xref: /linux/fs/gfs2/quota.c (revision ba6e0e5cb5b2c2e736e16b4aead816450a8718e6)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
5  */
6 
7 /*
8  * Quota change tags are associated with each transaction that allocates or
9  * deallocates space.  Those changes are accumulated locally to each node (in a
10  * per-node file) and then are periodically synced to the quota file.  This
11  * avoids the bottleneck of constantly touching the quota file, but introduces
12  * fuzziness in the current usage value of IDs that are being used on different
13  * nodes in the cluster simultaneously.  So, it is possible for a user on
14  * multiple nodes to overrun their quota, but that overrun is controlable.
15  * Since quota tags are part of transactions, there is no need for a quota check
16  * program to be run on node crashes or anything like that.
17  *
18  * There are couple of knobs that let the administrator manage the quota
19  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
20  * sitting on one node before being synced to the quota file.  (The default is
21  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
22  * of quota file syncs increases as the user moves closer to their limit.  The
23  * more frequent the syncs, the more accurate the quota enforcement, but that
24  * means that there is more contention between the nodes for the quota file.
25  * The default value is one.  This sets the maximum theoretical quota overrun
26  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
27  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
28  * number greater than one makes quota syncs more frequent and reduces the
29  * maximum overrun.  Numbers less than one (but greater than zero) make quota
30  * syncs less frequent.
31  *
32  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
33  * the quota file, so it is not being constantly read.
34  */
35 
36 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 
38 #include <linux/sched.h>
39 #include <linux/slab.h>
40 #include <linux/mm.h>
41 #include <linux/spinlock.h>
42 #include <linux/completion.h>
43 #include <linux/buffer_head.h>
44 #include <linux/sort.h>
45 #include <linux/fs.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/kthread.h>
49 #include <linux/freezer.h>
50 #include <linux/quota.h>
51 #include <linux/dqblk_xfs.h>
52 #include <linux/lockref.h>
53 #include <linux/list_lru.h>
54 #include <linux/rcupdate.h>
55 #include <linux/rculist_bl.h>
56 #include <linux/bit_spinlock.h>
57 #include <linux/jhash.h>
58 #include <linux/vmalloc.h>
59 
60 #include "gfs2.h"
61 #include "incore.h"
62 #include "bmap.h"
63 #include "glock.h"
64 #include "glops.h"
65 #include "log.h"
66 #include "meta_io.h"
67 #include "quota.h"
68 #include "rgrp.h"
69 #include "super.h"
70 #include "trans.h"
71 #include "inode.h"
72 #include "util.h"
73 
74 #define GFS2_QD_HASH_SHIFT      12
75 #define GFS2_QD_HASH_SIZE       BIT(GFS2_QD_HASH_SHIFT)
76 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
77 
78 #define QC_CHANGE 0
79 #define QC_SYNC 1
80 
81 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
82 /*                     -> sd_bitmap_lock                              */
83 static DEFINE_SPINLOCK(qd_lock);
84 struct list_lru gfs2_qd_lru;
85 
86 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
87 
88 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
89 				 const struct kqid qid)
90 {
91 	unsigned int h;
92 
93 	h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
94 	h = jhash(&qid, sizeof(struct kqid), h);
95 
96 	return h & GFS2_QD_HASH_MASK;
97 }
98 
99 static inline void spin_lock_bucket(unsigned int hash)
100 {
101         hlist_bl_lock(&qd_hash_table[hash]);
102 }
103 
104 static inline void spin_unlock_bucket(unsigned int hash)
105 {
106         hlist_bl_unlock(&qd_hash_table[hash]);
107 }
108 
109 static void gfs2_qd_dealloc(struct rcu_head *rcu)
110 {
111 	struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
112 	struct gfs2_sbd *sdp = qd->qd_sbd;
113 
114 	kmem_cache_free(gfs2_quotad_cachep, qd);
115 	if (atomic_dec_and_test(&sdp->sd_quota_count))
116 		wake_up(&sdp->sd_kill_wait);
117 }
118 
119 static void gfs2_qd_dispose(struct gfs2_quota_data *qd)
120 {
121 	struct gfs2_sbd *sdp = qd->qd_sbd;
122 
123 	spin_lock(&qd_lock);
124 	list_del(&qd->qd_list);
125 	spin_unlock(&qd_lock);
126 
127 	spin_lock_bucket(qd->qd_hash);
128 	hlist_bl_del_rcu(&qd->qd_hlist);
129 	spin_unlock_bucket(qd->qd_hash);
130 
131 	if (!gfs2_withdrawn(sdp)) {
132 		gfs2_assert_warn(sdp, !qd->qd_change);
133 		gfs2_assert_warn(sdp, !qd->qd_slot_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 
141 static void gfs2_qd_list_dispose(struct list_head *list)
142 {
143 	struct gfs2_quota_data *qd;
144 
145 	while (!list_empty(list)) {
146 		qd = list_first_entry(list, struct gfs2_quota_data, qd_lru);
147 		list_del(&qd->qd_lru);
148 
149 		gfs2_qd_dispose(qd);
150 	}
151 }
152 
153 
154 static enum lru_status gfs2_qd_isolate(struct list_head *item,
155 		struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
156 {
157 	struct list_head *dispose = arg;
158 	struct gfs2_quota_data *qd =
159 		list_entry(item, struct gfs2_quota_data, qd_lru);
160 	enum lru_status status;
161 
162 	if (!spin_trylock(&qd->qd_lockref.lock))
163 		return LRU_SKIP;
164 
165 	status = LRU_SKIP;
166 	if (qd->qd_lockref.count == 0) {
167 		lockref_mark_dead(&qd->qd_lockref);
168 		list_lru_isolate_move(lru, &qd->qd_lru, dispose);
169 		status = LRU_REMOVED;
170 	}
171 
172 	spin_unlock(&qd->qd_lockref.lock);
173 	return status;
174 }
175 
176 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
177 					 struct shrink_control *sc)
178 {
179 	LIST_HEAD(dispose);
180 	unsigned long freed;
181 
182 	if (!(sc->gfp_mask & __GFP_FS))
183 		return SHRINK_STOP;
184 
185 	freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
186 				     gfs2_qd_isolate, &dispose);
187 
188 	gfs2_qd_list_dispose(&dispose);
189 
190 	return freed;
191 }
192 
193 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
194 					  struct shrink_control *sc)
195 {
196 	return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
197 }
198 
199 struct shrinker gfs2_qd_shrinker = {
200 	.count_objects = gfs2_qd_shrink_count,
201 	.scan_objects = gfs2_qd_shrink_scan,
202 	.seeks = DEFAULT_SEEKS,
203 	.flags = SHRINKER_NUMA_AWARE,
204 };
205 
206 
207 static u64 qd2index(struct gfs2_quota_data *qd)
208 {
209 	struct kqid qid = qd->qd_id;
210 	return (2 * (u64)from_kqid(&init_user_ns, qid)) +
211 		((qid.type == USRQUOTA) ? 0 : 1);
212 }
213 
214 static u64 qd2offset(struct gfs2_quota_data *qd)
215 {
216 	return qd2index(qd) * sizeof(struct gfs2_quota);
217 }
218 
219 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
220 {
221 	struct gfs2_quota_data *qd;
222 	int error;
223 
224 	qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
225 	if (!qd)
226 		return NULL;
227 
228 	qd->qd_sbd = sdp;
229 	qd->qd_lockref.count = 0;
230 	spin_lock_init(&qd->qd_lockref.lock);
231 	qd->qd_id = qid;
232 	qd->qd_slot = -1;
233 	INIT_LIST_HEAD(&qd->qd_lru);
234 	qd->qd_hash = hash;
235 
236 	error = gfs2_glock_get(sdp, qd2index(qd),
237 			      &gfs2_quota_glops, CREATE, &qd->qd_gl);
238 	if (error)
239 		goto fail;
240 
241 	return qd;
242 
243 fail:
244 	kmem_cache_free(gfs2_quotad_cachep, qd);
245 	return NULL;
246 }
247 
248 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
249 						     const struct gfs2_sbd *sdp,
250 						     struct kqid qid)
251 {
252 	struct gfs2_quota_data *qd;
253 	struct hlist_bl_node *h;
254 
255 	hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
256 		if (!qid_eq(qd->qd_id, qid))
257 			continue;
258 		if (qd->qd_sbd != sdp)
259 			continue;
260 		if (lockref_get_not_dead(&qd->qd_lockref)) {
261 			list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
262 			return qd;
263 		}
264 	}
265 
266 	return NULL;
267 }
268 
269 
270 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
271 		  struct gfs2_quota_data **qdp)
272 {
273 	struct gfs2_quota_data *qd, *new_qd;
274 	unsigned int hash = gfs2_qd_hash(sdp, qid);
275 
276 	rcu_read_lock();
277 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
278 	rcu_read_unlock();
279 
280 	if (qd)
281 		return 0;
282 
283 	new_qd = qd_alloc(hash, sdp, qid);
284 	if (!new_qd)
285 		return -ENOMEM;
286 
287 	spin_lock(&qd_lock);
288 	spin_lock_bucket(hash);
289 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
290 	if (qd == NULL) {
291 		new_qd->qd_lockref.count++;
292 		*qdp = new_qd;
293 		list_add(&new_qd->qd_list, &sdp->sd_quota_list);
294 		hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
295 		atomic_inc(&sdp->sd_quota_count);
296 	}
297 	spin_unlock_bucket(hash);
298 	spin_unlock(&qd_lock);
299 
300 	if (qd) {
301 		gfs2_glock_put(new_qd->qd_gl);
302 		kmem_cache_free(gfs2_quotad_cachep, new_qd);
303 	}
304 
305 	return 0;
306 }
307 
308 
309 static void qd_hold(struct gfs2_quota_data *qd)
310 {
311 	struct gfs2_sbd *sdp = qd->qd_sbd;
312 	gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
313 	lockref_get(&qd->qd_lockref);
314 }
315 
316 static void qd_put(struct gfs2_quota_data *qd)
317 {
318 	struct gfs2_sbd *sdp;
319 
320 	if (lockref_put_or_lock(&qd->qd_lockref))
321 		return;
322 
323 	BUG_ON(__lockref_is_dead(&qd->qd_lockref));
324 	sdp = qd->qd_sbd;
325 	if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
326 		lockref_mark_dead(&qd->qd_lockref);
327 		spin_unlock(&qd->qd_lockref.lock);
328 
329 		gfs2_qd_dispose(qd);
330 		return;
331 	}
332 
333 	qd->qd_lockref.count = 0;
334 	list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
335 	spin_unlock(&qd->qd_lockref.lock);
336 }
337 
338 static int slot_get(struct gfs2_quota_data *qd)
339 {
340 	struct gfs2_sbd *sdp = qd->qd_sbd;
341 	unsigned int bit;
342 	int error = 0;
343 
344 	spin_lock(&sdp->sd_bitmap_lock);
345 	if (qd->qd_slot_ref == 0) {
346 		bit = find_first_zero_bit(sdp->sd_quota_bitmap,
347 					  sdp->sd_quota_slots);
348 		if (bit >= sdp->sd_quota_slots) {
349 			error = -ENOSPC;
350 			goto out;
351 		}
352 		set_bit(bit, sdp->sd_quota_bitmap);
353 		qd->qd_slot = bit;
354 	}
355 	qd->qd_slot_ref++;
356 out:
357 	spin_unlock(&sdp->sd_bitmap_lock);
358 	return error;
359 }
360 
361 static void slot_hold(struct gfs2_quota_data *qd)
362 {
363 	struct gfs2_sbd *sdp = qd->qd_sbd;
364 
365 	spin_lock(&sdp->sd_bitmap_lock);
366 	gfs2_assert(sdp, qd->qd_slot_ref);
367 	qd->qd_slot_ref++;
368 	spin_unlock(&sdp->sd_bitmap_lock);
369 }
370 
371 static void slot_put(struct gfs2_quota_data *qd)
372 {
373 	struct gfs2_sbd *sdp = qd->qd_sbd;
374 
375 	spin_lock(&sdp->sd_bitmap_lock);
376 	gfs2_assert(sdp, qd->qd_slot_ref);
377 	if (!--qd->qd_slot_ref) {
378 		BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
379 		qd->qd_slot = -1;
380 	}
381 	spin_unlock(&sdp->sd_bitmap_lock);
382 }
383 
384 static int bh_get(struct gfs2_quota_data *qd)
385 {
386 	struct gfs2_sbd *sdp = qd->qd_sbd;
387 	struct inode *inode = sdp->sd_qc_inode;
388 	struct gfs2_inode *ip = GFS2_I(inode);
389 	unsigned int block, offset;
390 	struct buffer_head *bh;
391 	struct iomap iomap = { };
392 	int error;
393 
394 	mutex_lock(&sdp->sd_quota_mutex);
395 
396 	if (qd->qd_bh_count++) {
397 		mutex_unlock(&sdp->sd_quota_mutex);
398 		return 0;
399 	}
400 
401 	block = qd->qd_slot / sdp->sd_qc_per_block;
402 	offset = qd->qd_slot % sdp->sd_qc_per_block;
403 
404 	error = gfs2_iomap_get(inode,
405 			       (loff_t)block << inode->i_blkbits,
406 			       i_blocksize(inode), &iomap);
407 	if (error)
408 		goto fail;
409 	error = -ENOENT;
410 	if (iomap.type != IOMAP_MAPPED)
411 		goto fail;
412 
413 	error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
414 			       DIO_WAIT, 0, &bh);
415 	if (error)
416 		goto fail;
417 	error = -EIO;
418 	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
419 		goto fail_brelse;
420 
421 	qd->qd_bh = bh;
422 	qd->qd_bh_qc = (struct gfs2_quota_change *)
423 		(bh->b_data + sizeof(struct gfs2_meta_header) +
424 		 offset * sizeof(struct gfs2_quota_change));
425 
426 	mutex_unlock(&sdp->sd_quota_mutex);
427 
428 	return 0;
429 
430 fail_brelse:
431 	brelse(bh);
432 fail:
433 	qd->qd_bh_count--;
434 	mutex_unlock(&sdp->sd_quota_mutex);
435 	return error;
436 }
437 
438 static void bh_put(struct gfs2_quota_data *qd)
439 {
440 	struct gfs2_sbd *sdp = qd->qd_sbd;
441 
442 	mutex_lock(&sdp->sd_quota_mutex);
443 	gfs2_assert(sdp, qd->qd_bh_count);
444 	if (!--qd->qd_bh_count) {
445 		brelse(qd->qd_bh);
446 		qd->qd_bh = NULL;
447 		qd->qd_bh_qc = NULL;
448 	}
449 	mutex_unlock(&sdp->sd_quota_mutex);
450 }
451 
452 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
453 			 u64 *sync_gen)
454 {
455 	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
456 	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
457 	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
458 		return 0;
459 
460 	if (!lockref_get_not_dead(&qd->qd_lockref))
461 		return 0;
462 
463 	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
464 	set_bit(QDF_LOCKED, &qd->qd_flags);
465 	qd->qd_change_sync = qd->qd_change;
466 	slot_hold(qd);
467 	return 1;
468 }
469 
470 static int qd_bh_get_or_undo(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
471 {
472 	int error;
473 
474 	error = bh_get(qd);
475 	if (!error)
476 		return 0;
477 
478 	clear_bit(QDF_LOCKED, &qd->qd_flags);
479 	slot_put(qd);
480 	qd_put(qd);
481 	return error;
482 }
483 
484 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
485 {
486 	struct gfs2_quota_data *qd = NULL, *iter;
487 	int error;
488 
489 	*qdp = NULL;
490 
491 	if (sb_rdonly(sdp->sd_vfs))
492 		return 0;
493 
494 	spin_lock(&qd_lock);
495 
496 	list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
497 		if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
498 			qd = iter;
499 			break;
500 		}
501 	}
502 
503 	spin_unlock(&qd_lock);
504 
505 	if (qd) {
506 		error = qd_bh_get_or_undo(sdp, qd);
507 		if (error)
508 			return error;
509 		*qdp = qd;
510 	}
511 
512 	return 0;
513 }
514 
515 static void qdsb_put(struct gfs2_quota_data *qd)
516 {
517 	bh_put(qd);
518 	slot_put(qd);
519 	qd_put(qd);
520 }
521 
522 static void qd_unlock(struct gfs2_quota_data *qd)
523 {
524 	gfs2_assert_warn(qd->qd_sbd, test_bit(QDF_LOCKED, &qd->qd_flags));
525 	clear_bit(QDF_LOCKED, &qd->qd_flags);
526 	qdsb_put(qd);
527 }
528 
529 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
530 		    struct gfs2_quota_data **qdp)
531 {
532 	int error;
533 
534 	error = qd_get(sdp, qid, qdp);
535 	if (error)
536 		return error;
537 
538 	error = slot_get(*qdp);
539 	if (error)
540 		goto fail;
541 
542 	error = bh_get(*qdp);
543 	if (error)
544 		goto fail_slot;
545 
546 	return 0;
547 
548 fail_slot:
549 	slot_put(*qdp);
550 fail:
551 	qd_put(*qdp);
552 	return error;
553 }
554 
555 /**
556  * gfs2_qa_get - make sure we have a quota allocations data structure,
557  *               if necessary
558  * @ip: the inode for this reservation
559  */
560 int gfs2_qa_get(struct gfs2_inode *ip)
561 {
562 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
563 	struct inode *inode = &ip->i_inode;
564 
565 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
566 		return 0;
567 
568 	spin_lock(&inode->i_lock);
569 	if (ip->i_qadata == NULL) {
570 		struct gfs2_qadata *tmp;
571 
572 		spin_unlock(&inode->i_lock);
573 		tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
574 		if (!tmp)
575 			return -ENOMEM;
576 
577 		spin_lock(&inode->i_lock);
578 		if (ip->i_qadata == NULL)
579 			ip->i_qadata = tmp;
580 		else
581 			kmem_cache_free(gfs2_qadata_cachep, tmp);
582 	}
583 	ip->i_qadata->qa_ref++;
584 	spin_unlock(&inode->i_lock);
585 	return 0;
586 }
587 
588 void gfs2_qa_put(struct gfs2_inode *ip)
589 {
590 	struct inode *inode = &ip->i_inode;
591 
592 	spin_lock(&inode->i_lock);
593 	if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
594 		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
595 		ip->i_qadata = NULL;
596 	}
597 	spin_unlock(&inode->i_lock);
598 }
599 
600 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
601 {
602 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
603 	struct gfs2_quota_data **qd;
604 	int error;
605 
606 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
607 		return 0;
608 
609 	error = gfs2_qa_get(ip);
610 	if (error)
611 		return error;
612 
613 	qd = ip->i_qadata->qa_qd;
614 
615 	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
616 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
617 		error = -EIO;
618 		gfs2_qa_put(ip);
619 		goto out;
620 	}
621 
622 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
623 	if (error)
624 		goto out_unhold;
625 	ip->i_qadata->qa_qd_num++;
626 	qd++;
627 
628 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
629 	if (error)
630 		goto out_unhold;
631 	ip->i_qadata->qa_qd_num++;
632 	qd++;
633 
634 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
635 	    !uid_eq(uid, ip->i_inode.i_uid)) {
636 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
637 		if (error)
638 			goto out_unhold;
639 		ip->i_qadata->qa_qd_num++;
640 		qd++;
641 	}
642 
643 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
644 	    !gid_eq(gid, ip->i_inode.i_gid)) {
645 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
646 		if (error)
647 			goto out_unhold;
648 		ip->i_qadata->qa_qd_num++;
649 		qd++;
650 	}
651 
652 out_unhold:
653 	if (error)
654 		gfs2_quota_unhold(ip);
655 out:
656 	return error;
657 }
658 
659 void gfs2_quota_unhold(struct gfs2_inode *ip)
660 {
661 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
662 	u32 x;
663 
664 	if (ip->i_qadata == NULL)
665 		return;
666 
667 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
668 
669 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
670 		qdsb_put(ip->i_qadata->qa_qd[x]);
671 		ip->i_qadata->qa_qd[x] = NULL;
672 	}
673 	ip->i_qadata->qa_qd_num = 0;
674 	gfs2_qa_put(ip);
675 }
676 
677 static int sort_qd(const void *a, const void *b)
678 {
679 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
680 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
681 
682 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
683 		return -1;
684 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
685 		return 1;
686 	return 0;
687 }
688 
689 static void do_qc(struct gfs2_quota_data *qd, s64 change, int qc_type)
690 {
691 	struct gfs2_sbd *sdp = qd->qd_sbd;
692 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
693 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
694 	s64 x;
695 
696 	mutex_lock(&sdp->sd_quota_mutex);
697 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
698 
699 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
700 		qc->qc_change = 0;
701 		qc->qc_flags = 0;
702 		if (qd->qd_id.type == USRQUOTA)
703 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
704 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
705 	}
706 
707 	x = be64_to_cpu(qc->qc_change) + change;
708 	qc->qc_change = cpu_to_be64(x);
709 
710 	spin_lock(&qd_lock);
711 	qd->qd_change = x;
712 	spin_unlock(&qd_lock);
713 
714 	if (qc_type == QC_CHANGE) {
715 		if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
716 			qd_hold(qd);
717 			slot_hold(qd);
718 		}
719 	} else {
720 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
721 		clear_bit(QDF_CHANGE, &qd->qd_flags);
722 		qc->qc_flags = 0;
723 		qc->qc_id = 0;
724 		slot_put(qd);
725 		qd_put(qd);
726 	}
727 
728 	if (change < 0) /* Reset quiet flag if we freed some blocks */
729 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
730 	mutex_unlock(&sdp->sd_quota_mutex);
731 }
732 
733 static int gfs2_write_buf_to_page(struct gfs2_sbd *sdp, unsigned long index,
734 				  unsigned off, void *buf, unsigned bytes)
735 {
736 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
737 	struct inode *inode = &ip->i_inode;
738 	struct address_space *mapping = inode->i_mapping;
739 	struct page *page;
740 	struct buffer_head *bh;
741 	u64 blk;
742 	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
743 	unsigned to_write = bytes, pg_off = off;
744 
745 	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
746 	boff = off % bsize;
747 
748 	page = grab_cache_page(mapping, index);
749 	if (!page)
750 		return -ENOMEM;
751 	if (!page_has_buffers(page))
752 		create_empty_buffers(page, bsize, 0);
753 
754 	bh = page_buffers(page);
755 	for(;;) {
756 		/* Find the beginning block within the page */
757 		if (pg_off >= ((bnum * bsize) + bsize)) {
758 			bh = bh->b_this_page;
759 			bnum++;
760 			blk++;
761 			continue;
762 		}
763 		if (!buffer_mapped(bh)) {
764 			gfs2_block_map(inode, blk, bh, 1);
765 			if (!buffer_mapped(bh))
766 				goto unlock_out;
767 			/* If it's a newly allocated disk block, zero it */
768 			if (buffer_new(bh))
769 				zero_user(page, bnum * bsize, bh->b_size);
770 		}
771 		if (PageUptodate(page))
772 			set_buffer_uptodate(bh);
773 		if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
774 			goto unlock_out;
775 		gfs2_trans_add_data(ip->i_gl, bh);
776 
777 		/* If we need to write to the next block as well */
778 		if (to_write > (bsize - boff)) {
779 			pg_off += (bsize - boff);
780 			to_write -= (bsize - boff);
781 			boff = pg_off % bsize;
782 			continue;
783 		}
784 		break;
785 	}
786 
787 	/* Write to the page, now that we have setup the buffer(s) */
788 	memcpy_to_page(page, off, buf, bytes);
789 	flush_dcache_page(page);
790 	unlock_page(page);
791 	put_page(page);
792 
793 	return 0;
794 
795 unlock_out:
796 	unlock_page(page);
797 	put_page(page);
798 	return -EIO;
799 }
800 
801 static int gfs2_write_disk_quota(struct gfs2_sbd *sdp, struct gfs2_quota *qp,
802 				 loff_t loc)
803 {
804 	unsigned long pg_beg;
805 	unsigned pg_off, nbytes, overflow = 0;
806 	int error;
807 	void *ptr;
808 
809 	nbytes = sizeof(struct gfs2_quota);
810 
811 	pg_beg = loc >> PAGE_SHIFT;
812 	pg_off = offset_in_page(loc);
813 
814 	/* If the quota straddles a page boundary, split the write in two */
815 	if ((pg_off + nbytes) > PAGE_SIZE)
816 		overflow = (pg_off + nbytes) - PAGE_SIZE;
817 
818 	ptr = qp;
819 	error = gfs2_write_buf_to_page(sdp, pg_beg, pg_off, ptr,
820 				       nbytes - overflow);
821 	/* If there's an overflow, write the remaining bytes to the next page */
822 	if (!error && overflow)
823 		error = gfs2_write_buf_to_page(sdp, pg_beg + 1, 0,
824 					       ptr + nbytes - overflow,
825 					       overflow);
826 	return error;
827 }
828 
829 /**
830  * gfs2_adjust_quota - adjust record of current block usage
831  * @sdp: The superblock
832  * @loc: Offset of the entry in the quota file
833  * @change: The amount of usage change to record
834  * @qd: The quota data
835  * @fdq: The updated limits to record
836  *
837  * This function was mostly borrowed from gfs2_block_truncate_page which was
838  * in turn mostly borrowed from ext3
839  *
840  * Returns: 0 or -ve on error
841  */
842 
843 static int gfs2_adjust_quota(struct gfs2_sbd *sdp, loff_t loc,
844 			     s64 change, struct gfs2_quota_data *qd,
845 			     struct qc_dqblk *fdq)
846 {
847 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
848 	struct inode *inode = &ip->i_inode;
849 	struct gfs2_quota q;
850 	int err;
851 	u64 size;
852 
853 	if (gfs2_is_stuffed(ip)) {
854 		err = gfs2_unstuff_dinode(ip);
855 		if (err)
856 			return err;
857 	}
858 
859 	memset(&q, 0, sizeof(struct gfs2_quota));
860 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
861 	if (err < 0)
862 		return err;
863 
864 	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
865 	be64_add_cpu(&q.qu_value, change);
866 	if (((s64)be64_to_cpu(q.qu_value)) < 0)
867 		q.qu_value = 0; /* Never go negative on quota usage */
868 	qd->qd_qb.qb_value = q.qu_value;
869 	if (fdq) {
870 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
871 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
872 			qd->qd_qb.qb_warn = q.qu_warn;
873 		}
874 		if (fdq->d_fieldmask & QC_SPC_HARD) {
875 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
876 			qd->qd_qb.qb_limit = q.qu_limit;
877 		}
878 		if (fdq->d_fieldmask & QC_SPACE) {
879 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
880 			qd->qd_qb.qb_value = q.qu_value;
881 		}
882 	}
883 
884 	err = gfs2_write_disk_quota(sdp, &q, loc);
885 	if (!err) {
886 		size = loc + sizeof(struct gfs2_quota);
887 		if (size > inode->i_size)
888 			i_size_write(inode, size);
889 		inode->i_mtime = inode_set_ctime_current(inode);
890 		mark_inode_dirty(inode);
891 		set_bit(QDF_REFRESH, &qd->qd_flags);
892 	}
893 
894 	return err;
895 }
896 
897 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
898 {
899 	struct gfs2_sbd *sdp = (*qda)->qd_sbd;
900 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
901 	struct gfs2_alloc_parms ap = { .aflags = 0, };
902 	unsigned int data_blocks, ind_blocks;
903 	struct gfs2_holder *ghs, i_gh;
904 	unsigned int qx, x;
905 	struct gfs2_quota_data *qd;
906 	unsigned reserved;
907 	loff_t offset;
908 	unsigned int nalloc = 0, blocks;
909 	int error;
910 
911 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
912 			      &data_blocks, &ind_blocks);
913 
914 	ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
915 	if (!ghs)
916 		return -ENOMEM;
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(sdp, offset, qd->qd_change_sync, qd,
966 							NULL);
967 		if (error)
968 			goto out_end_trans;
969 
970 		do_qc(qd, -qd->qd_change_sync, QC_SYNC);
971 		set_bit(QDF_REFRESH, &qd->qd_flags);
972 	}
973 
974 out_end_trans:
975 	gfs2_trans_end(sdp);
976 out_ipres:
977 	gfs2_inplace_release(ip);
978 out_alloc:
979 	gfs2_glock_dq_uninit(&i_gh);
980 out_dq:
981 	while (qx--)
982 		gfs2_glock_dq_uninit(&ghs[qx]);
983 	inode_unlock(&ip->i_inode);
984 	kfree(ghs);
985 	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
986 		       GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
987 	if (!error) {
988 		for (x = 0; x < num_qd; x++)
989 			qda[x]->qd_sync_gen = sdp->sd_quota_sync_gen;
990 	}
991 	return error;
992 }
993 
994 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
995 {
996 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
997 	struct gfs2_quota q;
998 	struct gfs2_quota_lvb *qlvb;
999 	loff_t pos;
1000 	int error;
1001 
1002 	memset(&q, 0, sizeof(struct gfs2_quota));
1003 	pos = qd2offset(qd);
1004 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
1005 	if (error < 0)
1006 		return error;
1007 
1008 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1009 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1010 	qlvb->__pad = 0;
1011 	qlvb->qb_limit = q.qu_limit;
1012 	qlvb->qb_warn = q.qu_warn;
1013 	qlvb->qb_value = q.qu_value;
1014 	qd->qd_qb = *qlvb;
1015 
1016 	return 0;
1017 }
1018 
1019 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1020 		    struct gfs2_holder *q_gh)
1021 {
1022 	struct gfs2_sbd *sdp = qd->qd_sbd;
1023 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1024 	struct gfs2_holder i_gh;
1025 	int error;
1026 
1027 	gfs2_assert_warn(sdp, sdp == qd->qd_gl->gl_name.ln_sbd);
1028 restart:
1029 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1030 	if (error)
1031 		return error;
1032 
1033 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1034 		force_refresh = FORCE;
1035 
1036 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1037 
1038 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1039 		gfs2_glock_dq_uninit(q_gh);
1040 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1041 					   GL_NOCACHE, q_gh);
1042 		if (error)
1043 			return error;
1044 
1045 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1046 		if (error)
1047 			goto fail;
1048 
1049 		error = update_qd(sdp, qd);
1050 		if (error)
1051 			goto fail_gunlock;
1052 
1053 		gfs2_glock_dq_uninit(&i_gh);
1054 		gfs2_glock_dq_uninit(q_gh);
1055 		force_refresh = 0;
1056 		goto restart;
1057 	}
1058 
1059 	return 0;
1060 
1061 fail_gunlock:
1062 	gfs2_glock_dq_uninit(&i_gh);
1063 fail:
1064 	gfs2_glock_dq_uninit(q_gh);
1065 	return error;
1066 }
1067 
1068 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1069 {
1070 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1071 	struct gfs2_quota_data *qd;
1072 	u32 x;
1073 	int error;
1074 
1075 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON &&
1076 	    sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET)
1077 		return 0;
1078 
1079 	error = gfs2_quota_hold(ip, uid, gid);
1080 	if (error)
1081 		return error;
1082 
1083 	sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1084 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1085 
1086 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1087 		qd = ip->i_qadata->qa_qd[x];
1088 		error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1089 		if (error)
1090 			break;
1091 	}
1092 
1093 	if (!error)
1094 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1095 	else {
1096 		while (x--)
1097 			gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1098 		gfs2_quota_unhold(ip);
1099 	}
1100 
1101 	return error;
1102 }
1103 
1104 static bool need_sync(struct gfs2_quota_data *qd)
1105 {
1106 	struct gfs2_sbd *sdp = qd->qd_sbd;
1107 	struct gfs2_tune *gt = &sdp->sd_tune;
1108 	s64 value;
1109 	unsigned int num, den;
1110 
1111 	if (!qd->qd_qb.qb_limit)
1112 		return false;
1113 
1114 	spin_lock(&qd_lock);
1115 	value = qd->qd_change;
1116 	spin_unlock(&qd_lock);
1117 
1118 	spin_lock(&gt->gt_spin);
1119 	num = gt->gt_quota_scale_num;
1120 	den = gt->gt_quota_scale_den;
1121 	spin_unlock(&gt->gt_spin);
1122 
1123 	if (value <= 0)
1124 		return false;
1125 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1126 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1127 		return false;
1128 	else {
1129 		value *= gfs2_jindex_size(sdp) * num;
1130 		value = div_s64(value, den);
1131 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1132 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1133 			return false;
1134 	}
1135 
1136 	return true;
1137 }
1138 
1139 void gfs2_quota_unlock(struct gfs2_inode *ip)
1140 {
1141 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1142 	struct gfs2_quota_data *qda[2 * GFS2_MAXQUOTAS];
1143 	unsigned int count = 0;
1144 	u32 x;
1145 	int found;
1146 
1147 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1148 		return;
1149 
1150 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1151 		struct gfs2_quota_data *qd;
1152 		bool sync;
1153 
1154 		qd = ip->i_qadata->qa_qd[x];
1155 		sync = need_sync(qd);
1156 
1157 		gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1158 		if (!sync)
1159 			continue;
1160 
1161 		spin_lock(&qd_lock);
1162 		found = qd_check_sync(sdp, qd, NULL);
1163 		spin_unlock(&qd_lock);
1164 
1165 		if (!found)
1166 			continue;
1167 
1168 		if (!qd_bh_get_or_undo(sdp, qd))
1169 			qda[count++] = qd;
1170 	}
1171 
1172 	if (count) {
1173 		do_sync(count, qda);
1174 		for (x = 0; x < count; x++)
1175 			qd_unlock(qda[x]);
1176 	}
1177 
1178 	gfs2_quota_unhold(ip);
1179 }
1180 
1181 #define MAX_LINE 256
1182 
1183 static int print_message(struct gfs2_quota_data *qd, char *type)
1184 {
1185 	struct gfs2_sbd *sdp = qd->qd_sbd;
1186 
1187 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET)
1188 		fs_info(sdp, "quota %s for %s %u\n",
1189 			type,
1190 			(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1191 			from_kqid(&init_user_ns, qd->qd_id));
1192 
1193 	return 0;
1194 }
1195 
1196 /**
1197  * gfs2_quota_check - check if allocating new blocks will exceed quota
1198  * @ip:  The inode for which this check is being performed
1199  * @uid: The uid to check against
1200  * @gid: The gid to check against
1201  * @ap:  The allocation parameters. ap->target contains the requested
1202  *       blocks. ap->min_target, if set, contains the minimum blks
1203  *       requested.
1204  *
1205  * Returns: 0 on success.
1206  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1207  *                  quota must allow at least min_req blks for success and
1208  *                  ap->allowed is set to the number of blocks allowed
1209  *
1210  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1211  *                  of blocks available.
1212  */
1213 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1214 		     struct gfs2_alloc_parms *ap)
1215 {
1216 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1217 	struct gfs2_quota_data *qd;
1218 	s64 value, warn, limit;
1219 	u32 x;
1220 	int error = 0;
1221 
1222 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1223 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1224 		return 0;
1225 
1226 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1227 		qd = ip->i_qadata->qa_qd[x];
1228 
1229 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1230 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1231 			continue;
1232 
1233 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1234 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1235 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1236 		spin_lock(&qd_lock);
1237 		value += qd->qd_change;
1238 		spin_unlock(&qd_lock);
1239 
1240 		if (limit > 0 && (limit - value) < ap->allowed)
1241 			ap->allowed = limit - value;
1242 		/* If we can't meet the target */
1243 		if (limit && limit < (value + (s64)ap->target)) {
1244 			/* If no min_target specified or we don't meet
1245 			 * min_target, return -EDQUOT */
1246 			if (!ap->min_target || ap->min_target > ap->allowed) {
1247 				if (!test_and_set_bit(QDF_QMSG_QUIET,
1248 						      &qd->qd_flags)) {
1249 					print_message(qd, "exceeded");
1250 					quota_send_warning(qd->qd_id,
1251 							   sdp->sd_vfs->s_dev,
1252 							   QUOTA_NL_BHARDWARN);
1253 				}
1254 				error = -EDQUOT;
1255 				break;
1256 			}
1257 		} else if (warn && warn < value &&
1258 			   time_after_eq(jiffies, qd->qd_last_warn +
1259 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1260 					 * HZ)) {
1261 			quota_send_warning(qd->qd_id,
1262 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1263 			error = print_message(qd, "warning");
1264 			qd->qd_last_warn = jiffies;
1265 		}
1266 	}
1267 	return error;
1268 }
1269 
1270 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1271 		       kuid_t uid, kgid_t gid)
1272 {
1273 	struct gfs2_quota_data *qd;
1274 	u32 x;
1275 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1276 
1277 	if ((sdp->sd_args.ar_quota != GFS2_QUOTA_ON &&
1278 	    sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET) ||
1279 	    gfs2_assert_warn(sdp, change))
1280 		return;
1281 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1282 		return;
1283 
1284 	if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1285 				 ip->i_qadata->qa_ref > 0))
1286 		return;
1287 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1288 		qd = ip->i_qadata->qa_qd[x];
1289 
1290 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1291 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1292 			do_qc(qd, change, QC_CHANGE);
1293 		}
1294 	}
1295 }
1296 
1297 static bool qd_changed(struct gfs2_sbd *sdp)
1298 {
1299 	struct gfs2_quota_data *qd;
1300 	bool changed = false;
1301 
1302 	spin_lock(&qd_lock);
1303 	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1304 		if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
1305 		    !test_bit(QDF_CHANGE, &qd->qd_flags))
1306 			continue;
1307 
1308 		changed = true;
1309 		break;
1310 	}
1311 	spin_unlock(&qd_lock);
1312 	return changed;
1313 }
1314 
1315 int gfs2_quota_sync(struct super_block *sb, int type)
1316 {
1317 	struct gfs2_sbd *sdp = sb->s_fs_info;
1318 	struct gfs2_quota_data **qda;
1319 	unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1320 	unsigned int num_qd;
1321 	unsigned int x;
1322 	int error = 0;
1323 
1324 	if (!qd_changed(sdp))
1325 		return 0;
1326 
1327 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1328 	if (!qda)
1329 		return -ENOMEM;
1330 
1331 	mutex_lock(&sdp->sd_quota_sync_mutex);
1332 	sdp->sd_quota_sync_gen++;
1333 
1334 	do {
1335 		num_qd = 0;
1336 
1337 		for (;;) {
1338 			error = qd_fish(sdp, qda + num_qd);
1339 			if (error || !qda[num_qd])
1340 				break;
1341 			if (++num_qd == max_qd)
1342 				break;
1343 		}
1344 
1345 		if (num_qd) {
1346 			if (!error)
1347 				error = do_sync(num_qd, qda);
1348 
1349 			for (x = 0; x < num_qd; x++)
1350 				qd_unlock(qda[x]);
1351 		}
1352 	} while (!error && num_qd == max_qd);
1353 
1354 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1355 	kfree(qda);
1356 
1357 	return error;
1358 }
1359 
1360 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1361 {
1362 	struct gfs2_quota_data *qd;
1363 	struct gfs2_holder q_gh;
1364 	int error;
1365 
1366 	error = qd_get(sdp, qid, &qd);
1367 	if (error)
1368 		return error;
1369 
1370 	error = do_glock(qd, FORCE, &q_gh);
1371 	if (!error)
1372 		gfs2_glock_dq_uninit(&q_gh);
1373 
1374 	qd_put(qd);
1375 	return error;
1376 }
1377 
1378 int gfs2_quota_init(struct gfs2_sbd *sdp)
1379 {
1380 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1381 	u64 size = i_size_read(sdp->sd_qc_inode);
1382 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1383 	unsigned int x, slot = 0;
1384 	unsigned int found = 0;
1385 	unsigned int hash;
1386 	unsigned int bm_size;
1387 	u64 dblock;
1388 	u32 extlen = 0;
1389 	int error;
1390 
1391 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1392 		return -EIO;
1393 
1394 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1395 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1396 	bm_size *= sizeof(unsigned long);
1397 	error = -ENOMEM;
1398 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1399 	if (sdp->sd_quota_bitmap == NULL)
1400 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1401 						 __GFP_ZERO);
1402 	if (!sdp->sd_quota_bitmap)
1403 		return error;
1404 
1405 	for (x = 0; x < blocks; x++) {
1406 		struct buffer_head *bh;
1407 		const struct gfs2_quota_change *qc;
1408 		unsigned int y;
1409 
1410 		if (!extlen) {
1411 			extlen = 32;
1412 			error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1413 			if (error)
1414 				goto fail;
1415 		}
1416 		error = -EIO;
1417 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1418 		if (!bh)
1419 			goto fail;
1420 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1421 			brelse(bh);
1422 			goto fail;
1423 		}
1424 
1425 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1426 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1427 		     y++, slot++) {
1428 			struct gfs2_quota_data *qd;
1429 			s64 qc_change = be64_to_cpu(qc->qc_change);
1430 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1431 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1432 						USRQUOTA : GRPQUOTA;
1433 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1434 						      be32_to_cpu(qc->qc_id));
1435 			qc++;
1436 			if (!qc_change)
1437 				continue;
1438 
1439 			hash = gfs2_qd_hash(sdp, qc_id);
1440 			qd = qd_alloc(hash, sdp, qc_id);
1441 			if (qd == NULL) {
1442 				brelse(bh);
1443 				goto fail;
1444 			}
1445 
1446 			set_bit(QDF_CHANGE, &qd->qd_flags);
1447 			qd->qd_change = qc_change;
1448 			qd->qd_slot = slot;
1449 			qd->qd_slot_ref = 1;
1450 
1451 			spin_lock(&qd_lock);
1452 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1453 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1454 			atomic_inc(&sdp->sd_quota_count);
1455 			spin_unlock(&qd_lock);
1456 
1457 			spin_lock_bucket(hash);
1458 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1459 			spin_unlock_bucket(hash);
1460 
1461 			found++;
1462 		}
1463 
1464 		brelse(bh);
1465 		dblock++;
1466 		extlen--;
1467 	}
1468 
1469 	if (found)
1470 		fs_info(sdp, "found %u quota changes\n", found);
1471 
1472 	return 0;
1473 
1474 fail:
1475 	gfs2_quota_cleanup(sdp);
1476 	return error;
1477 }
1478 
1479 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1480 {
1481 	struct gfs2_quota_data *qd;
1482 	LIST_HEAD(dispose);
1483 	int count;
1484 
1485 	BUG_ON(test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
1486 
1487 	spin_lock(&qd_lock);
1488 	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1489 		spin_lock(&qd->qd_lockref.lock);
1490 		if (qd->qd_lockref.count != 0) {
1491 			spin_unlock(&qd->qd_lockref.lock);
1492 			continue;
1493 		}
1494 		lockref_mark_dead(&qd->qd_lockref);
1495 		spin_unlock(&qd->qd_lockref.lock);
1496 
1497 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1498 		list_add(&qd->qd_lru, &dispose);
1499 	}
1500 	spin_unlock(&qd_lock);
1501 
1502 	gfs2_qd_list_dispose(&dispose);
1503 
1504 	wait_event_timeout(sdp->sd_kill_wait,
1505 		(count = atomic_read(&sdp->sd_quota_count)) == 0,
1506 		HZ * 60);
1507 
1508 	if (count != 0)
1509 		fs_err(sdp, "%d left-over quota data objects\n", count);
1510 
1511 	kvfree(sdp->sd_quota_bitmap);
1512 	sdp->sd_quota_bitmap = NULL;
1513 }
1514 
1515 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1516 {
1517 	if (error == 0 || error == -EROFS)
1518 		return;
1519 	if (!gfs2_withdrawn(sdp)) {
1520 		if (!cmpxchg(&sdp->sd_log_error, 0, error))
1521 			fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1522 		wake_up(&sdp->sd_logd_waitq);
1523 	}
1524 }
1525 
1526 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1527 			       int (*fxn)(struct super_block *sb, int type),
1528 			       unsigned long t, unsigned long *timeo,
1529 			       unsigned int *new_timeo)
1530 {
1531 	if (t >= *timeo) {
1532 		int error = fxn(sdp->sd_vfs, 0);
1533 		quotad_error(sdp, msg, error);
1534 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1535 	} else {
1536 		*timeo -= t;
1537 	}
1538 }
1539 
1540 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1541 	if (!sdp->sd_statfs_force_sync) {
1542 		sdp->sd_statfs_force_sync = 1;
1543 		wake_up(&sdp->sd_quota_wait);
1544 	}
1545 }
1546 
1547 
1548 /**
1549  * gfs2_quotad - Write cached quota changes into the quota file
1550  * @data: Pointer to GFS2 superblock
1551  *
1552  */
1553 
1554 int gfs2_quotad(void *data)
1555 {
1556 	struct gfs2_sbd *sdp = data;
1557 	struct gfs2_tune *tune = &sdp->sd_tune;
1558 	unsigned long statfs_timeo = 0;
1559 	unsigned long quotad_timeo = 0;
1560 	unsigned long t = 0;
1561 
1562 	while (!kthread_should_stop()) {
1563 		if (gfs2_withdrawn(sdp))
1564 			break;
1565 
1566 		/* Update the master statfs file */
1567 		if (sdp->sd_statfs_force_sync) {
1568 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1569 			quotad_error(sdp, "statfs", error);
1570 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1571 		}
1572 		else
1573 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1574 				   	   &statfs_timeo,
1575 					   &tune->gt_statfs_quantum);
1576 
1577 		/* Update quota file */
1578 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1579 				   &quotad_timeo, &tune->gt_quota_quantum);
1580 
1581 		try_to_freeze();
1582 
1583 		t = min(quotad_timeo, statfs_timeo);
1584 
1585 		t = wait_event_interruptible_timeout(sdp->sd_quota_wait,
1586 				sdp->sd_statfs_force_sync ||
1587 				gfs2_withdrawn(sdp) ||
1588 				kthread_should_stop(),
1589 				t);
1590 
1591 		if (sdp->sd_statfs_force_sync)
1592 			t = 0;
1593 	}
1594 
1595 	return 0;
1596 }
1597 
1598 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1599 {
1600 	struct gfs2_sbd *sdp = sb->s_fs_info;
1601 
1602 	memset(state, 0, sizeof(*state));
1603 
1604 	switch (sdp->sd_args.ar_quota) {
1605 	case GFS2_QUOTA_QUIET:
1606 		fallthrough;
1607 	case GFS2_QUOTA_ON:
1608 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1609 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1610 		fallthrough;
1611 	case GFS2_QUOTA_ACCOUNT:
1612 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1613 						  QCI_SYSFILE;
1614 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1615 						  QCI_SYSFILE;
1616 		break;
1617 	case GFS2_QUOTA_OFF:
1618 		break;
1619 	}
1620 	if (sdp->sd_quota_inode) {
1621 		state->s_state[USRQUOTA].ino =
1622 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1623 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1624 	}
1625 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1626 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1627 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1628 	return 0;
1629 }
1630 
1631 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1632 			  struct qc_dqblk *fdq)
1633 {
1634 	struct gfs2_sbd *sdp = sb->s_fs_info;
1635 	struct gfs2_quota_lvb *qlvb;
1636 	struct gfs2_quota_data *qd;
1637 	struct gfs2_holder q_gh;
1638 	int error;
1639 
1640 	memset(fdq, 0, sizeof(*fdq));
1641 
1642 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1643 		return -ESRCH; /* Crazy XFS error code */
1644 
1645 	if ((qid.type != USRQUOTA) &&
1646 	    (qid.type != GRPQUOTA))
1647 		return -EINVAL;
1648 
1649 	error = qd_get(sdp, qid, &qd);
1650 	if (error)
1651 		return error;
1652 	error = do_glock(qd, FORCE, &q_gh);
1653 	if (error)
1654 		goto out;
1655 
1656 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1657 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1658 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1659 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1660 
1661 	gfs2_glock_dq_uninit(&q_gh);
1662 out:
1663 	qd_put(qd);
1664 	return error;
1665 }
1666 
1667 /* GFS2 only supports a subset of the XFS fields */
1668 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1669 
1670 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1671 			  struct qc_dqblk *fdq)
1672 {
1673 	struct gfs2_sbd *sdp = sb->s_fs_info;
1674 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1675 	struct gfs2_quota_data *qd;
1676 	struct gfs2_holder q_gh, i_gh;
1677 	unsigned int data_blocks, ind_blocks;
1678 	unsigned int blocks = 0;
1679 	int alloc_required;
1680 	loff_t offset;
1681 	int error;
1682 
1683 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1684 		return -ESRCH; /* Crazy XFS error code */
1685 
1686 	if ((qid.type != USRQUOTA) &&
1687 	    (qid.type != GRPQUOTA))
1688 		return -EINVAL;
1689 
1690 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1691 		return -EINVAL;
1692 
1693 	error = qd_get(sdp, qid, &qd);
1694 	if (error)
1695 		return error;
1696 
1697 	error = gfs2_qa_get(ip);
1698 	if (error)
1699 		goto out_put;
1700 
1701 	inode_lock(&ip->i_inode);
1702 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1703 	if (error)
1704 		goto out_unlockput;
1705 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1706 	if (error)
1707 		goto out_q;
1708 
1709 	/* Check for existing entry, if none then alloc new blocks */
1710 	error = update_qd(sdp, qd);
1711 	if (error)
1712 		goto out_i;
1713 
1714 	/* If nothing has changed, this is a no-op */
1715 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1716 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1717 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1718 
1719 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1720 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1721 		fdq->d_fieldmask ^= QC_SPC_HARD;
1722 
1723 	if ((fdq->d_fieldmask & QC_SPACE) &&
1724 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1725 		fdq->d_fieldmask ^= QC_SPACE;
1726 
1727 	if (fdq->d_fieldmask == 0)
1728 		goto out_i;
1729 
1730 	offset = qd2offset(qd);
1731 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1732 	if (gfs2_is_stuffed(ip))
1733 		alloc_required = 1;
1734 	if (alloc_required) {
1735 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1736 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1737 				       &data_blocks, &ind_blocks);
1738 		blocks = 1 + data_blocks + ind_blocks;
1739 		ap.target = blocks;
1740 		error = gfs2_inplace_reserve(ip, &ap);
1741 		if (error)
1742 			goto out_i;
1743 		blocks += gfs2_rg_blocks(ip, blocks);
1744 	}
1745 
1746 	/* Some quotas span block boundaries and can update two blocks,
1747 	   adding an extra block to the transaction to handle such quotas */
1748 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1749 	if (error)
1750 		goto out_release;
1751 
1752 	/* Apply changes */
1753 	error = gfs2_adjust_quota(sdp, offset, 0, qd, fdq);
1754 	if (!error)
1755 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1756 
1757 	gfs2_trans_end(sdp);
1758 out_release:
1759 	if (alloc_required)
1760 		gfs2_inplace_release(ip);
1761 out_i:
1762 	gfs2_glock_dq_uninit(&i_gh);
1763 out_q:
1764 	gfs2_glock_dq_uninit(&q_gh);
1765 out_unlockput:
1766 	gfs2_qa_put(ip);
1767 	inode_unlock(&ip->i_inode);
1768 out_put:
1769 	qd_put(qd);
1770 	return error;
1771 }
1772 
1773 const struct quotactl_ops gfs2_quotactl_ops = {
1774 	.quota_sync     = gfs2_quota_sync,
1775 	.get_state	= gfs2_quota_get_state,
1776 	.get_dqblk	= gfs2_get_dqblk,
1777 	.set_dqblk	= gfs2_set_dqblk,
1778 };
1779 
1780 void __init gfs2_quota_hash_init(void)
1781 {
1782 	unsigned i;
1783 
1784 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1785 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1786 }
1787