xref: /kernel/linux/linux-5.10/fs/gfs2/quota.c (revision 8c2ecf20)
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                              */
80static DEFINE_SPINLOCK(qd_lock);
81struct list_lru gfs2_qd_lru;
82
83static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
84
85static 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
96static inline void spin_lock_bucket(unsigned int hash)
97{
98        hlist_bl_lock(&qd_hash_table[hash]);
99}
100
101static inline void spin_unlock_bucket(unsigned int hash)
102{
103        hlist_bl_unlock(&qd_hash_table[hash]);
104}
105
106static 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
112static 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
145static 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
163static 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
180static 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
186struct 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
194static 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
201static 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
211static 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
235fail:
236	kmem_cache_free(gfs2_quotad_cachep, qd);
237	return NULL;
238}
239
240static 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
262static 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
300static 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
307static 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
318static 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;
334out:
335		qd->qd_slot_count++;
336	}
337	spin_unlock(&sdp->sd_bitmap_lock);
338
339	return error;
340}
341
342static 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
352static 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
365static int bh_get(struct gfs2_quota_data *qd)
366{
367	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
368	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
369	unsigned int block, offset;
370	struct buffer_head *bh;
371	int error;
372	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
373
374	mutex_lock(&sdp->sd_quota_mutex);
375
376	if (qd->qd_bh_count++) {
377		mutex_unlock(&sdp->sd_quota_mutex);
378		return 0;
379	}
380
381	block = qd->qd_slot / sdp->sd_qc_per_block;
382	offset = qd->qd_slot % sdp->sd_qc_per_block;
383
384	bh_map.b_size = BIT(ip->i_inode.i_blkbits);
385	error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
386	if (error)
387		goto fail;
388	error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, 0, &bh);
389	if (error)
390		goto fail;
391	error = -EIO;
392	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
393		goto fail_brelse;
394
395	qd->qd_bh = bh;
396	qd->qd_bh_qc = (struct gfs2_quota_change *)
397		(bh->b_data + sizeof(struct gfs2_meta_header) +
398		 offset * sizeof(struct gfs2_quota_change));
399
400	mutex_unlock(&sdp->sd_quota_mutex);
401
402	return 0;
403
404fail_brelse:
405	brelse(bh);
406fail:
407	qd->qd_bh_count--;
408	mutex_unlock(&sdp->sd_quota_mutex);
409	return error;
410}
411
412static void bh_put(struct gfs2_quota_data *qd)
413{
414	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
415
416	mutex_lock(&sdp->sd_quota_mutex);
417	gfs2_assert(sdp, qd->qd_bh_count);
418	if (!--qd->qd_bh_count) {
419		brelse(qd->qd_bh);
420		qd->qd_bh = NULL;
421		qd->qd_bh_qc = NULL;
422	}
423	mutex_unlock(&sdp->sd_quota_mutex);
424}
425
426static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
427			 u64 *sync_gen)
428{
429	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
430	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
431	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
432		return 0;
433
434	/*
435	 * If qd_change is 0 it means a pending quota change was negated.
436	 * We should not sync it, but we still have a qd reference and slot
437	 * reference taken by gfs2_quota_change -> do_qc that need to be put.
438	 */
439	if (!qd->qd_change && test_and_clear_bit(QDF_CHANGE, &qd->qd_flags)) {
440		slot_put(qd);
441		qd_put(qd);
442		return 0;
443	}
444
445	if (!lockref_get_not_dead(&qd->qd_lockref))
446		return 0;
447
448	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
449	set_bit(QDF_LOCKED, &qd->qd_flags);
450	qd->qd_change_sync = qd->qd_change;
451	slot_hold(qd);
452	return 1;
453}
454
455static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
456{
457	struct gfs2_quota_data *qd = NULL;
458	int error;
459	int found = 0;
460
461	*qdp = NULL;
462
463	if (sb_rdonly(sdp->sd_vfs))
464		return 0;
465
466	spin_lock(&qd_lock);
467
468	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
469		found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
470		if (found)
471			break;
472	}
473
474	if (!found)
475		qd = NULL;
476
477	spin_unlock(&qd_lock);
478
479	if (qd) {
480		gfs2_assert_warn(sdp, qd->qd_change_sync);
481		error = bh_get(qd);
482		if (error) {
483			clear_bit(QDF_LOCKED, &qd->qd_flags);
484			slot_put(qd);
485			qd_put(qd);
486			return error;
487		}
488	}
489
490	*qdp = qd;
491
492	return 0;
493}
494
495static void qd_unlock(struct gfs2_quota_data *qd)
496{
497	gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
498			 test_bit(QDF_LOCKED, &qd->qd_flags));
499	clear_bit(QDF_LOCKED, &qd->qd_flags);
500	bh_put(qd);
501	slot_put(qd);
502	qd_put(qd);
503}
504
505static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
506		    struct gfs2_quota_data **qdp)
507{
508	int error;
509
510	error = qd_get(sdp, qid, qdp);
511	if (error)
512		return error;
513
514	error = slot_get(*qdp);
515	if (error)
516		goto fail;
517
518	error = bh_get(*qdp);
519	if (error)
520		goto fail_slot;
521
522	return 0;
523
524fail_slot:
525	slot_put(*qdp);
526fail:
527	qd_put(*qdp);
528	return error;
529}
530
531static void qdsb_put(struct gfs2_quota_data *qd)
532{
533	bh_put(qd);
534	slot_put(qd);
535	qd_put(qd);
536}
537
538/**
539 * gfs2_qa_get - make sure we have a quota allocations data structure,
540 *               if necessary
541 * @ip: the inode for this reservation
542 */
543int gfs2_qa_get(struct gfs2_inode *ip)
544{
545	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
546	struct inode *inode = &ip->i_inode;
547
548	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
549		return 0;
550
551	spin_lock(&inode->i_lock);
552	if (ip->i_qadata == NULL) {
553		struct gfs2_qadata *tmp;
554
555		spin_unlock(&inode->i_lock);
556		tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
557		if (!tmp)
558			return -ENOMEM;
559
560		spin_lock(&inode->i_lock);
561		if (ip->i_qadata == NULL)
562			ip->i_qadata = tmp;
563		else
564			kmem_cache_free(gfs2_qadata_cachep, tmp);
565	}
566	ip->i_qadata->qa_ref++;
567	spin_unlock(&inode->i_lock);
568	return 0;
569}
570
571void gfs2_qa_put(struct gfs2_inode *ip)
572{
573	struct inode *inode = &ip->i_inode;
574
575	spin_lock(&inode->i_lock);
576	if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
577		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
578		ip->i_qadata = NULL;
579	}
580	spin_unlock(&inode->i_lock);
581}
582
583int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
584{
585	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
586	struct gfs2_quota_data **qd;
587	int error;
588
589	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
590		return 0;
591
592	error = gfs2_qa_get(ip);
593	if (error)
594		return error;
595
596	qd = ip->i_qadata->qa_qd;
597
598	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
599	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
600		error = -EIO;
601		goto out;
602	}
603
604	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
605	if (error)
606		goto out_unhold;
607	ip->i_qadata->qa_qd_num++;
608	qd++;
609
610	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
611	if (error)
612		goto out_unhold;
613	ip->i_qadata->qa_qd_num++;
614	qd++;
615
616	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
617	    !uid_eq(uid, ip->i_inode.i_uid)) {
618		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
619		if (error)
620			goto out_unhold;
621		ip->i_qadata->qa_qd_num++;
622		qd++;
623	}
624
625	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
626	    !gid_eq(gid, ip->i_inode.i_gid)) {
627		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
628		if (error)
629			goto out_unhold;
630		ip->i_qadata->qa_qd_num++;
631		qd++;
632	}
633
634out_unhold:
635	if (error)
636		gfs2_quota_unhold(ip);
637out:
638	return error;
639}
640
641void gfs2_quota_unhold(struct gfs2_inode *ip)
642{
643	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
644	u32 x;
645
646	if (ip->i_qadata == NULL)
647		return;
648
649	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
650
651	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
652		qdsb_put(ip->i_qadata->qa_qd[x]);
653		ip->i_qadata->qa_qd[x] = NULL;
654	}
655	ip->i_qadata->qa_qd_num = 0;
656	gfs2_qa_put(ip);
657}
658
659static int sort_qd(const void *a, const void *b)
660{
661	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
662	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
663
664	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
665		return -1;
666	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
667		return 1;
668	return 0;
669}
670
671static void do_qc(struct gfs2_quota_data *qd, s64 change)
672{
673	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
674	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
675	struct gfs2_quota_change *qc = qd->qd_bh_qc;
676	s64 x;
677
678	mutex_lock(&sdp->sd_quota_mutex);
679	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
680
681	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
682		qc->qc_change = 0;
683		qc->qc_flags = 0;
684		if (qd->qd_id.type == USRQUOTA)
685			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
686		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
687	}
688
689	x = be64_to_cpu(qc->qc_change) + change;
690	qc->qc_change = cpu_to_be64(x);
691
692	spin_lock(&qd_lock);
693	qd->qd_change = x;
694	spin_unlock(&qd_lock);
695
696	if (!x) {
697		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
698		clear_bit(QDF_CHANGE, &qd->qd_flags);
699		qc->qc_flags = 0;
700		qc->qc_id = 0;
701		slot_put(qd);
702		qd_put(qd);
703	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
704		qd_hold(qd);
705		slot_hold(qd);
706	}
707
708	if (change < 0) /* Reset quiet flag if we freed some blocks */
709		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
710	mutex_unlock(&sdp->sd_quota_mutex);
711}
712
713static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
714				  unsigned off, void *buf, unsigned bytes)
715{
716	struct inode *inode = &ip->i_inode;
717	struct gfs2_sbd *sdp = GFS2_SB(inode);
718	struct address_space *mapping = inode->i_mapping;
719	struct page *page;
720	struct buffer_head *bh;
721	void *kaddr;
722	u64 blk;
723	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
724	unsigned to_write = bytes, pg_off = off;
725	int done = 0;
726
727	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
728	boff = off % bsize;
729
730	page = find_or_create_page(mapping, index, GFP_NOFS);
731	if (!page)
732		return -ENOMEM;
733	if (!page_has_buffers(page))
734		create_empty_buffers(page, bsize, 0);
735
736	bh = page_buffers(page);
737	while (!done) {
738		/* Find the beginning block within the page */
739		if (pg_off >= ((bnum * bsize) + bsize)) {
740			bh = bh->b_this_page;
741			bnum++;
742			blk++;
743			continue;
744		}
745		if (!buffer_mapped(bh)) {
746			gfs2_block_map(inode, blk, bh, 1);
747			if (!buffer_mapped(bh))
748				goto unlock_out;
749			/* If it's a newly allocated disk block, zero it */
750			if (buffer_new(bh))
751				zero_user(page, bnum * bsize, bh->b_size);
752		}
753		if (PageUptodate(page))
754			set_buffer_uptodate(bh);
755		if (!buffer_uptodate(bh)) {
756			ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
757			wait_on_buffer(bh);
758			if (!buffer_uptodate(bh))
759				goto unlock_out;
760		}
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
786unlock_out:
787	unlock_page(page);
788	put_page(page);
789	return -EIO;
790}
791
792static 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
836static 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, NULL);
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
891static 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
975out_end_trans:
976	gfs2_trans_end(sdp);
977out_ipres:
978	gfs2_inplace_release(ip);
979out_alloc:
980	gfs2_glock_dq_uninit(&i_gh);
981out_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);
988out:
989	gfs2_qa_put(ip);
990	return error;
991}
992
993static 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
1018static 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
1026restart:
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
1059fail_gunlock:
1060	gfs2_glock_dq_uninit(&i_gh);
1061fail:
1062	gfs2_glock_dq_uninit(q_gh);
1063	return error;
1064}
1065
1066int 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
1101static 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
1137void 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
1188static 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 */
1217int 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
1274void 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
1300int 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
1346int 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
1364int 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			int new = 0;
1398			error = gfs2_extent_map(&ip->i_inode, x, &new, &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
1460fail:
1461	gfs2_quota_cleanup(sdp);
1462	return error;
1463}
1464
1465void 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
1502static 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
1513static 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
1527static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1528{
1529	struct gfs2_inode *ip;
1530
1531	while(1) {
1532		ip = NULL;
1533		spin_lock(&sdp->sd_trunc_lock);
1534		if (!list_empty(&sdp->sd_trunc_list)) {
1535			ip = list_first_entry(&sdp->sd_trunc_list,
1536					struct gfs2_inode, i_trunc_list);
1537			list_del_init(&ip->i_trunc_list);
1538		}
1539		spin_unlock(&sdp->sd_trunc_lock);
1540		if (ip == NULL)
1541			return;
1542		gfs2_glock_finish_truncate(ip);
1543	}
1544}
1545
1546void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1547	if (!sdp->sd_statfs_force_sync) {
1548		sdp->sd_statfs_force_sync = 1;
1549		wake_up(&sdp->sd_quota_wait);
1550	}
1551}
1552
1553
1554/**
1555 * gfs2_quotad - Write cached quota changes into the quota file
1556 * @sdp: Pointer to GFS2 superblock
1557 *
1558 */
1559
1560int gfs2_quotad(void *data)
1561{
1562	struct gfs2_sbd *sdp = data;
1563	struct gfs2_tune *tune = &sdp->sd_tune;
1564	unsigned long statfs_timeo = 0;
1565	unsigned long quotad_timeo = 0;
1566	unsigned long t = 0;
1567	DEFINE_WAIT(wait);
1568	int empty;
1569
1570	while (!kthread_should_stop()) {
1571
1572		if (gfs2_withdrawn(sdp))
1573			goto bypass;
1574		/* Update the master statfs file */
1575		if (sdp->sd_statfs_force_sync) {
1576			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1577			quotad_error(sdp, "statfs", error);
1578			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1579		}
1580		else
1581			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1582				   	   &statfs_timeo,
1583					   &tune->gt_statfs_quantum);
1584
1585		/* Update quota file */
1586		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1587				   &quotad_timeo, &tune->gt_quota_quantum);
1588
1589		/* Check for & recover partially truncated inodes */
1590		quotad_check_trunc_list(sdp);
1591
1592		try_to_freeze();
1593
1594bypass:
1595		t = min(quotad_timeo, statfs_timeo);
1596
1597		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1598		spin_lock(&sdp->sd_trunc_lock);
1599		empty = list_empty(&sdp->sd_trunc_list);
1600		spin_unlock(&sdp->sd_trunc_lock);
1601		if (empty && !sdp->sd_statfs_force_sync)
1602			t -= schedule_timeout(t);
1603		else
1604			t = 0;
1605		finish_wait(&sdp->sd_quota_wait, &wait);
1606	}
1607
1608	return 0;
1609}
1610
1611static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1612{
1613	struct gfs2_sbd *sdp = sb->s_fs_info;
1614
1615	memset(state, 0, sizeof(*state));
1616
1617	switch (sdp->sd_args.ar_quota) {
1618	case GFS2_QUOTA_ON:
1619		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1620		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1621		fallthrough;
1622	case GFS2_QUOTA_ACCOUNT:
1623		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1624						  QCI_SYSFILE;
1625		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1626						  QCI_SYSFILE;
1627		break;
1628	case GFS2_QUOTA_OFF:
1629		break;
1630	}
1631	if (sdp->sd_quota_inode) {
1632		state->s_state[USRQUOTA].ino =
1633					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1634		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1635	}
1636	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1637	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1638	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1639	return 0;
1640}
1641
1642static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1643			  struct qc_dqblk *fdq)
1644{
1645	struct gfs2_sbd *sdp = sb->s_fs_info;
1646	struct gfs2_quota_lvb *qlvb;
1647	struct gfs2_quota_data *qd;
1648	struct gfs2_holder q_gh;
1649	int error;
1650
1651	memset(fdq, 0, sizeof(*fdq));
1652
1653	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1654		return -ESRCH; /* Crazy XFS error code */
1655
1656	if ((qid.type != USRQUOTA) &&
1657	    (qid.type != GRPQUOTA))
1658		return -EINVAL;
1659
1660	error = qd_get(sdp, qid, &qd);
1661	if (error)
1662		return error;
1663	error = do_glock(qd, FORCE, &q_gh);
1664	if (error)
1665		goto out;
1666
1667	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1668	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1669	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1670	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1671
1672	gfs2_glock_dq_uninit(&q_gh);
1673out:
1674	qd_put(qd);
1675	return error;
1676}
1677
1678/* GFS2 only supports a subset of the XFS fields */
1679#define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1680
1681static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1682			  struct qc_dqblk *fdq)
1683{
1684	struct gfs2_sbd *sdp = sb->s_fs_info;
1685	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1686	struct gfs2_quota_data *qd;
1687	struct gfs2_holder q_gh, i_gh;
1688	unsigned int data_blocks, ind_blocks;
1689	unsigned int blocks = 0;
1690	int alloc_required;
1691	loff_t offset;
1692	int error;
1693
1694	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1695		return -ESRCH; /* Crazy XFS error code */
1696
1697	if ((qid.type != USRQUOTA) &&
1698	    (qid.type != GRPQUOTA))
1699		return -EINVAL;
1700
1701	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1702		return -EINVAL;
1703
1704	error = qd_get(sdp, qid, &qd);
1705	if (error)
1706		return error;
1707
1708	error = gfs2_qa_get(ip);
1709	if (error)
1710		goto out_put;
1711
1712	inode_lock(&ip->i_inode);
1713	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1714	if (error)
1715		goto out_unlockput;
1716	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1717	if (error)
1718		goto out_q;
1719
1720	/* Check for existing entry, if none then alloc new blocks */
1721	error = update_qd(sdp, qd);
1722	if (error)
1723		goto out_i;
1724
1725	/* If nothing has changed, this is a no-op */
1726	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1727	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1728		fdq->d_fieldmask ^= QC_SPC_SOFT;
1729
1730	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1731	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1732		fdq->d_fieldmask ^= QC_SPC_HARD;
1733
1734	if ((fdq->d_fieldmask & QC_SPACE) &&
1735	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1736		fdq->d_fieldmask ^= QC_SPACE;
1737
1738	if (fdq->d_fieldmask == 0)
1739		goto out_i;
1740
1741	offset = qd2offset(qd);
1742	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1743	if (gfs2_is_stuffed(ip))
1744		alloc_required = 1;
1745	if (alloc_required) {
1746		struct gfs2_alloc_parms ap = { .aflags = 0, };
1747		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1748				       &data_blocks, &ind_blocks);
1749		blocks = 1 + data_blocks + ind_blocks;
1750		ap.target = blocks;
1751		error = gfs2_inplace_reserve(ip, &ap);
1752		if (error)
1753			goto out_i;
1754		blocks += gfs2_rg_blocks(ip, blocks);
1755	}
1756
1757	/* Some quotas span block boundaries and can update two blocks,
1758	   adding an extra block to the transaction to handle such quotas */
1759	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1760	if (error)
1761		goto out_release;
1762
1763	/* Apply changes */
1764	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1765	if (!error)
1766		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1767
1768	gfs2_trans_end(sdp);
1769out_release:
1770	if (alloc_required)
1771		gfs2_inplace_release(ip);
1772out_i:
1773	gfs2_glock_dq_uninit(&i_gh);
1774out_q:
1775	gfs2_glock_dq_uninit(&q_gh);
1776out_unlockput:
1777	gfs2_qa_put(ip);
1778	inode_unlock(&ip->i_inode);
1779out_put:
1780	qd_put(qd);
1781	return error;
1782}
1783
1784const struct quotactl_ops gfs2_quotactl_ops = {
1785	.quota_sync     = gfs2_quota_sync,
1786	.get_state	= gfs2_quota_get_state,
1787	.get_dqblk	= gfs2_get_dqblk,
1788	.set_dqblk	= gfs2_set_dqblk,
1789};
1790
1791void __init gfs2_quota_hash_init(void)
1792{
1793	unsigned i;
1794
1795	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1796		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1797}
1798