&sdp->sd_bitmap_lock);
358	if (qd->qd_slot_ref == 0) {
359		bit = find_first_zero_bit(sdp->sd_quota_bitmap,
360					  sdp->sd_quota_slots);
361		if (bit >= sdp->sd_quota_slots) {
362			error = -ENOSPC;
363			goto out;
364		}
365		set_bit(bit, sdp->sd_quota_bitmap);
366		qd->qd_slot = bit;
367	}
368	qd->qd_slot_ref++;
369out:
370	spin_unlock(&sdp->sd_bitmap_lock);
371	return error;
372}
373
374static void slot_hold(struct gfs2_quota_data *qd)
375{
376	struct gfs2_sbd *sdp = qd->qd_sbd;
377
378	spin_lock(&sdp->sd_bitmap_lock);
379	gfs2_assert(sdp, qd->qd_slot_ref);
380	qd->qd_slot_ref++;
381	spin_unlock(&sdp->sd_bitmap_lock);
382}
383
384static void slot_put(struct gfs2_quota_data *qd)
385{
386	struct gfs2_sbd *sdp = qd->qd_sbd;
387
388	spin_lock(&sdp->sd_bitmap_lock);
389	gfs2_assert(sdp, qd->qd_slot_ref);
390	if (!--qd->qd_slot_ref) {
391		BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
392		qd->qd_slot = -1;
393	}
394	spin_unlock(&sdp->sd_bitmap_lock);
395}
396
397static int bh_get(struct gfs2_quota_data *qd)
398{
399	struct gfs2_sbd *sdp = qd->qd_sbd;
400	struct inode *inode = sdp->sd_qc_inode;
401	struct gfs2_inode *ip = GFS2_I(inode);
402	unsigned int block, offset;
403	struct buffer_head *bh;
404	struct iomap iomap = { };
405	int error;
406
407	mutex_lock(&sdp->sd_quota_mutex);
408
409	if (qd->qd_bh_count++) {
410		mutex_unlock(&sdp->sd_quota_mutex);
411		return 0;
412	}
413
414	block = qd->qd_slot / sdp->sd_qc_per_block;
415	offset = qd->qd_slot % sdp->sd_qc_per_block;
416
417	error = gfs2_iomap_get(inode,
418			       (loff_t)block << inode->i_blkbits,
419			       i_blocksize(inode), &iomap);
420	if (error)
421		goto fail;
422	error = -ENOENT;
423	if (iomap.type != IOMAP_MAPPED)
424		goto fail;
425
426	error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
427			       DIO_WAIT, 0, &bh);
428	if (error)
429		goto fail;
430	error = -EIO;
431	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
432		goto fail_brelse;
433
434	qd->qd_bh = bh;
435	qd->qd_bh_qc = (struct gfs2_quota_change *)
436		(bh->b_data + sizeof(struct gfs2_meta_header) +
437		 offset * sizeof(struct gfs2_quota_change));
438
439	mutex_unlock(&sdp->sd_quota_mutex);
440
441	return 0;
442
443fail_brelse:
444	brelse(bh);
445fail:
446	qd->qd_bh_count--;
447	mutex_unlock(&sdp->sd_quota_mutex);
448	return error;
449}
450
451static void bh_put(struct gfs2_quota_data *qd)
452{
453	struct gfs2_sbd *sdp = qd->qd_sbd;
454
455	mutex_lock(&sdp->sd_quota_mutex);
456	gfs2_assert(sdp, qd->qd_bh_count);
457	if (!--qd->qd_bh_count) {
458		brelse(qd->qd_bh);
459		qd->qd_bh = NULL;
460		qd->qd_bh_qc = NULL;
461	}
462	mutex_unlock(&sdp->sd_quota_mutex);
463}
464
465static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
466			 u64 *sync_gen)
467{
468	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
469	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
470	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
471		return 0;
472
473	/*
474	 * If qd_change is 0 it means a pending quota change was negated.
475	 * We should not sync it, but we still have a qd reference and slot
476	 * reference taken by gfs2_quota_change -> do_qc that need to be put.
477	 */
478	if (!qd->qd_change && test_and_clear_bit(QDF_CHANGE, &qd->qd_flags)) {
479		slot_put(qd);
480		qd_put(qd);
481		return 0;
482	}
483
484	if (!lockref_get_not_dead(&qd->qd_lockref))
485		return 0;
486
487	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
488	set_bit(QDF_LOCKED, &qd->qd_flags);
489	qd->qd_change_sync = qd->qd_change;
490	slot_hold(qd);
491	return 1;
492}
493
494static int qd_bh_get_or_undo(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
495{
496	int error;
497
498	error = bh_get(qd);
499	if (!error)
500		return 0;
501
502	clear_bit(QDF_LOCKED, &qd->qd_flags);
503	slot_put(qd);
504	qd_put(qd);
505	return error;
506}
507
508static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
509{
510	struct gfs2_quota_data *qd = NULL, *iter;
511	int error;
512
513	*qdp = NULL;
514
515	if (sb_rdonly(sdp->sd_vfs))
516		return 0;
517
518	spin_lock(&qd_lock);
519
520	list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
521		if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
522			qd = iter;
523			break;
524		}
525	}
526
527	spin_unlock(&qd_lock);
528
529	if (qd) {
530		error = qd_bh_get_or_undo(sdp, qd);
531		if (error)
532			return error;
533		*qdp = qd;
534	}
535
536	return 0;
537}
538
539static void qdsb_put(struct gfs2_quota_data *qd)
540{
541	bh_put(qd);
542	slot_put(qd);
543	qd_put(qd);
544}
545
546static void qd_unlock(struct gfs2_quota_data *qd)
547{
548	gfs2_assert_warn(qd->qd_sbd, test_bit(QDF_LOCKED, &qd->qd_flags));
549	clear_bit(QDF_LOCKED, &qd->qd_flags);
550	qdsb_put(qd);
551}
552
553static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
554		    struct gfs2_quota_data **qdp)
555{
556	int error;
557
558	error = qd_get(sdp, qid, qdp);
559	if (error)
560		return error;
561
562	error = slot_get(*qdp);
563	if (error)
564		goto fail;
565
566	error = bh_get(*qdp);
567	if (error)
568		goto fail_slot;
569
570	return 0;
571
572fail_slot:
573	slot_put(*qdp);
574fail:
575	qd_put(*qdp);
576	return error;
577}
578
579/**
580 * gfs2_qa_get - make sure we have a quota allocations data structure,
581 *               if necessary
582 * @ip: the inode for this reservation
583 */
584int gfs2_qa_get(struct gfs2_inode *ip)
585{
586	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
587	struct inode *inode = &ip->i_inode;
588
589	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
590		return 0;
591
592	spin_lock(&inode->i_lock);
593	if (ip->i_qadata == NULL) {
594		struct gfs2_qadata *tmp;
595
596		spin_unlock(&inode->i_lock);
597		tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
598		if (!tmp)
599			return -ENOMEM;
600
601		spin_lock(&inode->i_lock);
602		if (ip->i_qadata == NULL)
603			ip->i_qadata = tmp;
604		else
605			kmem_cache_free(gfs2_qadata_cachep, tmp);
606	}
607	ip->i_qadata->qa_ref++;
608	spin_unlock(&inode->i_lock);
609	return 0;
610}
611
612void gfs2_qa_put(struct gfs2_inode *ip)
613{
614	struct inode *inode = &ip->i_inode;
615
616	spin_lock(&inode->i_lock);
617	if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
618		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
619		ip->i_qadata = NULL;
620	}
621	spin_unlock(&inode->i_lock);
622}
623
624int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
625{
626	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
627	struct gfs2_quota_data **qd;
628	int error;
629
630	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
631		return 0;
632
633	error = gfs2_qa_get(ip);
634	if (error)
635		return error;
636
637	qd = ip->i_qadata->qa_qd;
638
639	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
640	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
641		error = -EIO;
642		gfs2_qa_put(ip);
643		goto out;
644	}
645
646	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
647	if (error)
648		goto out_unhold;
649	ip->i_qadata->qa_qd_num++;
650	qd++;
651
652	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
653	if (error)
654		goto out_unhold;
655	ip->i_qadata->qa_qd_num++;
656	qd++;
657
658	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
659	    !uid_eq(uid, ip->i_inode.i_uid)) {
660		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
661		if (error)
662			goto out_unhold;
663		ip->i_qadata->qa_qd_num++;
664		qd++;
665	}
666
667	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
668	    !gid_eq(gid, ip->i_inode.i_gid)) {
669		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
670		if (error)
671			goto out_unhold;
672		ip->i_qadata->qa_qd_num++;
673		qd++;
674	}
675
676out_unhold:
677	if (error)
678		gfs2_quota_unhold(ip);
679out:
680	return error;
681}
682
683void gfs2_quota_unhold(struct gfs2_inode *ip)
684{
685	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
686	u32 x;
687
688	if (ip->i_qadata == NULL)
689		return;
690
691	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
692
693	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
694		qdsb_put(ip->i_qadata->qa_qd[x]);
695		ip->i_qadata->qa_qd[x] = NULL;
696	}
697	ip->i_qadata->qa_qd_num = 0;
698	gfs2_qa_put(ip);
699}
700
701static int sort_qd(const void *a, const void *b)
702{
703	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
704	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
705
706	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
707		return -1;
708	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
709		return 1;
710	return 0;
711}
712
713static void do_qc(struct gfs2_quota_data *qd, s64 change, int qc_type)
714{
715	struct gfs2_sbd *sdp = qd->qd_sbd;
716	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
717	struct gfs2_quota_change *qc = qd->qd_bh_qc;
718	s64 x;
719
720	mutex_lock(&sdp->sd_quota_mutex);
721	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
722
723	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
724		qc->qc_change = 0;
725		qc->qc_flags = 0;
726		if (qd->qd_id.type == USRQUOTA)
727			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
728		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
729	}
730
731	x = be64_to_cpu(qc->qc_change) + change;
732	qc->qc_change = cpu_to_be64(x);
733
734	spin_lock(&qd_lock);
735	qd->qd_change = x;
736	spin_unlock(&qd_lock);
737
738	if (qc_type == QC_CHANGE) {
739		if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
740			qd_hold(qd);
741			slot_hold(qd);
742		}
743	} else {
744		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
745		clear_bit(QDF_CHANGE, &qd->qd_flags);
746		qc->qc_flags = 0;
747		qc->qc_id = 0;
748		slot_put(qd);
749		qd_put(qd);
750	}
751
752	if (change < 0) /* Reset quiet flag if we freed some blocks */
753		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
754	mutex_unlock(&sdp->sd_quota_mutex);
755}
756
757static int gfs2_write_buf_to_page(struct gfs2_sbd *sdp, unsigned long index,
758				  unsigned off, void *buf, unsigned bytes)
759{
760	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
761	struct inode *inode = &ip->i_inode;
762	struct address_space *mapping = inode->i_mapping;
763	struct folio *folio;
764	struct buffer_head *bh;
765	u64 blk;
766	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
767	unsigned to_write = bytes, pg_off = off;
768
769	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
770	boff = off % bsize;
771
772	folio = filemap_grab_folio(mapping, index);
773	if (IS_ERR(folio))
774		return PTR_ERR(folio);
775	bh = folio_buffers(folio);
776	if (!bh)
777		bh = create_empty_buffers(folio, bsize, 0);
778
779	for (;;) {
780		/* Find the beginning block within the folio */
781		if (pg_off >= ((bnum * bsize) + bsize)) {
782			bh = bh->b_this_page;
783			bnum++;
784			blk++;
785			continue;
786		}
787		if (!buffer_mapped(bh)) {
788			gfs2_block_map(inode, blk, bh, 1);
789			if (!buffer_mapped(bh))
790				goto unlock_out;
791			/* If it's a newly allocated disk block, zero it */
792			if (buffer_new(bh))
793				folio_zero_range(folio, bnum * bsize,
794						bh->b_size);
795		}
796		if (folio_test_uptodate(folio))
797			set_buffer_uptodate(bh);
798		if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
799			goto unlock_out;
800		gfs2_trans_add_data(ip->i_gl, bh);
801
802		/* If we need to write to the next block as well */
803		if (to_write > (bsize - boff)) {
804			pg_off += (bsize - boff);
805			to_write -= (bsize - boff);
806			boff = pg_off % bsize;
807			continue;
808		}
809		break;
810	}
811
812	/* Write to the folio, now that we have setup the buffer(s) */
813	memcpy_to_folio(folio, off, buf, bytes);
814	flush_dcache_folio(folio);
815	folio_unlock(folio);
816	folio_put(folio);
817
818	return 0;
819
820unlock_out:
821	folio_unlock(folio);
822	folio_put(folio);
823	return -EIO;
824}
825
826static int gfs2_write_disk_quota(struct gfs2_sbd *sdp, struct gfs2_quota *qp,
827				 loff_t loc)
828{
829	unsigned long pg_beg;
830	unsigned pg_off, nbytes, overflow = 0;
831	int error;
832	void *ptr;
833
834	nbytes = sizeof(struct gfs2_quota);
835
836	pg_beg = loc >> PAGE_SHIFT;
837	pg_off = offset_in_page(loc);
838
839	/* If the quota straddles a page boundary, split the write in two */
840	if ((pg_off + nbytes) > PAGE_SIZE)
841		overflow = (pg_off + nbytes) - PAGE_SIZE;
842
843	ptr = qp;
844	error = gfs2_write_buf_to_page(sdp, pg_beg, pg_off, ptr,
845				       nbytes - overflow);
846	/* If there's an overflow, write the remaining bytes to the next page */
847	if (!error && overflow)
848		error = gfs2_write_buf_to_page(sdp, pg_beg + 1, 0,
849					       ptr + nbytes - overflow,
850					       overflow);
851	return error;
852}
853
854/**
855 * gfs2_adjust_quota - adjust record of current block usage
856 * @sdp: The superblock
857 * @loc: Offset of the entry in the quota file
858 * @change: The amount of usage change to record
859 * @qd: The quota data
860 * @fdq: The updated limits to record
861 *
862 * This function was mostly borrowed from gfs2_block_truncate_page which was
863 * in turn mostly borrowed from ext3
864 *
865 * Returns: 0 or -ve on error
866 */
867
868static int gfs2_adjust_quota(struct gfs2_sbd *sdp, loff_t loc,
869			     s64 change, struct gfs2_quota_data *qd,
870			     struct qc_dqblk *fdq)
871{
872	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
873	struct inode *inode = &ip->i_inode;
874	struct gfs2_quota q;
875	int err;
876	u64 size;
877
878	if (gfs2_is_stuffed(ip)) {
879		err = gfs2_unstuff_dinode(ip);
880		if (err)
881			return err;
882	}
883
884	memset(&q, 0, sizeof(struct gfs2_quota));
885	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
886	if (err < 0)
887		return err;
888
889	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
890	be64_add_cpu(&q.qu_value, change);
891	if (((s64)be64_to_cpu(q.qu_value)) < 0)
892		q.qu_value = 0; /* Never go negative on quota usage */
893	qd->qd_qb.qb_value = q.qu_value;
894	if (fdq) {
895		if (fdq->d_fieldmask & QC_SPC_SOFT) {
896			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
897			qd->qd_qb.qb_warn = q.qu_warn;
898		}
899		if (fdq->d_fieldmask & QC_SPC_HARD) {
900			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
901			qd->qd_qb.qb_limit = q.qu_limit;
902		}
903		if (fdq->d_fieldmask & QC_SPACE) {
904			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
905			qd->qd_qb.qb_value = q.qu_value;
906		}
907	}
908
909	err = gfs2_write_disk_quota(sdp, &q, loc);
910	if (!err) {
911		size = loc + sizeof(struct gfs2_quota);
912		if (size > inode->i_size)
913			i_size_write(inode, size);
914		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
915		mark_inode_dirty(inode);
916		set_bit(QDF_REFRESH, &qd->qd_flags);
917	}
918
919	return err;
920}
921
922static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
923{
924	struct gfs2_sbd *sdp = (*qda)->qd_sbd;
925	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
926	struct gfs2_alloc_parms ap = {};
927	unsigned int data_blocks, ind_blocks;
928	struct gfs2_holder *ghs, i_gh;
929	unsigned int qx, x;
930	struct gfs2_quota_data *qd;
931	unsigned reserved;
932	loff_t offset;
933	unsigned int nalloc = 0, blocks;
934	int error;
935
936	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
937			      &data_blocks, &ind_blocks);
938
939	ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
940	if (!ghs)
941		return -ENOMEM;
942
943	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
944	inode_lock(&ip->i_inode);
945	for (qx = 0; qx < num_qd; qx++) {
946		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
947					   GL_NOCACHE, &ghs[qx]);
948		if (error)
949			goto out_dq;
950	}
951
952	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
953	if (error)
954		goto out_dq;
955
956	for (x = 0; x < num_qd; x++) {
957		offset = qd2offset(qda[x]);
958		if (gfs2_write_alloc_required(ip, offset,
959					      sizeof(struct gfs2_quota)))
960			nalloc++;
961	}
962
963	/*
964	 * 1 blk for unstuffing inode if stuffed. We add this extra
965	 * block to the reservation unconditionally. If the inode
966	 * doesn't need unstuffing, the block will be released to the
967	 * rgrp since it won't be allocated during the transaction
968	 */
969	/* +3 in the end for unstuffing block, inode size update block
970	 * and another block in case quota straddles page boundary and
971	 * two blocks need to be updated instead of 1 */
972	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
973
974	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
975	ap.target = reserved;
976	error = gfs2_inplace_reserve(ip, &ap);
977	if (error)
978		goto out_alloc;
979
980	if (nalloc)
981		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
982
983	error = gfs2_trans_begin(sdp, blocks, 0);
984	if (error)
985		goto out_ipres;
986
987	for (x = 0; x < num_qd; x++) {
988		qd = qda[x];
989		offset = qd2offset(qd);
990		error = gfs2_adjust_quota(sdp, offset, qd->qd_change_sync, qd,
991							NULL);
992		if (error)
993			goto out_end_trans;
994
995		do_qc(qd, -qd->qd_change_sync, QC_SYNC);
996		set_bit(QDF_REFRESH, &qd->qd_flags);
997	}
998
999out_end_trans:
1000	gfs2_trans_end(sdp);
1001out_ipres:
1002	gfs2_inplace_release(ip);
1003out_alloc:
1004	gfs2_glock_dq_uninit(&i_gh);
1005out_dq:
1006	while (qx--)
1007		gfs2_glock_dq_uninit(&ghs[qx]);
1008	inode_unlock(&ip->i_inode);
1009	kfree(ghs);
1010	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
1011		       GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
1012	if (!error) {
1013		for (x = 0; x < num_qd; x++)
1014			qda[x]->qd_sync_gen = sdp->sd_quota_sync_gen;
1015	}
1016	return error;
1017}
1018
1019static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
1020{
1021	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1022	struct gfs2_quota q;
1023	struct gfs2_quota_lvb *qlvb;
1024	loff_t pos;
1025	int error;
1026
1027	memset(&q, 0, sizeof(struct gfs2_quota));
1028	pos = qd2offset(qd);
1029	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
1030	if (error < 0)
1031		return error;
1032
1033	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1034	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1035	qlvb->__pad = 0;
1036	qlvb->qb_limit = q.qu_limit;
1037	qlvb->qb_warn = q.qu_warn;
1038	qlvb->qb_value = q.qu_value;
1039	qd->qd_qb = *qlvb;
1040
1041	return 0;
1042}
1043
1044static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1045		    struct gfs2_holder *q_gh)
1046{
1047	struct gfs2_sbd *sdp = qd->qd_sbd;
1048	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1049	struct gfs2_holder i_gh;
1050	int error;
1051
1052	gfs2_assert_warn(sdp, sdp == qd->qd_gl->gl_name.ln_sbd);
1053restart:
1054	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1055	if (error)
1056		return error;
1057
1058	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1059		force_refresh = FORCE;
1060
1061	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1062
1063	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1064		gfs2_glock_dq_uninit(q_gh);
1065		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1066					   GL_NOCACHE, q_gh);
1067		if (error)
1068			return error;
1069
1070		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1071		if (error)
1072			goto fail;
1073
1074		error = update_qd(sdp, qd);
1075		if (error)
1076			goto fail_gunlock;
1077
1078		gfs2_glock_dq_uninit(&i_gh);
1079		gfs2_glock_dq_uninit(q_gh);
1080		force_refresh = 0;
1081		goto restart;
1082	}
1083
1084	return 0;
1085
1086fail_gunlock:
1087	gfs2_glock_dq_uninit(&i_gh);
1088fail:
1089	gfs2_glock_dq_uninit(q_gh);
1090	return error;
1091}
1092
1093int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1094{
1095	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1096	struct gfs2_quota_data *qd;
1097	u32 x;
1098	int error;
1099
1100	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1101		return 0;
1102
1103	error = gfs2_quota_hold(ip, uid, gid);
1104	if (error)
1105		return error;
1106
1107	sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1108	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1109
1110	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1111		qd = ip->i_qadata->qa_qd[x];
1112		error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1113		if (error)
1114			break;
1115	}
1116
1117	if (!error)
1118		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1119	else {
1120		while (x--)
1121			gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1122		gfs2_quota_unhold(ip);
1123	}
1124
1125	return error;
1126}
1127
1128static bool need_sync(struct gfs2_quota_data *qd)
1129{
1130	struct gfs2_sbd *sdp = qd->qd_sbd;
1131	struct gfs2_tune *gt = &sdp->sd_tune;
1132	s64 value;
1133	unsigned int num, den;
1134
1135	if (!qd->qd_qb.qb_limit)
1136		return false;
1137
1138	spin_lock(&qd_lock);
1139	value = qd->qd_change;
1140	spin_unlock(&qd_lock);
1141
1142	spin_lock(&gt->gt_spin);
1143	num = gt->gt_quota_scale_num;
1144	den = gt->gt_quota_scale_den;
1145	spin_unlock(&gt->gt_spin);
1146
1147	if (value <= 0)
1148		return false;
1149	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1150		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1151		return false;
1152	else {
1153		value *= gfs2_jindex_size(sdp) * num;
1154		value = div_s64(value, den);
1155		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1156		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1157			return false;
1158	}
1159
1160	return true;
1161}
1162
1163void gfs2_quota_unlock(struct gfs2_inode *ip)
1164{
1165	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1166	struct gfs2_quota_data *qda[2 * GFS2_MAXQUOTAS];
1167	unsigned int count = 0;
1168	u32 x;
1169	int found;
1170
1171	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1172		return;
1173
1174	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1175		struct gfs2_quota_data *qd;
1176		bool sync;
1177
1178		qd = ip->i_qadata->qa_qd[x];
1179		sync = need_sync(qd);
1180
1181		gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1182		if (!sync)
1183			continue;
1184
1185		spin_lock(&qd_lock);
1186		found = qd_check_sync(sdp, qd, NULL);
1187		spin_unlock(&qd_lock);
1188
1189		if (!found)
1190			continue;
1191
1192		if (!qd_bh_get_or_undo(sdp, qd))
1193			qda[count++] = qd;
1194	}
1195
1196	if (count) {
1197		do_sync(count, qda);
1198		for (x = 0; x < count; x++)
1199			qd_unlock(qda[x]);
1200	}
1201
1202	gfs2_quota_unhold(ip);
1203}
1204
1205#define MAX_LINE 256
1206
1207static void print_message(struct gfs2_quota_data *qd, char *type)
1208{
1209	struct gfs2_sbd *sdp = qd->qd_sbd;
1210
1211	if (sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET) {
1212		fs_info(sdp, "quota %s for %s %u\n",
1213			type,
1214			(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1215			from_kqid(&init_user_ns, qd->qd_id));
1216	}
1217}
1218
1219/**
1220 * gfs2_quota_check - check if allocating new blocks will exceed quota
1221 * @ip:  The inode for which this check is being performed
1222 * @uid: The uid to check against
1223 * @gid: The gid to check against
1224 * @ap:  The allocation parameters. ap->target contains the requested
1225 *       blocks. ap->min_target, if set, contains the minimum blks
1226 *       requested.
1227 *
1228 * Returns: 0 on success.
1229 *                  min_req = ap->min_target ? ap->min_target : ap->target;
1230 *                  quota must allow at least min_req blks for success and
1231 *                  ap->allowed is set to the number of blocks allowed
1232 *
1233 *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1234 *                  of blocks available.
1235 */
1236int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1237		     struct gfs2_alloc_parms *ap)
1238{
1239	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1240	struct gfs2_quota_data *qd;
1241	s64 value, warn, limit;
1242	u32 x;
1243	int error = 0;
1244
1245	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1246	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1247		return 0;
1248
1249	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1250		qd = ip->i_qadata->qa_qd[x];
1251
1252		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1253		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1254			continue;
1255
1256		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1257		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1258		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1259		spin_lock(&qd_lock);
1260		value += qd->qd_change;
1261		spin_unlock(&qd_lock);
1262
1263		if (limit > 0 && (limit - value) < ap->allowed)
1264			ap->allowed = limit - value;
1265		/* If we can't meet the target */
1266		if (limit && limit < (value + (s64)ap->target)) {
1267			/* If no min_target specified or we don't meet
1268			 * min_target, return -EDQUOT */
1269			if (!ap->min_target || ap->min_target > ap->allowed) {
1270				if (!test_and_set_bit(QDF_QMSG_QUIET,
1271						      &qd->qd_flags)) {
1272					print_message(qd, "exceeded");
1273					quota_send_warning(qd->qd_id,
1274							   sdp->sd_vfs->s_dev,
1275							   QUOTA_NL_BHARDWARN);
1276				}
1277				error = -EDQUOT;
1278				break;
1279			}
1280		} else if (warn && warn < value &&
1281			   time_after_eq(jiffies, qd->qd_last_warn +
1282					 gfs2_tune_get(sdp, gt_quota_warn_period)
1283					 * HZ)) {
1284			quota_send_warning(qd->qd_id,
1285					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1286			print_message(qd, "warning");
1287			error = 0;
1288			qd->qd_last_warn = jiffies;
1289		}
1290	}
1291	return error;
1292}
1293
1294void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1295		       kuid_t uid, kgid_t gid)
1296{
1297	struct gfs2_quota_data *qd;
1298	u32 x;
1299	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1300
1301	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ||
1302	    gfs2_assert_warn(sdp, change))
1303		return;
1304	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1305		return;
1306
1307	if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1308				 ip->i_qadata->qa_ref > 0))
1309		return;
1310	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1311		qd = ip->i_qadata->qa_qd[x];
1312
1313		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1314		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1315			do_qc(qd, change, QC_CHANGE);
1316		}
1317	}
1318}
1319
1320static bool qd_changed(struct gfs2_sbd *sdp)
1321{
1322	struct gfs2_quota_data *qd;
1323	bool changed = false;
1324
1325	spin_lock(&qd_lock);
1326	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1327		if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
1328		    !test_bit(QDF_CHANGE, &qd->qd_flags))
1329			continue;
1330
1331		changed = true;
1332		break;
1333	}
1334	spin_unlock(&qd_lock);
1335	return changed;
1336}
1337
1338int gfs2_quota_sync(struct super_block *sb, int type)
1339{
1340	struct gfs2_sbd *sdp = sb->s_fs_info;
1341	struct gfs2_quota_data **qda;
1342	unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1343	unsigned int num_qd;
1344	unsigned int x;
1345	int error = 0;
1346
1347	if (!qd_changed(sdp))
1348		return 0;
1349
1350	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1351	if (!qda)
1352		return -ENOMEM;
1353
1354	mutex_lock(&sdp->sd_quota_sync_mutex);
1355	sdp->sd_quota_sync_gen++;
1356
1357	do {
1358		num_qd = 0;
1359
1360		for (;;) {
1361			error = qd_fish(sdp, qda + num_qd);
1362			if (error || !qda[num_qd])
1363				break;
1364			if (++num_qd == max_qd)
1365				break;
1366		}
1367
1368		if (num_qd) {
1369			if (!error)
1370				error = do_sync(num_qd, qda);
1371
1372			for (x = 0; x < num_qd; x++)
1373				qd_unlock(qda[x]);
1374		}
1375	} while (!error && num_qd == max_qd);
1376
1377	mutex_unlock(&sdp->sd_quota_sync_mutex);
1378	kfree(qda);
1379
1380	return error;
1381}
1382
1383int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1384{
1385	struct gfs2_quota_data *qd;
1386	struct gfs2_holder q_gh;
1387	int error;
1388
1389	error = qd_get(sdp, qid, &qd);
1390	if (error)
1391		return error;
1392
1393	error = do_glock(qd, FORCE, &q_gh);
1394	if (!error)
1395		gfs2_glock_dq_uninit(&q_gh);
1396
1397	qd_put(qd);
1398	return error;
1399}
1400
1401int gfs2_quota_init(struct gfs2_sbd *sdp)
1402{
1403	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1404	u64 size = i_size_read(sdp->sd_qc_inode);
1405	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1406	unsigned int x, slot = 0;
1407	unsigned int found = 0;
1408	unsigned int hash;
1409	unsigned int bm_size;
1410	u64 dblock;
1411	u32 extlen = 0;
1412	int error;
1413
1414	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1415		return -EIO;
1416
1417	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1418	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1419	bm_size *= sizeof(unsigned long);
1420	error = -ENOMEM;
1421	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1422	if (sdp->sd_quota_bitmap == NULL)
1423		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1424						 __GFP_ZERO);
1425	if (!sdp->sd_quota_bitmap)
1426		return error;
1427
1428	for (x = 0; x < blocks; x++) {
1429		struct buffer_head *bh;
1430		const struct gfs2_quota_change *qc;
1431		unsigned int y;
1432
1433		if (!extlen) {
1434			extlen = 32;
1435			error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1436			if (error)
1437				goto fail;
1438		}
1439		error = -EIO;
1440		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1441		if (!bh)
1442			goto fail;
1443		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1444			brelse(bh);
1445			goto fail;
1446		}
1447
1448		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1449		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1450		     y++, slot++) {
1451			struct gfs2_quota_data *qd;
1452			s64 qc_change = be64_to_cpu(qc->qc_change);
1453			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1454			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1455						USRQUOTA : GRPQUOTA;
1456			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1457						      be32_to_cpu(qc->qc_id));
1458			qc++;
1459			if (!qc_change)
1460				continue;
1461
1462			hash = gfs2_qd_hash(sdp, qc_id);
1463			qd = qd_alloc(hash, sdp, qc_id);
1464			if (qd == NULL) {
1465				brelse(bh);
1466				goto fail;
1467			}
1468
1469			set_bit(QDF_CHANGE, &qd->qd_flags);
1470			qd->qd_change = qc_change;
1471			qd->qd_slot = slot;
1472			qd->qd_slot_ref = 1;
1473
1474			spin_lock(&qd_lock);
1475			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1476			list_add(&qd->qd_list, &sdp->sd_quota_list);
1477			atomic_inc(&sdp->sd_quota_count);
1478			spin_unlock(&qd_lock);
1479
1480			spin_lock_bucket(hash);
1481			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1482			spin_unlock_bucket(hash);
1483
1484			found++;
1485		}
1486
1487		brelse(bh);
1488		dblock++;
1489		extlen--;
1490	}
1491
1492	if (found)
1493		fs_info(sdp, "found %u quota changes\n", found);
1494
1495	return 0;
1496
1497fail:
1498	gfs2_quota_cleanup(sdp);
1499	return error;
1500}
1501
1502void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1503{
1504	struct gfs2_quota_data *qd;
1505	LIST_HEAD(dispose);
1506	int count;
1507
1508	BUG_ON(!test_bit(SDF_NORECOVERY, &sdp->sd_flags) &&
1509		test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
1510
1511	spin_lock(&qd_lock);
1512	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1513		spin_lock(&qd->qd_lockref.lock);
1514		if (qd->qd_lockref.count != 0) {
1515			spin_unlock(&qd->qd_lockref.lock);
1516			continue;
1517		}
1518		lockref_mark_dead(&qd->qd_lockref);
1519		spin_unlock(&qd->qd_lockref.lock);
1520
1521		list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru);
1522		list_add(&qd->qd_lru, &dispose);
1523	}
1524	spin_unlock(&qd_lock);
1525
1526	gfs2_qd_list_dispose(&dispose);
1527
1528	wait_event_timeout(sdp->sd_kill_wait,
1529		(count = atomic_read(&sdp->sd_quota_count)) == 0,
1530		HZ * 60);
1531
1532	if (count != 0)
1533		fs_err(sdp, "%d left-over quota data objects\n", count);
1534
1535	kvfree(sdp->sd_quota_bitmap);
1536	sdp->sd_quota_bitmap = NULL;
1537}
1538
1539static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1540{
1541	if (error == 0 || error == -EROFS)
1542		return;
1543	if (!gfs2_withdrawing_or_withdrawn(sdp)) {
1544		if (!cmpxchg(&sdp->sd_log_error, 0, error))
1545			fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1546		wake_up(&sdp->sd_logd_waitq);
1547	}
1548}
1549
1550static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1551			       int (*fxn)(struct super_block *sb, int type),
1552			       unsigned long t, unsigned long *timeo,
1553			       unsigned int *new_timeo)
1554{
1555	if (t >= *timeo) {
1556		int error = fxn(sdp->sd_vfs, 0);
1557		quotad_error(sdp, msg, error);
1558		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1559	} else {
1560		*timeo -= t;
1561	}
1562}
1563
1564void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1565	if (!sdp->sd_statfs_force_sync) {
1566		sdp->sd_statfs_force_sync = 1;
1567		wake_up(&sdp->sd_quota_wait);
1568	}
1569}
1570
1571
1572/**
1573 * gfs2_quotad - Write cached quota changes into the quota file
1574 * @data: Pointer to GFS2 superblock
1575 *
1576 */
1577
1578int gfs2_quotad(void *data)
1579{
1580	struct gfs2_sbd *sdp = data;
1581	struct gfs2_tune *tune = &sdp->sd_tune;
1582	unsigned long statfs_timeo = 0;
1583	unsigned long quotad_timeo = 0;
1584	unsigned long t = 0;
1585
1586	set_freezable();
1587	while (!kthread_should_stop()) {
1588		if (gfs2_withdrawing_or_withdrawn(sdp))
1589			break;
1590
1591		/* Update the master statfs file */
1592		if (sdp->sd_statfs_force_sync) {
1593			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1594			quotad_error(sdp, "statfs", error);
1595			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1596		}
1597		else
1598			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1599				   	   &statfs_timeo,
1600					   &tune->gt_statfs_quantum);
1601
1602		/* Update quota file */
1603		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1604				   &quotad_timeo, &tune->gt_quota_quantum);
1605
1606		t = min(quotad_timeo, statfs_timeo);
1607
1608		t = wait_event_freezable_timeout(sdp->sd_quota_wait,
1609				sdp->sd_statfs_force_sync ||
1610				gfs2_withdrawing_or_withdrawn(sdp) ||
1611				kthread_should_stop(),
1612				t);
1613
1614		if (sdp->sd_statfs_force_sync)
1615			t = 0;
1616	}
1617
1618	return 0;
1619}
1620
1621static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1622{
1623	struct gfs2_sbd *sdp = sb->s_fs_info;
1624
1625	memset(state, 0, sizeof(*state));
1626
1627	switch (sdp->sd_args.ar_quota) {
1628	case GFS2_QUOTA_QUIET:
1629		fallthrough;
1630	case GFS2_QUOTA_ON:
1631		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1632		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1633		fallthrough;
1634	case GFS2_QUOTA_ACCOUNT:
1635		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1636						  QCI_SYSFILE;
1637		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1638						  QCI_SYSFILE;
1639		break;
1640	case GFS2_QUOTA_OFF:
1641		break;
1642	}
1643	if (sdp->sd_quota_inode) {
1644		state->s_state[USRQUOTA].ino =
1645					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1646		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1647	}
1648	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1649	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1650	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1651	return 0;
1652}
1653
1654static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1655			  struct qc_dqblk *fdq)
1656{
1657	struct gfs2_sbd *sdp = sb->s_fs_info;
1658	struct gfs2_quota_lvb *qlvb;
1659	struct gfs2_quota_data *qd;
1660	struct gfs2_holder q_gh;
1661	int error;
1662
1663	memset(fdq, 0, sizeof(*fdq));
1664
1665	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1666		return -ESRCH; /* Crazy XFS error code */
1667
1668	if ((qid.type != USRQUOTA) &&
1669	    (qid.type != GRPQUOTA))
1670		return -EINVAL;
1671
1672	error = qd_get(sdp, qid, &qd);
1673	if (error)
1674		return error;
1675	error = do_glock(qd, FORCE, &q_gh);
1676	if (error)
1677		goto out;
1678
1679	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1680	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1681	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1682	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1683
1684	gfs2_glock_dq_uninit(&q_gh);
1685out:
1686	qd_put(qd);
1687	return error;
1688}
1689
1690/* GFS2 only supports a subset of the XFS fields */
1691#define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1692
1693static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1694			  struct qc_dqblk *fdq)
1695{
1696	struct gfs2_sbd *sdp = sb->s_fs_info;
1697	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1698	struct gfs2_quota_data *qd;
1699	struct gfs2_holder q_gh, i_gh;
1700	unsigned int data_blocks, ind_blocks;
1701	unsigned int blocks = 0;
1702	int alloc_required;
1703	loff_t offset;
1704	int error;
1705
1706	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1707		return -ESRCH; /* Crazy XFS error code */
1708
1709	if ((qid.type != USRQUOTA) &&
1710	    (qid.type != GRPQUOTA))
1711		return -EINVAL;
1712
1713	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1714		return -EINVAL;
1715
1716	error = qd_get(sdp, qid, &qd);
1717	if (error)
1718		return error;
1719
1720	error = gfs2_qa_get(ip);
1721	if (error)
1722		goto out_put;
1723
1724	inode_lock(&ip->i_inode);
1725	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1726	if (error)
1727		goto out_unlockput;
1728	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1729	if (error)
1730		goto out_q;
1731
1732	/* Check for existing entry, if none then alloc new blocks */
1733	error = update_qd(sdp, qd);
1734	if (error)
1735		goto out_i;
1736
1737	/* If nothing has changed, this is a no-op */
1738	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1739	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1740		fdq->d_fieldmask ^= QC_SPC_SOFT;
1741
1742	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1743	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1744		fdq->d_fieldmask ^= QC_SPC_HARD;
1745
1746	if ((fdq->d_fieldmask & QC_SPACE) &&
1747	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1748		fdq->d_fieldmask ^= QC_SPACE;
1749
1750	if (fdq->d_fieldmask == 0)
1751		goto out_i;
1752
1753	offset = qd2offset(qd);
1754	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1755	if (gfs2_is_stuffed(ip))
1756		alloc_required = 1;
1757	if (alloc_required) {
1758		struct gfs2_alloc_parms ap = {};
1759		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1760				       &data_blocks, &ind_blocks);
1761		blocks = 1 + data_blocks + ind_blocks;
1762		ap.target = blocks;
1763		error = gfs2_inplace_reserve(ip, &ap);
1764		if (error)
1765			goto out_i;
1766		blocks += gfs2_rg_blocks(ip, blocks);
1767	}
1768
1769	/* Some quotas span block boundaries and can update two blocks,
1770	   adding an extra block to the transaction to handle such quotas */
1771	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1772	if (error)
1773		goto out_release;
1774
1775	/* Apply changes */
1776	error = gfs2_adjust_quota(sdp, offset, 0, qd, fdq);
1777	if (!error)
1778		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1779
1780	gfs2_trans_end(sdp);
1781out_release:
1782	if (alloc_required)
1783		gfs2_inplace_release(ip);
1784out_i:
1785	gfs2_glock_dq_uninit(&i_gh);
1786out_q:
1787	gfs2_glock_dq_uninit(&q_gh);
1788out_unlockput:
1789	gfs2_qa_put(ip);
1790	inode_unlock(&ip->i_inode);
1791out_put:
1792	qd_put(qd);
1793	return error;
1794}
1795
1796const struct quotactl_ops gfs2_quotactl_ops = {
1797	.quota_sync     = gfs2_quota_sync,
1798	.get_state	= gfs2_quota_get_state,
1799	.get_dqblk	= gfs2_get_dqblk,
1800	.set_dqblk	= gfs2_set_dqblk,
1801};
1802
1803void __init gfs2_quota_hash_init(void)
1804{
1805	unsigned i;
1806
1807	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1808		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1809}
1810